Lecture Notes in

Mathematics
A collection of informal reports and seminars
Edited by A. Dold, Heidelberg and B. Eckmann, ZUrich

237

Bo StenstrOm
University of Stockholm, Stockholm/Sweden

Rings and Modules

of Quotients

$
Springer-Verlag
Berlin. Heidelberg- New York 1971
AMS Subject Classifications (1970): 16A 08

ISBN 3-540-05690-4 Springer-Verlag Berlin. Heidelberg. New York

ISBN 0-387-05690-4 Springer-Verlag New York- Heidelberg. Berlin

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned,
specifically those of translation, reprinting, re-use of illustrations, broadcasting, reproduction by photocopying machine
or similar means, and storage in data banks.
Under § 54 of the German Copyright Law where copies are made for other than private use, a fee is payable to the publisher,
the amount of the fee to be determined by agreement with the publisher.

© by Springer-Verlag Berlin • Heidelberg 1971, Library of Congress Catalog Card Number 70-180692. Printed in Germany.
Offsetdruck: Julius Bdtz, Hemsbach
Contents

Chapter I. Torsion theory

I. Preradicals I
2. Torsion theories 4
3. Topologies 12
4. Stable torsion theories 20
5. Topologies for a commutative noetherian ring 23
6. ~-injective modules 29

Chapter 2. Categories of modules of quotient s

7. Construction of rings and modules of quotients 33

8. Modules of quotients and F..-injeetive envelopes 41
9. Coreflective subcategories of Mod-A 44
I0. Giraud subeategories and the Popescu-Gabriel
theorem 48

Chapter 3. General properties of rin~s of quotients

II. Lattices eli-pure submodules 58

12. Finiteness conditions on topologies 68
13. Flat epimorphisms of rings 72
14. Maximal flat epimorphic extension of a ring 82
15. 1-topologies and rings of fractions 86

Chapter 4. Self-injective rings

16. The endomorphism ring of an injective module 93

17. Coperfect rings 97
18. Quasi-Frobenius rings I01

Chapter 5. Maximal and classical rin~s of quotients

19. The maximal ring of quotients IIO

20. The maximal ring of quotients of a non-singular
ring 113
21. The maximal ring of quotients of a reduced ring 118
22. The classical ring of quotients 123
References 130
Introduction

These notes are intended to give a survey of the basic, more

or less well-known, results in the theory of rings of quotients.

An effort has been made to make the account as self-contained and

elementary as possible. Thus we assume from the reader only a

knowledge of the elements of the theory of rings and of abelian

categories.

We will briefly describe the contents of the notes. Chapter 1

treats the necessary preliminaries on torsion theory. The main

result here is the establishing of a I-I correspondence between

hereditary torsion theories and topologies on a ring (Gabriel [31]

and Maranda [51] )

In Chapter 2 we construct rings and modules of quotients with

respect to an additive topology, following the approach of

briel ([10],bl]). This construction is a special instance of

the construction of associated sheaves of presheaves for an

additive Grothendieok topology! in these notes, however, we will

not pursue that course. Rings and modules of quotients are then

described in terms of injective envelopes (Johnson and Wong [114]

and Lambek [46]). The main result of this chapter is the theorem

of Popescu and Gabriel [62] which asserts that every Grothendieck

category 1
C is the category of modules of quotients for a

suitable topology on the endomorphism ring of a generator for C .

The proof we give for this theorem is a simplified version of

Popescu's proof [12], due to J. Lambek [48] and J.E. Roos [un-

published].
 VI

In Chapter 3 we treat some aspects of rings of quotients

related to finiteness conditions. In particular we prove a

theorem of Popescu and Spircu [104] which characterizes flat

epimorphisms in the category of rings as a special class of

rings of quotients. In this context we also discuss rings of

fractions, i.e. rings obtained by inverting elements of some

multiplicatively closed subset of a ring.

Chapter 4 contains some material on self-injective rings.

Various well-known characterizations of quasi-Frobenius rings

are given here. These results are used in Chapter 5 where ws

discuss maximal rings of quotients and classical rings of

quotients. Necessary and sufficient conditions for these rings

to be regular, semi-simple or quasi-Frobenius are given (results

due to Gabriel [ 3 ~ , Goldie [91], Mewborn and Winton [541,

Sandomierski [681, and others).

The references at the end of each section are intended to tell

the reader where he can find a further discussion of the treated

topics. Their purpose is not to record credit for the results of

the section.

I am grateful to J.E. Roos for allowing me to use his preiimi-

na~j manuscripts for [66 I.

Stockholm, December 1970

Bo StenstrSm
Some notation

All rings have identity elements. A denotes a ring and all

modules are right A-modules, unless otherwise is stated.

The category of right A-modules is denoted by Mod-A , and

we write MA to indicate that M is in this category.

E(M) is the injective envelope of MA .

If L is a submodule of M and x ¢ M , then (L:x) =

J(A) ~ or simply J , denotes the Jacobson radical of A .

 Chapter I. Torsion theory

§ I. Preradieals.

A preradioal r of Mod-A assigns to each module M a submodule

r(M) in such a way that every homomorphism M-~N induces

r(M)--~r(N) by restriction. In other words, a preradical is a

subfunctor of %he identity functor of Mod-A .

The class of all preradicals of Mod-A is a complete lattice,

because there is a partial ordering in which rl~r 2 means that

rI(M)Cr2(M ) for all modules M , and any family (ri) of pro-

radicals has a supremum ~ ri and an infimum ( ~ r i , defined in

the obvious ways. If rI and r 2 are preradioals, one defines

preradicals rlr 2 and rl:r 2 as rlr2(M ) = rl(r2(M)) and

(rl,r2)(M)Irl(M) r2( Irl( )) . A preradical r is called

idempotent if rr = r and is called a radical if r:r = r , i.e.

if r(M/r(M)) = 0 for every module M .

Proposition I.I. For every preradioal r there exists a largest

A
idempotent preradical r smaller than r , and there exists a

smallest radical ~ larger than r .

A
Proof. To obtain r , we define a sequence of preradicals by trans-

finite induction as follows: if 8 is not a limit ordinal we put

r ~ = rr 8-I , and if ~ is a limit ordinal we put r8 = N ra .

In this way we obtain a decreasing sequence of preradicals r~

A
and we put ~ =Nr ~ . It is easy to see that the preradical r is

~dempotent, and that i% in fact is the largest idempotent preradical

smaller than r .

is obtained in a dual fashion. If ~ is not a limit ordinal we

set r = r_l:r , i.e. r is given by r (M)/r _I(M ) = r ( M / r _ l ( M ) ) ,

and for a limit ordinal ~ we set r8 = ~ ra(M) . This ~ v e s an

ascending sequence of preradicals ~r , and we set ~ -~r .

Lemma 1.2. If r is a radical and LOt(M) , then r(M/L) = r(M)/L .

Proof. The canonical homomorphism M-¢M/L induces r(M)--~r(M/L)

with kernel r(M)nL = L , so r(M)/LCr(M/L) . On t h e o t h e r hand,

the canonical homomorphism M/L-~M/r(M) maps r(M/L) into zero,

so r(MIL)Cr(M)IL .

Proposition 1.3. (i) If r is idempotent, then so is also ~ .

(ii) If r is a radical, then so is also r .

Proox. To prove (i) we have to shows

(I) If rI and r2 are idempotent, then so is also rl,r 2 .

(2) If ri (i~I) are idempotent, then so is also Z ri •

For (ii) we have to show:

(3) If rI and r2 are radicals, then so is also rlr 2 .

(4) If r. (i~I) are radicals, then so is also ~r..

i 1

(i): We have rl(M) C (rlzr2)(M) C M , and since rI is idempoten~

this gives rl(M) C r l ( r l z r 2 ) ( M ) C rI(M ) , so rl(rlsr2) = r I .

Therefore we obtain from (rl:r2)(rl~r2)(M)/rl(rl:r2)(M) =

r2((rl:r2)(M)/rl(rl:r2)(M)) that (rlzr2)(rl:r2)(M)/ rl(M ) =

= r2((rl:r2)(M)/rl(M)) = r2r2(M/rl(M)) = r2(M/rl(M)) = (rl:r2)(M)Irl(M).

Consequently rlzr 2 is idempotent.

(2): From ri(M)C ~r~(M) C M we get ri(M ) = ri(~jrj(M)) , so

j~

3
(3): Since rlr2(M) C r2(~) , we get from Lemma 1.2 ~hat

rlr2CM/rlr2(M)) = rlCr2(M)/rlr2CM)) = 0 .
(4): Since j~rj(M) C ri(M) , we get from Lemma 1.2 that

Nri(~Inj
rj(~))i = Nri(M)l~
rS(N~i " o.

Proposition 1.4. The following assertions are equivalent for a

preradical r :

(a) r is a left exact functor.

(b) If L CM , then r(L) = r ( M ) n L .

The proof is easy. A consequence of this result is that a left

exact preradical always is idempotent.

To each preradical r we associate two classes of modules, namely

L = (MI r(~') = M }
~r = T~ I rCM) = 0 }.
Clearly:

Proposition I.~. ~r is closed under quotient modules, while

is closed under submodules.

=r
Corol!ar ~ 1. 6 . If M ¢ ~r and N c ~r ' then HomA(M,N) = 0 .

Examples:

I. Let A be an integral domain. For every module M we let t(M)

denote the torsion submodule of M , and let d(M) denote the

maximal divisible submodule of M . Beth t and d are idempotent

radicals, t is also left exact, which is not the case for d .

2. Let A be an arbitrary ring. For each right module M we let

s(M) denote the socle of M , i.e. the sum of all simple submodules

of M , and let r(M) denote the intersection of all maximal proper

submodules of M . s is a left exact preradical, while r is a

radioal.
Exercisesz

I. Show that every preradical commutes with direct sums, and that

a left exact preradical commutes with directed unions of

submodules.

2. Suppose r is an idempotent preradical. Show that ~(M) is

the largest submodule L of M such that r(L/L') $ 0 for

every L' ~ L .

3. Show that if r is a left exact preradical, then also

is left exact.

4. Show that a preradical r is left exact if and only if r is

idempotent and T
=~ is closed under submodules.

References: Radicals have been studied by many authors, usually

in the context of general categories or the category of rings.

The case of module categories has been treated e.g. by Amitsur[~|,

Kurosch [ ~ and Maranda [ S ~ . For some further references to

the russian literature, see Math. Reviews 35, no. 234.

§ 2. Torsion t h e o r i e s

Definition. A torsion theo,r~ for Mod-A is a pair (~,~) of

classes of right A-modules such that

(i) ~om(e,~) = 0 for all ~ 6T, F ~F,

(ii) Tm and F are maximal classes having property (i).

The modules in T are called torsion modules and the modules

in F are torsion-free.
 Any given class C
w
of modules generates a torsion theory In

in the following way:

= I H (c,F) o all C c

rorall F

Clearly this (~, F) is a torsion theory, and we also note that

T
S
is the smallest class of torsion modules containing C .

Dually, the class ~ o o6enerates a torsion theory (T, F)

such that F is the smallest class of torsion-free modules

containing ~ .

We can characterize those classes of modules which may appear

as the class of torsion modules for some torsion theory! such a

class of modules will be called a torsion class . (Classes will

henceforth be assumed to he closed under isomorphisms).

Proposition 2.1. A class of modules is a torsion class if and

only if it is closed under quotients, direct sums and extensions.

Proof. A class C is said to be "closed under extensions" if

for every exact sequence 0-bL-eM -- N - e 0 with L and N

in ~ , also M e ~ .
Suppose (~, ~) is a torsion theory. ~ is obviously closed

under quotients and direct sums, since Hom(QTi?F) =~Hom(Ti,F) •

Let O-~L-oM-#N-vO be axaot with L and N in T . If F

is torsion-free and =sM-VF , then ~ is zero on L , so =

factors over N . But also Hom(N,F) = 0 , so a = 0 . Hence

MGT •

Conversely, assume C to be closed under quotients, direct sums

and extensions. Let (~, ~) be the torsion theory generated by

C . We want to show that _C = T , so suppose Horn (T, F) = 0

for all F s F . Let C be the sum of all submodules of T

belonging to ~ . C then also belongs to ~ , since ~ is closed

under direct sums and quotients. To show that C = T , it suffices

to show that T/C e F . Suppose we have a:C'-eT/C where C' • C .

The image of a is a module in C , and if a = O , then we would

get a submodule of T which strictly contains C and belongs to

C , since C is closed under extensions. This would contradict

the maximality of C , and thus we must have a = 0 , and T/C e F .

Dually one shows ( by working in an abelian category one obtains

this immediately by duality):

Proposition 2.2. A class of modules is a torsion-free class if

and only if it is closed under submodules, direct products and

extensions.

Let (7, ~) be a torsion theory. It follows from 2.1 that

every module M has a unique largest submodule t(M) belonging

to T ~ called the torsion submodule of M . In this way we obtain

a preradical t of Mod-A , and it is easily verified that t

is an idempotent radical. Conversely, given any idempotent

radical t of Mod-A ~ the pair of classes (~t' ~t ) defined

in § I is a torsion theory. As a result we obtain:

Proposition 2.3. There is a I-I correspondence between torsion

theories and idempotent radicals.

Proposition 2.4. Let r be an idempotent preradical. ~ is the

idempotent radical associated with the torsion theory generated

by
Proof. We know from I.I and 1.3 that ~ is the smallest idempotent
radical containing r . It must therefore correspond to the

smallest torsion class containing ~r ' which is just ~ .

Proposition 2.~. Let C be a class of modules closed under

quotients. The torsion class generated by C consists of all

modules M such that each non-zero quotient of M has a

non-zero submodule in C .

Proof. Let T(~) be the c l a s s of modules M satisfying the

announced condition. T(~) obviously contains ~ , and if

is a torsion class containing ..~, then it is easily seen that

~T(~) . It therefore remains for us to show that T(~) is

a torsion class. T(C_) is clearly closed under quotients.

Suppose (Mi) is a family of modules in T(~) . If a:QMi-~N

is a non-zero epimorphism, choose Mi such that ~ is non-zero

on M i . Since a(Mi) contains a non-zero submodule in ~ , so

does also N . T(~) is thus closed under direct sums.

Let O-eL~ M-eNdO be exact with L , N in T(~) . Suppose

azM-~M' is a non-zero epimorphism. We get an exact commutative

diagram
0 0 0

0 --~ K ~ L - - ' 4 K --~ K t ~ 0

L L L
0 -'~ L --~ M --J N ---D O
L l I
0-'~L/Kf~L-~ M'-* N' --m 0
t J t
0 0 0

If K f~L # L , then L/Kf~L has a non-zero submodule in C and

so has then also M' . If LCK , then M' ~ N' and so M' still

has a non-zero submodule in C . Hence M e T(_C) .

 Of particular interest for us will be the torsion theories

(T, F) where ~ is closed under submodules.

Pro~sition 2 6 ~et (~, ~) be a torsion thsory and let t

be the corresponding radical. The following assertions are

equivalent:

(a) T is closed under submodules.

(b) F= is closed under injective envelopes.

(c) t is left exact.

(d) If LCM , then t(L) = L N t ( M ) .

Proof. (c)4~(d) was remarked in 1.4.

(a)~(d): If LCM , then certainly t(L)CLNt(M) . But L~t(M)

is a torsion module, since it is a submodule of t(M) , so we

must have t(L) = L ~ t ( M ) .

(d) ~ ( b ) : Suppose F e F . Then t(E(F))f%F = t(F) = 0 . But

F is essential in E(F) , so this implies t(E(F)) = 0 .

(b)~a): Suppose T c T and CCT . There is a commutative

diagram
C t ; T

o/t(c),,,, ~(c/t(c))

But E(C/t(C)) is torsion-free, so ~ = 0 . This implies a = 0

and hence C = t(C) c T .

A torsion theory with the properties stated in 2.6 is called

hereditary. Applying 2.6 to Proposition 2.3~ we obtain:

Corollary 2.7. There is a I-i correspondence between hereditary

torsion theories and left exact radicals.

Proposition 2.8. A hereditary torsion theory is generated by

the family of those cyclic modules A/I which are torsion modules.

Proof. A module M is torsion-free if and only if Hom(A/I,M) = 0

for all A/I c T .

A hereditary torsion theory is thus uniquely dstermined by the

family of right ideals I for which A/I is a torsion module.

In the next § we will characterize these families of right ideals.

Prp~ositi0n 2. 9 . Let ~ be a class of modules closed under

quotients and submodules. The torsion theory generated by

is hereditary.

Proof. We will show that the class of torsion-free modules is

closed under injective envelopes. Suppose F is torsion-free

and there exists a non-zero a:C-~E(F) with C ~ ~ . Then

Im a ¢ C and Ff%Im a is a non-zero submodule of F belonging

to C , which is a contradiction.

Example s:

1. Let A be an integral domain. The "standard" torsion theory

is hereditary. An example of a non-hereditary torsion theory is

given by ~ = [divisible modules), ~ = ~reduced modules}.

The next two examples generalize the standard torsion theory to

arbitrary rings.

2. Let A be an arbitrary ring and let ~ be the class of

modules of the form M/L where L is an essential submodule

of M . The torsion theory generated by ~ is called the Goldie

torsion theory. Since ~ is closed under quotients and submodules,

the Goldie torsion theory is hereditary and is generated by the

 10

family of cyclic modules A/I where I is an essential right

ideal of A . M is a Goldie torsion module if and only if each

non-zero quotient of M has a submodule @ 0 in ~ which we

may assm, e cyclic (2.5). Hence:

Proposition 2.10. M is a Goldie torsion module if and only if

for each submodule L~M there exists 0 @ x ¢ M such that

(L:x) is an essential right ideal.

3. The torsion theory cogenerated by E(A) will be called the

Lambek torsion theory. It is hereditary, because we have:

Proposition 2.11. M is a Lambek torsion module if and only if

Hom(L,A) = 0 for every submodule L of M .

Proof. If LCM and a:L-~A is a non-zero homomorphism, then

a may be extended to M--*E(A) . So every Lambek torsion module

has the mentioned property. The converse follows from the fact

that if a:M-~E(A) is a non-zero homomorphism, then Im a ~ A = 0

-I
so a (A)"~A is non-zero.

Exercises:

I. Show that in the definition of a torsion theory (7, [) one

may replace the axiom (ii) by:

(ii)': Each module M has a submodule T c T such that

M/T ~ ~ ! moreover =T and F= are closed under

isomorphisms.

2. Let ~ be any class of modules. Let T(~) be the class of

of modules M such that each non-zero quotient module of M

has a non-zero submodule in C . Show that:

 11

(i) T(C_) is a torsion class!

(ii) if ~ is a torsion class containing ~ , then ~DT(~)

(note that ~T(~) in general).

3. Let T
I
be a hereditary torsion class, generated by a class

C
--
closed under quotients and submodules. Show that if T'
Z
is

a hereditary torsion class contained in T , then T' is

generated by C~' . (Hint: use 2.5).

4. (i) State and prove the dual of Proposition 2.9.

(ii) Show that a hereditary torsion theory (~, ~) is cogenerated

by the family of injective modules E(A/I) such that

A/I e F .

(iii) Let E be an injective module. Show that the torsion

theory cogenerated by ~ E~ is hereditary.

5. A module M is called torsionless if for every 0 $ x e M

there exists a:M-eA such that a(x) $ 0 .

(i) Show that M is torsionless if and only if M is a

submodule of a direct product of copies of A .

(ii) Show that the class L of torsionless modules is closed

under submodules and direct products.

(iii) Let T
W
be the class of modules M for which Hom(M,A) = O

Show that the following assertions are equivalents

(a) E(A) is torsionless.

(b) L is closed under injective envelopes.

(c) %) is a hereditary torsion theory.

(d) (~, ~) is the Lambek torsion theory.

If direct products of projective modules are projective,

show that these conditions are equivalent to:

 12

(e) E(A) is projective.

(Hint: show 3o show

suppose given O~ L~ M ~M/L'O0 with L , M/L in ~ !

choose maximal K with K~L = 0 and remark that KCM/L

and K + L is essential in M ).

References: Alin [11, Dickson ~21], Mishina and Skornjakov [~].

Exercise 5 is based on a paper by Wu, MOohizuki and Jans [~].

§ 3. Topologies

In this § we will establish a i-i correspondence between

hereditary torsion theories and certain families of right ideals,

called "topologies".

Definition. A right additive topology (or Just topology) is a

non-empty family ~ of right ideals of A , satisfying:

T I. If I ~ ~ and a s A , then (l:a) C ~ .

T 2. If I is a right ideal and there exists J ~ ~ such that

(l:a) 6 ~ for every a ~ J , then I ~ ~ .

Note that T i , together with the asst~mption that ~ is non-empty,

implies A C F .
i

Lemma 3.1. A t o p o l o ~ F is a filter, i.e. satisfies:

T 3. If J ¢ F and JCI , then I s F .

T 4. If I and J are in F , then l(~J e F .

Proof. T 3- If a ¢ J , then (l:a) = A ¢ F , so I s F by T 2.

T 4: If a c J , then ((IHJ):a) = (l:a)i~(J,a) = (l:a) s F_

by T I, so IftJ ~ F by T 2.
 IS

The axioms T I, 3 and 4 state that A is a topological ring

in which F is a fundamental system of neighborhoods of zero.

A non-empty family of right ideals satisfying T I, 3 and 4 is

called a pretopology.

Proposition 3.2. There is a i-I correspondence between:

I) Non-empty families of right ideals of A satisfying T I.

2) Non-empty classes u
C of modules such that M ~ C
m
if and

only if every cyclic submodule of M is in C .

Proof. If F is a family of right ideals satisfying T I, we

define _C as the class of modules M for which Ann(x) ~ F_

for every x c M . C obviously satisfies (2). COnversely, if

C is the given class of modules, we put F--~I~A/I ~ CI.

This F satisfies T I, for if I c F and a e A , then left

multiplication by a induces an exact sequence

0 -~ (I:a) --0 A .-e A/I

which shows that A/(l:a) C A/I .

It remains to verify that these correspondences are the inverses

of each other. Starting with F we get C = ~ M~ Ann(x) s F for

all x S M~, and from this we get (I~ A/I s C ~ = ~ I ~ (I:a) ~ Z

for all a S A~ = F by T I. On the other hand, if we start with

_c, we f i r s t get Z and then A (x)

~ M | each cyclic submodule e C ~ = C .

Proposition 3.3. There is a I-I correspondence between:

l) Pretopologies on A .

2) Classes of modules closed under submodules, quotients and

direct sums.

3) Left exact preradicals of Mod-A .

 14

Proof. (I) 4-~(2): This is a particular case of 3.2. For it is

easy to see that if F is a pretopology, then the class of

modules M for which Ann(x) ¢ ~ , all x e M , is closed under

submodules (by 3.2), quotients (by T 3) and direct sums (by T 4).

Conversely, if ~ is the given class of modules, then ~ ={I

A/I e ~ is a pretopology, for T I is satisfied by 3.2, T 3 is

obviously satisfied and so is also T 4 because A/I~J is a

submodule o f A/I(~A/J .

(2)*=#(3): If ~ is the given class of modules, we let r(M)

be the larges~ submodule of M belonging to C . r is a left

exact preradical. Cbnversely, if r is a left exact preradical,

then the class ~r is closed under quotients and submodules.I%

is also closed under direct sums, for if M i c ~r ' then

r ( ~ M i ) ~ M j = r(Mj) = ~j f o r each J , so r(~Mi) =@M i

(or use § I, Exercise I). The correspondence between 2 and 3 is

clearly I-i.

Note in particular that given the pretopology ~ , the preradical

r is obtained as r(M) ffi{ x C IAnn(x) ¢ }.

Theorem 3.4. There is a I-I correspondence between:

I) Topologies on A .

2) Hereditary torsion theories for Mod-A .

3) Left exact radicals of Mod-A .

Proof. We have already established the I-I correspondence (2)4-@(3)

in Corollary 2.7. ~f F
m
is a topology, then the corresponding

left exact preradical t , defined above as t(M) ={ x c M

Ann(x) s [~, is a radical. In fact, if ~ S M/t(M) and Ann(E) s [ ,

then Ann(~) = { a i xa ¢ t(M)~ = { a ~ Ann(xa) ¢ [ ~ = I a~ (Ann(x):a)

¢~}, and i% follows from T 2 that Ann(x) c ~ and hence ~ ffi 0 .

 15

O~ t h e o t h e r hand, i f (~, ~) i s a hereditary t o r s i o n t h e o r y ,

then the corresponding pretopology Z'( I~A/I e T} satisfies

T 2. For if I is a right ideal such that (Isa) e ~ for all

a ¢ J for some J c F , we consider the exact sequence

0 --@~/l~J ---eA/l--~ A/lu J --~ 0

where A/IuJ ¢ ~ s i n c e i t i s a q u o t i e n t module o f A/J e ~ ,

and also ~/I~J e T since a e J implies ((I~J):a) =

= (Isa) ~ F . Since T is closed under extenaions, it follows

that A/I e ~ and hence I c ~ . This concludes the proof of

t h e theorem.

If F__i and F_2 are topologies on A , we say that FI is

,eakor than F_2 (and is stronger Zl)if Zl C

It is clear that any intersection of topologies is a topology,

so if E is any family of right ideals of A , there exists a

weakest topology 3(E) oontaining ~ . The c o r r e s p o n d e n c e

between p r e t o p o l o g i e s and c l a s s e s o f modules d e s c r i b e d i n 3.3

preserves inclusions. It follows in particular that if _E is

a pretopology, then J(E_) is the topology corresponding to the

hereditary torsion theory generated by ~ A / I ~ I 6 E }.

Examples,

I. The family E of essential right ideals is a pretopology,

but does not always satisfy T 2. The corresponding left exact

preradioal is usually denoted by Z , and one calls Z(M) the

sin~ular submodule of M . The hereditary torsion theory

 16

corresponding to J(~) is the Goldie torsion theory (§ 2,

Example 2). The Goldie torsion radical ~ is the smallest radical

containing Z . The transfinite process which leads from Z to G

(Proposition i.i) gets stationary at an early stage! in fact we

have, in the notation of I.I:

Proposition 3.~. G = Z2 .

Proof. We first remark that Z(M) is an essential submodule of

G(M) for every module M . For if L C G(M) and L N Z(M) = 0 ,

then Z(L) = L , Z(M) = 0 and L is thus torsion-free, which

implies L = 0 . We now have Z2(M)/Z(M ) = Z ( M / Z ( M ) ) D G(M)/Z(~) ,

so z2(M) ~ G(M) .

~ 0 ~ o e i t i c n !.6. If ~ is the family of essential right ideals,

then J(~) = { I ~ there exists J e ~ such that ICJ and

(I:a) e ~ for all a e J}.

Proof. If I satisfies the condition, then r ¢ J(~) by T 2.

If I e J(~) , then A/I is a Goldie torsion module, so A/I =

Z2(A/I) . Ws R i t e Z(A/I) = J/I . The f o = u l a defi~ing Z2

~ves A/J = Z(A/J) a~ hence J c ~. ~e also have Z(J/I) =

= Z2(A/I) = Z(A/I) = J/I , so (l:a) s ~ for all a e J .

2. T"he right ideals of the topology corresponding to the Lamhek

torsion theory are called dense. We denote this toplogy by ~ .

Proposition 3.7. ~ is the strongest topology on A such that

A is torsion-free.

Proof. A is trivially torsion-free for ~ . If (~, ~) is any

hereditary torsion theory such that A 6 F , then T contains

the Lambek torsion modules and the corresponding topology is

weaker than u
D .
 17

A more explicit description of D is given by:

Proposition 3.8. A right ideal I is dense if and only if

(~r:a) has no left annihilators for any a c A .

Proof. Suppose (Iza) never has left annihilators. ~f there

exists a non-zero homomorphism A/I-@E(A) , then there exists

0 @ x c E(A) such that xI = 0 . Let a @ A be such that

0 ~ xa @ A , If b ¢ (l:a) , then xa~ - 0 , and hence xa is

a left annihilator of (I:a) , a contradiction.

Suppose conversely that I is dense, and suppose there exist

a , b in A with b(l:a) = 0 . We get a commutative diagram

f
A/(I:a) ~ A

h:
where f(~) = bc , g{~) = a. and h e x i s t s because E(A) is

injective. I dense implies h = 0 and henoe b = 0 .

Corollary 3.~. Every dense right ideal is essential in A .

The ring A is called right n o , s i n g u l a r if the singular right

ideal Z(A) is zero.

Proposition 3.10. If A is right non-singular, then every essential

right ideal is dense, and the Lambek torsion theory coincides

with the Goldie torsion theory.

Proof. If I is an essential right ideal, then also (l:a) is

essential for every a c A , and so non-singularity implies that

(Y:a) can have no non-zero left annihilators.

 18

3. Let F be the family of those right ideals which contain a

non-zero-divisor. F is a topology if and only if A satisfies

the "right Ore condition", i.e. for every non-zero-divisor s

and arbitrary a ~ A there exist a non-zero-divisor t and

b C A such that at = sb .

4. Let S be a multiplioatively closed subset of A (<by this

we also understand that I c S ). The family F_(S) = { right

ideals I i for every a c A there exists s ~ S with as e ]~}

is easily verified to he a topology.

5. Let C be the class of semi-simple modules, which is closed ~

under submodules, quotients and direct sums. The corresponding

pretopology consists of all finite intersections of maximal right

ideals, and the corresponding left exact preradical s associates

to each module M its socle s(M) . The corresponding torsion

radical is as usual denoted by ~ . Note that the torsion-free

modules are those modules which have zero socle.

Pr0position 3.1.!,. When A is a right noetherian ring, ~(M) is

the sum of all artinian submodules of M .

Proof. A module is called artinian if it satisfies DCC (descending

chain condition) on submodules. Put t(M) = sum of all artinian

submodules of M . t is clearly a left exact preradical. It is

not quite so obvious that it is a radical, and here we need the

noetherian hypothesis. We write t(M/t(M)) = L/t(M) , and want

to show that every finitely generated submodule L' of T. is

artinian. We have ,.'It(I, ') -- I , ' / L ' n t(M) = Z,' + t ( I ~ ) / t ( l ~ ) C

CLIt(M) and h e n c e L'It(L') is a r t i n i a n . Since A is n o e t h e r i a n ,

 39

also t(L') is finitely generated and hence artinian. Since the

class of artinian modules is closed under extensions, it follows

that L' is artinian.

t thus defines a torsion theory, and this torsion theory is the

same as the one given by ~ , because the torsion-free modules

are clearly the same in both cases.

The ring A is called right semi-artinian if every non-zero

module has non-zero socle. In that case we have no torsion-free

modules # 0 ~ and thus every module is a torsion module, i.e.

~(M) = M for all M . In particular we see from 3.11 that a

right noetherian and semi-artinian ring is right artinian.

Proposition 3.12. A is right semi-artinian if and only if

every hereditary torsion theory is generated by a class of

simple modules.

Proof. If A is right semi-artinian, then the torsion class

generated by all simple modules is Mod-A . It follows from § 2,

Exercise 3 that any hereditary torsion theory is generated by a

class of simple modules.

The converse follows from the trivial fact that Mod-A is a

hereditary torsion class.

Exercises:

I. Show that if F is a topology and I , J c F , then also

II~ F .
B

2. Suppose the ring A is commutative and F_. is a pretopology

such that every ideal in F

w
contains a finitely generated

ideal in F . Show that if the product of two ideals in F

 20

always is in F , then
m
F is a topology.

3. (i) Show that A (as a right A-module) is a Goldie torsion

module if and only if Z(A) is an essential right ideal in A .

(ii) Show that the ring Z_/p2~ (where p is a prime number)

is a Goldie torsion module over itself.

4. (i) Let I be a right ideal. Show that the family of right

ideals containing I is a topology if and onl~ if I is a

two-sided ideal and 12 = I .

(ii) Show that the following assertions are equivalent for

a hereditary torsion class T :

m

(a) T is closed under direct products.

(b) The corresponding topology is of the form {J~ J D I

for some right ideal I .

(C) ~ ={M~ MI = O ~ for some ideal I •

References, Alin Ill , Bourhaki ~o] (p. 157), Gabriel [Zl] (p. 411),

Xar da , ishina and sko S ov [Sb], Roos l).

§ 4. S~able torsion theories

In this and the following sections of this chapter, we will

consider some more particular aspects of torsion. The results

of these sections are not necessary for the understanding of the

following chapters.

A hereditary torsion theory (5' ~) is called stable if

is closed under injective envelopes. Alternative characterizations

of stable torsion may be given by using the concept of "comple-

mented" (or "closed") submodules. A submodule L of M is

 21

complemented if there exists a suhmodule K of M such that

L is a maximal submodule having the property that K f~ L = 0 .

In such a case, K + L is an essential submodule of M .

Lemma 4.1. L is complemented in M if and only if L = E(L)~IM

(i.e. L has no essential extension within M ).

Proof. Suppose L is maximal such that L ~ K - 0 . There is an

isomorphism E(M) • E(K) • E(L) which reduces to the identity

map .on K + L . Therefore E(L)f~MDL and E(L)~ M ~ K = O ,

and by maximality of L it follows that E(L)(~M = L .

Suppose conversely that L is "essentially closed" in M .

By Zorn's lemma there exists a maximal submodule K of M such

that K~L = 0 . Then also L is maximal with respect to

L~K = 0 , for if L' D L and L'~K - 0 , then L+K is an

essential submodule of L' + K , which implies that L is essential

in L' and hence L = L' .

It should be remarked here that E(L) has in general no unique

imbedding in E(M) .

Proposition 4.2. The following properties of a hereditary torsion

theory are equivalents

(a) The theory is stable.

(b) The torsion submodule t(M) is complemented in M , for

every module M .

(o) Every injective module is the direct sum of a torsion module

and a torsion-free module.

 22

Proof. ( a ) @ (b): Since t(M) is a torsiom module, we have

E(t(M))C t(E(M)) by stability. But t(M) is an essential

submodule of t(E(M)) , for if 0 @ L C t(E(M)) , then L is

a torsion module and LNM ~ 0 , so L ~ t(M) ~ 0 . Therefore

we must have E(t(M)) = t(E(M)) . Left exactness of t now

implies t(M) = t ( Z ( M ) ) n M = Z~t(~))nM , and ~em~a ~ 4.1

~ves (h).

(b) ~ ( O ) : If E is an injective module, then t(E) is also

injective by Lemma 4.1, so t(E) splits off.

(c) ~ ( a ) : If M is a torsion module, then E(M) = T ( D F where

T is a torsion module and F is torsion-free. Since T D M ,

we must have F = 0 .

Proposition 4.3. Let r be a left exact preradical such that

the torsion theory associated to ~ is stable. Then ~(M) is

the unique maximal essential extension of r(M) in M , for

every module M .

Proof. We first show that r(M) is an essential submodule of

~(M). Suppose L~(M). If L~r(M)= 0, then r(L)= 0 by

left exactness. This implies 0 = ~(L) = L .

Suppose r(M) is essential in some submodule L of M . By 4.2

there exists K CL such that ~(L) is maximal with respect to

~(L)NK = O . But then r(K) = 0 so we must have KNr(M) = O .

Hence K = 0 and ~(L) = L by the maximality condition. This

&,ives Ln~(M) = ~ ( ~ ) =L, and so ~C~(M) .

 23

An i m p o r t a n t example of s t a b l e t o r s i o n is g i v e n byt

Proposition 4.4. The G o l d i e t o r s i o n t h e o r y i s s t a b l e .

Proof. If M is a torsion module, consider the exact sequence

0 N E(K) --. E(.)/M o

where also E(M)/M is a torsion module by definition of the

Goldie theory. It follows that E(M) is a torsion module.

Reference, Gabriel [~I] (P. 426).

§ 5. T o p o l o g i e s f o r a commutative n o e t h e r i a n r i n ~

Our purpose i s t o d e t e r m i n e a l l t o p o l o g i e s f o r a commutative

n o e t h e r i a n r i n g . The f i r s t s t e p i n t h a t d i r e c t i o n i s t a k e n by

d e t e r m i n i n g a l l i n J e o t i v e modules.

Proposition ~.1. If A is a right noetherian ring, then every

d i r e c t l i m i t o f i n j e o t i v e modules i s i n J e c t i v e .

Proof. Let (Ea) be a d i r e c t f a m i l y o f i n j e c t i v e modules and

let fzl-~ l~m E~ be a homomorphimm from a right ideal I .

Since I is finitely generated, there exists an index 8 such

that f(I) lies in the image of E in lim..~E= . Choose a

finitely generated submodule L of E~ which maps onto f(I) .

The kernel K of L f(I) is then also finitely generated.

Since K goes to zero in lira E= , there exists ~~ ~ such that

K goes to zero already in E~ . We may then factor f over the

inJeotive module E~ , and it follows that f may be extended

to A
 24

Diagram for 5. I: I

K ~ L ~ f(I) /

r / '
I ~: E C.@ lira E
E i P -"* "

E
Y
Coro!!ar~ }.2. If A is right noetherian, then every direct sum

of injective modules is inJective.

Prcpositicn ~. 3. If A is right ncetherian, then every injective

module is a direct sum cf indecomposable injective modules.

Proof. Let E be an inJective mcdule. Consider all families

(Ei) of indecomposable injective submodules such that the su~

~E i in E is direct. By Zorn,s lemma there exists a maximal

such family (Ei) I . From 5.2 we know that the sum (D E i is an

I
injective module, sc we can write E = (v~Ei) (D E' . We want to
~M

show that E' = 0 , and fcr this it suffices tc show that every

injeotive module E' $ 0 contains an indecomposable direct summand.

Let x be any element • 0 in E' . Consider all injective sub-

modules E" of E such that x ~ E" . By Zcrn's lemma and 5.1

there exists a maximal such submodule E" . We have E' = E" ~ L ,

and we assert that L is indecomposable. For if L = L' (~ L"

with L' and L" $ 0 , then (E"~L')n (E"eL") = ~' sc one

of the modules E"(~L' or E"~L" does nct contain x , which

contradicts the maximality of E" .

 25

It is possible to prove the converse statement, namely that if

every inJeotive module is a direct sum of indecomposable modules,

then A must be right noetherian (see e.g. Faith-Walker [30])? )

Lemma ~.4. An injective module E is indeoomposable if and only

E = E(A/I) where I is an irreducible right ideal.

Proof.Easy exercise.

For the res% of this § we assume A t o be a commutative

noetherian rin 6. We write Spec(A) for the family of prime ideals

of A . For each module M we define

Ass(M) - { 2 ~ Speo(A) I ~ = A ~ ( x ) for some x , o in M~.

Lemma ~.~. If M % 0 , then Ass(M) , ~ .

Proof. The family of ideals Ann(y) for y , 0 has a maximal

member Ann(x) since A is noetherian. We assert that Ann(x)

is a prime ideal. Suppose a, b ¢ A with abe Ann(x) but

b ~ Ann(x) . Then bx , O and since Ann(bx)~ Ann(x) , we mus~

have Ann(x) = Ann(bx) . NOW a e Ann(bx) implies a e Ann(x) ,

and Ann(x) is p r i m e .

Proposition ~.6. There is a i-I correspondence between prime

ideals of A and isomorphism classes of indeoomposable inJeotive

A-modules, g i v e n by 2 ~ E(A/2) •

~)Our "proof' of this result in a paper in Arkiv fSr Matematik,

vol. 7 (Theorem 2 on p. 428) is insufficient! one has to use some

kind of "counting' argument, as in [~O].

 26

Proof. If ~ is a prime ideal, then ~ is irreducible and

E(A/~) is indecomposable by 5.4. If ~ and E are prime ideals

such that E(A/~) ~ E(A/~) , we consider both A/~ and A/~ as

submodules of E(A/2 ) , and we then have A / 2 n A / a ~ 0 since

E(A/~) is indecomposable. But Ann(x) = ~ for every non-zero

x ¢ A/~ and likewise Ann(y) = ~ for non-zero y e A/~ , so

it follows that ~ = E .

Finally, let E be an indecomposable injective module. By 5.5

there exists ~ e Ass(E) , and since E is indecomposable we

must have E = E(A/~) .

Let Z be a topology on A , and let (~, ~) be the oorreB-

ponding torsion theory.

Proposition ~.7. (~, ~) is generated by the cyclic modules of

the form A/~ where ~ e Spec(A)~ .

Proof. We must show that if a module M has the property that

Hom (A/~, M) = 0 for all prime ideals ~ in ~ , then M is

torsion-free. We decompose the injeotive envelope of M as

E(M) = ~ E ( A ~ ) . For each stsnmand we have A/~M @ 0 , and

for each non-zero x e A/~ we have Ann(x) = ~ since

is prime. By the hypothesis on M this implies ~ #~ , and

hence each A/~ is torsion-free. Then E(M) = e Z ( A / £ ) is

torsion-free by heredity, and so M is torsion-free.

Lemma 5.8. For each finitely generated module M there exists

a sequence of submodules M = Mn~Mn_ 1 D ..... ~ M 0 = (0)

such that M i/Mi_ 1 ~= A/~ i where ~i e Spec(A) .

M is a torsion module if and only if each ~i e ~ .

 27

Proof. Suppose we have constructed M o , ..... , Mi_ I . If

Mi_ I 4 M , we have Ass(M/Mi_l) % ~ by 5.5, so there exists

H iC M such that Mi/Mi_ I =" A / ~ . Since M is noetherian, the

chain will eventually stop. It is clear that M is a torsion

module if and only if each Mi/Mi_ I is a torsion module.

The0rem ' ~.~. A family F of ideals of A is a topology if

and only if there exists P_ C Spec(A) such that

_• . { I ! I~ 20"..-.'20 where ~ c P}.

Proof. Suppose F is a topology. If I c F , it follows from

5.8 that there exist ~o ''''' ~n c S p e o ( A ) N F such that

~o" "'" "~n annihilates A/I , i.e. 20-....-20 C I . Conversely,

if I contains a product of ideals in Spec(A)N_F , then I ¢ F

by § 3, Exercise I.

On t h e o t h e r hand, if PC Speo(A) i s g i v e n and F = { I I

I ~ 20......20 with ~ c _P } , then _r i s c l e a r l y a p r ~

t o p o l o g y and i s i n f a o t a t o p o l o g y by § 3, E x e r c i s e 2.

LmRa p.lO. If P 9 Spec(A) and x ¢ E(A/~) , x # 0 , then

Ann(x) is an irreducible l~-primary ideal.

Proof. Since E(A/~) is indeooaposahle, it is an injeotive

envelope of A/Ann(x) . Lemma 5.4 gives that Ann(x) is an

irreducible ideal, hence primary (~g~], p. 209). Let ~ be

the prime ideal associated to Ann(x) . Let n be the smallest

integer s ~ h that ~¢A~(x). Considering AI2 and AIA~(x)

as submodules of E(A/2 ) , we have A/pN~n-I/Ann(x) ~ 0 . Hence

there exists a e E n'l such that E % 0 in A/Ann(x) and

 28

Ann(Z) = 2 • We have ~aCAnn(x) , so Ann(~) D ~ • But on the

other hand, if bE = 0 , then ba e Ann(x) and b ~ ~ since

Ann(x) is ~-primary. Hence ~ = Ann(F) = ~ .

Corollary ~.iI. If x c E(A/~) , then j~nx = 0 for some n>O .

Proof. Since Ann(x) is ~-primary, n C Ann(x) for some n .

Propositiqn 5.12. Every hereditary torsion theory on A-Nod

is stable.

Proof. Let M be a torsion module. We decompose E(M) as

E(M) = e E ( A / , ~ ) • S i n c e each A/~. has non-zero intersection

with the torsion module M , we must have ~i s ~ " rt follows

from 5.9 and 5.11 that each E(A/~i) is a torsion module,

and hence E(M) is a torsion module.

Example:

Let A be a commutative noetherian ring and let s be the

preradical assigning to each module M its socle s(M) (§ 3,

Example 5). From Proposition 4.3 we get that the corresponding

torsion submodule ~(M) is the maximal essential extension of

s(N) in N

Exercises:

Let A be a commutative noetherian ring with a topology ~.

i. Show that a module M is a torsion module if and only if

Ass(N) C ~ •

2. L e t M be a f i n i t e l y g e n e r a t e d m o d u l e and l e t I be a n y

ideal. Show that InN = O for some n> 0 if and only if

 29

I is contained in every ~ c Ass(M) .

References: Gabriel ~1~ (oh. 5, §4 and 5), Matlis [g2], [$3].

§ 6. ~-in~eetive modules

Let ~ be a topology on A .

Definition. A module E is F_-in~ective if ExtI(A/I,E) = 0

for every I ¢ F .

Before giving some alternative descriptions of F_-injectivity,

we will introduce a convenient terminology. If M is a module,

a submodule L of M is called an F-submodule if (Lzx) e

for every x ~ M , i.e. if M/L is a torsion module. The family

of all F_-submodules of M is a filter (as a consequence of T3

and T4) which we denote by ~(M) . Note that T 1 implies E(A) = ~ .

Lemma 6 . 1 . If K c F_(L) and L ~ ~(M) , t h e n K ¢~(M) .

Proof. We have an exact sequence O-~L/K-eM/K-eM/L-eO where

L/K and M/L are torsion modules. Then also M/K is torsion.

Proposition 6.2. The following properties of a module E are

equivalent:

(a) E is ~-injective.

(b) ExtI(M,E) = 0 for every torsion module M °

(c) If L ¢ ~(M) , then every homomorphism L--~E may be extended

to a homomorphism M--*E .

Proof. (a) =@ (o), Let L c F_(M) and f:L--,E be given. In the

usual way we may assume that there is a maximal extension

f':L'--~E of f , where L C L' C M . Then also L' e ~(M) by

T 3. Suppose there exists x c M such that x ~ L' . Let I =

 SO

= (L':x) C ~ , and let ¢:I-9E be the homomorphism a(a) = f,(xa) .

By (a) we may extend ~ to A , i.e. there exists y ¢ E such

that a(a) = ya . We may then define g:L' + x A - ~ E as g(z + xa) =

= f'(z) + ya . g extends f' , which is a contradiction.

(c)~ (b): We choose an exact sequence O-~K-~F-~M-,O where

F is free. Since Ann(x) ¢ ~ for every x c M , we have K e ~(M) .

The exact sequence

Hom(F,E)---, ~om(K,E) ;ExtZ~M,E) , 0 ,

together with (c), shows that ExtI(M,E) = 0 .

(b) =~(a) is trivial.

Definition. An F_-in~ective envelope of M is a monomorphism

M C, E such that E is [-inJective and M ¢ ~(E) .

Propositi0n 6.3. The su~od~e {x ~ E(M) I (,,x) c ~ } of E(M)

is an ~-injective envelope of M .

Proof. Clearly the subset in question is a suhmodule of E(M) .

It suffices to show that E' = { x C E(M)~ (M:x) C ~ } is [-injecti,.

Suppose we are given f:l-~E' with I ¢ F . f may in any case be

extended %o a homomorphism g:A-*E(M) , so we get a commutative

diagram

O ---~ I A : A/I ~ 0

o ~ E'~E(M) ;E(~)/E, , 0

o-.,
L
E,/~ ~
t
E(~)/M ~
U
E(,)/E, , 0

If h % 0 , then E(M)/E' contains a non-zero torsion submodule,

and from the lowest row we see that this contradicts the fact that
 3!

E' is the maximal submodule of E(M) containing M as an

F-submodule. Hence h = 0 , and g actually maps A into E'

It follows from this result that the F_-injective envelope of

M is ~ i q u e up to isomorphism! it A l l he denoted as ~(M) .

We will give another u s e f ~ description of the F_-injective

envelope. Recall first that a hereditary torsion t h e o ~ is

cogenerated by some injective m o d ~ e (§ 2, Exercise 4).

~oposition 6.4. If C is an injective m o d ~ e cogenerati~ the

torsion theory a s s o o i a t ~ to F , then

~CM) = { x ~ E(H) I f(x) = o for all f,~(M)~0 with fCM) = O ~

for e v e ~ mod~e M .

~oof. Suppose x c ~(M) and consider any f:E(M)--~C with

f(M)} = 0 . There exists I ~ ~ such that xI C M . Then f(x)I =

= f(xI) = 0 and f(x) = 0 since C is t o r s i o ~ f r e e .

Conversely, a s s ~ e x has the p r o p e r ~ that f(x) = 0 for

f:E(M)~C with f(M) = 0 . We want to show that the right ideal

I = (M:x) is in f , i.e. to show that Hom(A/I,C) = 0 . If

g:A/I~C is given, we may extend it to give a commutative dia~Tam

A/~ i ~ E(M)/M

where i(~) = xa . Since hf(M) = 0 , we have hf(x) = 0 which

implies g = 0 .
 $2

Examples:

i. Proposition 6.2 holds also when F is only a pretopology.

If e.g. ~ is the family of essential right ideals, then the

filter ~(M) consists of the essential submodules of M , while

the [-injective modules are precisely the injective modules.

2. Let D be the family of dense right ideals. The members of

D_(M) are usually called dense submodules of M .

Exercises:

1. An e x a c t sequenoe 0--~ L ~ M-, N---0 is called F-~ure if

for e~ex-J x G N with Ann(x) c F there exists y c M mapping

onto x such that Ann(y) = Ann(x) . L is then an F_-pure

submodule of M . Show that the following properties of a module

L are equivalent:

(i) L is F-injective.

(ii) Every exact sequence 0 -, L -~ M -~ N-, 0 is F-pure.

(iii) L is an F-pure submodule of E(L) .

References: Walker and Walker [8|].

Chapter 2. Categories of modules of quotients

§ 7. Construction of rin~s and modules of ~uotients

Let F be a right additive topology on the ring A . For each

right module M we will define its module of quotients with

respect to F . This we will do in two steps. The first step we

take is to define

M(F) = ~ HomA(I,M ) , I e F ,

where %he direct limit is taken over the downwards directed family

of right ideals. We want to give A([) the structure of a

ring and M([) that of a right A(~)-module. For this we need:

Lemma 7.1. If I , J c F and =:Ir-~A is a homomorphism,

then a"l(J) c F .
Proof. For each a ¢ I we have (a-l(j):a) = { h l=(ah) ¢ J } =

=(j,,(a)) c Z hyTl. so -l(j) c Z by T2.

We define a pairing M(~) " A(~)---~ M(~) as follows: suppose

x c M(~) , a ¢ A(~) are represented by ~:J-~M and a:I--~A !

we then define xa c M(~) to be represented by the composed map

-l(j) ~ J-"~M ,

using Lemma 7.1. It is easy to see that xa is well-defined,

i.e. is independent of the choice of the representing homomorphisms

, a . One also easily verifies that the pairing M([)- A(Z)--~M([ )

is biadditive. When M = A , this makes A(~) into a ring, and

in the general case it makes M(~) into an A(~)-module.

 34

There are canonical homomorphisme

~M ' ~ ~" H°mA(A'M) ---" lira HomA(I,M ) M(F_) •

In particular~ ~A is a ring homomorphism. By pullback along

~A we may consider each M(F_) as an A-module. ~M is then

A-linear. The assignment M~-P M(F_) is a fm~ctor Mod-A-'~Mod-A(F_)

which is left exact, because Hom is left exact and direct limits

are exact in Mod-A . We denote by LsMod-A--~Mod-A this functor

followed by the forgetful functor Mod-A(F_)--@ Mod-A .

Let t denote the torsion radical associated to F .

Lemma 7,.2.,, Ker(M ~M-~-~M(F)) = t(M) .

Proof. If ~M(x) = 0 ~ ~hen x goes to zero already in some

Hom(I,M) , I ~_F . This means that the map a~xa is zero on I ,

i.e. xl = 0 and x e t(M) . This arguzent may he reversed.

Lemm~ 7.3. M is a torsion module if and only if M(F ) . 0 .

Proof. If M(F ) = 0 , then t(M) = K e r ~ M = M by 7.2. Suppose on

the other hand that M is a torsion module. Let x e M(F ) be

represented by ~:J-PM . If we can show that Ker~ c F , it

will follow that x = 0 . For each a c J there exists Ia ¢ F

such that ~(a)I a = 0. Put K = ~" aI a . Then KCKer~ , and

aeJ
for each a e J we have (K:a) D I a , so (K:a) C F and it

follows that K e _F by T 2. Hence Ker ~ e F .

Lemma 7 . 4 . If x G M(F__) is represented by ~ : I - e M , I e F ,

then the diagram
i c---p A

M
I
commutes, where ~(a) = xa .
 35

Proof. If a e I , then xa is represented by the composed morphism

A = (I:a)-@l-@M given by b~(ab) = l(a)b , so xa = ~ M l ( a ) .

An immediate consequence of this isl

Lemma 7.~. Coker~M is a torsion module.

As the second step in our construction of modules of quotients we

apply the functor L once more to M([) . This results in a ring

AF (called the ring of quotients of A with respect to [ ) and

an AF-mOdule MF for each A-module M . The description of this

is considerably simplified by,

Lemma 7.6. L carries the monomorphism M/t(M)C~ M(_F) into an

isomorphism (Kit(M))(~) ~ ~ .
Proof. Apply the left exact funotor L to the exact sequence

0 --* MIt(M) --* M(_F)--~ C o k e r ~ M --~ 0

and apply Lemmas 7.5 and 7.3.

We have thus obtained the formula

M F = lim H O m A ( I , M/t(M)) , I S F .

One verifies that the ring structure of AF and the module

structure of MF aregiven by the following pairing MFXAF--~M F :

let x e MF be represented by ~:J-~M/t(M) ,

a e AF -,, - a:I--~A/t(A) ,

induces J/t(J)~/t(X) and we have ~/t(J)~A/t(A) by left

exactness of % ! xa e ~ is represented by
u

.-ICJltCJ)) --~ItCJ)--~MltCM)
 36

For each f:M-~N in Mod-A one ge%s fF:MF-eNF in Mod-A F ,

w ~ - - m

which gives a funotor q:Mod-A-~Mod-A F . There are canonical

J

homomorphisms ~M:M-*MF of A-modules! in particular

TA:A'@A F is a ring homomorphism. For each f e HomA(M,N) we

m

g e t a commutative diagTsm
f
M ~ N

MF - .~ N F
-- fF --

Note t h a t Ker TM = t(M) and %hat Coker ~M also is a torsion

module.

When the torsion theory is stable (§ 4), the formula for

simplifies somewhat:

~oposition 7.7. When the torsion theory is stable, one has

= I~ HomA(I,M ) for every module M .

Proof. Since I~ is an exact functor, the sequence

O---*~(M) - ~ M -~ M/t(M) - ~ 0

induces an exact sequence

0-~ lim
---@ Hom(I,t(M))--~ li~ Hom(I,MI---~lim Hom(I,M/t(M))--~
- lim tl(1,t(M)) .

The first term is zero by Lemma 7.3. If E is an injective envelope

of t(M) , then E is a torsion module by hypothesis. The sequence

0- t(M) ,E :o

induces the exact sequence

lim Hom(I, EItCM))----~ lim Extl(I,t(M)) ; 0

where the first term is zero, again by Lemma 7.3. Hence the last

term of the long exact sequence is zero.

 37

We w a n t t o s t u d y the image category of the fm~ctor q . For this

purpose we i n t r o d u c e r

Definition. MA is F-olosed if the oanonioal maps

M ~ Ho~A(A,M ) ~ HomA(I,M )

are isomorphisms for all I c F .

Thus M is F-closed if and only if M is both torsion-free

and ~-injective (as defined in § 6). For every F__-elosed module

M we get an isomorphism T M Z M ~---~MF . Conversely we have:

Propositiom7.8. MF is F-closed for every module MA .

Proof. To show that MF is torsion-free, it suffices to show

that if M is torsion-free, then M(_F) is torsion-free. Suppose

x e M(F_) and xJ = 0 for some J ¢ F_ . L e t x be represented by

:I - ~ M . BY Lemma 7.4 we have a commutative diagram

I ( r A

= xa

M ?K P N(F-)

so ~M~ is zero when restricted to If~J s F . But ~M is

a monomorphism, so ~|INJ = 0 and x = 0 .

Next we show that MF is F_-injective. Suppose we are given

f-I-~M F with I e F . Consider the pullback diagram

j c ~ I

H/t(xl ' ",-

w h e r e also J is a right ideal. We have I/J ~= Coker ~ =

 38

= Coker~M , which is a torsion module, so it follows from T 2

~at J ¢ F . Lemma 7.4 now tells us that we may extend g to

a homomorphism h:A-~M F . h is also an extension of f , because

m

h~I and f are equal on J , and therefore their difference

factors over the torsion module I/J , and ~ torsion-free

then implies hJI = f .

Corollar~ 7.~. The f u l l subcategory of Mod-AF consisting of

m

modules o f t h e form MF is equivalent to the full subcategory

of Mod-A consisting of F-closed modules.

Let m
C be the full subcategory of Mod-A consisting of

F-closed modules. We have a number of interesting functorsz

m

q -e.

is the forgetful functor,

q is the functor M~M F ,

~t is t h e f u n c t o r M~MQA AF '

i is the inclusion functor,

is the functor M~ M?

considers each F_-closed module as an AF-mOdule and is full

a n d faithful.

We h a v e Ja = q and i = 'T~,J • ~ induces a natural equivalence

a i ~ - - I d C . On t h e o t h e r h a n d , i a = LL = T m q .
 39

PTo~osition 7.10. a is a left adJoint of i .

Proof. We must show that the canonical map

HomA(MF,N ) --# HOmA(M,N )

m

is an isomorphism when N is F-closed.

~M
M • r y~
!
i l
f~ IfF

N ~ P N~

It clearly is an epimorphism, so it remains to show that it is a

monomorphism. The exact sequence

o --, M F --P C o k e r T M ~ 0

induces

where the first term is zero since Coker~M is a torsion module.

The desired conclusion now follows from the observation that

Hom( It(M),N) Hom(M, ) .

We will show in the following sections that the category _C~ of

F-closed modules is very well-behaved, in fact it is an abelian

category with exact direct limits, although the inclusion functor

i is not exact.

Examples Let D be the family of dense right ideals. The ring

AD is called the maximal (or complete) ring of quotients of A

and will he denoted by Qm " Since A is _D-torsion-free, we

have ~ = ~ Hom(I,A) , I ¢ D .
 40

Exercises,

I. Show t h a t if E CF are topologies, there is a ring homo-

morphism A E --~ A F .

2. Show that if A is a commutative ring, then also every AF

is commutative. (Hint: one can reduce the problem to showing

that if I ~ F and =, ~ : I - ~ A , then a~(x) = ~=(x) for

xcI2).

3. Let A = K[X,Y] where K is a field, and let m_ be the

maximal ideal _m = (X,Y) . Consider the topology F w (I

I D m_n for some n~. Show that,

(i) AF = A .

(ii) If M = A/(X), then NF = KIT,I/T], where r is the

class of Y in M . (So if I:A-~M is the canonical

epimorphism, then fF is not an epimorphism, and the

functor q is n o t e x a c t ) .

4. L e t F b e a t o p o l o g y on A and assume A F-torsion-free

(so A is a suhring of AF). For each right A-submodule I

of ~ , put I~=~q c AF ~ qICA}.

(i) If I ¢ F(AF) (of. § 6), show that I~ HomA(I,A ) .

m

(ii) Call I F-invertible if there exist al,..,a n e I

and ql'""" 'qn ~ I4t such that 1 - ~ aiq i . Show that the

following properties of I are equivalent:

(a) I is _F-invertible.

(C) I is a finitely generated projective module and I e _F(~).

m

(Hint, of. [131, p.

 41

5. Show that for any ring A and module MA one has

E(M)IZ(M) = lim Hom(I,M)

where I runs through the downwards direo%ed family of

essential right ideals of A .

References, Bourbaki [lO], (p. 157 and following), Gabriel [ 31],

(p 4n and fonowing), Gol~an [33], ~oos [66] (oh. 1).

§ 8. Modules °f quotients andS-in'co rive envelopes

Let F be a topology on A .

Pro~qsition 8.!. ' If M is a torsion-free module, then

MF~ EF(M ) as AF-mOdules.

Proof. MF is ~-injective and M~M is a torsion module by

7.5, so MF is an ~-injective envelope of M . If EF(M) =

= ~x e E(M) I (M:x) e ~ } for a fixed injective envelope E(M)

of M , then the isomorphism M F C EF(M ) is AF-linear by 7.9.

Let us desribe explicitly the AF-mOdule structure of ~ (M )

for a torsio~free module M S~se x ~ ~(.) , and q ~

is represented by =:I-~A/t(A) . Since ~(M) is F_-closed,

there exists a unique y e ~(M) such that xa(a) = ya for all

a e I (note that EF(M ) is a module over A/t(A) ), and then

xq = y . This description of the module structure is applicable

also for the F-closed module E(M) .

Proposition 8. 2. If M is torsion-free, then E(M) is an

injeotive envelope of MF in Mod-~ .

m
 42

Proof. The inclusion map MF= EF(M)--~E(M ) is AF-linear by 7.8.

Since it is an essential monomorphism in Mod-A , it is obviously

essential also in Mod-A F . It only remains to show that E(M)

is injective as an ~-module. This follows from:

Le~aa 8.3. Every torsion-free injective A-module is inJeotive

over AF .
M

Proof. Let E be torsion-free injective over A . Suppose

N'--~N is any monomorl~ism in Mod-AF and f:N'--~E is

~-linear. f extends to a homomorphism g,N--~E in Mod-A .

For each x C N , consider the two maps AF-aeE given by

g'(q) = g(xq) and ~'(q) = g(x)q . They are both A-linear

and coincide on A . Hence g'-g" factors over a homomorphism

Coker ~ A - - V E . But Coker ~ A is a torsion module (Lemma 7.5)

and E is torsion-free, so g, = ~' , and g is ~-linear.

For the remaining part of this § we will assume that AA is

torsion-free. Thus F is contained in the the topology D of

dense right ideals, and the torsion theory is cogenerated by

an inJeotive module C = E(A) ~ F . Let H be the endomorphism

ring of the module CA , and consider C as bimodule HCA .

The ring HomH(C,C ) is usually called the double centralizer

of C . Since C is F_-closed, H is also the endomorphism

ring of C as an AF-module. It follows that there is a commuta-

tive diagram of canonical ring homomorphisms

i ""'* Ho
(C,c)
 43

Theorem 8.4. ~ is an isomorphism between AF and the double

centralizer of the cogenerating inJective C .

Proof. We w i l l exhibit an inverse /* of ~ . Suppose

e HomH(C,C ) . Let psC-~E(A) be the canonical projection.

We want to show that p~(1) e ~ ( A ) by using Proposition 6.4.

Let ftE(A)--pC be any A-linear map such that f(1) = 0 .

Extend f to ~sC-~C by defining T(F) - 0 , where C~=

. ~(A) ~F . Then f(p~(1)) . ~(~(1)) ~ ~(f(1)) ~ 0 since

is H-linear.,H~ce ps(1) e ~ ( A ) = ~ . We may now define

as ~C~) = p~(1) c ~

i s an a d d i t i v e map. C l e a r l y ~k= id.. Then ~X~-~ ,

and i t o n l y r e m a i n s t o show t h a t ~ i s a monomorl0hism . Thus

we m u s t show t h a t if 8(1) e F , then 8 = O . For every x e C

there exists h e H such that h(1) = x and h(F) = 0 . Then

~(x) = ~(h(1)) . h(~(1)) = 0 .

Exaaples.

I. Let _F be the Goldie t o p o l o g y (§ 3, Example l). For each

torsion-free (i.e. non-singular) module we g e t I ~ - E(M) .

2. The maximal right ring of quotients of A is the double

centralizer of E(A) .

E x e r o i see:

Let E be an inJectiva module. The E-dominant dimension of

MA is said to be ~n (notation: E--dora.dim M ~ n ) if there

exists an exact sequence

 44

0 --* M .......:..... E l - - - - + ....... ----*E n

where each E. is a direct product of copies of E . Consider

1
the hereditary torsion theory cogenerated by E . Show that:

(i) M is torsion-free if and only if E-dom.dlm M ~ I .

(ii) M is closed if and only if E-dom.dim M $ 2 . (Hint: M

is ~-injective if and only if E(M)/M is F_-torsion-free).

References: Lamhek [471,[48], [49], Morita ['57], Tachikawa [74],

Turnidge [79], Wong and Johnson [114].

§ 9. C0reflective Subcate6ories of Mod-A.,

We prepare the study of the category of F_-closed modules by

a consideration of a more general situation:

Definition. A full subcategory C of Mod-A is called oorefleo~ive

if the inclusion functor i:C-PMod-A has a left adjoint a .

In such a case there exists a natural transformation ~ :l-~ ia

such that the bijection

where N ¢_C , is given by a~a~M .

Examples:

I. If F
m
is a toplogy on A , then the category of F-closed

modules is coreflective in Mod-A .

2. If (T, =F) is a torsion theory for Mod-A , then the category

of torsion-free modules is coreflective in Mod-A , with

~,(M) = M/t(M) .
 45

Let C be corefleotive in Mod-A . If M is a module in C ,

we may choose a(M) = M with the identity map as ~M "

Lemma 9.1. If there exists =:a(M)-@M such that =~M = IM '

then ~M is an isomorphism.

Proof. From the preceding remark it follows that ~a(M) is the

identity map. The morphism a induces by naturality of ~ a

commutative diagram

l='a(M) [ ['M

2(~) -- , a(~)

H,noe ~M~ = a(a)~a(M ) . la(M) , and so a is the invers of ~ M

Proposition 9.2~ A corefleotive subcategory of Mod-A has

arbitrary limits and colimits.

Proof. Let ~ he a small category and G:~-~ a functor. Then

lim iG exists and we denote it by M , and let =d:M-~G(d)

denote the canonical projections. Since G(d) 6 ~ , there exist

~d:afM) -~ G(d) such that ~ d ~ M = =d "

X~ I "~ G(d)
a(~)
The family ~ 8d}dsD is compatible with the morphisms in D ,

for if X-d-.~, i. _D, then G(k)~dT~ = G(k)=d = ~d' =

= ~d' ~ M and hence G(k )Sd = 8d' " It is therefore induced a

 46

map 8:a(M)-~M such that ~d ~ = ~d for all d c ~ . Then

~d ~ M = ~d~M = ad ' so 8~M = IM " From 9.1 it follows that

~M is an isomorphism, and it is then clear that a(M) is a

limit for G in ~ . Note that we have obtained the formula

i( m G) . lim
__ iG
which also follows from the fact that a right adJoint functor

always commutes with limits (when these exist).

To prove that lim

__~ G exists in C is easier, because the

left adjoint a preserves colimits and we therefore have

a(lim iG) = lim aiG = lim G , since ai ~ I .

In other words, limits in C may be computed in Mod-A ,

while colimits in C are taken in Mod-A and then coreflected

into C .

Proposition 9.3. If ~ is coreflective in Mod-A and

a:Mod-A-*~ preserves kernels, then ~ is an abelian category

with exact direct limits and a generator.

Proof. C is preadditive since it is a full subcategory of

Mod-A . We have proved that C has limits and colimits. To prove

that C is abelian, it only remains to show that if a:M--~N

is a homomorphism in ~ , the~ the canonical map

~:Coker(ker a ) - - - ~ K e r ( c o k e r ~)

is an isomorphism. If we denote kernels and cokernels ~aken in

Mod-A by underlining them~ we have Coker(ker a) = a(Coker(ke F a))

and Ker(coker a) = Ker(a(coker a)) = a(Ker(coker a)) since a

preserves kernels. ~ is therefore an isomorphism.

 47

Next we show that direct limits are exact. Let u

D be a small

directed category and G , G':~-~ two functors with a mono-

morphism G-~G' . The induced morphism lim i G - e l i m iG' is

a monomorphism in Mod-A , and since a preserves monomorphisms,

it follows that ~ G-elim__e G' is a monomorphism in C .

Finally, it is easy to see that a(A) will be a generator

for ~.

Definition. A coreflective subcategory of Mod-A is called a

Giraud subcategory if the left adjoint of the inclusion functor

preserves kernels.

Thus if i:C-*Mod-A is a Giraud suboategory, then the left

adjoint a of i is an exact functor. It is important to notice

that the inclusion functor i is in general not exact! epi-

morphisms in C are not necessarily surjective maps. An

abelia~: category -with exact direct limits and a generator is

usually called a Grothendieck category. Proposition 9.3 thus

states that every Giraud subcategory is a Grothendieck category.

Conversely, the Popescu-Gabriel theorem (Theorem 10.3) states

that every Grothendieck category is a Giraud subcategory of

Mod-A , where A is the endomorphism ring of some generator of

the category.

References: Mitchell [102] (ch. V:5).

 48

§ i0. Giraud subcate~ories and the Popescu-Gabriel theorem

Proposition !0.i. if [ is a topology on A , then the ~-closed

modules form a Giraud subcategory of Mod-A.

Proof. The category of F-closed modules is coreflective by

Proposition 7.10. Since the functors L and i preserve kernels

and i is full and faithful, the relation ia = LL implies

that a preserves kernels.

Theorem 10.2. There is a i-I correspondence between topologies

on A and equivalence classes of Giraud subcategories of Mod-A .

Proof. We already know how to associate a Giraud subcategory to

a topology. Conversely, let C be a Giraud subcategory of

Mod-A and let a be the left adjoint of the inclusion functor

i:C--#Mo~-A . Let T be the class of modules M for which

a(M) = 0 . We verify that T is a hereditary torsion class.

Since a is exact, it is clear that T is closed under

I

extensions, submodules and quotient modules. Since a has a

right adjoint, a commutes with direct sums and hence T is

closed also under direct sums. To T there corresponds the

topology {I ~ A/I C =T~, which is the topology we associate

with the given Giraud subcategory C .

We will now show that the two maps

P
{topologies on A} --'--=* {Giraud subcategories of Mod-A )

are the inverses of each other. We first show that ~ ~ = id..

Let ~ be a toplogy and let ~ be the category of F-closed

modules. We must show that if a(M) = 0 , then M is an

F_-torsion module. But Y~ = 0 certainly implies M = t(M) .

 49

It remains to show ~= id.. Let ~ be a Giraud suheategory

of Mod-A with the inclusion i':~-~Mod-A and its left adjoint

a' . Let [ he the toplogy of right ideals I for which

a'(A/l) = 0 . We wan% to show that the category ~ of ~-closed

modules is equivalent to _D.

Mod-A
a'i i ai'

We first note that a'ia ~ a' , for if M ¢ Mod-A then the

exact sequence

0 • t(M)--~ M--~M F ~ CokerTM--*O

m

gives a'(M) ~ a ' ( ~ ) . Similarly we have ai'a' = a , beoause

the adjointness transformation ~:id.--~i'a' gives the exact

sequence

~M i'd'

and a'(~M) is an iso.o~his., so Ker ~ M and Coker ~ M

are F-torsion modules, and it follows that a(~M) is an

isomorphism a(M) ~ ai'a'(M) .

From these two natural equivalenoes we obtain a'i -ai'

a'i' ~ id. and ai'.a'i ~ ai ~ id. , and ~ and ~ are

thus equivalent.

We may now state the Popescu-Gabriel theorem.

 50

Theorem 10.3. Let ~ be a Grothendieck category with a generator

U . Put A = HOmc(U,U ) an~ let T:~-IpMod-A be the functor

T(C) = HOmc(U,C ) . Then:

(i) T is full and faithful.

(ii) T has a left adjoint S:Mod-A-@~ which is exact.

(iii) T induces an equivalence between ~ and a Giraud

subcategory of Mod-A .

Proof. Let us first see how (i) and (ii) imply (iii). Let Im T

be the full suboategory of Mod-A consisting of modules of the

form T(C) , C e C . We have a commutative diagram

T
C .~ Mod-A

Im T

and (i) states that T' is an equivalence. Define a = T'S:

Mod-A-@Im T . a is exact and is a left adjoint of i by (ii).

Im T is thus a Giraud subc~tegory of Mod-A .

Proof of (i): T = Hom(U,.) is faithful since U is a generator.

To see that it is full, we must show that if C , D 8 C and

~:HOmc(U,C)--~HOmc(U,D ) is A-linear, then ~ is of the form

@(f) = ~ f for some ~ C * D . Let (fi)I be the set of all

morphisms U - ~ C . There is a corresponding exact sequence

o_~K ~_a~ f
-------~ C ~0

where ~ is the direct sum of I copies of U . The morphisms

~(fi)'u ~ D induce a mo~hism h, ~--~ D . ~or each s ~ n d Ui

of UI we set K i = Ker fi = K f%U i .

 51

gi f"I
0--.-,K. 1
, ) U ~ C ,0

0 )K L
t g ) ~ ~ c - - o
JI
1,
D
For every s e Hom(U, Ki) we get by the A-linearity of ~ that

o. #(figi s) _ - ~ ( f i ) g i ~ , ~o ~ ( f i ) g i = o beca~, u i~ a
generator. It follows that hg = 0 , and h factors as h =~f

for some ~:C-~D . For each fi:U~ C we then have ~(fi) =

= hui = ~ f i "

Proof of the easy part of (ii), namely that a left adjoint S

exists: For this we max either use general existence theorems

for adjoints ([102], oh. V,§3), or we max proceed as follows.

Consider A as a preadditive category A with only one object)

A
and define in the obvious way a functor u:A-*C with image

object U . By a standard result in elementary category theory

([I02], p.106) we can extend u to a colimit preserving

functor S:Mod-A--~C . Since every module M is the colimit

of a functor i~A i (i ¢ I), where each Ai is A A , we get

HomA(M,T(C)) = HOmA(I~Ai,Hom~(U,C)) = *---limHomA(Ai,Hom~(U,C)) =

" 4--lim Hom~(Ui,C ) - lime.. H°m~(S(Ai)'C) " Hom,(lim~._~ S(Ai),C )
= ~om£(S(~Ai),C) = ~om~(S(M),C) , and t h ~ s S i~ a left

adjoint of T .

Before we go o n a n d prove the exactness of S , we m a k e the

following observation:
 52

Proposition 10. 4. Let M be any class of modules. There exists

a strongest topology ~ such that all modules in M are

~-closed. A module L is a torsion module for this topology

if and only if Hom(A,M) ~ P Hom(Ann(x),M) for every x c L

and M cM .

Proof. If ~ is any topology, then a module M is _E-closed

if and only if M is _E-torsion-free and _E-injective, which

is equivalent to requiring M and E(M)/M to be _E-torsion-

free(Proposition 6.3). It follows that the torsion theory

cogenerated by ~ E(M) ~ E(E(M)/M) I M C M__~, which is hereditary

by § 2, Exercise 4, defines the strongest topology for which all

modules in M are closed.

It remains to determine the torsion modules for this strongest

topology F_~ . If L is a torsion module, then Ann(x) e ~ for

each x ¢ L and hence Hom(A,M) ~ Hom(Ann(x),M) for all M 6 M .

Conversely, if a module L satisfies this later condition, we

may restate this as

Hom(C,M) = ExtI(c,M) = 0 for every cyclic submodule C of L ,

as one sees from the exact sequence

O-*Hom(A/Ann(x),M)--~Hom(A,M).-~Hom(Ann(x),M)-~Extl(A/Ann(x),M)-'tO
But if Hom(C,M) = 0 , then ExtI(c~M) ~ Hom(C,E(M)/M) .

Consequently we have Hom(C,M~E(M)/M) = 0 for every cyclic

submodule C of L and M e Mu
. This implies that L is a

torsion module, because of the following easily verified fact:

Lemma IO.~. If L and M are modules, then Hom(L,E(M)) = O if

and only if Hom(C,M) = 0 for every cyclic suhmodule C of L .

 53

We continue the proof of Theorem 10.3, where it remains to show

that S is exact. Let F be the strongest topology for which

all modules T(C) , C c ~ , are F_-closed. Let ~ be the

corresponding category of F - c l o s e d m o d u l e s . We h a v e a d i a g r a m

of f u n o t o r s

i
Im T ~ ~ Mod-A

a'i i'

where T' is an equivalence. We have i'a'i ~ i by the

definition of F . It now suffices to show that a'iT' = a'T

is an equivalence, because T = i'-a'T implies that the left

adjoint S of T is the composition of the left adjoint a'

of i' and the left adjoint of the equivalence a'T ! since

both these two adjoints are exact, S will he exact. Since

i'a'T = T is full an faithful, also a'T is full and faithful,

and i t r e m a i n s f o r u s t o show t h a t e v e r y module i n D is

i s o m o r p h i c t o a m o d u l e o f t h e form a'T(C) , i.e. that every

F-closed module M is isomorphic to T(C).

Thus let M be an F-closed module. Choose an exact sequence

m

(~) eA
("i j) ~A ~M ~ 0
I J
in Mod-A . Since the functor S has a right adjoint, it preserves

oolimits and t h e r e f o r e carries (~) into an e x a c t sequence

 54

s(alj)
(.~) e u ,ou ~s(M) ~0
I J
in C .

Lemma 10.6. The functor a'T,~-~ is exact and preserves

direct su~s.

We conclude the proof of the theorem before we prove the lemma.

By applying the Lemma to ( ~ ) and noting that A - T(U) is

~-closed, we obtain the upper exact row of the following diagram

in D :
a'Ts(~ij)
OA ....... ~@A a,TS(N) ,o
I J

fl u
M ~ 0
ei A ......... a,faij~, ~ ej A

The lower row is obtained by applying a' to ( ~ ) , and is also

exact. The diagram commutes because A - T(U) implies that aij

has the form ~iJ = i(~ij) ' and one has a'TSi = a'iT'Si =

= a'iai = a'i . We conclude from the diagram that M ~ a'TS(M) .

Proof of Laama 10.6, We already know the funotor a'T to be left

exact, so to prove exactness it will suffice to show that it

preserves epimorphisms. This means that if f,C'~ C" is an

epimorphism in C , then we should show that Coker T(f) is

an F-torsion module. By Proposition 10.4 this is equivalent

to showing that for each x ¢ T(C") we have

Hom(A,T(C)) -~ Hom((Im T(f) ,x),T(C))

for all C s _C . Define hsU-~C" such that T(h).A-~T(C")

maps i into x . From the pullback in C

 55

0 --~ K k p g ~ U --* 0

0--* K ~ C ' ------e C " - - t O

f

we get a pullback diagram with exact rows in Mod-A

0 ---~T(K) T(k) ~ T(P) ....T(g) ~ A

o---,
1 l 1
T(C")

and (Im T(f) :x) = Im T(g) . Note that A is the cokernel of

T(k) in the subcategory Im T , because T',~-~Im T is an

equivalence. Y% follows that if we have a homomorphism

Im T ( g ) - * T ( C ) , for some C c ~ , then it factors uniquely over

A = Coker T(k) , and this is precisely what we wanted to show.

It remains to show that a'T preserves direct sums. Actually

we prove a little more, namely that a'T preserves directed

~ions. Let (Ca) be a directed family of subobjects of C c ~ .

We must show that the cokernel of the monomorphism

f,U (ca)
a

is a torsion module. By Proposition 10.4 this means that for each

x C T(U Ca ) we shall show that Hom(A,T(C')) ~ Hom((Im f :x),

T(C')) for all C' 6 ~ . Define hzU--~UC a such that

T(h):A-~T(~Ca) maps 1 to x . From the pullback diagram

Pa ~ U

Ca ~ U Ca
 56

we obtain a commutative diagram

UT( ) - g .- A
t T(h)
UT(C.) T(U C=)
f
which is a pullback diagram because pullbacks are preserved both

by T and when taking direct limits in Mod-A . Therefore we

have (Im f :x) = Im g ~ U T ( P a ) . Now Hom(~JT(Pa),T(C'))

lim HomCT(P~),TCC')) ~ lim HomCP ,C') : H o m C ~ P a , C' ) =

4--
= Hom(U,C') ~ Hom(A,T(C')) , where we have utilized the fact that

exactness of lim
--~ implies ~Pa = U .

Example: Let A be any ring. Proposition 10.4 provides A with

a strongest topology ~ for which A is F_-closed. This top@-

logy is called the canonical topology on A .

Exercises:

I. Let ~ be a topology on Mod-A and let T:Mod-A-P~ be

an exact functor into an abelian category ~ , such that

T(M) = 0 for all F-torsion modules. Show that T has a

unique factorization T = T'a over the category of F_-olosed

modules. (The category of F_-closed modules is thus a

solution of a universal problem).

2. Let ~ be a Giraud subcategory of Mod-A and let a be the

left adjoint of the inclusion functor i . Show that:

(i) If E is an injective object in ~ , then ~ i(E) is

an injective module.

(ii) i preserves injective envelopes.

 57

(iii) I f the t o r s i o n theory corresponding to C is stable

(§ 4) and E is an inJeotive module, then a(E) is

injeetive in C .

Referencess Bucur-Deleanu [12] (oh. 6 , written by N. Popescu),

G~h~i.1 [31] (oh. 3), ~ b ~ [~] (§ 4), P o ~ o u - G ~ b ~ l [62],

Roos [66] (oh. I), ~akeuohi [112].
Chapter 3. General properties of rings of cuotients

§ II. Lattices of ~-p~e submodules

We will assume F to be a topology on A . For each module

M and submodule L of M , we define

={,. }.
The operation L ~ Lc is a closure operation on the lattice of

all submodules of M . Those submodules L for which Le = M

were called F-submodules of M in § 6. On the other hand, we

Note that L° = L if and only if M/L is torsion-free! in

particular, if M is torsion-free, then CF(M ) is the family

of ~-pure submodules of M (§ 6, Exercise).

Proposition II.i. C_F(M) is a complete modular lattice.

Proof. From the fact that L~L ° is a closure operation it

follows that C~(M) is a complete lattice with intersection

as meet ([18], Ch. 2.1). The join is given by V L i = (~ Li)C .

It remains to verify modularity. Let H , K and L be

members of C~(M) with H C K . Then Kn(H~ L) = K ° O (H + L) c =

= (K~(H + L)) ° = (H + ( K ~ L ) ) ° = K V ( K ~ L ) , using the

modularity of the lattice of all submodules of M .

Propqsition 11.2. If K c ~(M) , then there is a lattice

i~omorphism $ ( M ) - ~ ( K ) ~ven by L~ ~n K .
m m

Proof. If L ~ $(M) , then clearly L~ K ~ $(K) . The inverse

m

map is defined as L~L c , where the closure is taken in M .

 59

For if L C C_~(K) , then LO~K = { x c K 1 (L:x) ¢ ~ = L , while

if L e ~(M) , then (L~K) ° = LONK c = LnM = L .

In the following we will mainly be concerned with ~(M) when

is torsion-free. In case M is also F_-injective, we have:

Proposition 11.3. Let M be F_-closed. A submodule L of M

is F-closed if and only if L c ~(M) .

Proof. If L CM , then M/L is torsion-free if and only if

is F_-closed, as is seen from the exact sequence

0 = Hom(A/I,M)-~
E x t I ( A / I ,Hom(A/I,M/L)-~
L ) . . . ~~E x t.I ( A / I , M. )-. . . . . - - - = 0 .

Recall that a submodule L of M is called oomp!emented if

L is maximal with respect to L~K = O for some KC M (§ 4).

Proposition 11.4. Every complemented submodule of a torsion-free

module M is a member of C_F(M ) .

Proof. If L is maximal such that L~K = 0 , then LC~K c =

= ( L ~ K) ° = 0 c = 0 since M is torsion-free. We must then

have L° = L by maximality.

Proposition 11.~. The following properties of a torsion-free

module M are equivalent:

(a) C_F(M ) is a complemented lattice.

(b) C_~(M) consists of the complemented submodules of M .

(o) Every essential submodule of M is an F_-submodule.

Proof. (a)~ (b): Every complemented submodule is in ~(M) by

11.4. Suppose conversely L c Ca(M)

-L • By hypothesis there exists
 60

K CM such that KNL = 0 and (K + L) c = M . Let L' D L be

maximal with respect to L'NK = 0 . For each x ¢ L' we have

xlcK + L for some I ~ F . But xANK = 0 , so xI¢ K .

L C_~F(M ) implies x c L and hence L = L' is complemented.

(b)=~ (c): Let L be an essential submodule of M . Lc is

then both complemented and essential in M . Hence Lc = M ,

and L is an F_-submodule.

(C) =~ (a): If L C ~(M) , choose K maximal with respect to

KNL = 0 . K + L is then an essential submodule of M , hence

an F-submodule. Thus we have K~L = M and KNL i 0 .

Proposition 11.6. Let M be an F-closed module such that

C_F(M ) is complemented. Then:

(i) Every F_-closed submodule of M is a direct smnmand.

(ii) The endomorphism ring of M is regular (in the sense

of yon Nemnann).

Proof. (i). If L CM is F_-closed~ then L c ~_F(M ) by II.3.

Hence there exists KC M such that K~L = 0 and K + L is

essential, hence an F-submodule, in M . BY Proposition 6.2

we may extend the canonical projection K + L--~L to a homo-

morphism M - ~ L . This makes L into a direct su~mand of M .

(ii): Let f:M-, M be an endomorphism. Then Kerf c ~(M)

m

because if x e M and xlC Kerf for some I c F_ , then

f(x)l = 0 and M torsion-free implies x c Kerf . Kerf is

thus a direct su~,nand of M . Write M = Ker f~ K . f induces

an isomorphism f~K:K-*Im f , so also Im f is an F_-closed

 61

module. Hence Im f is a direct s,nmand of M , and we may

therefore extend (fJK) -I to a homomorphism h:M-PM . Then

f = fhf , and we have established the regularity of the

endomorphism ring.

The lattice C_~(A) has an interesting description as the

D

set of annihilators of subsets of an injective module. Recall

that .CF(A) = [ 1% A/I is torsion-free ~ . In the following

three propositions we let E be an injective module which

cogenerates the torsion theory corresponding t o F , i.e.

F = { I~ Hom(A/I,E) = 0 } . If S is a subset of any module,

we p u t Ann(S) = ~ a S A i Sa ,, 0 7 ,

Proposition 11. 7 . CF(A ) = ~Ann(S) ~ subsets S C E ~.

Proof. Suppose S C E . Then Ann(S) = N Ann(x) , where each

xsS
AnnCx) ¢ CF(A ) since A/Ann(x) C E is torsion-free. But C_F(A)
m

is closed under intersections, so also Ann(S) e CF(A ) .

Suppose on the other hand that I e ~(A) and put

s o{x Elxl - 0h Then Ann(S)= I . To show that every

a C Ann(S) belongs to I , it suffices to show that (l:a) 8 _F ,

i.e. that Hom(A/(I:a),E)= 0. Let a:A/(l:a)-~A/I be the

monomorphism a(~) = a-~ . For every f:A/(I:a)-@ E we get a

commutative diagram

A/(I:a) ~ A/I

f ~
~ g
E ~
where g(~) = xb for some x c E . Then necessarily xl = 0 ,
 62

so x c S . But then also xa = 0 , so g~ = 0 . Hence f = 0 .

Before the statement of the nex% result we need to make two

definitions. A lattice is called noetherian if every ascending

chain is stationary. A n inJective module is called ~-in~ective

if every direct sum of copies of the module is injective.

.Prgl0osition 11.8. The following assertions are equivalent:

(a) Every direct sum of torsion-free injective modules is

injective.

(b) E is Z - i n j e c t i v e .

(o) C_F(A ) is a noetherian lattice.

Proof. (a) ~ ( b ) is trivial.

(b) =~(c): By 11.7 it suffices to show that every strictly

ascending chain IlCl 2 C .... of annihilators of subsets of

E must be finite. If the chain were not finite, we could choose

for each n an element xn c E such that _.xnI

n = 0 but
w
.x• In+ 1 $ 0 . Put I = U I n and define f : I - ~ Q E as f(a) =
i i
= (xla , x2a,... ) . Note that f is well-defined! Since E is

-injective, f has the form f(a) = ya for some

Y = (Yl ~ Y2 ''''~ Ym' 0~...) e ~ E , which contradicts the

I
choice of the elements x .
n

(C)~ (a): Let (Ea) be a family of torsion-free injective

modules. Suppose we are given a homomorphism f:I-~QE= . It

suffices to show that f maps I into the sum of finitely

many E a . Suppose on the contrary that there exists an infinite

sequence of indices = , which we write as a = I, 2,... , such

 63

that Im f has non-zero coordinates in each E . Put

n
I n = f-l(E I @ .... ~) En) . The ascending chain IICC12 c C ...

is by hypothesis finite, which implies that for some n we

have I k C In c for all k Let a e Ik be some element for

which f(a) has non-zero coordinate xk in ~ , for some k> n .

o
Since a e In , we have aJ C I n for some J ¢ ~ . This gives

f(a)J C E 1 ~ .... • E n • But then xkJ = 0 , which contradicts

the assumption that Ek is torsion-free.

Proposition II.9: If C_F(A ) is noetherian, then [ contains

a cofinal family of finitely generated right ideals.

Proof. For each right ideal I of A we set IA = { x c E

xI = 0 } . Note that I 8 ~ if and only if I A = 0 . The

operations Ann and A define an order-inverting bijsction

between the set of right ideals of the form Ann(S) and the

set of submodules of E of the form I A . Since we have ACC

on the former set (by 11.7), we must have DCC on the sub-

modules I A . Let I be any right ideal in F . Consider the

family { jA IJ finitely generated right ideal ¢ I~ and let

J be a minimal member of this family. For each a e I , the

right ideal Jl = J + aA is also finitely generated C I and

satisfies Jl~ C j A . By minimality we must have Jl~ = J~ so in

particular JAa = 0 . Since this holds for all a e I , we have

JAI = 0 and thus J ~ C l A , which implies that also jA = 0

and JeF .
 64

We will now show that C~(A) is isomorphic to a corresponding

lattice of right ideals in A F , assuming for simplicity that

m

A is torsion-free, so that A is a subring of A F. Define

ZF = [right ideals J of AF I J N A S [}.

Proposition I!.I0. F_e is a topology on AF .

Proof. Suppose J ~ F_~ and q c $ . Then (J,q)n A --{a ~ A l

qa ~ J} ~ {a ~ A iqa ~ J - A ~ ~ Z, because Jn X ~ F_ and A is

an F_-submodule of AF , so we may use Lemma 6.1. Thus T I is

satisfied. Next we verify T 2. Let I be a right ideal of AF

such that there exists J e ZF for which (l:q) e F e for all

q ¢ J . Then Jf~A ¢ F and for each a e Jf~A we have {b e A

ah C I h A ~ ={ b s A F ~ab s I } G A S F . T 2 for F implies

I•A S F and hence I ¢ Fe .

Proposition II.II. AF is equal to its ring of quotients with

m

respect to Fe .

Proof. The ring of quotients of ~ is isomorphic to the

~e-injective envelope of A F , and is a submodule of E(AF) =

= E(A) (Propositions 8.1 and 8.2). But the F_e-injective

envelope of AF coincides with its F--injective envelope, as

m

one immediately verifies, and AF is F_-olosed.

Proposition 11.!2. If M is a torsion-free A-module, then

CF(M ) =~ C_Fe(MF) -

Proof. Since M is an F-submodule of MF , we have CF(M ) =~

~= CF(MF)_
_ by 11.2. It remains to see that ~(MF)_
_ = .CFe(MF) .
D
 65

Suppose L e ~F(MF) . L is then an F_-closed module by ii.3 and

is therefore an AF-mOdule. If x e MF has the property that

xJC L for some J e F_e , then x(JNA) C L and hence x e L .

Consequently L e _CFe(MF_)

Suppose conversely that L e _CFe(MF_) . If x e MF_ and xIC L

m

for some I ¢ ~ , then xlA F C L and hence IAF ¢ F..

e implies
m

. It follows that .

Examples,

I. Let ~ be the Goldie topology, i.e the topology generated

by the family of essential right ideals (§ 3, Example I). From

Propositions 11.5 and 11.6 we obtain:

Proposition I!.13. The endomorphism ring of a non-singular

injective module is regular.

For the Goldie topology one can prove the converse of Propo-

sition 11.9, namelys

Proposition !!.!4. The following assertions are equivalent for

the Goldie topology F :

(a) Every direct sum of non-singular injective modules is

injeutive.

(h) lattice $(A) of complemented right ideals is

noetherian.

(c) ~ contains a oofinal family of finitely generated right

ideals.

Proof. It remains to prove (c) ~ (a). Let (Ea) be a family of

 66

no~-singular injective modules, and let f : l - * ~ ) E a he a homo-

a
morphism, for an arbitrary right ideal I . Choose a right

ideal J such that l~J = 0 and I + J is essential in A .

By (c) there exists a finitely generated right ideal K e ~

contained An I + J . Extend f to I ~ J-'~ E~ by f~J = 0

and then restrict to a homomorphism g:K-e6 E ~ . Since K is

finitely generated~ g maps i n t o t h e sum o f f i n i t e l y maw

E a , and is therefore extendable to h:A--~6 E a • Since (~E a

a a
is torsion-free, the usual argument shows that hll = f (of.

the proof of Lemma 8.3).

2. Taking ~= { A} , we otain as a special case of Proposition

11.8 that A is right noetherian if and only if every direct

s~n of injeutive modules is injective.

3. Let D be the family of dense right ideals of A (§ 3,

Example 2). Then D_e is the family of dense right ideals of

Qm ' as one easily verifies by means of Proposition 3.8.

Exercises:

i. Let A be a regular ring and let ~ be the family of

essential right ideals. Show that:

(i) A is non-singular.

(ii) ~(A) is noetherian if and only if A is semi-simple.

2. Show that the following two properties of a topology ~ are

equivalent:

(a) If I I C 12 C .... is a countable ascending chain of

 67

righ~ ideals such that U I n e ~ , then some In e ~ .

(b) If I I C 1 2 C .... is a oountahle ascending chain in

~F(A) , then also t J I n S C_~(A) .

Show that these properties are satisfied if every I ¢

contains a finitely generated J e ~ .

3. Show that the following properties of a right self-injective

ring A are equivalent:

(a) A satisfies ACC on right annihilators of subsets of A .

(b) AA is ~-injective.

(c) Every projective module is injective.

4. The ring A is called right finite-dimensional ' if no right

ideal can be written as a direct sum of infinitely many

non-zero right ideals of A . Show that:

(i) A is right finite-dimensional if and only if every

right ideal is an essential extension of a finitely

generated right ideal.

(ii) Every right finite-dimensional ring satisfies the

conditions of Proposition 11.14. (Hint: use 3.6 to

verify ll.14(c) ).
5. Show that if E is an injective module and M is non-

singular, then every exact sequence O-*K-*E-*M--~O

splits. Using this, show that if E and E' are injective

submodules of a non-singular module, then also E + E' is

inJective.
 68

6. Show that the conditions of Proposition 11.14 are equivalent

to:

(d) Every non-singular module contains a unique maximal

injective submodule.

7. Let ~ be a topology on A . Show that F_e is the strongest

topology ~' on AF such that all F_-closed modules

i

(considered as AF-mOdules ) are ~'-closed.

~eferencss, A=endaris [85], Faith [27], C28] (§'7 ~d 8),

JTohnson [40], Teply [75],[76], utumi [8o].

§ 12. Finiteness conditions on to~ologies

In this § we will consider two kinds of finiteness conditions

on the topology F . The first one is introduced in the next

proposition, where C as usual denotes the category of F-closed

modules and i:C-~ Mod-A is the inclusion functor~ while

q:Mod-A--PMod-A F is the functor M~-~M F .

Proposition 12.1. The following assertions are equivalent:

(a) If I I C 12 C... is a countable ascending chain~ of right

ideals such that t ~ I k ¢ F , then In ¢ F_ for some n .

I
(b) Every direct sum of F--closed modules is F_-closed.

(b') Every direct sum of coumtably many F_-closed modules is

F--closed.

(c) The functor i commutes with direct sums.

(d) The functor q commutes with direct sums.

Proof. (a) ~ (b): Let { M } be a family of F_-closed modules.

 69

~M~ is of course torsion-free, and we must show that it also

is ~-injective. Let f:l-~$ Ma be any homomorphism with I e ~ .

Considering $ Ma as a suhmodule of the F_-injective module

Mm , there exists x = (xa) e ~ M a such that f(a) = xa for

all a e I . We only have to show that x e ~ M a . If this were

not true, there would exist an infinite set (al' m2'''" ) of

indices ~ for which xai @ 0 . Put In = { a e I ~ xaia = 0

for i~n~. Since f(1) C ~ M m , we have

I = U In . But (a)
n
then implies that I = In for some n . Now M a is torsion-free,
n
so x I = 0 implies xa = 0 , which is a contradiction.
n n
(h)~(c): Let (Ma) be a family of S-closed m o d u l e s . The

direct sum o f t h e m o d u l e s Ma in the category ~ of F_-closed

modules i s a(ei(Ma)) hy oposition 9 . 2 . So i f

e i(M=) is F_-closed, then i(~)Ma) =(Bi(M=) .

(o) ~ ( d ) = ~ ( b ) and (b) ~ ( b ' ) are rather o b v i o u s .

(b')~(a): Let I 1 C 12 C . . . be an a s c e n d i n g c h a i n w i t h

I = WI n c ~ . There is a well-defined canonical map I--tO A/I n .

Since I e F , one obtains a commutative diagram

I - ~ A

(D A/I n - ' @ (A/In)F_

where f(a) - xa f o r some x = (Xn) ~ e ( A / I n ) ~ • There e x i s t s

m such that xn = 0 for n)m . The image of I in A/I m

then lies in the kernel of A/Im--@(A/Im) F , so I/I m is a

torsion m o d u l e . The e x a c t s e q u e n c e
 70

0 ---* I/I m ....~ A/I m .....~ A/I ~ 0

where also A/I is torsion, shows that A/I m is torsion, and

hence I c F .
m --

We strengthen the finiteness condition somewhat by considering

topologies with the following properties:

P ~ p o s i t i o n 12.2. The following assertions are equivalent for

a topology F :

(a) ~ contains a cofinal family of finitely generated right

ideals.

(b) Every directed union of F_-closed modules is F_-closed.

(c) The functor i commutes with directed unions.

(d) The torsion radical t commutes with direct limits.

Proof. (a)=~(b): Let (Ma) be a directed family of F_-closed

submodules of some module. ~.~Ma is then torsion-free, so it

a
remains to verify that it is F_-injective. Consider any homo-

morphiem f:l--~tJM where I c [ . I contains a finitely

generated J c ~ . f maps J into some M a , so by the

F_-injeetivity of Ma there exists x c M= such that f(a) =

= xa for a c J . Since ~M= is torsion-free, one then has

f(a) = xa also for all a c I (by the same argument as in the

proof of Proposition 7.8).

(b)4~(c) similarly to the preceding Proposition.

(b)~(a): Write I c ~ as the directed union of finitely

generated right ideals I a . Then ia(A) = ia(1) = ia(t)la) =

 7!

=Uia(la) , and so the canonical homomorphism ~:A-~ AF factors

as
A ""' ia(A)

ia(I a) ¢ Uia(Ia)
for some = . Let g:ia(l=)-~ia(A) be the canonical map. We

want to show that g is an isomorphism, because this would imply

Ia ¢ ~ . g is obviously a monomorphism, since ~g = j . We have

~ g f = jf = ~ , so gf = ~ . It follows that Im g is an

~ - c l o s e d submodule of A? containing Im~ , and we conclude

that g is an epimorphism~

(a) ~ ( d ) : Let (Ma) be a direct system of modules. The inclusions

t(Mm)-+M a i n d u c e i n t h e l i m i t an i n c l u s i o n lim t(Ma)--+lim M..

The class of torsion modules is closed under direct limits,

since it is closed under direct sums and quotients, lim t(Ma)

is therefore a submodule of t(l~ Ma) . To show that we actually

have equality, suppose x C t(l~ M ) . Then xl = 0 for some

finitely generated I e mF . Since I is finitely generated, it

is clear that we may represent x by some xa e Ma such that

still xaI = 0 . Then x a c t(M ) , and x e lim

--@ t(Ma) -

(d)~(a): Write I e ~ as the directed union of finitely

generated right ideals I= . A/I = lim A/I a is a torsion module,

so A/I = t(A/I) = lim t(A/la) . In particular the generator

T e A/I comes from some t(A/la) , which means that there exist

a e A and J e ~ such that aJC Ia and l-a e I . We may

choose a so that l-a e I a , and then J C I a . Hence Ia e ~ .

 72

It is clear that every topology satisfying Proposition 12.2

also satisfies 12.1 (cf. Exercise 2 of § Ii). The converse holds

e.g. when all right ideals in A are countably generated.

References; Goldman t331, Roos [661 (oh. l ) .

§ 13. Fl,at epimorphisms of rings

In many examples of rings of quotients one obtains the module

ME of quotients as M F = M ~ A AF . Here we will prove that this

is equivalent to several other nice properties of the localization,

e.g. that AF is obtained by a kind of generalized calculus of

fractions.

Let ~ be a topology on A and let ~ :A--~AF be the canonical

m

ring homomorphism. We have the diagram of functors (of. § 7):

q
Mod-k < ,~ : Mod-A F

where in ~artio~ar q(~) . ME ana ~*(M) = M ® A ~ • ~here i~

a natural transformation G : ~ ~-~ q where O M : M @ A F-~ NF is

defined as 9 M ( x @ q ) = ~M(x)q . The diagram

ool~utes.
 Y3

Theorem 13.I. The following properties of F are equivalent:

(a) The functor j:~-~Mod-A F is an equivalence.

(b) The functor i:~-*Mod-A has a right adjoint.

(c) The functor i:~-~Mod-A is exact and commutes with direct

sums.

(d) ~ contains a cofinal family of finitely generated right

ideals and i is exact.

(e) @ :~-~q is a natural equivalence of functors.

(f) KerCM~M®A ~ 3 = t(~) for a l l MA .

(~) ,I,(I)~=~ for every Le~.

Proof. (a) =~(b): i = ~4 j , where j has a right adjoint by

hypothesis~ and ~ has the right adjoint HOmA(AF,. ) .

(b) ~ ( c ) : Clear, since i always is left exact.

(c)~(d): Follows from Proposition 12.2.

(c) =~(e): We have ~a q = ia , which by hypothesis preserves

arbitrary colimits. Since the forgetful functor T~ also pre-

serves colimits, it follows that q preserves colimits. A well-

known argumen't ([I021, p.157) then shows that (~ is a natural

equivalence between " ~A AF and q .

(e) ~ (f): K e r ( M - * M O A F ) -- Ker(M-~MF) = t(M) .

u

(f) ~ (g): If I e F , then A/I is a torsion module and hence

the map A/I--~A/IQA F = A F / ~ ( 1 ) A F is zero. This implies

~(~),
(g) =~ (a): We must show that every right AF-module M is F_-closed

as an A-module. M is torsion-free, for if x e M and xl = 0

for some I c F , then x~ = xlA F = 0 and hence x = 0 .

 74

We next verify that M is F_-injective. Let f:I--FM be a

homomorphism with I c ~ . Write I c AF in the form

i =~T(ai)qi with qi c AF and a i c I . Then J =

m

= N ( ~ ( A ) : q i) ~ ~ and I ~ J c ~ . Since M is torsion-free,

i
we may factor f over ~ : ~ ( I ) - , M . For every a 6 I~ J we

get f(a) = ~ ( ~ (a)) = ~(~ ?(ai)qia ) = ~ ~ ( ~ (ai))qia =

= ~ f(ai)qia . Thus the restriction of f to I~ J has an

extension to h:A-~M , and as usual we must then necessarily

have h~l = f (cf. proof of Proposition 7.8).

Definition. A topology is called ~erfect if it has the properties

listed in the Theorem.

Co rollar ~ !3.2. If ~ is a perfect topology, then ~ is

flat as a left A-module.

Proof. The functor • ~A F = q is left exact.

The main drawback of Theorem 13.1 is that none of the conditions

(a) - (g) are internal, i.e. they do not give conditions for

to be perfect solely in terms of F and the ideal structure

of A . So e.g. when one is applying (d), it is usually difficult

to tell from F
m
whether i is exact. In one special case this

is however possible:

Pr0~ositi0 ~ 13.3. If A is right hereditary, then i is exact

for all topologies on A .

Proof. Exactness of i means that if f:L--~M is a homomorphism

between F-closed modules~ then also Coker f is F-closed.

Since Ker f is ~-closed, it clearly suffices to consider

 75

monomorphisms f:L-~M . M/L is then torsion-free, as one sees

from the exact sequence

HomCAII,M)---~ Hom(AII,MIL) ~ ExtI(A/I,L) .

Let g:I-~M/L be any homomorphlsm with I c [ . Since I is

a projective module, we may lift g to h:l-@M . Since M is

~-closed. we may extend h to A-~M . The composition

A-~M-~M/L then extends g , and M/L is F_-injective.

Corollar[ 13.4. If A is a right noetherian hereditary ring,

then every topology is perfect.

There is an interesting abstract characterization of those

ring hcmomorphisms which are of the form A--~A F for a perfect

topology ~ , namely they are precisely the "flat epimorphisms"

in the category of rings. As a preparation for this result, we

are going to review some facts about epimorphisms of rings.

Let ~ : A - - ~ B be a ring homomcrphism. Recall that $ is an

e~imorphism (in the category of rings) if for any ring C and

homomorphisms a , ~:B--~C , a S = ~ implies a = ~ . More

generally we say that b ¢ B is dominated by $ if ~? = ~

always implies ~(b) = ~(b) . The set of elements of B

dominated by ~ is a subring of B , called the dominion of ~ .

is an epimorphism if and only if its dominion equals B .

Proposition 13.~. ~ : A - , B dominates b ~ B if and only if

h®l = l®b in B® A B .

Proof. Suppose b@l = l~b and let a ,~IB-~C be homo-

morphisms such that aS = ~. Define a homomorphism of

 76

A-A-bimodules T:B~) A B-~C as T(beb') = a(b)~(b') . b@l =

= l@b implies a(b) = W ( b ® l ) =r(l®b) = ~(b) .

Suppose o n the other hand that @~ dominates b . The assertion

now follows by applying the following lemma to x = I@I c B@ A B .

Lemma 13.6. Let ~ :A--)B be a ring homomorphism and M a

B-B-bimodule. If x c M has the property that T(a)x = x T(a)

for all a c A , then bx = xb for all b in the dominion of ~ .

Proof. We make BX M into a ring by defining

(b,y) + (b,,y,) = (b + b',y + y~)

(b,y) • (b',y') = (bb',by' + yb') .

The ring axioms are easily verified, in particular BK M has

an identity, namely (I,0) . Define two maps a ~ ~ :B-~B~ M

as =(b) = (b,O) and ~(b) = (b,bx-xb) . Both a and ~ are

ring homomorphisms, and a~ = ~ . So if b is dominated by

, then a(b) = o(b) and thus bx = xb .

If one is interested in dominions in the category of commutative

rings, one may prove an analogue of Proposition 13.5. The proof

in the commutative case has to be done separately (for Lemma 13.6

would introduce non-commutative rings), but is easy since

B~ A B is a ring whenever A is commutative.

We will elaborate on the consequences of Proposition 13.5. Let

Mod-A 4 Mod-B

be the funotors ~*(M) = M A ~ ' ~,(N) = N . Recall that ~% is

a left adjoint of ~ .
 77

Proposition 13.7~ The following properties of a ring homo-

morphism ~:A--~B are equivalent:

(a) ~ is an epimorphism.

(b) The canonical map B@ A B-*B is bijectiva.

(c) The adjunction transformation ~mT~-@id. is a natural

equivalence of functors.

(d) The fm~ctor ~:Mod-B-@Mod-A is full.

Proof. ( a ) ~ (d), Suppose M and N ar~ B-modules and

=:M--~N is A-linear. For each x e M oonsider the map

~ : B ® A B--~N @~iven by~ ~(b@b,) = a(xb)b, . Note that this

really is a welldefined map. Since l®b = b®l , we have

a(xb) = a(x)b , and therefore a is B-linear.

(d)~(c): We have to show that the homomorphism ~ : M O A B - @ M

given by x@b~xb , is an isomorphism for every B-module M .

The map M--,MQ A B given by x~xQl is clearly A-linear,

and by hypothesis therefore B-linear. It is the desired inverse

of ~ .

(c)~(b) is clear, while ( b ) ~ ( a ) follows from Proposition 13.5.

An immediate consequence of (d) is:

Corollar~ 13.8. Let ~ :A--~B be an epimorphism. If M is a

right B-module such that MA is injective, then MB is

injective.

For the proof of the next theorem we need the following result:
 78

Lemma 13. 9 . Let there be given modules LA and A M . The

following assertions a~e equivalent:

(a) xA@M = 0 for every x s L .

(b) For every y e M and X1,...,X" e L , there exist

m
yl,...,y m e M and al,...,a m e A such that y =5 aiY i
I
and x.a. = 0 for all i and j .
i S
(c) For every y s M and x e L , there exist yl,...,ym e M
m
and a.,..i"'am s A such that y = ~ a_v_ivl and xa. = 0
I J
for all j .

Proof. The equivalence of (a) and (c) is well-known ([i0], oh. I,

§ 2, Lemma I0). (a) implies that C®M = 0 for every submodule

C of L , and one easily proves that this implies C~M = 0

for every submodule C of a direct sum Ln of copies of L .

(b) is then obtained from (a) by considering (Xl,..,xn) as an

element of Ln .

Theorem 13.10. Let ~ :A-~ B be a ring homomorphism. The

following assertions are equivalent:

(a) ~ is an epimorphism and makes B into a flat left A-module.

(b) The family F of right ideals I of A such that ~(1)B = B

is a topology, and there exists a ring isomorphism ~ :B--*A F

such that G~= ~A "

(c) The following two conditions are satisfied:

(i) for every b s B there exist Sl,...,s n c A and

bl,...,b n e B such that b~(si) e~(A) and Z $ ( s i ) b i = 1 !

(ii) if $ ( a ) = 0 , then there exist Sl...,s n e A and

bl,...,b n e B such that as i = 0 and ~(si)b i = i .

 79

Proof. (a) :=~ (b): The f o r g e t f u l functor ~:~t:Mod-B--,Mod-A

makes Mod-B equivalent to a full suboategory of Mod-A by

Proposition 13.7(d). ~ has a left adjoint ~*-- " ~ A B which

is exact, since AB is flat. We may thus consider Mod-B as

a Giraud subcategory of Mod-A . From Theorem 10.2, combined

with Theorem 13.1, we may conclude that the family ~ of right

ideals I such that ~m(1) ~ ~4(A) , i.e. such that ~o (1)B = B ,

~s a t o p o l o g y and that AF = B .

(b),~ (c): If b s B = A F , there exists I c ~ such that

bl C ~ ( A ) . By Theorem 13.1 we may assume I finitely generated,

say generated by Sl,..,s n . Then I =~(si)b i for some

b i s B , and (i) is verified.

If ~ ( a ) = 0 , then a S Ker(A-~AF) , so there exists I c

such that al = 0 . As before one may assume I generated by

Sl, .... ,sn , and i =~ (si)b i . Thus also (ii) is verified.

(c)~(a): First of all we note:

Lemma 13.11. If ~ :A-@B is a ring homomorphism satisfying

condition (i) of (c), then C~ A B = 0 for every submodule

C of B/~O(A)

Proof. (i) implies that 13.9(c) is satisfied.

We now prove (c)~(a). Since we have just seen that

B/~(A)~ A B = 0 , we o b t a i n epimorphisms

B ~ A® A B ) ~(A)®A B , B(Z)A B .

The canonical map B@ A B-~B must then be a bijection, and

 80

thi;s shows that ~ is an epimorphism (Proposition 13.7).

It remains to see AB is flat. For this we use Proposition 1 9

of [101, oh. I, § 2. Suppose we have Sl,...,s n c A and

bl,...,hn ¢ B such that ~(si)b i = 0 . Applying Lemma 13.9(b)

tO 1 ~ B and ~l,...,~n c B / ~ ( A ) , we obtain al,..,am • A

and h~,..,h'm c B such that

a
h i ~ ( % ) =~(oiS) for some cij c A

hen Z (sici ) = Z?(si)bi (aj> - 0 and so sicij o

8 Ker T for each J . We now make use of condition (ii), and

obtain for each J elements tjl,...,tjr c A and bBl,..,h'jr 6 B

such that

sicijtjk = 0
i
'e ( t ) ~ , = 1
Jk Jk "

We then have

hi = ~ b .
a j,k

and E sicijtjk = 0 , i.e. the given relation comes from a

i
relation in A , as was to be shown.

Remark I: Note that condition (c) of the theorem implies in

particular that each b e B may be wTitten as

h = ~b~(si)h i =~(ai)h i with ~(si)b i = I . B is thus

obtained by a sort of generalized calculus of fractions (of. §15).

 8i

Remark 2. As was noted in the proof of ( c ) ~ (a) , condition

(i) of (o) may be restated as:

(i)' For every family bl,..~b r c B there exist Sl,..,s n c A

and '
bl''''' b'n c B such that hj%O(si) c ~ ( A ) and

Z ~(si)b ~ : 1
i

Corol!ar ~ 13.12. There is a I-I correspondence between perfect

topologies on A and equivalence classes of ring epimorphisms

A--~B such that AB is flat.

Exercise s:

I. Show that a topology F is perfect if and only if all

AF-modules are torsion-free as A-modules.

2. Suppose m
F is a perfect topology. Show that:

(i) MA isatorsio, module if and only if M ® A ~ - - 0

(ii) t(M) -- TOrl(M,~JA) for ~ l mod~es ~ , ass~ing
A torsion-free.

3. Let A C B be commutative rings. Show that A c~ B is a flat

epimorphism if and only if (A:b)B = B for every b c B .

(Hint: note that if laB = B for a finite family of

ideals I= , then (~la)B = B ).

References: Goldman [33], Lambek [48] (§ 2), Popesou and

Spircu [I04~ , Roos [66] (ch. I), Silver [70], Walker and

Walker [81].

For commutative rings: Akiba [84], Nastasescu and Popescu [59],

S@m. Samuel [67] (Expos@ 6 by Olivier).

 82

§ 14. Maximal flat epimor~hic extension of a ring

Theorem 14.1. For every ring A there exists a ring M(A) and

a ring homomorphism ~ : A - ~ M ( A ) such that

is an injective epimorphism and M(A) is a flat left

A-module.

(ii) For every injective epimorphism a:A--~B of rings such

that AB is fiat, there exists ~ unique ring homomorphism

~:B-*M(A) such that ~a = ~ ! moreover, ~ is also

injective.

Proof. Every flat epimorphism is obtained as the canonical homo-

morphism ~ : A - * ~ for a perfect topology ~ (Theorem 13.10).

is injective if and ordy if A has no ~-torsion, i.e. if

and only if F CD m
(the family of dense right ideals). 8o if

M(A) exists, it should be a subring of the maximal ring of

quotients AD = ~ . This leads us to consider the family

of all subrings B of ~ such that A C B and A~B is

a flat epimorphism. Note that inclusion of subrings in

corresponds to inclusion of the corresponding perfect topologies.

Lemma 14.2. The family w

P is directed under inclusion.

Proof. We will show that if B and C are members of m

P ,

then the smallest subring D of Qm containing both B and

is also a member of P . Every element of D is a sum of

elements of the form

(~) d = blC I .... bnC n with b i c B , ci c C ,

 83

and we may assume that each d appearing in a given sum has the

same length n . We will verify condition (i) of 13.10(c), i.e.:

given dl,...~d r of length r~ , there exist d l' ' ' ' " d'm in D

and Sl,...,s m c A such that dis k C A for all i , k and

We do this by induction on the length n . For n = 0 , i.e.

d = I , the condition is clearly satisfied. Suppose n> I and

the condition has been verified for n-1 . To simplify the

notation somewhat, we only consider the case r = 1 ! it is

easy to see that our argument extends immediately to the case

of a family dl,...,d r c D . So suppose we are given d of the

form ( ~ ) . By the induction hypothesis there exist 4'''" d'

m

in D and t l,...,t m in A such that

x i = b2c 2 .... bnCnt i c A for all i , and ~tid i

t t
By the remark 2 of § 13~ applied to C , there exist Cl,..,c p

in C and s l,...,sp in A such that

xij = ClXiS j ¢ A for all i , j , and ~sjc~ = I .

Similarly there exist hl,...,bq in B and rl,...,rq in A

such that

blXijr k ~ A for all i , j , k , and ~rk~ = i

t t t
We have thus got elements bkCjdii ¢ D and tisjr k c A such that

dtisjr k = blClXiSjrk c A and ~- t.s.rkb~Ct.d.~ = I .

i,j,k I J J

This finishes the proof of the Lemma, and we may continue the

proof of the Theorem. Define M(A) as the union of all rings

 84

in ~ . It is obvious that ~:A--pM(A) is an epimorphism, and

M(A) is flat as a left A-module since it is a direct limit of

flat A-modules.

Suppose m:A-~B is any other injective flat epimorphism. There

is a corresponding perfect topology [ , and as we noticed above,

we have F C D and hence a commutative diagram

A t ~ Qm

But also ~:B-@Q m must be injective, because ~(b) = 0 would

mean that there exists a homomorphism f:l--~A with I c F

such that f~J = 0 for some J C ~ ! for each a ¢ I we have

(J:a) c ~ and is D-torsion-free, so it follows that f = 0 .

We conclude that the image of ~ lies within M(A) . Since

obviously is unique ( a is an epimorphism), we have proved the

Theorem.

We will investigate the properties of M(A).

Propositi0n 14.3. If f:A--~B is a ring epimorphism such that

AB is flat, then J = f-l(j)B for every right ideal J of B .

Proof. Put I = f-l(j) . Tensoring the inclusion A/I c.@ B/J

by B , we obtain B/IB ~--~B/JeA B = B/J by 13.7(o). Since

B/IB-*B/J obviously is surjective~ it is an isomorphism and

J=IB .

Corollar~ !4.4. If A is right noetherian% then so is also M(A) .

~opositi..on 14.~. If A is von Neumann regular, then M(A) = A .

Proof. By Lemma 13.11 we have M(A)/A ® A M(A) = 0 . But since

 85

A is regular, A c_~ M(A) induces a monomorphism

MCA)/A® A A~ ~(A)/Ae A ~(A) , so ~(A)/A -- 0

Further results on M(A) may be obtained by using homolcgical

methods. When A--~ B is a ring homomcrphism such that AB is

flat, one has the following formulas ([13], ch. 6~ § 4):

(1) ~t~CM,~) "= ~ t ~ ( M ~ A B,~) for MA , N B ,

(2) TornA(M,N) ~ Torn(M, A B,N) for MA , ~ .

Combining these formulas with 13.7(o) we obtain:

Proposition 14.6. If ~ :A--~ B is a ring epimorphism such

that AB is flat, then

~t~(M,~) = ~ ( M , ~ for ~ , ~ ,

•or~<M,~> "=Tor~(M,~> for 5 ' ~ "

Corollar[ 14.[. r.gl.dim M(A)@ r.gl.dim A and

w.gl.dim M(A)6 w.gl.dim A ,

where r.gl.dim is the right global dimension and w.gl.dim

is the weak global dimension.

The ring A is called right coherent if every direct product

of flat left A-modules is flat [17].

Proposition 14.8. If A is right coherent, then so is also M(A) .

Proof. Let (N=) be a family of flat left M(A)-modules. Then

each N is flat as an A-module by (2). Thus ~ N~ is flat

a
over A , but is then flat also over M(A) by 14.6.
 86

Example:

It follows from Corollary 13.4 that when A is right noetherian

hereditary, then M(A) = Qm " This result will be improved in

§ 20 (Theorem 20.2).

Exercises:

I. Let ~ be the Goldie topology (§ 3, Example i). Show that

if ~ is perfect, then AF is right noetherian.

n

2. Suppose A is a ring for which Qcl exists and such that

every finitely generated right ideal of A is principal.

Show that Qcl = M(A) .

References: Knight [991, Popescu and Spircu [1041. For the

commutative case: Akiba [841, Lazard [I00] (ch. 4).

§ 15. l--topologies and rings of fractions

A l-topolog~ on A is a topology containing a cofinal family

of principal right ideals. A 1-topology F is determined by the

set ZC_F) = { s A l sA }.
Proposition I~.I. The map ~ defines a I-I correspondence

between I-topologies on A and subsets S of A satisfying:

SI. I eS .

$2. s, t e S implies st e S .

$3. If s e S and a ¢ A , then there exist t e S and b e A

such that sb = at ( S is right permutable).

$4. If ab e S , then a e S .
 87

Proof. Let F be a 1-topology. ~ (F) obviously satisfies S I.

S 2 follows from axiom T 2 for topologies, because for each

sa s s A we have (stA:sa) D (tA:a) ¢ F by T I. S 3 is clear

from T I, because we have (sA:a) D tA for some tA S F . S 4

is immediate from T 3.

Conversely, if S satisfies S I-3 and one sets F ={ I I IDsA

for some s e S t , then F is easily verified to be a 1-topology.

S 4 is a saturation axiom which makes the correspondence F_,@S_

one-to-one.

For a 1-topology one may describe the modules of quotients in

a rather explicit way.

Proposition 15.2. If F is a 1-topology and M c Mod-A , then

-{ (x,s) s L sa. 0 in A implies zat 0 f o r some

where ~0 i s the e q u i v a l e n c e r e l a t i o n given by (x,s)~ (y,t) if

there exist a , b ¢ A such that sa = t b e S and xa = yb .

Proof. Recall that we have MF = ~ m HomA(SA,M/t(M)) with

m

s e S . A homomorphism c~:sA--~M/t(M) is dtermined by an

element x e M such that sa = sb in A implies xa-xb e t(M) ,

i.e. x~t = xbt for some t e S . In the limit, ~0 gives the

same element of MF as ~ : t A - @ M / t ( M ) , determined by y e M ,

if and only if cp and ~ coincide on some uA c sAf~tA with

u G S , i.e. if and only if there exist a , b G A such that

u = sa = tb c S and xa-yb c t(M) . This clearly corresponds

to the relation ~J.

 88

One easily verifies that under the isomorphism described in

the Proposition, the module operations in ~ take the form:

(x,s) + (y,t) = (xa+yb,u) where sa = tb = u c S !

(x,s) • (a,t) = (xb, tv) for some b ¢ A , v c S such that

a V = sb .

Proposition I~.3. A 1-topology ~ is perfect if and only if

S = ~ (~) satisfies:

$5'. F o r every s e S there exists a ~ A such that sac S

and such that sab = 0 implies ahu = O for some u c S .

Proof. Perfectness of F means that for every s 6 S there

exists q c ~ such that sq = 1 (Theorem 13.1(g)). Suppose

that so is the case. Using 15.2, we write q = (a,t) . Then

(s,l)(a,t) = (sa, t ) ~ (I,I) , so there exist a', b' ~ A such

that ta' = l.b' 8 S and saa' = l.b' . We may then take aa'

as the element a in S 5', for saa' = b' 6 S and if

saa'b = 0 , then ta'b = b'h = saa'b = 0 and (a,t) c A F

implies aa'bu = 0 for some u c S (by 15.2).

Conversely, if for each s 8 S there exists a 8 A such that

as in S 5' , then (a, sa) represents an element in ~ and

one has (s,l)(a, sa) = 1

The axiom S 5' is a weakened form of the perhaps more well-

known condition:

$5. If sa = 0 with s c S , then at = 0 for some t C S

( S is right reversible).
 89

The most important examples of 1-topologies are those given

by the rings of fractions. Let S be a multiplioatively closed

subset of A . A right ring of fractions of A with respect

tO S is a ring A[S -I] and a ring homomorphism ~ : A - * A[S -I]

satisfying:

FI. ~(s) is invertible for every s C S .

F2. every element in A[S-1] has the form ~(a)~(s) -1

with s e S .

F3. ~(a) = 0 if and only if as = 0 for some s e S .

Similarly one defines the left ring of fractions [S-I]A of A

with respect to S . It is not immediately clear that the axioms

f ~
F 1-3 dte~ine ALs-~J uniquely, but that so is the case

follows from the fact AIS -I] is a solution of a universal

problem:

Proposition I~. 4. If A[S-I~ exists, it has the following

property: for every ring homomorphism ~:A-*B such that ~(s)

is invertible in B for every s ¢ S , there exists a unique

homomorphi~ ~ , B ~ A [ S -1] such t h a t ,~-~

Proof. We define ~ as ~ ( c ~ ( a ) C ~ ) ( s ) - 1 ) = Ut(a)~(s) -1 . We

then have to verify that this is well-defined. So suppose

C~(a)~(s) -I = ~ ( b ) ~ ( t ) -I . Then ~(a) = cp(b)%O(t)-l~(s) =

= ~ (b)~(c)~(u) -1 f o r some c 9 A , u c S , by F 2. So

@~(a)$(u) = ~ (h)~(c) , and by F 3 this implies that auv =

= bcv f o r some v e S . Then ~(a)~(u) = ?,(b) 7(c) since

(V) is invertible, and we may go backwards to obtain

 90

~ ( a )- ~ C "sI) "" " = ~ ' ( b -) ~ ( t")I "" " . We leave %0 the reader to verify

that ~ is a homomorphism. It is clear that ~c~ . ~ and that

is unique.

Corollar~ I~.~. A[S -I] is unique up to isomorphism.

The unicity of the solution of a universal problem also implies:

Corollary 1~.6. If both A[S -I] and [S-I]A exist, then they

are isomorphic.

Since there are examples of A , S such that [S-I]A exists

but A[S -I] does not exist ([I0], p. 163), a ring may satisfy

the universal property of 15.4 without being a right ring of

fractions with respect to S . We now turn to the question of

the existence of A[S -I] .

Proposition i~.7. Let S be multiplicatively closed in A .

A[S -1] exists if and only if S is both right permutable and

right reversible. If A[S-~ exists, then A[S -I] = A F where

the topology ~ = {I~ I D sA for some s ~ S} is perfect.

Proof. If S is right permutable and right reversible, ~ e n

is a perfect topology by Proposition 15.3. Let ~:A--~A F be

the canonical homomorphism, and use the description of A? in

Proposition 14.2. S 5 guarantees that (l,s) represents an

element of AF whenever s e S . (l,s) will be an inverse

of ~(s) . Every element (a,s) c A F has the form (a,l)(l,s) =

= ~ ( a ) ~ ( s ) -I. Finally, if (a,l)~(O,l) , then as = 0 for

some s c S . Hence ~ satisfies F I-3.

Suppose conversely A[S -I] exists. S is then right permutable,

 91

for if a ¢ A , s c S are given, then ~ ( s ) - l ~ "( a ) " " " = ~ ( b )-~ ( t ")I "" "

by F 2 . i.e. ~p(at) -- T ( s b ) . ByF 3 this means t h a t atu-- sbu

for some u c S . Since tu c S , we have S 3. If sa = 0 with

s c S , then ~(a) = 0 by F I and at = 0 for some t c S by

F 3~ so we have S 5.

Corollary I~.8. There is a I-i correspondence between right

rings of fractions of A and subsets of A satisfying S 1-5.

For AL$-I] one may simplify the formula of Proposition 15.2

somewhat, in that

MF = M~ S/~ , where N is defined as before.

Of course one has M F = M ~ A A[S -I] .

Examples:

I. When A is commutative~ S 3 and S 5 are automatically

satisfied. The theory of rings of fractions is well-known

in that case [I0].

2. Let S be the set of non-zero-divisors of A . A[S-II is

called the classical right ring of quotients of A , and will

be denoted by Qcl " It is a subring of the maximal right ring

of quotients Qm of A , and is in fact a suhring of M(A) .

From Proposition 15.7 we get:

Pro~osition I~. 9. The classical right ring of quotients of A

exists if and only if A satisfies the "Ore condition", i.e.

for a c A and a non-zero-divisor s there exist b ¢ A and

a non-zero-divisor t such that sb = at .

 92

Note that if A has both a classical right ring of quotients

and a classical left ring of quotients, then these two rings

coincide, by Corollary 15.6.

Exercises:

I. Show that if A has no nilpotent elements ~ 0 , then S 5 is

always satisfied.

2. Let AtS -I] be a ring of fractions. Show that an A-module

M is F-closed for the corresponding topology F if and

only if M is torsion-free and divisible (i.e. M = Ms

for every s ¢ S ).

3. Let A be a regular ring and let S =~a e A~ ba = 0 implies

b = 0~. Show that S satisfies S 1-5' and that S satisfies

S 5 only if all elements in S are invertible. (llluetration:

A is the endomorphi~n ring o£ an infinite-dimensional vector

space).

References.- Als,~vist [3], Bourbaki [I0] (p. 162-163), Eriksson

t261, G a b r i e l [31],Gabriel-Zism [89].

Chapter 4. Self-in~ective rings

§ 16. The endomorphism rin~ of an in~ective module

A ring A is called regular (in the sense of von Neumann) if

for every a s A there exists x s A such that axa = a . We

recall the following alternative characterizations of regular

rings (of. [I0], p. 64):

Pro~gsition 16.!. The following properties of A are equi-

valent:

(a) A is regular.

(b) Every finitely generated right ideal of A is generated

by an idempotent element.

(c) Every right A-module is flat.

It follows from (b) that every right noetherian regular ring

is semi-simple. More generally, (b) implies that if A is

regular and has no infinite family of orthogonal idempotents,

then A is semi-simple.

Theorem 16.2. Let E be an injective A-module with endo-

morphism ring H and let J be the Jacobson radical of H .

Then:

(i) is re.nat.
(ii) Idempotents may be lifted modulo J .

The last assertion means that if e is an idempotent in

H/J ~ then there exists an idempotent in H mapping canonically

 9~

onto e . It is well-known that if idempotents may be lifted,

then one may lift any countable orthogonal family of idem-

potents in H/J so that orthogonality is preserved ([47],

§ 3.6). The proof of the Theorem will be broken up into several

steps. We define

N = ~h c H ~ Ker h is an essential submodule of E}.

Lemma 16.3. N is a two-sided ideal of H , and H/N is a

regular ring.

Proof. If f, g e N , then f+g e N since Ker(f+g)

Ker f C~ Ker g . If f e N and h c H , then fh c N since

Ker fh = h-l(Ker f) , and hf e N since Ker hf D K e r f . N is

thus a two-sided ideal. It remains to show regularity of H/N .

Let h ~ H . Choose a submodule K of E which is maximal

with the property that K~Ker h = 0 . K + Ker h is then an

essential submodule of E . Since the restriction of h to K

is a monomorphism, and since E is injective, there exists

g:E-*E such that gh(x) = x for x e K .

K h~K .~ E

c~n,1
E

If y e K + Ker h , we write y = x+z with x e K and h(z) -- 0 .

Then hgh(y) : hgh( ) : h(x) h(y) , hgh-h c . This

shows that H/N is regular.

Lemma 16.4. N = J .

Proof. Suppose h c N o Since K e r ( 1 - h ) ~ Ker h = 0 and Ker h

 95

is essential in E , it follows that l-h is a monomorphism.

Im(l-h) must then be a direct stmmand of E , because E is

injective. If h(x) = 0 , then x = (l-h)(x) , so Im(l-h)

Ker h . Im(l-h) is therefore essential in E , and we must

have Im(l-h) = E . l-h is thus invertible for every h e N ,

and so N C J .

On the other hand, since the Jacobson radical is defined as the

intersection of all maximal right ideals, it is clear that the

radical of H/N is J/N . But H/N is regular by 16.3, so

J/N = 0 and J = N .

To conclude the proof of the Theorem, it remains to show that

idempotents may be lifted mod J . Assume h c H and h-h 2 e J ,

which means that L = Ker(h-h 2) is essential in E . The

injective envelope of h(L) is a direct st~mand of E and

thus has the form Im e for some idempotent e in H . Then

eh = h on L , so eh-h e N . Put f = e + eh(l-e) and note

that f = ~ . Put L' = Im(l-e) + h(L) and note that L' is

an essential submodule of Im(l-e) + Im e = E . f-eh = e-ehe ,

and therefore f = eh on L' , so f-eh e N . Since we already

have eh-h c N , it follows that f-h e N = J .

C o r o l l a r ~ ! 6 . ~. If A is a right self-injective ring with

Jacobson radical J , then:

(i) J is the right singular ideal of A .

(ii) A/J is regular.

(iii) Idempotents may be lifted modulo J .

 96

We remark that the ring H/J of the Theorem may be shown to

be right self-injective (Osofsky [61~, Renault [107], Roos [109]),

but this is a fact which we will not need. Instead we are inte-

rested in the case when H/J is semi-simple.

Definition. The ring A is semi-perfect if A/J is semi-simple,

where J is the Jacobson radical of A , and idempotents may be

lifted mod J .

It is well-known that one can always lift idempotents when J

is a nilideal ([47], § 3.6), and therefore every right or left

artinian ring is semi-perfect. From 16.5 follows immediately:

PropQsition 16.6. A right self-injective ring is semi-perfect

if and only if it has no infinite family of orthogonal idem-

potents.

A module M is called finite-dimensional if it does not

contain any infinite family of non-zero submodules M. such

i

that their s~n [M. is direct. It is clear that an injective

I

module is finite-dimensional if and only if its endomorphism

ring has no infinite family of orthogonal idempotents. Hence

Theorem 16.2 gives the following generalization of 16.6:

Pro~qsition 16.7. The endomorphism ring of an injective module

E is semi-perfect if and only if E is finite-dimensional.

Examples:

I. If E is an indecomposable injective module, then H/J is

a division ring ([52]).

 97

2. The ring A is called right finite-dimensional if AA is

a finite-dimensional module (cf. § II, Exercise 4). Such a ring

cannot have any infinite family of orthogonal idempotents.

Exercises:

I. Show that a commutative ring is semi-perfect if and only if

it is a product of finitely many local rings.

2. Show that a semi-perfect ring is a.direct st,n of indecompos-

able right ideals.

3. Show that a module M is finite-dimensional if and only if

E(M) is finite-dimensional.

References: Faith [28~, § 5 , Lambek [471, § 4.4.

§ 17. Coperfect rin~s

As a natural generalization of the class of artinian rings we

define:

Definition. A is a semi-primary ring if the Jacobson radical

J is nilpotent and A/J is semi-simple.

If A is semi-primary and right noetherian, then A is right

artinian~ by a classical argument ([9], § 6, Prop. 12).

A further generalization leads to:

Definition. A is a right coperfect ring if it satisfies DCC

on finitely generated right ideals.

The usual Zorn's lemma argument shows that the DCC is equi-

valent to the condition that every non-empty family of finitely

 98

generated right ideals contains a minimal member. It is known

that a ring is right coperfect if and only if it is left

p e r f e c t in the sense cf Bass [7] (Bj rk [86]), but this fact

we do not have tc use.

Proposition 17.1. Every semi-primary ring is right and left

coperfect.

Proof. We use inducticn ~on the smallest integer n>~ I such that

jn = 0 . When n -- 1 , A is semi-simple and obvicusly right and

left coperfect. Suppose the assertion has been proved for all

semi-primary rings such that jm : 0 for some m < n . Let A

be semi-primary with jn = 0 but ~-I @ 0 . Suppose

I I D 12 D .... is a descending chain of finitely generated

right ideals of A . The canonical map A - * A/J n-I takes this

chain into a chain ~i D T 2 3 ... of finitely generated right

ideals of A/J n-l. But the radical cf A/Jn-z is j/jn-I , so

this later chain is stationary by the induction hypothesis.

Hence we have Ik C I r + jn-I for all r)k , for some k .

Since jn = 0 , this gives IkJ = IrJ for all r>k . Now

consider the chain Ii/IkJ D 12/IkJ D .. of right ideals over

the semi-simple ring A/J . This chain is stationary, sc for

some s we have I s C I t + IkJ = I t + IsJ (all t) s ) since

we may assume s~ k . The Nakayama lemma ([9], § 6, Cot. 2 of

Prcp. 6) gives Is = It for t>,s .

Some basic properties of ccperfect rings are summarized in the

following statement:
 99

Propositi0n !~.2. If A is right coperfect, then:

(i) A is right semi-artinian.

(ii) J is a nilideal.

(iii) A is semi-perfect.

Proof. (i): It suffices to show that every cyclic right module

A/I contains a simple submodule. Let I' be a minimal finitely

generated right ideal ~I . !+I'/I is then a simple submodule

of A/I .

(ii): Let s be the preradical associating to each module its

socle. From § 3, Example 5, we recall that J = ~(J) , i.e.

J = s0(J ) for some ordinal 0 • For each a s J we define

o(a) as the smallest 0 such that a c s0(J ) . o(a) is never

a limit ordinal, for if a s~sa(J ) , then a s sa(J ) for

a~ 0
some a . Hence we may write o(a) = a+l for some ordinal a .

Recall that s +l(J)/sa(J ) = s(J/sa(J)) and is therefore anni-

hilated by J , so s + I ( J ) . J C sa(J ) . This shows that for any

a, b c J we have h(ab)(h(a) . If a c J were not nilpotent,

the sequence h(a n) would be an infinite strictly decreasing

sequence of ordinals, which is impossible.

(iii): Since J is a nilideal, we may lift idempotents. It

remains to see that A/J is semi-simple. It is easy to see that

A/J also is right semi-artinian. If I is a minimal right

ideal of A/J , then there exists a maximal ideal I' such

that I~I' = 0 since the Jacobson radical of A/J is zero.

I is therefore generated by an idempotent. Since we can lift

idempotents modulo J , and since a right coperfect ring

 100

obviously cannot have any infinite family of orthogonal idem-

potents, the right eocle of A/J is a finite direct stun of

minimal right ideals, and is therefore itself a direct summand

of A/J . Its complementary summand must then have zero socle

and is therefore zero. Hence A/J is semi-simple.

It should be remarked here that it can be shown that conditions

(i) and (iii) of 17.2 imply conversely that A is left perfect

(Bass ~7]) and hence right coperfect (BJGrk ~86]).

PropQsition 17.3. If A is right noetherian and right or left

coperfect, then A is right artinian.

Proof. If A is right noetherian and right coperfect, then A

is obviously right artinian. Suppose A is right noetheriau and

left coperfect. J is a nilideal, but every nilideal in a

noetherian ring is nilpotent ([47], P. 70). Since A/J is

semi-simple by 17.2, A is a semi-primary ring and hence right

artinian by a previous remark.

This result may be generalized somewhat. For this we introduce

s o m e terminology. For each subset S of A we put

r(S) = { a ¢ A I sa = o } , l(S) = { a ¢ A I aS = 0 } .

A right ideal is said to be a right annihilator if it has the

form r(S) for some S CA .

Proposition 17.4. If A satisfies ACC on right annihilators

and is left coperfect, then A is semi-primary.

Proof. Consider the ascending chain of right annihilators

 I01

r(J) C r ( j 2 ) C .... By hypothesis we have r(J n) = r(J n+l)

for some n . If r(J n) ¢ A , then the left A-module

A/r(J n) has non-zero soole, which is of the form I/r(J n)

for some left ideal I D r(J n) . But the semi-simple module

I/r(J n) is annihilated by J . so Jl C r(J n) , which gives

I C r ( J n+l) = r(J n) . Then I = r(J n) , which is a contradiction,

and hence r(J n) = A , so jn = 0 .

References: Bass /.7~, Faith t27J.

§ 18. Quasi-Frobenius rings

In this § we discuss three classes of rings: S-rings, PF-rings

and @Y-rings (in order of d e c r e a s i ~ generality). An injective

module E is a co6enerator if Hom(M~E) ~ 0 for every module

M @ 0 (of course it suffices to consider cyclic modules M , so

E is a cogenerator if and only if for every right ideal I @ A

there exists x @ 0 in E with xl = 0 ). In the following

Proposition we let E(A) denote the injective envelope of AA .

Proposition 18.1. The following properties of A are equivalent:

(a) E(A) is an injective oogensrator.

(b) Every simple right module is isomorphic to a minimal right

ideal of A .

(c) Hom(C,A) @ 0 for every cyclic module C @ 0 .

(d) l(I) ~ 0 for every right ideal I ~ A .

(e) A has no proper dense right ideals.

Proof. (a) ~ (b): If S is a simple module, there exists a non-

 102

zero f : S - @ E(A) . Since A is essential in E(A) , the image

of f must lie in A .

(b) ~ ( c ) : Clear, because C has a simple quotient module.

(c) (d): Ob ous.

(d) ~ ( e ) : Immediate from Proposition 3.8.

(e) ~=~(a): A right ideal I is by definition dense if and

only if Hom(A/I,E(A)) = 0 .

Definition. A is called a right S-ring if it has the properties

of 18.1. (We have reversed the terminology of Morita [56], who

calls these rings "left S-rings" and furthermore asstunes

minimum conditions on A ).

Proposition 18.2. When A is a right self-injective ring, the

following conditions are equivalent:

(a) A is a right S-ring.

(b) AA is an injeotive cogenerator.

(c) r(l(1)) = I for every right ideal I .

Proof. (a) ~ ( b ) : Clear, since AA is an injeotive module.

(b) ~ ( c ) : Suppose r(l(I)) @ I . Then there exists a non-zero

homomorphism f:r(l(1))/l-,A . Since A is injective, the

composed homomorphism g:r(l(1))-~ r(l(I))/l-* A must be of

the form g(a) = ba for some b C A • Since g(1) = O , we

have b s I(I) . But then g(a) = 0 for every a ¢ r(l(1))

which is impossible.

(c) =>(a): Obvious by 18.1(d).

Definition. A is a right PF-ring if AA is an injective

cogenerator.
 103

Proposition 18.3. The following properties of A are equivalent:

(a) A is a right PF-rin~.

(b) A is right self-injective and an essential extension of

its right socle~ and A/J is semi-simple.

(c) Every faithful right module is a generator for Mod-A .

Proof. We will only prove the implication ( b ) ~ ( a ) , since this

is the only one we will need in the following. For the proof of

(a)~(b) and ( b ) ~ ( c ) we refer to Azumaya [61 and Osofsk~ [60]

(cf. also Kato [42]).

(b) ~ ( a ) : We write A/J as a direct sum of indecomposable

right ideals YI,..,Yn . We can lift each ~k to an indecompos-

able right ideal Ik of A so that A = Ii@ ... ~ I n $ I' .

Then I' G J and hence I' = 0 . Each Ik contains a unique

minimal right ideal Jk " Every simple A-module S may also

be considered as a simple A/J-module, and is therefore iso-

morphic to some Yk " In particular, each of the minimal right

ideals Jk must be isomorphic to some . If Ji = Jk ' then

I i ~ Ik since Ik is an injective envelope of Jk " It follows

that the number of non-isomorphic right ideals Jk is equal to

the number of non-isomorphic modules ~k " Hence each ~ , and

then every simple module, is isomorphic to some Jk ' and A

is thus a right S-ring.

Ne next prove a very usefu~ property of self-injective rings:

 104

Proposition 18. 4. The following properties of A are equivalent:

(a) Every homomorphism f:l-@A , where I is a finitely

generated right ideal, has the form f(a) = ca for some c e A.

(b) A satisfies (i) l(IlnI2) = 1(Ii) + I(I2) for all

finitely generated right ideals Ii, 12 .

(ii) l(r(a)) = Aa for every a e A .

Proof. (a) =~ (b): If II and 12 are finitely generated right

ideals, then obviously i(II) + 1(12) C l(Ii(~ 12) . Suppose on

the other hand that a e I(Ii(~ 12) . We can define a homo-

I b b e II
morphism a-I I + 12--~A as a(b) = (l+a)b b e 12

since these two expressions coincide on I i ~ I 2 . By (a) there

exists c e A such that =(b) = cb . For b e II we thus

have cb = b , i.e. (c-l)b = 0 . As a consequence we may write

a = (o-I) + (l+a-o) e l(Ii) + 1(12) , which proves (i).

For every a e A we have Aa C l(r(a)) . If b e l(r(a)) ,

we can define a homomorphism A a - ~ Ab as xa,-~xb . It must be

given by left multiplication with some c ¢ A , so in particular

b = ca and b e Aa .

(b) = ~ (a): Consider f:I--~A where I is finitely generated,

say I = alA + . . . + a n A . We use induction on n , the case n = 0

being obvious. 8o we may assume there exist c and c' in A

such that f(a) =~oa for a e I' = alA +...+an_iA

!
~c'a for a e ariA

For a e I'~anA we must have (o-c')a = 0 , so o-c' e

e l(I'(~anA ) = l ( I ' ) + l(anA ) by ( i ) . We accordingly write

 105

c-c' = b-b' with bl' = 0 and b'a = 0 . Then left multi-

n
plication by c-b = c'-h' coincides with f on I = I' + a n A .

Proposition 18.~. If A is right self-injective, then:

(i) l(lln12) = I(Ii) + 1(12) for all right ideals I I , 12 .

(ii) l(r(1)) = I for every finitely generated left ideal I .

Proof. (i): The argument used in the preceding proof works

equally well in the present situation.

(ii): Write I = Aa I + ... +Aa n . Then l(r(Aa I +...+Aan)) =

= l(r(Aal)a...nr(Aan)) = l(r(Aal)) + ... + l(r(Aan)) =

Aa I + ... + Aa n by (i) and 18.4(ii).

Definition. A i s a QF-ring ( o r q u a s i - F r o b e n i u s r i p ~ ) i f i t is

both right and l e f t artinian and r i g h t and l e f t self-injective.

We will show, however, that much weaker conditions suffice to

make A a QF-ring. Our first step is to show that the conditions

may be made one-sided. For this we need the following reformu-

lation of the definition:

Lemma 18.6. A right and left artinian ring is a QF-ring if and

only if it satisfies

r(l(1)) = I , l(r(I')) = I'

for all right ideals I and left ideals I' .

Proof. A QF-ring satisfies the double annihilator conditions

by 18.5, Conversely, these conditions imply that the operations

r and 1 define an anti-isomorphlsm between the ordered sets

of left resp. right ideals. They therefore reverse the lattice

operations, so conditions (i) of 18.4 are satisfied, and hence

the ring is right and left self-injective.

 106

Proposition 18.[. If A is right or left artinian and is right

or left self-injective, then A is a QF-ring.

Proof. There are two cases we must consider:

I) A is right artinian and right self-injective!

2) A is left artinian and right self-injective.

We can reduce case I) to 2): If IIC I2C ... is an ascending

chain of finitely generated left ideals, then the descending

chain r(l I) D r(12)D ... is stationary since A is right

artinian. The ascending chain is then also stationary by 18.5,

and A is thus left noetherian. But a ring which is both left

noetherian and right artinian, is also left artinian.

We now consider case 2). Since A is left artinian, it is

right coperfect and therefore right semi-artinian (Propositions

17.1 and 17.2). It then follows from 18.3 that A is a right

PF-ring, and this implies r(l(1)) = I for every right ideal I ,

by 18.2. We also have l(r(I)) = I for every left ideal I by

18.5. By an argument similar to the previous reduction of I) to

2), we finally have that A is right artinian, and by applying

Lemma 18.6 we may conclude that A is QF.

We may weaken the conditions for A to be QF quite a bit more:

Proposition 18.8. If A is right or left noetherian and is

right or left self-injective, then A is a QF-ring.

Proof. There are two cases to be considered:

I) A is left noetherian and right self-injective!

2) A is right noetherian and right self-injective.

 107

Case I): To show that A is left artinian, it suffices to show

that A is semi-primary (cf. remark at the beginning of § 17).

A/J is a left noetherian regular ring by 16.5, hence semi-

simple. It remains to show that J is nilpotent. The ascending

chain of two-sided ideals r(J) C r(j2) C ... is stationary

since A is left noetherian. Let r(Jn) = r(J n+l) . From 18.5

follows that jn = / n + l which implies jn = 0 by the Nakayama

lsmma §6, Cor.2 de op.6).

Case 2): We prove a slightly stronger statement:

Theorem ! 8 . 9 . If A satisfies ACC o n right or o n left annihi-

lators and is right or left self-injective, then A is a

QF-ri~g.

Proof. Case I)- Suppose A satisfies ACC on left annihilators

and is right self-injective. Every finitely generated left ideal

is a left annihilator by 18.5, so A is left noetherian. A is

then QF by case I) of 18.8.

Case 2): Suppose A satisfies ACC on right annihilators and is

right self-injective. If I I ~ 12 ~ ... is a descending chain

of finitely generated left ideals, then r(l I) C r(12)C ... is

stationary by hypothesis, and so A is left coperfect and even

semi-primary by 17.4. A is then a right PF-ring by 18.3, which

implies that every right ideal is a right annihilator (18.2).

A is therefore right noetherian. But since we have just proved

that A is semi-primary, A must then be right artinian. We

conclude from 18.7 that A is QF.

 108

An interesting property of QF-rings is:

Proposition 18.10. Every module over a @y-ring is a submodule

of a free module.

Proof. It suffices to show that every injective module is

projective when A is QF. Since A is noetherian, every

injective module is a direct sum of indecomposable injective

modules, so it clearly suffices to show that E(S) is

projective, where S is any simple module. But S is iso-

morphic to a right ideal of A , and since A is self--injeotive,

E(S) is a direct summand of A .

Combining this result with § II, Exercise 3, we find that

over a QF-rin~, a module is injective if and only if it is

projective.

Example: The group ring of a finite group over a field is a

~-ring ([201, p. 402).

Exercises:

I. Show that an injective module is a cogenerator if and only

if it contains a copy of each simple module.

2. Show that a right PF-ring with zero Jacobson radical cannot

have essential right ideals @ A and must therefore be semi-

simple.(This is a special case of the implication (a) ~ (b)

of 18.3, which we did not prove).

3. Show that if A is a right PF-rin~ and I is a maximal

right ideal, then l(I) is a minimal left ideal.

 109

References; Azumaya [6], BjSrk [8], Eilenberg-Nakayama [23~,

Faith [27], ~ e a ~ x ~ [37], Kato t421.
Cha~_ter 5- Maximal and classical rings of qu_qtients

§ 19. The maximal rin~ of quotients

We recall that a right ideal I of A is dense if (l:a)

has no left annihilators for any a c A , and that the family

of dense right ideals is a topology, corresponding to the

hereditary torsion theory cogenerated by E(A) . The ring AD

is the maximal right rin G of quotients of A , and is denoted

by % .

P~PPqsiti°n 19"1" Qm is its own maximal right ring of quotients.

Proof. This follows from Proposition Ii.ii since De is the

family of dense right ideals of ~ .

Let H be the endomorphism ring of E(A) . We recall from § 8

that there is a commutative diagram

~ ~ HomH(E(A ) ,E(A))

E(A)
where k is a ring isomorphism, ~ is the canonical inclusion

(Proposition 8.2) and £(f) = f(1) .

Proposition 19.2. The following assertions are equivalent:

(a) ~ is a right self-injeotive ring.

(b) I,Qm_-~E(A ) is an isomorphism.

(c) The right ideal (A:x) is dense for every x ¢ E(A) .

(d) There is a ring isomorphism H-@Q m such that the diagram

 111

commutes, where e(h) = h(1) .

~oof. The equivalence of (b) and (c) is clear from Proposition

8.1, and the equivalence of (a) and (b) from Proposition 8.2.

(b)@(d): ~ Hom%(E(A),E(A)) = HomA(E(A),E(A)) .

(d)~(b): • is ob~ously s~jective, so if ~ exists, then

is also s~Jective.

Lemma 19.3. If A is a r i ~ t finit~dimensional ring, then so

is also ~ .

Proof. If a r i ~ t ideal J of ~ is a direct s t of non-zero

r i ~ t ideals J= of ~ , then each JaVA is a non-zero right

i d e ~ of A . It follows that ~ e family ( ~) m~t be fibre.

Proposition 19.4. Suppose ~ is r i ~ t self-injective. ~ is

a semi-perfect ring if ~ d only if A is r i ~ t finite-

dimensional.

Proof. ~ ~ HomA(E(A),E(A)) by 19.2, so ~ is semi-perfect if

~d only if E(A) is a finite-dimensional mod~e, by 16.7. But

E(A) is finite-dimensional if ~ d only if A is so.

Proposition 19.~. The following assertions are ~ u i ~ l e n t :

(a) % = M(A)
(h) ~ is f l a t as a left A-mod~e and A ~ is a ring

epimorphism.

(c) ~ is a right S-ring.

 112

Proof. (a)@~p(b) is obvious since one always has M(A) C Qm .

(a) ~ ( c ) : If Qm = M(A) , then the topology ~ of dense right

ideals of A is perfect. If J is a dense right ideal of Qm '

then JN A c ~ by Example 3 of § Ii. Therefore Qm = (J~A)Qm C.

C J , and J = Qm "

(c)~(a): Zf I C2 and I% ~ % , then 1(I~) %0 by 18.1.

Thus there exists O # q e Qm such that ql = 0 . But this is

impossible, because Qm is _D-torsion-free.

Corollar~ 19.6 ,. Suppose Qm is right self-injective. ~ is

a right PF-ring if and only if Qm = M(A) .

PropQsition I~. T. Suppose ~ is right self-injective. The

following assertions are equivalent:

(a) Qm is a quasi-Frobenius ring.

(b) Qm is a ~-injective right A-module.

(c) Every direct sum of D-torsion-free injective A-modules

is injective.

(d) The lattice ~D(A) of _D-pure right ideals is noetherian.

(e) A satisfies ACC on right annihilators of subsets of E(A)

Proof. The equivalence of (b), (c), (d) and (e) was proved in

Propositions 11.7 and 11.8. Since every QF-ring is noetherian,

(a) ~l~(d) by Proposition 11.12.

(e)=)(a): By Theorem 18.9 it suffices to show that the ring

Qm satisfies ACC on right annihilator ideals. In view of

 113

condition (e), it will be enough to show that if SI and S2

are subsets of Qm ' then ~q c Qm I Slq = 0 } C I q s Qml S2q = 0

if and only if {a s A~ Sla = 0~ C I a c A ~ S2a = 0 } . The first

inclusion trivially implies the second one. Suppose the second

inclusion holds, and Slq = 0 for some q s Qm " Choose I c

such that qlCA . Slql = 0 implies S2ql = 0 , which gives

S2q = 0 since Qm is torsion-free.

References: Lambek [47], Mewborn-Winton [54], StenstrGm [72].

§ 20. The maximal rin 6 of quotients of a non-singular rin~

For non-singular rings there are more precise results on the

structure of Qm than we were able to obtain in the general

case. Recall that when ~ is right non-singular, the t o p o l o ~

of dense right ideals coincides with the family of essential

right ideals, which we denote by E . Also the Goldie t o p o l o ~

(§ 3, Example I) reduces to E in this case.

Proposition 20.1. The following assertions are equivalent:

(a) A is right non-singular.

(b) Qm is a regular ring.

(c) Qm is a right self-injective regular ring.

(d) The Jacobsen radical of HomA(E(A),E(A)) is zero.

Proof. Suppose A is right non-singular. If x c E(A) , then

(A:x) is essential. From Proposition 19.2 fellows that Qm is

right self-injective and that Qm = H°mA(E(A)'E(A)) " If we prove

(a) @ ( d ) , then also (a)@(c) will follow, as a consequence of

 114

Theorem 16.2.

(a)~ (d): Recall that the Jacobson radical of Hom(E(A),E(A))

consists of those ~ :E(A)--~E(A) which have Ker~ essential

in E(A) (lemma 16.4). But if so is the case, then for each

x ¢ E(A) we have xl C K e r ~ for some essential right ideal I ,

and ~(x)I = 0 implies ~(x) = 0 since E(A) is non-singular!

hence ~= 0 .

(d)~ (a): Suppose aI = 0 where a c A and I is an essential

right ideal. The homomorphism A-*E(A) given by b~ab

extends to an endomorphism ~ of E(A) . Then Ker~ D I is

essential in E(A) , so ~ is in the Jacobson radical of the

endomorphism ring, and hence ~ = 0 . This implies a = 0 .

(b) ~ (a): Suppose again al = 0 where 0 # a ¢ A and I is

essential. We may write a = aqa for some q c Qm " Since

qa @ 0 and I is essential in ~ , there exists b ~ A such

that 0 % qab e I . But then ab = aqab C al = O which contra-

dicts qab # 0 .

We see in particular from this Proposition that if ~ is a

semi-simple ring, then A is~on-singular. Conversely we have:

Theorem 20.2. Let A be right non-singular. The following

assertions are equivalent:

(a) qm is semi-simple.

(o) Qm is flat as a left A-module and A-~ is a ring epi-

morphism.

(d) Every essential right ideal contains a finitely generated

 115

essential right ideal.

(e) A is right finite-dimensional.

(f) The lattice ~(A) of complemented right ideals is

noetherian.

(g) Every direct sum of non-singular injective modules is

injective.

(h) E(A) is a -injeotlve module.

Proof. We know to begin with that Qm is right self-injective

and regular. Therefore the equivalence of (a), (f), (g) and (h)

is immediate from Proposition 19.7. We have (a) =~(b) by 19.6,

(b)@~(c) trivially, (c) $ ( d ) by Theorem IB.l.

Ne get (d)~=p(e) from Propositions 19.4 and 19.3, and (d)4=~(f)

from Proposition 11.14.

Corollar~ 20.3. If Qm is semi-simple, then:

(i) evelv right Qm-module is injective as an A-module.

(ii) every left Qm-module is flat as an A-module.

Proof. Follows from the formulas (i) and (2) of § 14.

Corol!ar~ 20.4. If Qm is semi-simple and M is a non-singular

A-module, then M ® A Qm is an injective envelope of M .

Proof. We have ME-- E(M) by 8.1 , and M~ = M S A Qm by 13.1.

Proposition 20.~. Let A be right finite-dimenslonal non-

singular. Qm is then also a maximal left ring of quotients of

A if and only if Qm is flat as a right A-module.

Proof. If Qm is a maximal left ring of quotients of A , then

 116

it is right flat over A by Theorem 20.2. On the other hand,

if Qm is right flat, then A--~Q m is a right flat epimorphism

of rings. This implies that Qm is a subring of MI(A) (i.e.

the maximal right flat epimorphic extension ring of A ) by

Theorem 14.1. But MI(A ) is a subring of the maximal left ring

of quotients of A . Since Qm is left self-injective, it must

then coincide with the maximal left ring of quotients of A

(of. Proposition 19.1).

We will take a closer look at the lattices C_E(M) of _E-pure

submodules. We assume A to be right non-singular. As a special

case of Proposition 11.5 we have:

Proposition 20.6. If M is a non-singular module, then ~E(M)

is a complemented modular lattice consisting of the complemented

submodules of M .

Ne may describe ~E(A) to a certain extent, because this

lattice is isomorphic to the lattice of complemented right ideals

of Qm (Proposition 11.12). Since Qm is right self-injective

regular, this second lattice consists of the direct summands

of qm " We thus have:

Proposition 20.7. If A is right non-singular, then ~E(A)

is isomorphic to the lattice of principal right ideals of Qm "

Corollary 20.8. If A is right finite-dimensional non-singular,

then ~E(A) is a modular lattice of finite length.

 117

Examples:

I. If A is a Boolean ring (i.e. a commutative regular ring

with all elements idempotent), then Qm is the "completion"

of A in the sense of Boolean algebra theory.

2. It may happen that Qm is flat as a left A-module even

if Qm is not semi-simple. In fact, one can just take A to

be a regular ring whieh is not semi-simple. More generally:

Proposition 20. 9 . If A is right semi-hereditary, then A

is right non-singular and Qm is flat as a left A-module.

Proof. If 0 $ a c A , then A/r(a) ~ aA is a projective

module, so r(a) is a direct summand of A and can therefore

not be essential in A . Hence A is non-singular.

To show tha% Qm is flat over A , it suffices to show that

the map I@Qm--~AQQm is a monomorphism for every finitely

generated right ideal I . Since I is projective, IQQ m is

a projective Qm-module and is therefore non-singular as a

right A-module. $uppose ~ai@qi ~ I®Qm and ~ aiq i = 0 .

There is an essential right ideal J such that qiJCA for

all i , and for each a s J we get (~ a i @ q i ) a =~ai(qia)®l =

= 0 . l®Qm non-singular implies ~ a i ~ qi = 0 .

E x e r c i se s :

I. Show that if A is regular and Qm is projective as a right

A-module, then A = Qm " (Hint: an imbedding E(rA)C-pF , F

free, induces AC-~F' , F' finitely generated free! then

use the fact that a finitely presented flat module is pro-

jective).
 118

2. Let K be a skew-field and A the ring of upper triangular

2 ~ 2-matrices over K . Show that:

(i) The matrices of the form

constitute a minimal essential right ideal of ~ .

(ii) The ring M2(K ) of all 2.2-matrices is the maximal

right (and left) ring of quotients of A .

(iii) ~ ( K ) is projective as a right (and as a left)

A-modul e.

References: Cateforis [15],[16], Faith [28], Lambek [47], Sando-

mierski [681, [69], Utumi [80], Walker and Walker [81].

§ 21. The maximal tin6 of quotients of a reduced ring

A ring is called reduced if there are no nilpotent elements % 0 .

In this § we want to find out when the maximal ring of quotients

of a reduced ring also is reduced. For a commutative ring, this

is always the case, because if q c ~ and qn = 0 , then

0 ~ aq e A for some a e A and also (aq) n = 0 .

Lemma 21.1. If A is reduced and S CA , then:

(i) r(S) is a two-sided ideal.

(ii) S~r(S) = 0 .

(iii) Every idempotent is central.

(iv) A is non-singular.
 119

Proof. (i): It suffices to show that r(S) = l(S) . But Sa = 0

implies aS = 0 because (aS) 2 = aSaS = 0 , and similarly

aS = 0 implies Sa = 0 .

(if): ( S n r ( S ) ) 2 C Sr(S) = 0 , so S~r('S) = 0 .

(iii): If e2 e and a c A , then (ea(l-e)) 2 ea(l-a)ea(l-e)

= 0 , so ea = eae . Similarly ae = eae , so ea = ae .

(iv): r(a) cannot be essential in A since a A n r(a) =

= aA~r(aA) = 0 by (i) and (ii).

When A is commutative, the converse of (iv) holds, because

a non-singular ring is a subring of a commutative regular ring,

which obviously cannot have nilpotent elements % 0 . In the non-

commutative case, this is no longer true (consider e.g. matrix

rings). Since A reduced implies Qm regular, we should first

determine under what conditions a regular ring is reduced.

Proposition 21.2. The following properties of a ring A are

equival en%:

(a) A is a reduced regular ring.

(b) Every principal right ideal is generated by a central

idempotent.

(c) A is regular and every right ideal is two-sided.

(d) A is strongly regular, i.e. for every a c A there exists

2
x c A such that a -- a x .

Proof. (a)~ (b): Clear from Lemma 21.1(iii).

(b)~(c): Since every principal right ideal is two-sided, all

right ideals are two-sided.

 120

(c)$(d): F o r every a e A there exists x such that a = axa .

Since axA also is a left ideal, axa = bax for some b . Then
2 2
a = a x a . a = baxa = ba , so a = a x .
2
(d)~(a): A is o b v i o u s l y reduced since a = a x implies a =
n n-I 2 2
= a x for each n . We also have (a-axa) 2 = a + axa xa -

2 2
- a x a - axa = 0 , so a = axa and A is regular.

If Qm is a s t r o n g l y r e g u l a r ring, then A m u s t be reduced.

C O n v e r s e l y we have:

Proposition 21.3. The f o l l o w i n g properties of a r e d u c e d r i n g A

are equivalent:

(a) Qm is s t r o n g l y regular.

(b) Every complemented right ideal of A is a t w o - s i d e d ideal.

(c) aA~bA = 0 implies ab = 0 , for all a , b s A .

(d) InJ = 0 implies IJ = 0 , for all r i g h t ideals I , J

of A .

F o r the p r o o f w e need:

L e m m a 21.4. If A is any r i n g w i t h a t o p o l o g y ~ such that

A C A F , then for each I s~(A) one has I = IAFNA .

Proof. Suppose a = ~aiq i C A with ai 6 1 and qi c A F .

Choose J c [ such that qi J C A for all qi " Then aJ =

=ZaiqiJCI , and Z c$(A) implies a C Z .

P r o o f of 21.3: (a)@(b): Each complemented right ideal of A

has the form I = I~NA by the Lemma. Since I~ is a two-

 121

sided ideal of ~ , I is a two-sided ideal in A .

(b)$(d): If l~J = 0 , let K D J be a right ideal maximal

with respect to INK = 0 . K is then two-sided, so IJC INK = O

(d)~ (c): Trivial.

(c)=~(a): We know that Qm is the endomorphism ring of E(A)

(Proposition 19.2). Suppose f:E(A)-@ E(A) is nilpotent, say

f ~ 0 but f2 = 0 . Since the Jacobson radical of the regular

ring Qm is zero, Ker f is not an essential submodule of E(A) .

So there exists O # a s A such that aA~Ker f = 0 . But

f(a) s Ker f , so (c) implies that f(a)a = 0 . Hence a 2 S Ker f ,

2
and we must have a = 0 . Since A is reduced, this is a contra-

diction.

Propositio n 21.~. Let A be reduced with Qm strongly regular.

Then:

(i) C_F(A ) is the family of right annihilator ideals.

m

(ii) ~F(A) is a complete Boolean algebra with l~-~r(1) as a

w

complementation.

Proof. (i): Every ideal of the form r(S) , S C A , is complemented

by Proposition 11.7. Suppose conversely I is complemented, say

I is maximal with respect to I n J = 0 . Then fOr(J) by

21.3(d), and since also r(J)~J = 0 by 21.i, we must have

z = r(J) .

(ii): A Boolean algebra is the same as a complemented distributive

lattice [18]. That r(1) is a complement of I follows from

21.1 and 21.3(d). Since CF(A ) is isomorphic to the lattice of

m
 122

idempotents in Qm ' it suffices to show that this second lattice

is distributive, i.e. eA(fvg) = (eA f) v ( e A g ) . But eaf =

= ef and e ~ f = e+f-ef , and one has e(f+g-fg) = ef+eg-efg .

Example: Let A be strongly regular and right self-injective,

A is then also left self-injective. For let B be the maximal

left ring of quotients of A . B is then also strongly regular

(e.g. by 21.3(b)). For each b ¢ B there exists an idempotent

e s A such that 0 @ eb e A . Sirrce every idempotent in ~ is

central, we also have 0 ~ be c A , so A is an essential right

submodule of B . Since A is right self-injective, B must

equal A .

Exercises:

I. Show that every strongly regular ring is a subring of a

product of skew-fields.

2. Let K be a skew-field and I an arbitrary set. Let A be

the subring of 11K i consisting of (xi) such that almos~

all (i.e. all except a finite number) xi are equal. Show :

(i) A is strongly regular.

(ii) ~K. is a minimal essential ideal of A .

i
I
(iii) N K. is the maximal ring of quotients of A .
I i

Referenoes: Renault [106], Utt~i [113].

§ 22. The classical rin~ of ~uotients

When the classical ring of quotients Qcl exists, it is a

subring of Qm ~ in fact a subring of N(A) . If Qcl is right

self-injective, we have Qcl = N(A) = Qm "

Proposition 22.1. The following properties of A are equivalent:

(a) Qcl exists and is a right self-injective semi-perfect ring.

(b) A is right finite-dimensional and (A:x) cor~tains a non-

zero-divisor for each x S E(A) .

Proof. If Qcl is right self-injective, then Qcl = E(A) .

Every q c Qcl may be written q = ab -I where a s A and b

is a non-zero-divisor in A . H~nce (A:q) contains a non-

zero-divisor. If Qcl furthermore is semi-perfect, then A is

finite-dimensional by Proposition 19.4.

Suppose corr~ersely that A is right finite-dimensional and

(A:x) contains a non-zero-divisor for every x s E(A) . Every

(A:x) is then dense, because for any a c A we have

((A:x):a) = (A:xa) , which has no non-zero left annihilators.

Qm is right self-injective and semi-perfect by Propositions

19.2 and 19.4. We will show that ~ is a classical right ring

of quotients of A . Since for each q s Qm we have qa s A

for some non-zero-divisor a of A , it suffices to show that

every non-zero-divisor a of A is invertible in ~ . Let

~a:Qm--*Qm be the map q~-~aq . Since the kernel of ~a has

zero intersection with the essential submodule A of Qm ' ~ a

is a monomorphism. Im~a is thus an injective submodule of

 124

Qm ' so Qm = ~ a ( Q m ) ~ K . Iterating this, one gets Qm = ~ ( Q m )(~

~a(K) ~ K , using the fact that ~a is a monomorphism. More

generally~ for each n one gets Qm = ~ ( Q m ) ~ @ P ~ ~I(K) ~ ' " ~ K .

Since A is finite-dimensional and Qm = E(A) , we must there-

fore necessarily have K = (0) . ~ a is thus also an epimorphism,

so a has a right inverse. To conclude the proof, we use:

Lemma 22.2. Suppose A has no infinite family of orthogonal

idempotents. If a s A has a right inverse, then a is

invertible.

Proof. Suppose ab = I but ba # 1 . Define elements

bia i bi+lai+l
ei = - for i = I, 2, ....

Each ei is different from zero, because bia i = bi+lai+l

would give bia i = abi+lai+ib = ablalb = bi-la i-I , using

repeatedly the fact that ab = i , and this would finally lead

to ba = I One verifies immediately that the e. are ortho-

i
gonal idempotents, contrary to our assumption.

Lemma 22.3. If A satisfies ACC on right annihilators, then

the singular right ideal is nilpotent.

Proof. Ne will show that the ascending chain r(Z) C r(Z 2) C ...

would be strictly ascending if Z were not nilpotent. If

Z n ~ 0 , choose an element a s Z with zn-la # 0 and largest

possible right annihilator. For each b s Z we have r ( b ) n aA = 0

since r(b) is essential in A . So there exists c c A such

that ac¢ 0 but bac = 0 , which means that r(ba) is strictly

 125

larger than r(a) , and by the choice of a we must therefore

have zn-lba = 0 . Since b G Z is arbitrary, we get Zna = 0 ,

and hence r(~: -I) is strictly oon%ained in r(Z n) .

Proposition 22. 4. Suppose A satisfies:

I) K is right finite-dimensional!

2) A satisfies ACC on right annihilators!

3) (A:x) contains a non-zero-divisor for each x c E(A) .

Then Qcl exists and is a semi-primary right self-injective

ring.

Proof. In view of Proposition 22.1 it remains to show that the

Jacobsor~radical J of Qcl is nilpotent. From Lemma 16.4 we

know that that J is the singular submodule of the right

A-module Qcl ' and Z(A) = J ~ A . Lemma 22.3 says that Zn = 0

for some n , and from this it follows that also jn = 0 . In

fact, if ql,..,qn c J , write qi = aibi -I where ai and bi

are in A and bi are non-zero-divisors. Note that a i c Z(A) .

t
We may write ql....-qn in the form al.. .a~b -I , with a~I ¢

Z(A) and b a non-zero-divisor, by repeated use of the Ore

condition. Hence ql...qn = 0 .

By strengthening the condition (2) slightly we obtain a quasi-

Frobenius classical ring of quotients.

 126

Propositign ' 22.~. The following two assertions are equivalent:

(a) A satisfies

I) A is right finite-dimensional!

2) A satisfies ACC on right annihilators of subsets of

~(A) ,

3) (A:x) contains a non-zero-aivisor for each x e E(A) .

(b) Qcl exists aria i s a O~-ring.

Proof. (a)@(b): Qcl exists and coincides with ~ by 2 2 . 1 .

~t i s QF by 19.7.
(b)=~(a): This follows also from Propositions 19.7 and 22.1.

We may now characterize those rings for which Qcl is semi-

simple ("Goldie's theorem").

Theorem 22.6. The following properties of A ~re equivalent:

(a) Qcl exists and is a semi-simple ring.

(b) ~ is right finite-dimensional and right non-singular, and

has no nilpotent two-sided ideals @ 0 .

(c) A is right finite-dimensional, satisfies A C C on right

annihilators, and has no nilpotent two-sided ideals # 0 .

(d) A is right finite-dimensional and every essential right

ideal contains a non-zero-divisor.

Proof. (a)@(d): Since Qcl = Qm is semi-simple, we know from

Theorem 20.2 that for every essential right ideal I one has

Qcl = IQcl " Write I = ~aiq i with al,...,a n c Y . Write

-I
ql = blCl where
b I , cI ~ A and cI is a non-zero-divisor.
n
Then c I = alb I + ~ aiqic I . Write q2cl = b2c2 -I with b 2 ,
2
 127

02 ¢ A and 02 a non-zero-divisor. Then ClC 2 = alblC 2 + a2b 2 +

n
+ ~ aiqiclc 2 . Continuing in this manner, we get non-zero-divisors
3
o l,...,c n such that c I .... .cn ¢ ~ a i A C I . Thus I contains

a non-zero-divisor.

(d)~(a): Qcl exists and is a right self-injective ring by

22.1. Since A is right non-singular and right finite-dimensional,

it follows from Theorem 20.2 that Qc~ = Qm is semi-simple.

(a)~(o). From Proposition 22.5 follows that we only have to

show that every nilpotent two-sided ideal I is zero. l(I) is

an essential right ideal, for if a @ 0 is any element of A ,

let n be the smallest integer such that al n = 0 ! then there

exists b ¢ I n-I such that ab @ 0 and abc i(I) . But since

(a)~ (d), every essential right ideal contains a non-zero-

divisor, and therefore I must be zero.

(c)~(b): Z(A) is a nilpotent two-sided ideal by Lemma 22.3.

By hypothesis it must be zero, so A is right non-singular.

(b)=~(c): Qm is semi-simple by Theorem 20.2 and this implies

ACC on right annihilators by Proposition 19.7.

(c)-~(d): We first prove:

Lemma 22.7. If A satisfies ACC on right annihilators and has

no nilpotent two-sided ideals @ 0 , then every right or left

nilideal is zero.

Proof. Since Aa is nil if and only if aA is nil, it suffices

to consider a nilideal Aa . Assume Aa # 0 . Among the non-zero

elements of Aa , choose one b c Aa with maximal right annihi-

 128

lator. For each c c A , let (cb) k = 0 , (cb) k-I @ 0 . Since

r(b) C r((cb) k-l) , we must have equality by maximality. Hence

cb c r(b) and bAb = O , which gives (AbA) 2 = 0 . Since every

nilpotent ideal is zero, we get b = 0 . This is a contra-

diction, and so Aa = 0 .

We return to the proof of ( c ) ~ ( d ) : Let I be an essential

right ideal. Since A has ACC' on right annihilators and I is

not a nilideal (by the Lemma), there exists aI @ 0 in I such

that r(al) = r(al2 ) . If l~r(al) @ 0 , we continue and choose

a 2 c I n r ( a I) such that a2 @ 0 and r(a2) = r(a22 ) . Yf then

l~r(al) ~r(a2) % 0 , we go on and get a 3 ¢ Inr(al)N r(a2) ~ and

so on. At each step we obtain a direct sum alA ~ ... @ akA . This

is proved by induction: suppose alA ~ ... (~ ak_iA is direct and

akb = a l b l + . . . + S k _ l ~ _ l ! since for each i ~ k we have aia k = 0 ,

k-I
we get b i c r(ai 2) = r(ai) and hence ~ a.b. = 0 . But A is
i I z
right finite-dimensional, so the process must stop at some stage,

where we have l~r(al)~...~r(ak) = 0 . Then r(al)~...~r(a k) =

-- 0 , so if c = al+...+a k C I , then r(c) = 0 . We need:

Lemma 22.8. If A is right finite-dimensional and r(c) = 0 ,

then cA is an essential right ideal.

Proof. If I is any right ideal and cA(~l = 0 , then it is

easy to see that we get a direct sum I ( ~ cI ~ ... ~ cnI (~ . . . .

We may now conclude the proof of the Theorem. Since A is right

 129

non-singular, it follows from the Lemma that 1(o) = 0 , and c

is thus a non-zero-divisor belonging to I .

Exercises

Let A be the ring of matrioes

0 b

with a , b e ~ and q c Q . Show that the ring of upper tri-

angular matrices over Q is a two-sided classical ring of

quotients of A , while M2(Q) is the maximal right (and left)

ring of quotients of A . (~ote that A has nilpotent two-

sided ideals ~ 0 ).

~sferenoes, Q~briel [31], Goldie [ ~ ] , Jans [391, ~ b o = ~d

Winton !>41, P~ocesi and Small [ 6 4 , S~domier~i [68].
REFERENCES:

1. J . S . A l i n , S t r u c t u r e o f t o r s i o n modules, Ph.D. T h e s i s ,

U. of Nebraska 1967.

2. J.S. Alin and S.E. Diokson, Goldie's torsion theory and its

derived functor, Pacific J. Math. 24, 1968, 195-203.

3. G. Almkvist, Fractional categories, Arkiv f. Mat. 7, 1968,

449-476.

4. S.A. Amitsur, General theory of radicals II, American J. Math.

76, 1954, 100-125.

5- K. Asano, Uber die Quotientenbildung yon Sohiefringen,

J. Math. Soo. Japan I, 1949, 73-78.

6. G. Azumaya, Completely faithful modules and self-injeotive

rings, Nagoya Math. J. 27, 1966, 697-708.

7. H. Bass, Finitistio dimension and a homologioal generalization

of semi-primary rings, Trans. Amer. Math. Soo. 95, 1960,

466-488.
8. J.-E. Bj~rk, Rings satisfying certain chain conditions, to appes~

9. N. Bourbaki, Algbbre, oh. 8, Hermann 1958.

I0. - , - , Alg6bre commutative, oh. I and 2, Hermann 1961.

II. B. Brainerd and J. Lambek, On the ring of quotients of a

Boolean ring, Canad. Math. Bull. 2, 1959, 25-29.

12. I. Buour and A. Deleanu, Introduction to the theory of oategorie~

and functors, J. Wiley 1968.

13. H. Cartan and $. Eilenberg, Homological algebra, Princeton 1956.

14. V.C. Cateforis, Flat regular quotient rings, Trans. Amer. Math.

Soc. 138, 1969, 241-250.

 131

15. V.C. Cateforis, On regular self-inJeotive rings, Pacific J.

Math. 30, 1969, 39-45.
16. - , - , Two-sided semisimple maximal quotient rings~
Trans. Amer. Math. Soo. 149, 1970, 339-349.
17. S.U. Chase, Direct products of modules, Trans. Amer. Math.
Soo. 97, 1960, 457-473.
18. P.M. Cohn, Universal a l g e b r a , Harper and Row 1965.
19. R.S. C~ningham, E.A. R u t t e r and D.R. Ttu~nidge, Rings o f
q u o t i e n t s o f endomorphism r i n g s o f p r o j e c t i v e modules,
to appear.
~O. C.W. C u r t i s and I . Reiner, R e p r e s e n t a t i o n t h e o r y o f f i n i t e
groups and a s s o c i a t i v e a l g e b r a s , J . Wiley 1962.
21. S.E. Diokson, A t o r s i o n t h e o r y f o r a b e l i a n c a t e g o r i e s ,
Trans. Amer. Math. Soo. 121, 1966, 223-235.
22. M. D J a b a l i , Annenu de f r a c t i o n s d,un J-anneau, Canad. J . Math.
17, 1965, 1041-1052.
23. S. Eilenberg and T. Nakayama, O~ the dimension of modules and
algebras II, Nagoya Math. J. 9, 1955, 1-16.
24. V.P. E1izarov, On quotient rings of associative rings, Izv.
Akad. Nauk SSSR 24, 1960, 153-170. (Aaer. Math. Soo.
Transl. 52, 1966).
2~5, - . - , Flat extensions of rings, Soviet Math. Dokl. 8,
1967, 905-907.
26. H. Eriksson, Fraktionkategorier och Grothendieoktopologier
(mimeographed), Stockholm 1967.
27. C. Faith, Rings with ascending condition on annihilators,
Nagoya Math. J. 27, 1966, 179-191.
28. - " - t Lectures on inJeotive modules and quotient rings,
Springer Lecture Notes 49, 1967.
30. C. Faith and E.A. Walker, Direot-sm, representations of inJectiw
modules, J. Algebra 5, 1967, 203-221.
31. P. Gabriel, Des oat6gories ab@liennes, ~ 1 . Soc. Math. France
90, 1962, 323-448.
32. A.W. Goldie, Some aspects of ring theory, Bull. London Math.
Sos. 1, 1969, 129-154.
 132

33. O. Goldman, Rings and modules o f q u o t i e n t s , J. Algebra 13,

1969, 10-47.
34. R.N. Gupta, Self-inJeotive quotient rings and inJectlve quotient
modules, Osaka J. Math. 5, 1968, 69-87.
35. M. Hacque, Loealisations exaotes et localisations plates,
Pub1. D6p. Math. Lyon 6, 1969, 97-117.
36. G. Helzer, On divisibility and inJectivit¥, Canad. J . Math.
18, 1966, 901-919.
37. M. Iked~ and T. Nakayama, On some characteristic properties
of quasi-FTobenius and regular rings, Prec. Amer. Math.
See. 5, 1954, 15-18.
38. J.P. Jans, Some aspects of torsion, Pacific J. Math. 15, 1965,
1249-1259.
39. - ,, -- ~ On. o r d e r s in quasi-Frobenius r i n g s , J. Algebra 7~
1967, 35-43.
40. R.E. Johnson, Structure theory of faithful rings, I and II,
Trans. Amer. Math. Soc. 84, 1957, 508-544.
41. A.I. Ka~u, Closed classes of left A-modules and closed sets
of left ideals of the ring A, Mat. Zametki 5, 1969,
381-39o.
42. T. Kate, Self-injective rings, Tohoku Math. J. 19, 1967, 485--495.
43. - , - , Torsionless modules, Tohoku Math. J. 20, 1968, 234-243.
44. A.G. Kurosoh, Radicals in rings and algebras, Mat. Sb. 33,
1953, 13-26.
45. J. Lambek, On the structure of semi-prime rings and their
rings of quotients, Canad, J. Math. 13, 1961, 392-417.
46. -,- , On Utuli's ring of quotients, Canad. J. Math. 15,
1963, 363-370.
47. - . - , Lectures on rings and modules, Blaisdell 1966.
48. - . - , Torsion theories, additive semantics and rings of
quotients, Springer Lecture Notes 177, 1971.
49. - , - , Bioommutators of nice inJectives, to appear.
50. L. Levy, Torsion-free and divisible modules over non-integral
domains, Canad. J. Math. 15, 1963, 132-151.
 133

51. J.-M. Ma~anda, InJeotive structures, Trans. Amer. Math. Soo.

llO, 1964, 98-135.
52. E. Matlis, InJeotive modules over noetherian rings, Pacific
J. Math. 8, 1958, 511-528.
53. - , - , Modules with descending chain condition, Trans.
Amer. Math. Soo. 97, 1960, 495-508.
54. A.C. Mewborn and C.N. Winton, Orders in self-inJective semi-
perfect rings, J. Algebra 13, 1969, 5-9.
55. A.P. Mishina and L.A. SkornJakov, Abelian groups and modules
(in R u s s i a n ) , Moscow 1969.
56. K. Morita, On S-rings, Nagoya Math. J. 27, 1966, 687-695.
57. - ,, - , Localization in oategories of modules I, Math. Z.
114, 1970, 121-144.
58. - " - , Localization in categories of modules II, J. Reine
Angew. Math. 242, 1970, 163-169.
59. C. N~st~sescu and N. Popescu, On the localization ring of a
ring, J. Algebra 15, 1970, 41-56.
60. B.L. Osofsky, A generalization of quasi-Frobenius rings,
J. Algebra 4, 1966, 373-387.
61. - , - , Endomorphism rings of quasi-injective modules,
Canad. J . Math. 20, 1968, 859-903.
62. N. Popescu and P. Gabriel, Caraot6risation des oat6gories
ab@lienmes avee g@n@rateurs et limites inductives
exactes, C.R. Aoad. Soi. Paris 258, 1964, 4188-4190.
63. N. Popescu and T. Spircu, Sur les @pimorphismes plats d, anneaux,
C.R. Aoad. Soi. Paris 268, 1969, 376-379.
64. C. Procesi and L. Small, On a theorem of Goldie, J. Algebra 2,
1965, 80-84.
65. J.-E. Roos, Sur l,anneau maximal de fractions des AW~-alg~bres
et des anneaux de Baer, C.R. Acad. Sci. Paris 266, 1968,
449-452.
66. - ,, - , On the structure of abelian categories with
generators and exact direct limits. Applications,
to appear.
 134

67. S@minaire P. Samuel, Les @pimorphismes d, anneaux, Paris 1968.

68. F.L. Sandomierski, Semisimple maximal quotient rings, Trans.
Amer. Math. Soc. 128, 1967, 112-120.
69. -., - , Non-singular rings, Proo. Amer. Math. Soc.
19, 1968~ 225-230.
70. L. Silver, Non-commutative localization and applications,
J. Algebra 7, 1967, 44-76.
71. L.A. Skornjakov, Elizarov's quotient ring and the localization
principle, Mat. Zametki I, 1967, 263-268.
72. B. Stenstram, On the completion of modules in an additive
topology, J.Algebra 16, 1970, 523-540.
73. - ,, - , Flatness and localization over monoids, to
appear in Math. Naohrichten.
74. H. Tachikawa, Double centralizers and dominant dimensions,
Math. Z. 116, 1970, 79-88.
75. M.L. Teply, Torsion-free injective modules, Pacific J. Math.
28, 1969, 441-453.
76. - , - , Some aspects of Goldie's torsion theory, Pacific
J. Math. 29, 1969, 447-459.
77. M.L. Teply and J.D. Fuelberth, The torsion submodule splits
off, to appear.
78. D.R. Turnidge, Torsion theories and semihereditary rings,
Prom. Amer. Math. Soo. 24, 1970~ 137-143.
79. - , - , Torsion theories and rings of quotients of
Morita equivelent rings, to appear.
80. y. Utuli, On quotient rings, Osaka Math. J. 8, 1956, 1-18.
81. C.L. Walker and E.A. Walker, Quotient categories and rings
of quotients, mimeographed, 1963.
82. L.E.T. Wut H.Y. Moohizuki and J.P. Jans, A characterization
of QF-3 rings, Nagoya Math. J. 27~ 1966, 7-13.

83. O. Zariski and P. Samuel, Commutative algebra I, van Nostrand

1958.
 I35

Additional referen~esz
84. T. Akiba, Remarks on generalized rings of quotients, III,
J. of Math. Kyoto 9, 1969, 205-212.
85. E.P. Armendariz, On finite-dimensional torsion-free modules
and rings, Proc. Amer. Math. Soc. 24, 1970, 566-571.
86. J.E. BJ~rk, Rings satisfying a minimum condition on principal
ideals, J. Reine Angew. Math. 236, 1969, 112-119.
87. A. Cailleau and G. Renault, Sur l,enveloppe inJective des
anneaux semi-premiers & l,id6al singulier nul,
J. Algebra 15, 1970, 133-141.
88. V.P. Elizarov, Rings of quotients, Algebra i Logika Seminar 8,
1969, 381-424.
89. P. Gabriel and M. Zisman, Calculus of fractions and homotopy
theory, Springer 1967.
90. A.W. Goldie, The structure of prime rings under ascending
chain conditions, Proo. London Math. Soc. 8, 1958,
589-608.
91. - - - , Semi-prime rings with maximum condition, Proo.
London Math. Soc. 10, 1960, 201-220.
92. R.N. Gupta, Self-inJective quotient rings and injective
quotient modules, Osaka J. Math. 5, 1968, 69-87.
93. N. Jaoobson, Structure of rings, Amer. Math. Soc. Coll.
Pub1. 37, revised ed. 1964.
94. R.E. Johnson, Prime rings, Duke Math. J. 18, 1951, 799-809.
95. - , - , Extended centralizer of a ring over a module,
Proc. Amer. Math. Soc. 2, 1951, 891-895.
96. - , - , Quotient rings of rings with zero singular
ideal, Pacific Math. J. II, 1961, 1385-1392.
97. - ,, - , Rings with zero right and left singular ideals,
Trans. Amer. Math. Soo. 118, 1965, 150-157.
98. R.E. Johnson and E.T. Wong, Quasi-injective modules and
irreducible rings, I. London Math. Soo. 36, 1961,
260-268.
 136

99. J.T. Knight, On epimorphisms of non-commutative rings,

Proc. Cambr. Phil. Soc. 68, 1970, 589-601.
I00. D. Lazard, Autour de la platitude, Bull. Soo. Math. France
97, 1969, 81-128.
I01. A.C. Mewborn, Some conditions on commutative semiprime
rings, J. Algebra 13, 1969, 422-431.
102. B. Mitchell, Theory of categories, Academic Press 1965.
103. K. Morita, Localization in categories of modules IIY,
Math. Z. 119, 1971, 313-320.
104. N. Popescu and T. Spircu, Quelques observations sur les @pi-
morphismes plats (8 gauche) d'anneaux, J. Algebra 16,
1970, 40-59.
105. N. Popescu and D. Spulber, Sur les quasi-ordres ('~ gauche)
dans un anneau, .J. Algebra 17, 1971, 474-481.
106. G. Renault, Anneaux r@duits non commutatifs, J. Math. Pures
Appl. 46, 1967, 203-214.
107. - ,, - , Anneau associ@ ~ un module injectif, Bull. Sci.
Math. 92, 1968, 53-58.
108. J.C. Robson, Artinian quotient rings, Proc. London Math.
Soc. 17, 1967, 600-616.
109. J.E. Roos, Locally distributive spectral categories and
strongly regular rings, Reports Midwest Category
Seminar, Springer Lecture Notes 47, 1967, 156-181.
II0. L. Small, Orders in artinian rings, J. Algebra 4, 1966,
13-41 and 505-507.
Iii. H. Tachikawa, Localization and artinian quotient rin~s,
Math. Z. 119, 1971, 239-253.
112. M. Takeuchi, A simple proof of Gabriel and Popescu's theorem,
J. Algebra 18, 1971, 112-113.
113. Y. Ut~i, On rings of which any one-sided quotient rings are
two-sided, Proc. Amer. Math. Soc. 14, 1963, 141-147.
114. E.T. Wong and R.E. Johnson, Self-injective rings, Canad.
Math. Bull. 2, 1951, 167-173.