Martin Mathieu

Classically
Semisimple
Rings
A Perspective Through Modules
and Categories
Classically Semisimple Rings
Martin Mathieu

Classically Semisimple
Rings
A Perspective Through Modules
and Categories
Martin Mathieu
School of Mathematics and Physics
Queen’s University Belfast
Belfast, UK

ISBN 978-3-031-14208-6 ISBN 978-3-031-14209-3 (eBook)

https://doi.org/10.1007/978-3-031-14209-3

Mathematics Subject Classification: 16-01, 16B50, 16D10, 16D60, 16D70, 16D90

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland
AG 2022
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse
of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my children,
who never cease
to amaze me.
Le savant n’étudie pas la nature
parce que cela est utile;
il l’étudie
parce qu’il y prend plaisir
et il y prend plaisir
parce qu’elle est belle.

Henri Poincaré, 1908

Preface

This book is written for advanced undergraduate and beginning graduate students. It
tells the story of a classical classification problem, about rings that are semisimple in
the sense of Wedderburn and Artin. One of the first major achievements of modern
algebra, the Wedderburn–Artin theorem, paved the way for many like theories in the
first half of the twentieth century. We present this theory from a modern viewpoint
choosing an approach via modules. At the same time, the reader is allowed a
glimpse into category theory, which partly finds its origins in the theory of modules.
Category theory is highly abstract and therefore sometimes hard to digest in a first
helping. By interweaving the basic concepts with more concrete examples coming
from modules and rings, we aim to make the abstract ideas more accessible and, at
the same time, to present a classical beauty in an attractive modern setting.

Belfast, UK Martin Mathieu

Spring 2022

ix
Contents

1 Motivation from Ring Theory .. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

1.1 Basics on Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
1.1.1 Reducing Complicated Rings to Simpler Ones . . . . . . . . . . . . 5
1.1.2 One-Sided Ideals in Noncommutative Rings . . . . . . . . . . . . . . 6
1.1.3 Images of Ideals Under Homomorphisms . . . . . . . . . . . . . . . . . . 6
1.1.4 Embedding into the Endomorphism Ring and
General Representations . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 6
1.1.5 Group Representations . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 7
1.2 Categories of Modules . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 8
1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 12
2 Constructions with Modules. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
2.1 Some Special Morphisms . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 15
2.2 Quotient Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 16
2.3 Generating Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.4 Direct Sums and Products of Modules . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 17
2.5 Free Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 19
2.6 Special Objects in a Category . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 21
2.6.1 Free Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 21
2.6.2 Products and Coproducts . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 21
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 23
3 The Isomorphism Theorems. . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
3.1 Isomorphisms Between Modules . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 25
3.2 Functors and Natural Transformations . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 27
3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 32
4 Noetherian Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
4.1 Permanence Properties of Noetherian Modules .. . . . . . . . . . . . . . . . . . . 37
4.2 Exact Categories and Exact Functors . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 39
4.2.1 Kernels and Cokernels . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 40
4.2.2 Exact Categories . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 41
4.2.3 Exact Functors . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 42
4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43

xi
xii Contents

5 Artinian Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
5.1 Finitely Cogenerated Modules . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 48
5.2 Commutative Artinian Rings . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 50
5.3 Artinian vs. Noetherian Modules . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 52
5.4 Abelian Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 54
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 56
6 Simple and Semisimple Modules. . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
6.1 Decomposition of Modules . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 59
6.2 Projective and Injective Modules . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 64
6.3 Projective and Injective Objects . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 68
6.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 70
7 The Artin–Wedderburn Theorem . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
7.1 The Structure of Semisimple Rings . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 73
7.2 Maschke’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 80
7.3 The Hopkins–Levitzki Theorem . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 81
7.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 89
8 Tensor Products of Modules . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 93
8.1 Tensor Product of Modules . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 93
8.2 Tensor Product of Algebras . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 101
8.3 Adjoint Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 102
8.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 105
9 Exchange Modules and Exchange Rings . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 109
9.1 Basic Properties of Exchange Modules . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 110
9.2 Exchange Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 114
9.3 Commutative Exchange Rings . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 119
9.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 128
10 Semiprimitivity of Group Rings . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
10.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 131
10.2 Some Analytic Structure on C[G] . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 133
10.3 The Semiprimitivity Problem . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 137
10.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 141

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 145
Index of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 147
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 149
Introduction

In order to understand a mathematical object—in our case, a ring—one introduces

a variety of properties the object may have and distinguishes different objects
according to these properties. In pursuing this, it is often helpful to let the object
‘act’ on another kind of object which may have a simpler structure—in our case,
the ring acts on an abelian group. This births the concept of a module. A slightly
different, but related, point of view is to ‘represent’ the object (the ring) as a
subobject of a prototypical object (here, as a subring of an endomorphism ring).
Thus, by placing the object of interest into a variety of settings one obtains a better
view of it than by just ‘staring at it’ from one perspective.
One may want to think of all possible representations (modules) of our given
object (ring), put them into a ‘big set’ and study this set. The process starts again:
one introduces a variety of properties the modules may have . . . , and one ends up
with the need to bring some order into the situation. What can we do with the ‘set of
all modules’ of a given ring and how can we relate different subsets to each other,
or even relate those to sets made up of different structures? In this way, the notion
of a functor between categories is created.
This book comprises material which formed part of lecture courses by the author
given at the Eberhard Karls Universität Tübingen (a long time ago) and at Queen’s
University Belfast (far more recently). The aim is to introduce undergraduate
students with some background in abstract algebra or beginning postgraduate
students to the idea of investigating certain rings by exploring their categories
of modules. We have three goals in mind: firstly to study, in all detail, a very
specific class of (non-commutative) rings which became prominent in the early part
of the twentieth century. Secondly, to employ the theory of modules in order to
achieve this goal. Upon first impression, modules look like a slight generalisation
of vector spaces, but one quickly realises that they are, in fact, a vast generalisation.
For instance, the notion of a basis and the invariance of the length of a basis,
available and fundamental for vector spaces (even in infinite dimensions, cf., e.g.,
[22, p. 241]), lose its significance. In the second half of the twentieth century, a more
abstract approach to mathematics became predominant, and it emerged that the best
way to handle the multitude of different types of modules is to put them into various
classes and to study the relation between these classes. The theory of categories
and functors which grew out of this endeavour has found its way into every niche

xiii
xiv Introduction

of mathematics, as a universal language in which the essential qualities that make

things work can be exhibited. The downside of such an abstract approach, however,
is that one can easily be overwhelmed by terminology. Categories of modules
have the advantage that we can keep our feet on the ground whilst remaining
very comprehensive (in a sense that can be made precise, see Sect. 5.4 on abelian
categories). Finally, therefore, the third goal of this book is to guide the reader into
and through this abstract world in small, manageable steps.
The prerequisites to use this book successfully are minimal. The reader should
have studied Linear Algebra (the first few chapters of [20] or [33], for example, will
suffice; see also the treatment in [9]) as well as some basics in Ring Theory (e.g.,
[7], [23] or [34]); even the commutative theory may be sufficient. A good test is to
look at the list of examples of rings on page 2. If the reader can understand (most of)
them, they are well prepared for reading the book. A modest familiarity with groups
(such as contained in [34, Chapter 4], for example) without any deeper theory is
also helpful.
We begin with a motivation for the concept of a module in a leisurely manner
in Chap. 1, by observing how modules appear in a natural way in various instances.
We then have a first glimpse into the world of categories and look at a substantial
list of examples in the hope of convincing the reader that they are ubiquitous in
mathematics.
Chapter 2 is devoted to the basic constructions that one wants to, and has to,
perform with modules in order to build up a viable theory. These include quotients,
direct sums and direct products. We also have a look at the special class of free
modules; these are the ones that resemble vector spaces the most. This discussion
naturally leads us to consider objects with similar attributes in categories; often
these are defined via universal properties. The rather short Chap. 3 discusses the
fundamental isomorphism theorems for modules. Functors and mappings between
functors (‘natural transformations’) are introduced in the second part of this chapter.
Finiteness conditions on submodules are the subject of the following two
chapters. In Chap. 4, we discuss Noetherian modules and rings including Hilbert’s
basis theorem. The various permanence properties of Noetherian modules are dealt
with using exact sequences; this also allows for a smooth transition into the second
half of the chapter where a setting for exact sequences in more general categories is
introduced and explored, together with the concept of exact functors. In this chapter
the reader will start to realise that the morphisms are the really decisive parts of a
category; this is another advantage of using the general setting of categories which
emphasises this feature in contrast to the importance that usually is given to the
objects in a theory, whether it is simple groups or Banach spaces, for example.
Close relatives of the Noetherian modules are the Artinian ones, which are
investigated in Chap. 5. Much of their theory runs in parallel, but there are
differences. For example, the notion of a finitely cogenerated module cannot be
expressed in terms of elements, in contrast to the concept of finite generation of a
module. Another is that commutative Artinian rings, studied in Sect. 5.2, allow for a
more detailed structure theory. The highlight of this chapter is the Hopkins–Levitzki
theorem (Theorem 5.3.2) stating that every left Artinian ring is left Noetherian. The
Introduction xv

converse statement is in general false: the ring Z is the most obvious example.
The proof of this fundamental result is deferred to Sect. 7.3, but Exercise 5.5.8
already invites a proof in the commutative case. In Sect. 5.4 abelian categories are
introduced which are the most prominent generalisation of module categories, and
various typical properties and techniques are displayed.
The heart of the theory is Chap. 6, where simple and semisimple modules are
defined and studied. Finitely generated semisimple modules turn out to be both
Noetherian and Artinian (Theorem 6.2.4). The concept of a classically semisimple
ring is introduced, and these rings are characterised by the fact that their category of
(left) modules entirely consists of semisimple modules (Theorem 6.1.12). This, in
consequence, leads to other important classes of modules, projective and injective
modules (Sect. 6.2) and their generalisations to projective and injective objects in
arbitrary categories (Sect. 6.3). The description via the Hom-functors opens up the
connection to Homological Algebra which we do not pursue further here; see [21]
or [29] for a full development.
The Artin–Wedderburn theorem is the focus of Chap. 7. This was one of the
major achievements of the early theory of non-commutative rings and still is a cor-
nerstone of basic algebra. It forms the model for many structure theorems that were
obtained during the twentieth century. An important example of a semisimple ring
is the group ring of certain finite groups; this is Maschke’s theorem (Theorem 7.2.1)
discussed in Sect. 7.2. A full proof of the Hopkins–Levitzki theorem, Theorem 5.3.2,
is finally given in Sect. 7.3. This section also contains a discussion of the Jacobson
radical for general unital rings and the notion of a semiprimitive ring. Our proof
differs from the one commonly given, which relies on composition series, and uses
module techniques instead. We do not have anything to say about categories in this
chapter aside from Remark 7.1.13.
A construction of paramount importance is the tensor product of modules to
which Chap. 8 is devoted. The fundamental Hom-tensor adjointness relation is
established in Theorem 8.1.9 and further studied in Sect. 8.3 on adjoint functors
between general categories. The special situation of tensor products of algebras over
a field is the theme of Sect. 8.2.
An interesting generalisation of semisimple rings was introduced in 1972 by
Warfield, based on previous work on modules by Crawley and Jónsson. In order
to provide the reader with some more recent development in Ring Theory, we have
included Chap. 9 which gives an introduction to the theory of exchange modules
and exchange rings. More comprehensive treatments can be found in [13] and [16].
In Sect. 9.1, we compile a number of basic properties of exchange modules, and in
Sect. 9.2, exchange rings are introduced via a separation property using idempotents
(Definition 9.2.1). It is shown that a unital ring R is an exchange ring if and only
if the standard module R R is an exchange module (Theorem 9.2.6) which in turn is
equivalent to the fact that every projective module in the category R−mod has the
exchange property (Corollary 9.2.7). Commutative exchange rings are investigated
in great detail in Sect. 9.3 and the reader will discover a number of equivalent
properties characterising exchange rings (Theorem 9.3.10). We also make contact
with various notions of dimension of a commutative ring and with topology; e.g., it
xvi Introduction

is shown in Proposition 9.3.2 that a compact Hausdorff space X is zero-dimensional

if and only if the ring CR (X) of all continuous real-valued functions on X is an
exchange ring.
The problem when a group ring is semiprimitive has challenged algebraists
for a long time. It will be studied in Chap. 10, where we highlight some of the
major techniques and results. It is interesting to see how analytic tools, explored
in Sect. 10.2, lead to Rickart’s result from 1950 stating that the group ring with
complex coefficients is semiprimitive independent of the properties of the group
(Theorem 10.3.1). We do not aim for a comprehensive discussion of the problem
which would lead us too far afield but content ourselves with some sample results.
The historical development of non-commutative ring theory over the last
100 years is fascinating, and there are good accounts on it. We have therefore
confined ourselves to historical remarks on three major players, Artin, Noether and
Wedderburn, contained in the appropriate chapters. Each chapter is completed by
a section of exercises, of varying difficulty. The vast majority of them merely test
the understanding of the concepts introduced prior and are meant for the reader to
get their hands on something concrete, and not as a serious challenge. In category
theory proper one needs to work with ‘classes’ which we left undefined (as is
common practice in basic texts, just think of a class as a set which may not be
contained in another). A keen reader will find possible axiomatic approaches to
category theory in the appendix of [30] or in [25], but there are others.
Finally, I would like to thank all my students who diligently attended my lectures
and tried to learn what I had to offer. Several of them pointed out oversights
in the online notes, and I am particularly indebted to my former students Linda
Mawhinney and Michael Rosbotham for useful discussions on Chaps. 9 and 10,
respectively.
Motivation from Ring Theory
1

In this chapter, the reader will become familiar with our notation and basic
assumptions on the one hand; on the other, they are invited to various instances
where the setting of Ring Theory in its strictest sense turns out to be too narrow,
or at least somewhat cumbersome, in order to provide us with enough tools for
studying certain properties of rings. We shall take this as part of our motivation to
introduce the more comprehensive setting of modules below.
While the first section will be more informal in that we do not provide every detail
of each argument, the second section will devoted to a few first steps in Category
Theory, where we will introduce the fundamental concepts and illustrate these with
a number of examples.

1.1 Basics on Modules

We begin this section with a reminder on the concept of a ring.

Definition 1.1 A ring R consists of a non-empty set together with two binary
operations “+” and “ · ”, which are usually referred to as the addition and the
multiplication in R, such that

(i) (R, +) is an abelian group;

(ii) (R, ·) is a semigroup (which simply means · is associative);
(iii) the two distributivity laws

(x + y)z = xz + yz and z(x + y) = zx + zy (x, y, z ∈ R)

hold.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 1

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_1
2 1 Motivation from Ring Theory

The ring R is called unital if (R, ·) has an identity element, i.e., an element 1 such
that 1 · x = x = x · 1 for all x ∈ R. The ring R is called commutative if (R, ·) is
commutative.

Note that we will not automatically assume that every ring is unital but this
assumption will be made explicit whenever necessary; the theory of unital rings
runs more smoothly in general. We list a number of examples that are useful to
understand the general theory.

Examples 1.2

(i) Z, the ring of integers, is the most basic but still an interesting ring. Clearly it is
both unital and commutative. Any new concept introduced should be checked
for Z first.
(ii) Zn = Z × . . . × Z, for a natural number n, with coordinate-wise operations.
This gives another unital commutative ring which already has some different
features. E.g., it is no longer an integral domain.
(iii) Every field K— such as Q, R and C—provides us with an extremely well-
behaved commutative unital ring. To some extent, Ring Theory is about how
far one can relax the nice multiplicative properties in a field and still get useful
and interesting examples.
(iv) H, the quaternions, almost forms a field but for the non-commutativity of
multiplication. Thus, they yield the first interesting example of a non-trivial
division ring. (See, for example, [34, Sect. 5.2].)
(v) R[x], the ring of polynomials in the indeterminate x over a commutative unital
ring R. When iterated, this construction yields the ring R[x1 , . . . , xn ], n ∈ N of
polynomials in n indeterminates which is highly useful in Algebraic Geometry.
(vi) Mn (R), where R is a unital ring and n ∈ N, is the ring of n × n matrices with
entries in R. Even if R is commutative, this ring is only commutative in case
n = 1. In general, the matrix rings are a source of useful examples for many
ring theoretic properties. E.g., Mn (R) is simple (that is, has no non-trivial two-
sided ideals) if and only if R is simple. (See Proposition 7.1.1 in Chap. 7.)
(vii) Tn (R), for a unital ring R and n > 1, is the ring of strictly upper diagonal
matrices, that is, an n × n matrix (aij ) belongs to Tn (R) if aij = 0 for all i ≥ j
and aij ∈ R is arbitrary otherwise. This is a non-unital noncommutative ring.

In the above examples, the operations of addition and multiplication are the
canonical ones; hence there is no need to discuss these in further detail as they
are treated in every undergraduate textbook on ring theory such as [7] or [34], for
instance. The next two important examples may be less familiar; for that reason, we
spend a little more time on them. The first is fundamental for the development of
module theory.

(viii) End(G), the endomorphism ring of an abelian group G. An endomorphism of

an abelian group G is a mapping ϕ : G → G such that ϕ(g+h) = ϕ(g)+ϕ(h)
1.1 Basics on Modules 3

for all g, h ∈ G. (We write abelian groups additively, as is customary.) It

follows that ϕ(0) = 0 and ϕ(−g) = −ϕ(g). Composing two endomorphisms
ϕ and ψ yields a new one ψ ◦ ϕ which gives the multiplication in End(G).
The addition is defined via (ϕ + ψ)(g) = ϕ(g) + ψ(g), g ∈ G, and it is
straightforward to verify the distributivity laws. The identity in End(G) is of
course the identity mapping, and in general, End(G) is noncommutative.
(ix) R[G], the group ring of anarbitrary group G. The elements of R[G] are finite
formal sums of the form g∈G rg g where rg ∈ R and only finitely many rg
are non-zero. The addition is defined via
  
rg g + sg g = (rg + sg )g
g∈G g∈G g∈G

and the multiplication is given via convolution, that is,

    
rg g · sh h = tk k,
g∈G h∈G k∈G


with tk = k=gh rg sh . (Here, we write the group multiplicatively as it does
not have to be abelian.) If R is unital then G is canonically a subgroup of
R[G] via g → 1 · g, and if e ∈ G is the identity then 1 · e is the identity
in R[G]. If both R and G are commutative, so is R[G]. In the case when K is
a field, the group ring K[G] in addition is a K-vector space with basis G (and
the scalar multiplication simply defined via multiplication of the coefficients
in K), so it is a K-algebra.
In this way, methods from Ring Theory can be used to study groups. For
instance, one of the deep open and longstanding problems is whether a group
G is torsion free if and only if the group ring K[G] has no non-trivial divisors
of zero. See also Exercise 1.3.6. (Quite a lot on this problem can be found
in [31].) We shall discuss certain aspects of group rings in Chap. 10.

The aim of Ring Theory is to understand the structure and the properties of these
and many other classes of rings. An important tool to achieve this aim is the concept
of a module which allows us to look at a given ring from various perspectives.

Definition 1.3 Let R be a ring. Let M be an abelian group. We say that M is a

left R-module, and denote this by writing R M, if there is a bi-additive mapping
R × M → M, (r, m) → r · m such that (rs) · m = r · (s · m) for all r, s ∈ R and all
m ∈ M. Here, bi-additive means

(r + s) · m = r · m + s · m, r · (m + n) = r · m + r · n (r, s ∈ R, m, n ∈ M).

In the case when R is unital and 1 · m = m for each m ∈ M, the left module R M is
called unital (in the literature, unitary is used as well).
4 1 Motivation from Ring Theory

There is an obvious symmetric way to define a (unital) right R-module MR . If both

sides work together we obtain the notion of a bimodule.

Definition 1.4 Let R and S be rings, and let M be an abelian group. We say that
M is an R-S-bimodule, written as R MS , if M is a left R-module as well as a right
S-module and the compatibility condition r · (m · s) = (r · m) · s holds for all r ∈ R,
s ∈ S and m ∈ M.

Remark Let R M be a left R-module. Sometimes the mapping (r, m) → r · m is

referred to as the “scalar multiplication” and the elements in R as the “scalars”.
This mapping is also called the “action” of R on R M.

If one does not want to specify which kind of (one-sided) module one speaks of,
one says “a module M over the ring R”. We next come to the important concept of
a substructure.

Definition 1.5 Let R M be a left R-module. Let N be a subgroup of M which is

closed under the R-action, that is, r · n ∈ N for every r ∈ R and n ∈ N. Then N is
called a left R-submodule of R M; we denote this fact by writing R N ≤ R M.
Similarly, NR ≤ MR and R NS ≤ R MS are defined.

The submodules of a module are viewed as the (smaller) building blocks;

the connections between different modules are given by the structure preserving
mappings defined below.

Definition 1.6 Let R M and R N be left R-modules. A mapping f : R M → R N is

called a left R-module map (also: a left R-module homomorphism) if it is additive,
that is, f (m1 + m2 ) = f (m1 ) + f (m2 ) for all m1 , m2 ∈ M, and r · f (m) = f (r · m)
for all r ∈ R and m ∈ M.
In a similar fashion, right module maps and bimodule maps between right
modules and bimodules, respectively are defined.

We have now reached a point in our discussion where it is convenient to introduce

a notation for the collection of all modules over a given ring. In the second part of
this chapter, we shall, however, realise that this is much more than a mere shorthand.

Notation In order to save writing, we shall denote the collection of all left R-
modules over a fixed ring R by R-Mod. Analogous meaning is given to the symbols
Mod-R and R-Mod-S (where S is another ring). If R (and S) is unital, this
notation will automatically refer to the collection of all unital modules.

Since the module maps are of paramount importance in the theory, we also give the
sets comprising all of them special symbols.
1.1 Basics on Modules 5

Notation Let R M, R N ∈ R-Mod. Then

HomR (R M, R N) = {f : R M → R N | f is a left R-module map}.

Analogous meanings are given to HomR (MR , NR ) and HomR-S (R MS , R NS ).

Rather than delving further into the general theory of modules, let us now pause
to ask the question “Why study modules at all?” The following instances of the
appearance of modules shall provide us with motivation to develop a comprehensive
theory as well as with some first examples.

1.1.1 Reducing Complicated Rings to Simpler Ones

Suppose R is a commutative unital ring. In the case when R is a field, every

multiplicative equation ax = b, a = 0 can be solved uniquely. If R is merely an
integral domain, at least every equation of the form ax = 0, a = 0 has the unique
solution x = 0. This makes computations much easier: “Non-zero common factors
can be cancelled.” Unfortunately, the property of being an integral domain is all but
too easily destroyed (Example 1.2 (ii) above). Is there hope that this trivial way of
taking direct products might be the only method to destroy this good property of a
ring? For certain classes of rings this is indeed (almost) true. A natural assumption,
however, must be that there are no non-trivial nilpotent elements (which means that,
if x n = 0 for some n ∈ N, then x = 0). Rings satisfying this property are called
reduced.
The following is a well-known result in commutative Ring Theory; compare also
[24, Sect. 12].

Theorem Every commutative unital reduced ring is a subdirect product of integral

domains.

Let us indicate the argument for the sake of illustration. If R is such a ring
 then
there is an injective ring homomorphism ϕ from R into the direct product α Rα
of integral domains Rα (with a possibly large index set of α’s) such that Rα =
πα ◦ ϕ(R) for each α, where πα is the canonical projection onto Rα . The key to
this result is that, for every x ∈ R \ {0}, one can find a proper prime ideal Px of R
with the property x ∈ / Px . The integral domain Rα is then defined by R/Px (so that
α = x and the index set is all of R !). See also Exercise 1.3.11.
The above method of taking appropriate homomorphic images and then embed
the ring into a product of those is already very much in the spirit of module theory:
we will see that quotient rings have a canonical module structure right away.
6 1 Motivation from Ring Theory

1.1.2 One-Sided Ideals in Noncommutative Rings

Suppose R is a noncommutative ring. One-sided ideals play an important role in this

setting as R easily may have no non-trivial two-sided ideals (see Example 1.2 (vi)
above). Say, L is a left ideal of R. Unfortunately the quotient group R/L does not
carry a well-defined multiplication so it is not a ring. Nevertheless we can define a
left R-action on R/L via

r · (a + L) = ra + L (r, a ∈ R).

In this way, R/L turns into a left R-module which is unital if R is unital. Note also
that L ∈ R-Mod too!
We shall see that these two types of modules are very popular.

1.1.3 Images of Ideals Under Homomorphisms

Suppose R and S are rings, and ρ : R → S is a ring homomorphism. Let I be an

ideal in R. (It does not matter here whether it is one-sided or two-sided so suppose
that I is at least a left ideal.) In general, ρ(I ) will not be an ideal of S. However,
if we consider R as a left module R R in the obvious way (the left action is just
the ordinary multiplication), which is called the standard R-module, and we regard
S as a left R-module via r · s = ρ(r)s, r ∈ R, s ∈ S, then ρ(I ) becomes a left
R-submodule of R S. In fact, ρ(I ) is the image under the associated left R-module
map ρ.
We learn from this observation that modules allow for a much larger flexibility.

1.1.4 Embedding into the Endomorphism Ring and General

Representations

Every unital ring can be embedded into an endomorphism ring; thus, the latter may
be considered as “the mother of all rings”. Using the language of modules, this
process, reviewed below, becomes very natural.
Let R be a unital ring. We now formalise the construction of the standard R-
module R R indicated in the previous section. The abelian group (R, +) becomes a
unital left R-module in a natural way via

r ∈ R, s ∈ R : r · s = rs.

The endomorphism ring End((R, +)) contains R as a unital subring in a canonical

way:

λ : R → End((R, +)), r → λr ,
1.1 Basics on Modules 7

where λr (s) = rs, s ∈ R; this is the left regular representation of R. It is easy to

check that λ is an injective unital ring homomorphism.
This representation of R corresponds precisely to the unital left R-module R R.
More generally, a representation of R is a pair (π, G) consisting of an abelian group
G and a unital ring homomorphism π : R → End(G). In this situation, we can turn
G into a unital left R-module by defining

r ∈ R, g ∈ G : r · g = π(r)(g).

Conversely, suppose that R M ∈ R-Mod; then

πM : R → End(M), πM (r)(m) = r · m (r ∈ R, m ∈ M)

defines a representation of R on G. It is easily verified that these two procedures are

inverses to each other. Thus we have our first result.

Proposition 1.7 Let R be a unital ring. There is a one-to-one correspondence

between unital left R-modules and representations of R.

1.1.5 Group Representations

Let G be a group, written multiplicatively. Let K be a field and n ∈ N a natural

number. By GLn (K) we denote the group of invertible n × n matrices over K.
Suppose d : G → GLn (K) is a representation of G, that is, a group homomorphism
into GLn (K). Put R = K[G], the group ring of G (see Example 1.2 (ix) above). We
can endow the abelian group K n with a module structure as follows
  
ag g · v = ag d(g)(v), v ∈ K n.
g∈G g∈G

In this way, K n ∈ K[G]-Mod.

Conversely, suppose V ∈ K[G]-Mod is given. Since K ⊆ K[G] as a unital
subring via a → ae, V is a left K-module too, i.e., a K-vector space. Suppose
that dimK V = n ∈ N. As explained above in Sect. 1.1.4, this amounts to a
representation ρ : K[G] → End(V ) such that g · v = ρ(g)(v) for g ∈ G, v ∈ V . By
choosing a basis of V , we can identify End(V ) with Mn (K) and obtain in this way
a representation d : G → GLn (K), d(g) = ρ(g). A different choice of basis results
in a representation d : G → GLn (K) of G which is equivalent to d in the sense
that d (g) = S −1 d(g)S, g ∈ G, where S : V → V is the invertible transformation
implementing the change of basis.
On the other hand, one can show that any pair of equivalent representations of G
gives rise to isomorphic left K[G]-modules.
8 1 Motivation from Ring Theory

The above discussion shows that modules appear in many guises; therefore it
should be fruitful to have a well-developed theory at hand. To complete this section,
we record some (further) basic examples of modules.

Examples 1.8

(a) Every abelian group G becomes a Z-module in a natural way: for n ∈ N, g ∈ G

we define n · g = g + . . . + g, the sum of n copies of g. Furthermore, 0 · g = 0G
and −n · g = −(n · g). It is easy to verify the axioms of a unital Z-module.
Therefore, the theory of modules is a generalisation of the theory of abelian
groups.
(b) Let V be a vector space over the field K. Then V is a K-module via:

r ∈ K, v ∈ V : r · v = rv (the usual scalar multiplication).

Therefore, the theory of modules is also a generalisation of the theory of vector

spaces.
(c) Let V be a K-vector space with dimK V = n < ∞. Choose a basis for V and
put R = Mn (K). Then V ∈ R-Mod via:

a ∈ Mn (K), v ∈ V : a · v is the usual matrix-vector multiplication.

(d) Let V be a K-vector space, and let T : V → V be a linear mapping. Put

R = K[x]. For p(x) ∈ R, the linear mapping p(T ) is well defined. Hence,
V becomes a left R-module via

p(x) ∈ R, v ∈ V p(x) · v = p(T )(v).

To summarise this first section, we have encountered a number of instances where

modules over rings appear in a rather natural fashion. We will now take a look at
what we have achieved from a bird’s eye view.

1.2 Categories of Modules

It is now time to start introducing the language of categories. Category Theory itself
is nowadays a vast and highly developed area in Pure Mathematics with manifold
applications in other fields. We will only be able to touch upon the most basic
concepts and features; nevertheless this shall allow us a new view on the interplay
between rings and modules.
S. Eilenberg and S. MacLane established the foundations in 1945 in their article
“General theory of natural equivalences”, see [15]. Following their conviction that
“It should be observed first that the whole concept of a category is essentially an auxiliary
one; our basic concepts are essentially those of a functor and of a natural transformation.”
1.2 Categories of Modules 9

we will very soon include functors (as well as natural transformations) in our
discussion; see Sect. 3.2.
In the following, we will have to deal with “very large collections of sets” which
will be called classes. The reader familiar with Russell’s Paradox will know that
we cannot perform arbitrary operations with sets without creating set-theoretic
problems. The solution to this situation is a careful distinction between actual
sets and more flexible classes; this however needs an axiomatic approach to Set
Theory which would go far beyond the scope of this book. Theories which extend
the Zermelo–Fraenkel set theory together with the Axiom of Choice and allow a
rigorous treatment are the von Neumann–Bernays–Gödel axioms or the Morse–
Kelley set theory. An alternative is to work within Grothendieck universes. The
interested reader will find answers to their questions in [30], for example. The main
difference between sets and classes is that any class that is an element of another
class is a set.

Definition 1.9 A category C is a triple consisting of an object class ob(C); a

morphism class mor(C); and a law of composition ◦ ; that is,

(i) ob(C) is a class, its elements A, B, C, . . . are called the objects of C;

(ii) mor(C) is a class, its elements are pairwise disjoint sets Mor(A, B), for A, B ∈
ob(C); each element in Mor(A, B) is called a morphism or an arrow from A
to B; typically we write these as f, g, h, . . .;
(iii) ◦ yields a family of mappings

Mor(A, B) × Mor(B, C) → Mor(A, C), (f, g) → g ◦ f,

where A, B, C ∈ ob(C), such that the following two axioms hold

(a) (associativity)

∀ f ∈ Mor(A, B), g ∈ Mor(B, C), h ∈ Mor(C, D) :

h ◦ (g ◦ f ) = (h ◦ g) ◦ f ;

(b) (identities)

∀ A ∈ ob(C) ∃1 1A ∈ Mor(A, A) such that

∀ B ∈ ob(C) ∀ f ∈ Mor(A, B), g ∈ Mor(B, A) :
f = f ◦ 1A and 1A ◦ g = g.

1A is called the identity morphism of A.

The category C is called concrete if the elements of ob(C) have underlying sets and
the morphisms of mor(C) are mappings between these underlying sets. (A more
10 1 Motivation from Ring Theory

precise definition will be given in 3.2.2.) The category C is called small if ob(C) is
a set.

Notation Suppose f ∈ Mor(A, B) for two objects A, B in a category C. We say A

is the domain of f , and write A = dom(f ), and we say B is the codomain of f , and
write B = codom(f ) in this case. If there is a need to specify the category certain
morphisms belong to, we emphasise this by writing MorC (A, B), A, B ∈ C.
In order to simplify the notation, we shall sometimes write A ∈ C instead of
A ∈ ob(C) and f ∈ C instead of f ∈ Mor(A, B) ∈ mor(C) though f ∈ Mor(A, B)
will be the most common.

Let us have a look at some of the categories the reader may have encountered.

1.10 Examples of Categories

S the category of all sets where the objects are sets and the morphisms are
mappings between sets.
Top the category of all topological spaces; the objects are topological spaces
and the morphisms are continuous mappings between them.
Comp the category of compact Hausdorff spaces; the morphisms are again the
continuous mappings.
Gr the category of groups with group homomorphisms as morphisms.
AGr the category of abelian groups with group homomorphisms as mor-
phisms.
Ring the category of rings with ring homomorphisms as morphisms.
Ring1 the category of unital rings with unital ring homomorphisms as mor-
phisms.
R-Mod the category of (unital) left R-modules over a given (unital) ring R where
the morphisms are the left R-module maps.
Ban∞ the category of complex Banach spaces with bounded linear operators
as the morphisms.
Ban1 the category of complex Banach spaces with linear contractions as the
morphisms.
Lat the category of lattices and lattice homomorphisms as the morphisms.
HTop the category of all topological spaces and homotopy classes of continu-
ous mappings as the morphisms.

All of the above categories but for the last one are concrete categories; the objects
have an underlying set structure and the morphisms arise from mappings between
these sets. Moreover, the composition law is provided by the composition of
mappings. In HTop, this is not the case.
The reader will have no difficulties in defining the categories Mod-R and
R-Mod-S for themselves.
Although many of the above categories look alike on the surface, they have
rather different properties, as we shall see when moving on in the later chap-
1.2 Categories of Modules 11

ters. For example, for two objects R M, R N ∈ R-Mod the set of morphisms
Mor(R M, R N ) = HomR (R M, R N ) has a group structure (see Exercise 1.3.2)
whereas this is not the case for X, Y ∈ ob(Top): there is no canonical addition
for continuous mappings between X and Y .

Definition 1.11 A subcategory D of a category C consists of a subclass ob(D) ⊆

ob(C), a subclass mor(D) ⊆ mor(C) in the sense that Mor(A, B) ∈ mor(D) if
A, B ∈ ob(D), and ◦ is simply the restriction of the composition of morphisms
in C to D.
The subcategory D is said to be full if, for all A, B ∈ ob(D), every morphism
A → B in C is also a morphism in D, that is, MorD (A, B) = MorC (A, B).

For instance, Comp is a full subcategory of Top and AGr is a full subcategory
of Gr. On the other hand, Ring1 is a non-full subcategory of Ring and Ban1 is a
non-full subcategory of Ban∞ . Note that Ring is not even a subcategory of AGr
as an abelian group can carry several non-isomorphic rings structures.
Let R-mod denote the category whose objects are the finitely generated left
modules over a fixed ring R (i.e., those modules that have a finite set of generators,
see Definition 2.3.1) together with the left R-module maps. Then R-mod is a full
subcategory of R-Mod.

Remark 1.12 Even if the objects in a category have an underlying set (and are
specified by some additional structure on these sets), the morphisms do not have
to be mappings on the underlying sets. An example is HTop. Another one arises
from the open subsets of a given topological space X: the category OX has as objects
all open subsets of X and, for U, V ∈ ob(OX ), Mor(U, V ) = {→} if U ⊆ V and
Mor(U, V ) = ∅ otherwise. The composition law is defined through the transitivity
of set inclusion.

Let us have a look at some small categories.

1.13 Examples of Small Categories

(i) Let X be a non-empty set, and let ∼ be an equivalence relation on X. Define

a category EqX by taking as the objects in EqX the elements of X and for
x, y ∈ X set

{(x, y)} if x ∼ y
Mor(x, y) =
∅ otherwise.

The transitivity of ∼ yields the composition law ◦.

(ii) Let G be a group. We can consider G as a category whose single object is G
itself and whose morphism class (which is a set in this case) is G too, that is,
Mor(G, G) = G in the following sense: we identify g ∈ G with the morphism
h → gh, and the composition law is provided by the group operation.
12 1 Motivation from Ring Theory

Some of the morphisms in a category are of special importance; we shall

introduce these now.

Definition 1.14 Let C be a category. Let A, B ∈ ob(C). A morphism f ∈

Mor(A, B) is said to be a monomorphism if, for each C ∈ ob(C) and all g, h ∈
Mor(C, A), fg = f h implies g = h. The morphism f is said to be an epimorphism
if, for each C ∈ ob(C) and all g, h ∈ Mor(B, C), gf = hf implies g = h. (We
wrote gf for g ◦ f , etc. as is customary.) The morphism f is called an isomorphism
if there exists g ∈ Mor(B, A) such that gf = 1A and fg = 1B . (In this case, g is
unique.) The objects A and B are called isomorphic if there exists a isomorphism
between them.

We conclude this section with a basic but very useful construction.

Example 1.15 Let C be a category. The dual category C op is defined by ob(C op ) =

ob(C) and for A, B ∈ ob(C op ) we put MorCop (A, B) = MorC (B, A), that is,
we “reverse” all arrows and the composition in C op is the same as in C. This
dualisation procedure has many nice features; e.g., a monomorphism in C becomes
an epimorphism in the dual category.

1.3 Exercises

Exercise 1.3.1 Let R M and R N be in R-Mod. For a left R-module map

f : R M → R N , we define its kernel by

ker(f ) = {m ∈ R M | f (m) = 0}

and its image by

im(f ) = {f (m) | m ∈ R M}.

Show that ker(f ) is a left R-submodule of R M and im(f ) is a left R-submodule

of R N.

Exercise 1.3.2 For two module maps f and g defined on a module M with values
in a module N over the ring R define their sum by (f + g)(m) = f (m) + g(m)
for all m ∈ M. Show that HomR (R M, R N ) and HomR-S (R MS , R NS ) (where both
M and N are R-S-bimodules and S is another ring) are abelian groups under this
addition.

Exercise 1.3.3 Let ρ : R → S be a ring homomorphism between the two rings R

and S. Let S M ∈ S-Mod. Show that M becomes a left R-module under the action
r · m = ρ(r) · m, r ∈ R and m ∈ M.
1.3 Exercises 13

Exercise 1.3.4 Let R be a ring without identity. Show that we can embed R
into a unital ring in the following way: equip the abelian group R × Z with the
multiplication

(r, n) · (s, m) = (rs + ns + mr, nm) (r, s ∈ R, n, m ∈ Z).

Verify that we indeed get a ring multiplication in this way, and that (0, 1) serves
as an identity. The ring thus obtained is denoted by R × and is called the minimal
unitisation of R or the Dorroh superring of R.
Show further that the monomorphism r → (r, 0) allows us to view R as an ideal
in R × . (However, if performed for a ring which already has an identity, this process
will destroy the original identity!)

Exercise 1.3.5 Suppose R is a unital ring and that R M is a left R-module. Then

R M0 = {m ∈ M | r · m = 0 ∀ r ∈ R} and R M1 = {m ∈ M | 1 · m = m}

are both submodules of R M. Moreover, R M = R M0 ⊕ R M1 . Prove these statements!

The submodule R M1 is unital and the submodule R M0 carries only the trivial
R-action; so all information on R M is actually contained in R M1 .

Exercise 1.3.6 Let G be a group with non-trivial torsion; that is, there is g ∈ G
such that g n = e for some n ∈ N and g = e, where e is the neutral element in G.
Show that the group ring R[G] contains a non-trivial divisor of zero.

Exercise 1.3.7 Let H be a subgroup of a group

 G. Show that the mapping
τH : R[G] → R[H ] defined by τH g∈G rg g = g∈H rg g is a R[H ]-module
map.(Use the support of a ∈ R[G] defined by supp(a) = {g ∈ G | rg = 0}, where
a = g∈G rg g.)

Exercise 1.3.8 Let H be a subgroup of a group G. Let a ∈ R[H ]. Use the previous
exercise to show that, if a is invertible in R[G], then it is invertible in R[H ], and
that, if a is a zero divisor in R[G], then it is a zero divisor in R[H ].

Exercise
 1.3.9 Let
 G be a group and R be a unital ring. Define the trace on R[G]
by tr g∈G rg g = r1 , the coefficient at the identity 1 ∈ G. Show that tr is a left
R-module map and that tr(ab) = tr(ba) for all a, b ∈ R[G].

Exercise 1.3.10 Let R be a commutative unital ring, and denote by nil(R) the set
of all nilpotent elements in R. Show that nil(R) is an ideal of R; it is called the
nil radical of R. Show further that the quotient ring R/nil(R) does not contain any
non-zero nilpotent element.

Exercise 1.3.11 Let R be a unital reduced ring. A proper ideal P of R is said to be

prime if, whenever I1 , I2 are ideals of R such that I1 I2 ⊆ P , we have I1 ⊆ P or
14 1 Motivation from Ring Theory

I2 ⊆ P . Using Zorn’s Lemma, show that the intersection of all prime ideals of R is
zero.

Exercise 1.3.12 Let (π, G) be a representation of the unital ring R (that is, π : R →
End(G) is a unital ring homomorphism from R into the endomorphism ring of the
abelian group G). Show that the action of R on G defined by r ·g = π(r)(g), r ∈ R,
g ∈ G turns G into a left R-module.

Exercise 1.3.13 Let C be a concrete category. Show that a morphism in C which

is an injective map is a monomorphism and a morphism which is a surjective map
is an epimorphism. Does this imply that a bijective morphism is an isomorphism
in C?

Exercise 1.3.14 Let R be a unital ring. Show that every monomorphism in R-Mod
is injective and every epimorphism in R-Mod is surjective. Does a similar result
hold in the full subcategory R-mod of R-Mod?

Exercise 1.3.15 Show that the canonical inclusion Z → Q is an epimorphism but

not surjective.

Exercise 1.3.16 Let E and F be Banach spaces and let f : E → F be a morphism

from E to F in Ban1 . Show that f is a monomorphism if and only if f −1 (0) = 0
and that f is an epimorphism if and only if f (E) is dense in F . Are the same
statements true for f ∈ mor(Ban∞ )?
Constructions with Modules
2

This chapter is devoted to various basic constructions that we can perform within
R-Mod. Once we have found some interesting modules, we want to be able to
obtain new ones from them. Special classes of modules will play a distinguished
role here. In the second part of this chapter we shall review these constructions from
the viewpoint of category theory and thus realise that seemingly similar categories
can behave rather differently.

2.1 Some Special Morphisms

Let R be a ring, and let R M, R N ∈ R-Mod. We say that f ∈ HomR (R M, R N) is

a monomorphism if f is injective;
an epimorphism if f is surjective;
an isomorphism if f is bijective.

The two modules R M and R N are called isomorphic if there is an R-isomorphism

between them. Obvious modifications yield analogous notions for other types of
modules.
It follows from Exercises 1.3.13 and 1.3.14 that the above definition is consistent
with the one given in Definition 1.14.

Notation Let R be a ring, and let M be an R-module. Then

EndR (R M) = HomR (R M, R M) and EndR (MR ) = HomR (MR , MR ).

If S is another ring, then EndR-S (R MS ) = HomR-S (R MS , R MS ).

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 15

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_2
16 2 Constructions with Modules

In the last chapter, we already met submodules and homomorphic images of

modules. The connection between the two is provided by the following fundamental
construction.

2.2 Quotient Modules

Let R be a ring, and let R N , R M ∈ R-Mod with R N ≤ R M. On the (abelian)

quotient group M/N we define an R-action by

r ∈ R, m ∈ M : r · (m + N) = r · m + N.

This is well defined as, if m + N = n + N, then m − n ∈ N and thus r · (m − n) ∈ N.

It follows that

r · n + N = r · m − r · (m − n) + N = r · n + N.

In this way we obtain the quotient module R M/R N ; if R is unital, this construction
gives the correct object in R-Mod. We also have the canonical epimorphism

πN : R M → R M/R N , πN (m) = m + N.

Clearly, ker(πN ) = R N.

Proposition 2.2.1 Let R be a ring, and let R N, R M ∈ R-Mod with R N ≤ R M.

Then there is a one-to-one correspondence between the submodules of R M/R N and
the submodules of R M containing R N .

Proof It is easily checked that the inverse image of a submodule under a module
map is a submodule of the domain. Therefore, for a submodule R L ≤ R M/R N ,
−1
R K = πN (R L) = {m ∈ M | πN (m) ∈ L} is a submodule of R M containing R N .
Conversely, let R K ≤ R M be such that R N ⊆ R K. Then πN (R K) is a submodule
of R M/R N (compare Exercise
 1.3.1).
  
Evidently, R K ⊆ πN−1 πN (R K) . If m ∈ πN−1 πN (R K) then πN (m) ∈ πN (R K)
and so m differs from an element of R K by anelement in R N ⊆ R K. Consequently,
m ∈ R K and we obtain R K = πN−1 πN (R K) . On the other hand, we always have
 
πN πN−1 (R L) ⊆ R L for any R L ≤ R M/R N, and since πN is surjective, we get the
equality of the two sets. 


Definition 2.2.2 Let R be a ring, and let R M ∈ R-Mod. For a non-empty subset
S ⊆ R M we let

AnnR (S) = {r ∈ R | r · m = 0 for all m ∈ S}

be the annihilator of S in R.
2.4 Direct Sums and Products of Modules 17

The module R M is said to be faithful if AnnR (R M) = {0}.

Clearly, AnnR (S) is always a left ideal of R and AnnR (R M) is a two-sided ideal
of R. It goes without saying that all of the above can be defined analogously for
right modules and bimodules.

2.3 Generating Modules

In this section, we shall take a look at various ways to generate submodules from
subsets of a given module.
Let R M ∈ R-Mod for some ring R. Given a family {R M i | i ∈ I } of
submodules R M i ≤ R M, their intersection is easily seen to be a submodule
of R M. Suppose now that S ⊆ R M is a non-empty subset. The intersection of
all submodules of R M that contain S—which is evidently the smallest submodule
of R M containing S—is called the submodule generated by S and will be denoted
by R S.
Suppose that R is unital. For m ∈ M, R {m} = {r · m | r ∈ R} = Rm and for
any ∅ = S ⊆ R M, we have
 
R S = Rs = rj · sj | rj ∈ R, sj ∈ S, J finite .
s∈S j ∈J

Definition 2.3.1 Let R M ∈ R-Mod for a unital ring R. We say

(i) RM is cyclic if M = Rm for some m ∈ M;

(ii) R M is finitely generated if R M = Rm1 + . . . + Rmk for a finite subset
{m1 , . . . mk } ⊆ M.

The elements m and m1 , . . . , mk above are called generators.

2.4 Direct Sums and Products of Modules

Let {R M i | i ∈ I } be a family of left R-modules. On the cartesian product X R M i

i∈I
we define module operations as follows

(mi ) + (ni ) = (mi + ni ),

r · (mi ) = (r · mi )

for all (mi ), (ni ) ∈ X and r ∈ R. It is easy to check that we indeed obtain
RMi
i∈I 
a left R-module which is denoted by R M i and called the direct product of the
family {R M i | i ∈ I }. i∈I
18 2 Constructions with Modules

The submodule

R M i = (mi ) ∈ RMi | at most finitely many mi are non-zero
i∈I i∈I

is called the direct sum of the family {R M i | i ∈ I }.

There are some canonical module maps associated with these two modules:

pj : RMi → R M j , (mi ) → mj for some j ∈ I

i∈I

is called the canonical projection onto the j -th component; it is an R-epimorphism.

ιj : R M j → RMi, m → (mi ) where mj = m and mi = 0 for i = j

i∈I

is called the canonical injection from the j -th component; it is an R-monomorphism.

The two constructions above are characterised by some universal properties.

Proposition 2.4.1 Let {R M i | i ∈ I } be a family of left R-modules.

(i) For every R N ∈ R-Mod and every  family {fi | i ∈ I } ⊆ HomR (R N, R M i )

there is a unique f ∈ HomR (R N, i R M i ) such that pj f = fj , j ∈ I .

i R Mi
f

pj
RN

fj

R Mj

(ii) For every R N ∈ R-Mod and every family {fi | i ∈ I } ⊆ HomR (R M i , R N )

there is a unique f ∈ HomR ( i R M i , R N ) such that f ιj = fj , j ∈ I .

i RMi
f

ιj
RN

fj

R Mj
2.5 Free Modules 19

Proof

(i) For n ∈ N, we set f (n)i = fi (n) for each i ∈ I . Then pi f = fi and it is easy
to verify that f is a left R-module map.
(ii) For m ∈ i R M i , there are only finitely many i’s with mi = 0. Hence we can
define f (m) = i fi (mi ) and obtain a left R-module map satisfying f ιj = fj
for all j ∈ I .



2.5 Free Modules

This section deals with those modules that resemble vector spaces most closely. In
a way, they turn out to be the most general modules.
Let X be a non-empty set, and let R be a unital ring. We consider the set

{f ∈ R X | f (x) = 0 for at most finitely many x ∈ X}


and write its elements as (finite sums) x∈X rx x. We can regard X as a subset via
x → 1 · x.
An R-module structure can be defined by
  
rx x + sx x = (rx + sx )x,
x x x
 
r· rx x = rrx x (r ∈ R).
x x

In this way, we obtain a left R-module called the free left R-module on X which is
denoted by RX (sometimes also denoted as R (X) ). Note that, if we put Mx = R R
for each x ∈ X, then RX is nothing but Mx .
x∈X
Clearly, analogous constructions give us free right modules and free bimodules,
respectively.
Free modules are distinguished in several ways.

Definition 2.5.1 Let R M ∈ R-Mod. A subset S ⊆ M is called R-linearly inde-

pendent (or simply, linearly independent)
 if, for every finite subset {m1 , . . . , mk } ⊆
S and elements r1 , . . . , rk ∈ R, ki=1 ri · mi = 0 implies that ri = 0 for all
1 ≤ i ≤ k. The set S is a basis of R M (or, more precisely, an R-basis) if it is
linearly independent and R S = R M.

Considering the elements of X inside the free module RX, we see that X is a
basis of it.
20 2 Constructions with Modules

Proposition 2.5.2 Let R be a unital ring. Let R M ∈ R-Mod with basis S. Then
R M is isomorphic to RS as left R-modules.

Proof It is easy to check that the mapping S → RS, s → 1 · s extends to an

isomorphism between R M and RS. 


Remark Contrary to the situation of vector spaces over fields, the notion of the
‘length of a basis’ (and hence, the notion of ‘dimension’) is not well defined for free
modules. For example, let V = Q[x] and R = EndQ (Q V ). It can be shown that
∼ n ∼
R R = R R ⊕ R R and hence R R = R R for all n, m ∈ N. If R is commutative,
m

such a phenomenon cannot occur; see [8, Theorem 7.12].

Next we state the universal property of free modules.

Theorem 2.5.3 Let R be a unital ring, and let R M ∈ R-Mod. Let X be

a non-empty set. If η : X → M is a mapping then there is a unique f ∈
HomR (RX, R M) such that f (x) = η(x) for all x ∈ X.
  
Proof We set f x rx x = x rx η(x), x ∈ X. Evidently, f is a left R-module
map extending the mapping X → RX, x → 1 · x. Therefore f (x) = η(x) for
every x ∈ X. It is clear that f is uniquely determined by these properties. 


f
R X RM

Corollary 2.5.4 Every unital left module is a quotient of a free left module.

Proof Let R M ∈ R-Mod for a unital ring R. Let η : M → M be the identity. By

Theorem 2.5.3, there is a left R-module map f : RM → R M with f (m) = m
for every m ∈ M. Thus f is surjective. Let R N = ker(f ) which is a submodule
 of
RM and define fˆ : RM/R N → R M by fˆ(a + R N ) = f (a), a = m rm m ∈
RM. Then fˆ is an epimorphism and, since f (a) = 0 precisely when a ∈ R N, it
follows that fˆ is injective too. As a result, R M is isomorphic to a quotient of a free
module. 


We shall see in the next chapter that the technique used in the proof of the above
corollary allows us to identify any homomorphic image of a module as a quotient
module.
2.6 Special Objects in a Category 21

2.6 Special Objects in a Category

We now return to the viewpoint of categories. Motivated by the various construc-

tions with modules that we performed in the previous sections of this chapter,
we want to investigate when similar objects in a category might be available.
Constructions such as the direct product, for example, can be done in many of the
categories listed in Sect. 1.2; however, they often depend on manipulations with the
underlying sets. In a general category, we thus have to identify properties of objects
that can be solely expressed in terms of morphisms—this is where the “universal
properties” come in—and the perspective turns from construction to the one of
existence.
Our first special objects continue the line of thought of the very last section.

2.6.1 Free Objects

Let C be a concrete category. Given a non-empty set X, an object F ∈ ob(C) and

an injective mapping X → F we say F is free on X if, for every object A ∈ ob(C)
and every mapping η : X → A, there is a unique f ∈ MorC (F, A) such that η(x) =
f (x) for all x ∈ X; that is, the diagram below is commutative.

f
F A

Note that we do not use a special symbol for the underlying set of an object in C.
For example, for every unital ring R, if X is a non-empty set then RX is a free
object in R-Mod, by Theorem 2.5.3. On the other hand, the only free object in
Z-mod (the finitely generated Z-modules) is the zero module.

2.6.2 Products and Coproducts

Products and direct sums (called “coproducts”) are defined in a general category C
just as in R-Mod.
Let {Ai | i ∈ I } be a family of objects in a category C. A product of the family
{Ai | i ∈ I } consists of an object A ∈ ob(C) together with a family {pi | i ∈ I }
of morphisms in Mor(A, Ai ), i ∈ I , called projections, such that the following
diagram can be made commutative by a unique f ∈ Mor(C, A) for each C ∈ ob(C)
22 2 Constructions with Modules

and fi ∈ Mor(C, Ai ), i ∈ I :

A
f

C pi

fi
Ai

If such product exists, it is unique up to isomorphism and is denoted by i∈I Ai .
Dually, a coproduct of {Ai | i ∈ I } consists of an object A ∈ ob(C) together
with a family {ei | i ∈ I } of morphisms in Mor(Ai , A), i ∈ I , called injections, such
that the following diagram can be made commutative by a unique g ∈ Mor(A, B)
for each B ∈ ob(C), gi ∈ Mor(Ai , B), i ∈ I :

A
g

ei B

gi
Ai

If such coproduct exists, it is unique up to isomorphism and is denoted by i∈I Ai .

These two notions are defined in such a way that the following relations hold:
 
Mor(C, Ai ) ∼
= Mor C, Ai
i∈I i∈I
  (2.6.1)
Mor(Ai , B) ∼
= Mor Ai , B .
i∈I i∈I

Examples

(i) In the category S, the cartesian product Xi∈I Ai of a family {Ai | i ∈ I } of

sets together with the projections onto the coordinates
 forms the product. The
coproduct in S is the disjoint union i∈I Ai = i∈I (Ai × {i}) together with
the injections ei : Ai → i∈I Ai , a → (a, i), a ∈ Ai .
(ii) In R-Mod, the product is the direct product of a family of modules and the
coproduct is the direct sum, together with the canonical projections and the
canonical injections, cf. Proposition 2.4.1. In R-mod however, there exist only
finite products and coproducts (the index set of the family has to be finite).
2.7 Exercises 23

(iii) In Ban1 , the direct product of a family {Ei | i ∈ I } of complex Banach

spaces is the Banach space of all bounded functions (xi )i∈I from I into Xi∈I Ei
endowed with the sup-norm; the projections are the canonical projections
onto the coordinates. Often it is denoted by ∞ ((Ei )). The coproduct of
{Ei | i ∈I } is the linear subspace of Xi∈I Ei consisting of those (xi )i∈I
satisfying i∈I xi  < ∞ together with the componentwise injections. The
norm of (xi )i∈I is then defined by (xi ) = i∈I xi , and for that reason the
coproduct is typically denoted by 1 ((Ei )).

Two special situations for product and coproduct need to be singled out. Suppose
I = ∅. The “empty” product, if it exists, is the unique object A ∈ ob(C) with the
property that for each C ∈ ob(C) there is a unique morphism C → A; that is,
MorC (C, A) has precisely one element. In this case, A is called a final object. The
“empty” coproduct is the unique object A ∈ ob(C), if it exists, satisfying: for each
C ∈ ob(C) there is a unique morphism A → C; that is, MorC (A, C) has precisely
one element. In this case, A is called an initial object.
For instance in S, ∅ is the initial object and {∅} is the final object. In R-Mod,
the zero module 0 serves as both the initial and the final object. In the category of
fields, however, there are no initial or final objects (because a field homomorphism
preserves the characteristic of a field).
An object in a category that is both initial and final is called a zero object As such
an object is unique (cf. Exercise 2.7.12) it is often denoted as 0.
Some categories such as R-Mod or Ban∞ have an additional structure on their
morphism sets: they are abelian groups. In generalisation of this, one calls a category
C additive if it has zero, finite coproducts and all morphism sets Mor(A, B),
A, B ∈ ob(C) carry the structure of an abelian group such that the composition
of morphisms is bilinear. In this case, finite products exist as well and agree with the
finite coproducts.

2.7 Exercises

Exercise 2.7.1 Let R be a unital ring with centre Z(R). Show that the mapping
a → La , where, for a ∈ R, La : R → R is given by La (x) = ax, x ∈ R provides
isomorphisms between the rings R and EndR (RR ), respectively, and between Z(R)
and EndR (R RR ), respectively.

Exercise 2.7.2 Let R be a ring, and let R M ∈ R-Mod. Show that the annihilator
of R M in R,

AnnR (R M) = {r ∈ R | r · m = 0 for all m ∈ R M}

24 2 Constructions with Modules

is an ideal in R. Consider M as a left module over the quotient ring S =

R/AnnR (R M) in the canonical way. Show that S M is a faithful left S-module.

Exercise 2.7.3 Let R be a ring and R M ∈ R-Mod. If J is an ideal of R contained

in AnnR (R M) then M carries a natural structure as a left R/J -module. Show that a
subgroup of M is an submodule of R M if and only if it is a submodule of R/J M.

Exercise 2.7.4 Let R be a ring, and let R N, R M ∈ R-Mod with R N ≤ R M. Show

that, if both R N and R M/R N are finitely generated, then R M is finitely generated.

Exercise 2.7.5 Show that the direct product and the direct sum of a family of left
R-modules are uniquely determined up to isomorphism by the universal properties
stated in Proposition 2.4.1.

Exercise 2.7.6 Let R be a unital ring, and let R M ∈ R-Mod. Show that if R M is
finitely generated and free then it is isomorphic to R R n for some n ∈ N.

Exercise 2.7.7 Let R be a unital ring, and let R M, R N ∈ R-Mod. Suppose that
R N is free and let f : R M → R N be an R-epimorphism. Show that there exists
g ∈ HomR (R N , R M) such that f ◦ g = idN and that R M = ker(f ) ⊕ im(g) (that
is, R M = ker(f ) + im(g) and ker(f ) ∩ im(g) = 0).

Exercise 2.7.8 Prove the modular law of module theory:

R L ∩ (R K + R N) = R K + (R L ∩ R N)

whenever R K, R L, R N are submodules of R M ∈ R-Mod with R K ⊆ RL for

some ring R.

Exercise 2.7.9 Let R M ∈ R-Mod for a unital ring R. Show the following two
statements.

M, we have Rm ∼
(i) For each m ∈ R = R/AnnR (m).
(ii) AnnR (R M) = f ∈HomR (R,M) ker f .

Exercise 2.7.10 Prove the identities in Eq. (2.6.1).

Exercise 2.7.11 Show that, if they exist, (arbitrary, non-empty) coproducts in a

category are unique up to isomorphism.

Exercise 2.7.12 Let C be a category. Show that:

(i) All initial objects in C are isomorphic.

(ii) All final objects in C are isomorphic.
(iii) Hence a zero object is unique up to isomorphism, and we speak of the zero
object.
The Isomorphism Theorems
3

Just as in group theory and in ring theory, we need to be able to manipulate modules
freely under isomorphisms; this is done via the so-called isomorphism theorems
which are discussed in this short chapter. The second part of this chapter deals with
connections between categories, the functors, and with connections between the
functors, the natural transformations.

3.1 Isomorphisms Between Modules

The basis for all the canonical isomorphism theorems is the following result.

3.1.1 Canonical Factorisation of Module Homomorphisms Let R M, R N ∈

R-Mod for some ring R. For every f ∈ HomR (R M, R N) there exist a monomor-
phism ι ∈ HomR (im(f ), R N ), an epimorphism π ∈ HomR (R M, R M/ ker(f ))
and an isomorphism fˆ : R M/ ker(f ) → im(f ) such that f = ι ◦ fˆ ◦ π; that is, the
following diagram is commutative

f
RM RN

R M/ ker(f ) im(f )
f

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 25

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_3
26 3 The Isomorphism Theorems

Proof We define the R-module homomorphism fˆ by fˆ(m + ker(f )) = f (m),

m ∈ R M. Suppose m + ker(f ) = 0, that is, m ∈ ker(f ). Then f (m) = 0 and
therefore fˆ is well defined. Suppose fˆ(m + ker(f )) = 0; then m ∈ ker(f ) and
thus m + ker(f ) = 0. Therefore fˆ is injective. By construction, fˆ makes the above
diagram commutative, where π : R M → R M/ ker(f ) is the canonical epimorphism
and ι : im(f ) → R N is the canonical monomorphism, and it is surjective onto
im(f ). As a result, it is the desired isomorphism. 


We can now draw several nice consequences from the above theorem.

3.1.2 First Isomorphism Theorem Let R M, R N ∈ R-Mod for some ring R. For
every f ∈ HomR (R M, R N), we have R M/ ker(f ) ∼
= im(f ).

Proof The mapping fˆ in the canonical factorisation supplies the desired isomor-
phism. 


3.1.3 Second Isomorphism Theorem Let R M, R N , R K ∈ R-Mod for some

ring R. Suppose R N ≤ R M and R K ≤ R M. Then

(R N + R K)/R K ∼
= R N /(R N ∩ R K).

Proof Let f : R N → (R N + R K)/R K be the composition of the inclusion R N →

R N + R K and the canonical epimorphism from R N + R K onto (R N + R K)/R K.
For n ∈ R N and k ∈ R K we have n + R K = n + k + R K and hence, f is surjective.
Its kernel is ker(f ) = {n ∈ R N | n ∈ R K} = R N ∩ R K. The first isomorphism
theorem thus yields the result. 


3.1.4 Third Isomorphism Theorem Let R M, R K, R L ∈ R-Mod for some

ring R. Suppose R K ≤ R L ≤ R M. Then
  
R M/R K R L/R K
∼
= R M/R L.

Proof The mapping f : R M/R K → R M/R L is a well-defined epimorphism since

m + R K = 0 implies m + R L = 0, as R K ⊆ R L. Its kernel is ker(f ) = {m + R K |
m ∈ R L} = R L/R K; therefore, the first isomorphism theorem yields the statement.



In the subsequent chapters, we shall see manifold applications of these isomor-

phism theorems. For the moment, however, we want to address a similar question
for categories: Is there a convenient notion of “isomorphism” of categories? For this
question to make sense, we first have to introduce connections between two different
categories; these are the functors of the next section.
3.2 Functors and Natural Transformations 27

3.2 Functors and Natural Transformations

In this section we will discuss how ‘to move’ from one category to another; this
is done via ‘functions between categories’, the functors. We shall then see how to
move between functors; this is done via natural transformations.

Definition 3.2.1 Let C and D be categories. A functor F : C → D consists of

(i) a mapping A → F(A) between the object classes of C and D;

(ii) a family of mappings f → F(f ) between the sets of morphisms

MorC (A, B) → MorD (F(A), F(B)), A, B ∈ ob(C) (covariant functor), or

MorC (A, B) → MorD (F(B), F(A)), A, B ∈ ob(C) (contravariant functor)

such that
(a) ∀ A ∈ ob(C) : F(1A ) = 1F(A) ;
(b) ∀ A, B, C ∈ ob(C) ∀ f ∈ MorC (A, B), g ∈ MorC (B, C):

F(gf ) = F(g)F(f ) (covariant) or F(gf ) = F(f )F(g) (contravariant).

The functor F is called faithful if it is injective on morphism sets and it is called full
if it is surjective on morphism sets.

Let us have a look at a number of examples of functors.

Examples 3.2.2

1. The constant functor C → C, A → A0 where A0 is a fixed object in a

category C sends each morphism f to the identity 1A0 of A0 . It is both co- and
contravariant.
2. The identical functor C → C, A → A and f → f is an important covariant
functor on every category.
3. The forgetful functor: A concrete category is a pair (C, F) consisting of a
category C and a faithful functor F : C → S. Such a functor is an example of a
forgetful functor as it ‘forgets’ some of the structure the objects in C may have
by replacing them with the ‘underlying’ set. The morphisms in C correspond, via
F, to set mappings on the ‘underlying’ sets. Such a functor is clearly covariant.
Examples of concrete categories are R-Mod or Top.
Similar forgetful functors can be defined from Comp to Top or Ring to
AGr; in each case, part “of the structure on the objects” is forgotten and the
corresponding requirement on the morphisms ignored.
28 3 The Isomorphism Theorems

4. The inclusion functor: Let C0 be a subcategory of the category C. Then ob(C0 ) 

A → A ∈ ob(C), mor(C0 )  f → f ∈ mor(C) is a covariant functor, the
inclusion of the subcategory into the larger category.
5. The functor C: For a compact Hausdorff space X, let C(X) denote the set of
all continuous complex-valued functions on X. This set has a natural structure
of a commutative unital ring using the pointwise operations. For ϕ, ψ : X → C
we define ϕ + ψ and ϕψ via (ϕ + ψ)(x) = ϕ(x) + ψ(x) and (ϕψ)(x) =
ϕ(x)ψ(x), x ∈ X. The identity is the constant function x → 1. In this way,
every object X ∈ ob(Comp) is mapped onto an object C(X) ∈ ob(Ring1 ). Let
f ∈ Mor(X, Y ) for X, Y ∈ ob(Comp). Then C(f ) : ϕ → ϕ ◦ f defines a unital
ring homomorphism from C(Y ) to C(X). It is easy to check that this results in a
contravariant functor C.
Since C(X) in addition has the structure of a complex Banach space—the
norm of ϕ is defined as ϕ = sup{|ϕ(x)| | x ∈ X}—, there is a similar
contravariant functor C : Comp → Ban1 .
We now come to a class of functors that are of paramount importance for us.
6. The Hom-functors: Let R be a unital ring, and let C be a full subcategory
of R-Mod. Fix R M ∈ R-Mod. Then

HomR (R M, −) : C −→ AGr
RN −→ HomR (R M, R N ) (3.2.1)
HomR (R N , R L)  f −→ f∗ = HomR (R M, f )

given by f∗ (g) = fg for g ∈ HomR (R M, R N) yields the covariant Hom-

functor.

Note that f∗ : HomR (R M, R N ) → HomR (R M, R L) is a homomorphism of

abelian groups.
Similarly to (3.2.1) we get the contravariant Hom-functor, for fixed R M ∈
R-Mod,

HomR (−, R M) : C −→ AGr

RN −→ HomR (R N , R M) (3.2.2)
∗
HomR (R N , R L)  f −→ f = HomR (f, R M)

given by f ∗ (g) = gf for g ∈ HomR (R L, R M).

3.2 Functors and Natural Transformations 29

As above, f ∗ : HomR (R L, R M) → HomR (R N, R M) is a group homomor-

phism.
It is plain that the above concepts can be generalised to other categories; cf.,
e.g., the functor C above. A functor F : C → S is called representable if there is
an object A ∈ ob(C) such that F = MorC (A, −) in the covariant case and F =
MorC (−, A) in the contravariant case. In this situation, A is called a representing
object.

We now turn our attention to the ‘functions between functors’.

Let F and G be functors of the same variance between the categories C and D.
For each A ∈ ob(C), let ηA ∈ MorD (F(A), G(A)). Then η = (ηA ) is a natural
transformation from F to G if, for each A, B ∈ ob(C) and each f ∈ MorC (A, B),
the following diagram is commutative

F
F F

G G
G

covariant case

F
F F

G G
G

contravariant case

In the case that every ηA is an isomorphism, η is called a natural isomorphism.

A natural transformation shifts commuting triangles from one category to the
other; the diagram below illustrates the covariant case (where f ∈ MorC (A, B),
g ∈ MorC (B, C)).
30 3 The Isomorphism Theorems

Example 3.2.3 Let C be a full subcategory of R-Mod. Let R M, R N ∈ C. For

fixed m ∈ R M, a morphism f ∈ HomR (R M, R N) can be evaluated at m; this
gives f (m) ∈ R N . In this way, we obtain a mapping

RM × HomR (R M, R N ) −→ R N, (m, f ) → f (m);

this is called the evaluation map. By means of this, we have a mapping

 
ηM : R M −→ HomZ HomR (R M, R N ), R N

as each map f → f (m) is a group homomorphism from HomR (R M, R N) into R N .

The Hom-functors HomR (−,  R N) and HomZ (−, R N ) are contravariant; thus
their composition G = HomZ HomR (−, R N), R N is a covariant functor from C
into AGr. Let F : C → AGr be the forgetful functor. Then

ηM : F(R M) −→ G(R M) (R M ∈ C)
3.2 Functors and Natural Transformations 31

defines a natural transformation between F and G. To verify this statement, we need

to check the commutativity of the following diagram

Let m ∈ M, g ∈ HomR (R M, R L), f ∈ HomR (R L, R N ). Then

ηL (F(g)(m))(f ) = f (g(m)) = (fg)(m) = g ∗ (f )(m)

= ηM (m)(g ∗ (f )) = (ηM (m)g ∗ )(f ) = G(g)(ηM (m))(f )

which shows that the diagram is commutative.

Suppose C and D are categories and F : C → D and G : D → C are functors

such that F ◦ G is naturally isomorphic to the identity functor on D and G ◦ F
is naturally isomorphic to the identity functor on C. Then the two categories are
naturally equivalent; in this way we can compare categories of quite different ‘sizes’
with each other.

Example 3.2.4 A commutative unital C*-algebra A consists of a commutative

unital ring A which at the same time is a complex Banach space with a norm  · . In
addition, we require that the norm is submultiplicative, that is, for x, y ∈ A we have
xy ≤ x y, and that 1 = 1. Moreover, there exists an involution * on A, that
is, a conjugate-linear anti-multiplicative mapping * : A → A such that (x ∗ )∗ = x
and xx ∗  = x2 for all x ∈ A. It is easy to verify that, for every compact
Hausdorff space X, the space C(X) is a commutative unital C* -algebra, where the
involution is given by ϕ ∗ (x) = ϕ(x) for each x ∈ X; compare Example 3.2.2 above.
We define a morphism ρ between two commutative unital C* -algebras A and B
to be a linear mapping which is also a unital ring homomorphism and satisfies
ρ(x ∗ ) = ρ(x)∗ for all x ∈ A. It turns out that every such mapping is a bounded
linear operator of norm 1 and hence the isomorphisms are linear isometries. The
category built in this fashion will be denoted by AC1∗ . It is a non-full subcategory of
Ban1 . (For more information on C* -algebras, we refer the reader to [1] and [32].)
As a result, the above functor C can also be defined from Comp into AC1∗ .
Let A ∈ ob(AC1∗ ). Endowed with the hull-kernel topology the set of all maximal
ideals of A becomes a compact Hausdorff space, see [1] and [32], which we shall
denote by (A). Let ρ : A → B be a morphism in AC1∗ . For each maximal ideal
I ⊆ B the inverse image ρ −1 (I ) is a maximal ideal in A, we shall denote it by ρ ∗ (I ).
It can be shown that the mapping ρ ∗ : (B) → (A) so defined is continuous, i.e.,
32 3 The Isomorphism Theorems

a morphism in Comp. In addition, becomes a contravariant functor from Comp

into AC1∗ .
Gelfand theory now establishes that (C(X)) is naturally homeomorphic to X,
for every X ∈ ob(Comp), and that C( (A)) is naturally isomorphic to A, for
every A ∈ ob(AC1∗ ). By means of this, we obtain a natural equivalence between the
categories Comp and AC1∗ .

Far more detailed information on functors and natural transformations can be

found in [25] and [30], for example.

3.3 Exercises

We use this opportunity to introduce an important device from homological algebra.

A sequence

f j−1 fj f j+1 f j+2

... RMj RMj +1 RMj +2 ...
(3.3.1)

consisting of a sequence (R M j )j ∈N of modules in R-Mod and module maps fj ∈

HomR (R M j , R M j +1 ), j ∈ N is called exact at step j + 1 if ker(fj +1 ) = im(fj ).
The entire sequence is called an exact sequence if it is exact at every step j ∈ N.
In case the sequence consists of three (non-trivial) modules and the zero module
at either end,

f g
0 RM RN RL 0 (3.3.2)

and is exact we speak of a short exact sequence. Note that we do not have to specify
the module maps at either end (they can only be the zero maps) and that exactness
f g
in this situation means that R M R N is injective, R N R L is
surjective and ker(g) = im(f ). By the first isomorphism theorem, it follows that
∼
R L = R N/im(f ) so one can think of R L as a quotient of R N . The degenerate case
when R L = 0 means that f itself is an isomorphism.

Exercise 3.3.1 Let R N , R L be submodules of the module R M ∈ R-Mod. Define

canonical module mappings to obtain a short exact sequence

0 R M/( R N R L) R M/ R N R M/ R L R M/( R N + R L) 0.

Then apply this short exact sequence to obtain an isomorphism

(R N + R L)/(R N ∩ R L) ∼
= (R N + R L)/R N × (R N + R L)/R L.
3.3 Exercises 33

Exercise 3.3.2 Show that HomR (R M, −) for fixed R M ∈ R-Mod is a covariant

functor from R-Mod into the category of abelian groups.

Exercise 3.3.3 Show that a morphism f in R-Mod is a monomorphism if and

only if f∗ = HomR (R M, f ) is injective for every R M ∈ R-Mod and f is an
epimorphism if and only if f ∗ = HomR (f, R M) is injective for every R M ∈
R-Mod.

Exercise 3.3.4 In Chap. 2, p. 23 we introduced the notion of an additive category. A

functor F between two additive categories C and D is called additive if F(f + g) =
F(f ) + F(g) for any two morphisms f, g in the same morphism set in C. Show that
the Hom-functor HomR (R M, −) for fixed R M ∈ R-Mod is additive from R-Mod
into AGr.

Exercise 3.3.5 In Exercise 1.3.3 we saw how to turn an S-module into an R-module
given a fixed ring homomorphism ρ : R → S between two rings R and S. Does this
procedure give us a covariant functor Fρ : S-Mod → R-Mod?

Exercise 3.3.6 Let R be a unital ring and let R L, R N be submodules of the right R-
module R M. We define [R L : R N ] = {a ∈ R | a · R N ⊆ R L}. Show the following
statements:

(i) [R L : R N ] = R if R N ⊆ R L.
(ii) [R L1 ∩ R L2 : R N ] = [R L1 : R N ] ∩ [R L2 : R N ] for any R L1 , R L2 ≤ R M.
(iii) [R L : R L + R N ] = [R L : R N ].

Exercise 3.3.7 Let R be a unital commutative ring and let R M ∈ R-Mod. Using
the same notation as in the previous exercise, show that the following two conditions
are equivalent:

(a) For all 0 = R N ≤ R M, [0 : R N] = [0 : R M];

(b) For all ideals I  R and 0 = R N ≤ R M, I · R N = 0 implies that I · R M = 0.

Show further that, if I is an ideal of R, then the quotient module R R/R I has any of
the above two propeties if and only if I is a prime ideal.
Noetherian Modules
4

Finiteness conditions on the ideal structure of a ring allow us to develop a more

detailed structure theory. In the noncommutative framework, it is more natural to do
this for (one-sided) modules rather than for (one-sided) ideals, so we shall follow
this approach right from the beginning.
The following result is fundamental to this chapter.

Theorem 4.1 Let R M ∈ R-Mod for a unital ring R. The following conditions are
equivalent.

(a) R M satisfies ACC, the ascending chain condition:

Each ascending chain R N 1 ⊆ R N 2 ⊆ . . . ⊆ R N i ⊆ . . . of submodules R N i ≤
R M becomes stationary, i.e., there exists k such that R N i = R N k for all i ≥ k.
(b) Each strictly ascending chain R N 1 ⊂ R N 2 ⊂ . . . ⊂ R N i ⊂ . . . of submodules
R N i ≤ R M is finite.
(c) Every non-empty set of submodules of R M contains a maximal element.
(d) Every submodule of R M is finitely generated.

Proof The implication (a) ⇒ (b) is trivial. Towards (b) ⇒ (c) let N be a non-
empty set of submodules of R M without a maximal element. Take R N 1 ∈ N; then
there is R N 2 ∈ N with R N 1 ⊂ R N 2 . Let R N 1 , . . . , R N k ∈ N be chosen such that
R N 1 ⊂ R N 2 ⊂ . . . ⊂ R N k . Since R N k is not maximal in N, there is R N k+1 ∈ N
strictly containing R N k . By induction, we obtain an infinite strictly ascending chain
of submodules of R M. Thus (b) fails.
(c) ⇒ (d) Let R N ≤ R M. Let N be the set of all finitely generated submodules
of R N . As these are submodules of R M, N has a maximal element R N 0 . If there is
n ∈ R N \ R N 0 then R N 0 +Rn is a finitely generated submodule of R N strictly larger
than R N 0 . As this contradicts the maximality of R N 0 in N, we find that R N = R N 0
is finitely generated.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 35

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_4
36 4 Noetherian Modules

(d) ⇒ (a) Let R N 1 ⊆ R N 2 ⊆  . . . ⊆ R N i ⊆ . . . be an ascending chain

of submodules of R M. Set R N = i R N i which is a submodule of R M. By
hypothesis, R N is finitely generated. Let {m1 , . . . , mk } be a set of generators of R N .
Then there is j ∈ N such that {m1 , . . . , mk } ⊆ R N j . It follows that R N = R N j ,
equivalently, R N i = R N j for all i ≥ j . 


Definition 4.2 A left R-module R M ∈ R-Mod for a unital ring R is called

Noetherian if it satisfies any, and hence every, condition in Theorem 4.1 above.
The ring R is called left Noetherian if the module R R is Noetherian.

Historical Note Born in 1882 in Erlangen (Bavaria, Germany), Emmy Amalie

Noether was one of the first female students at a German university. After she
had become a certificated teacher of English and French in Bavarian girls schools,
she studied Mathematics in Erlangen and Göttingen where she attended lectures
by David Hilbert, Felix Klein and Hermann Minkowski, amongst others. In 1907
she received her doctorate under Paul Gordon. Many bureaucratic problems had to
be overcome before Emmy Noether achieved the position of Privatdozent at the
University in Göttingen in 1919. She worked on invariant theory and mathematical
physics and became instrumental in the development of abstract algebra. Her
influence on ring theory and especially finiteness conditions on ideals cannot be
overestimated. Twice she gave invited addresses to the International Congress of
Mathematicians, in Bologna in 1928 and in Zürich in 1932. She died in 1935 in Bryn
Mawr (Pennsylvania, USA) where she had to emigrate to because of her Jewish
ancestry.
A detailed account of Emmy Noether’s life and work, written by J. J. O’Connor
and E. F. Robinson can be found at
https://mathshistory.st-andrews.ac.uk/Biographies/Noether_Emmy/

There is, of course, an analogous concept of Noetherian right modules and of a

right Noetherian ring. These two concepts, however, differ from each other as the
following example illustrates.

Example 4.3 Let s and t be two symbols and let R = Zs, t | t 2 = ts = 0 be the
quotient of the free unital ring Zs, t by the stated relations. As every element of R
is a unique combination of words of the form s n and s n t, n ≥ 0, R can be written
as R = Z[s] ⊕ Z[s]t. The commutative ring Z[s] is Noetherian by Hilbert’s Basis
Theorem (see below). The ideal Z[s]t of R is considered as a Z[s]-module; therefore
the finitely generated Z[s]-module Z[s] ⊕ Z[s]t is Noetherian, by Corollary 4.1.4
below, in other words, R is left Noetherian.
In order to show that R is not right Noetherian, it suffices to verify that the
submodule Z[s]t ∈ Mod-R is not finitely generated. The monomials s n t, n ≥ 0
are Z-linearly independent; thus there cannot exist finitely many p1 , . . . , pk ∈ Z[s]t
such that an arbitrary element p = n≥0 mn s n t, where mn ∈ Z are non-zero for
k
at most finitely many n, can be written as p = j =1 mj pj as the latter would
4.1 Permanence Properties of Noetherian Modules 37

imply that Z[s]t is finitely generated as an abelian group. As a result, R is not right
Noetherian.

A unital ring is said to be Noetherian if it is both left and right Noetherian.

Evidently, every principal ideal domain is Noetherian. Far more general examples
are provided by the following fundamental result.

4.4 Hilbert’s Basis Theorem For every left Noetherian ring R, the polynomial
ring R[x1 , . . . , xn ] is left Noetherian.

Proof Since R[x, y] = R[x][y] it suffices to establish the result for polynomial
rings in one indeterminate and then proceed by induction. Suppose that R[x] is not
left Noetherian. Take a left ideal I in R[x] which is not finitely generated. Let f1 ∈ I
be a polynomial with minimal degree. Then I1 = R[x]f1  I . Take f2 ∈ I \ I1 of
minimal degree. Put I2 = R[x]f1 + R[x]f2  I . Since I is not finitely generated,
by induction we obtain an infinite strictly ascending chain (In )n∈N of left ideals
in R[x]. Let nk = deg fk , k ∈ N and let ank be the leading coefficient in fk (i.e.,
ank is the coefficient of x nk ). Observe that n1 ≤ n2 ≤ . . . ≤ nk ≤ nk+1 ≤ . . .. Let
Lk = Ran1 + . . . + Rank , k ∈ N. Clearly, (Lk )k∈N is an ascending chain of left
ideals of R. We claim that this sequence is strictly ascending (hence R is not left
Noetherian).
Suppose on the contrary that ank+1 ∈ Lk = Ran1 +. . .+Rank for some k. Hence,

ank+1 = ki=1 ri ani for some ri ∈ R. Put


k
f (x) = fk+1 (x) − ri x nk+1 −ni fi (x).
i=1

Then f ∈ Ik+1 , the left ideal generated by f1 , . . . , fk , fk+1 but f ∈ Ik . We note

that deg f < deg fk+1 :


n  
n 
ank+1 x nk+1 − ri x nk+1 −ni ani x ni = ank+1 − ri ani x nk+1 = 0.
i=1 i=1

This violates the choice of fk+1 being a polynomial in Ik+1 \ Ik of lowest degree.
As a result our initial assumption entails that R is not left Noetherian. 


4.1 Permanence Properties of Noetherian Modules

In this section, we shall study how Noetherian modules behave under various
canonical constructions, such as taking submodules or homomorphic images. The
key result is the following theorem in which we employ the concept of an exact
sequence of modules, see p. 32.
38 4 Noetherian Modules

Theorem 4.1.1 Let R be a unital ring. Let

f g
0 RM1 RM2 RM3 0

be a short exact sequence in R-Mod. Then R M 2 is Noetherian if and only if both

R M 1 and R M 3 are Noetherian.

Proof Suppose at first that R M 2 is Noetherian. Every ascending chain of sub-

modules of R M 1 becomes an ascending chain of submodules of R M 2 via the
injective module map f . Hence it becomes stationary, and thus R M 1 is Noetherian.
Every ascending chain of submodules of R M 3 is the image of an ascending chain
of submodules of R M 2 under the surjective module map g. Hence it becomes
stationary, and thus R M 3 is Noetherian.  
Suppose now that both R M 1 and R M 3 are Noetherian, and let R N (k) k∈N be an
ascending chain of submodules of R M 2 . Setting R N 1 = f −1 (R N (k) ) and R N 3 =
(k) (k)

g(R N (k) ), k ∈ N we obtain ascending chains of submodules of R M 1 and of R M 3 ,

(k) (m) (k)
respectively. By hypothesis, there is m ∈ N such that R N 1 = R N 1 and R N 3 =
(m)
RN3 for all k ≥ m. We want to show that R N (k) = R N (m) for all k ≥ m and
employ the commutative diagram below to this end.

where m ≤ k and f (k) and g (k) are the restrictions of f and g, respectively to the
corresponding submodules. By construction, all rows are short exact sequences. As
(m) ⊆ N (k) it remains to prove the reverse inclusion.
RN R
Let x ∈ R N (k) . Since R N (k) (m)
3 = R N 3 there is y ∈ R N
(m) such that g (k) (x) =
(k)
g (y). Since ker(g ) = im(f ) there is y ∈ R N 1 such that x − y = f (k) (y ).
(k) (k) (k)
(k) (m) (m)
As R N 1 = R N 1 , y ∈ R N 1 . It follows that

f (y ) = f (m) (y ) ∈ f (m) (R N 1 ) = f (f −1 (R N (m) )) ⊆ R N (m)

(m)

and hence x = y + f (y ) ∈ R N (m) as required. 


4.2 Exact Categories and Exact Functors 39

The above argument is an example of a ‘diagram chase’.

Applying Theorem 4.1.1 to the short exact sequence

0 RN RM R M/ R N 0,

where R N ≤ R M we immediately obtain the following result.

Corollary 4.1.2 Let R be a unital ring. Let R N be a submodule of R M ∈ R-Mod.

Then R M is Noetherian if and only if both R N and R M/R N are Noetherian.

From the above theorem we obtain a wealth of further Noetherian modules.

Corollary 4.1.3 Every finite direct sum of Noetherian modules is Noetherian.

Proof Using induction, it suffices to consider the case of two Noetherian modules
R L and R N . In this case, the statement follows directly from Theorem 4.1.1 applied
to the short exact sequence

0 RN RN RL RL 0.




Corollary 4.1.4 Every finitely generated unital left module over a left Noetherian
ring is Noetherian.

Proof Let R be a left Noetherian ring. By Corollary 4.1.3, the module R R n is

Noetherian for every n ∈ N. Let R M ∈ R-mod and suppose {m1 , . . . , mn } is a
set of generators of R M. The mapping


n
RR
n
−→ R M, (r1 , . . . , rn ) −→ ri · mi
i=1

is an epimorphism; thus Corollary 4.1.2 yields the claim. 



We observe therefore that a unital ring R is left Noetherian if and only if every
module in R-mod is Noetherian.

4.2 Exact Categories and Exact Functors

The concept of an exact sequence is fundamental in the theory of modules. In this

section we shall explore to what extent this tool can be employed in more general
categories.
40 4 Noetherian Modules

4.2.1 Kernels and Cokernels

Let C be a category with zero object, see Chap. 2, p. 23. We will denote this object
(which is unique up to isomorphism by Exercise 2.7.12) by 0. Let B, C ∈ ob(C).
The unique morphisms f : 0 → B and g : C → 0 have the properties

(i) h1 f = h2 f for all h1 , h2 ∈ MorC (B, A), A ∈ ob(C);

(ii) gh1 = gh2 for all h1 , h2 ∈ MorC (A, C), A ∈ ob(C);
(iii) fg : C → B is the unique morphism, denoted by 0(C,B) , satisfying

h 0(C,B) = 0(C,A) and 0(C,B) k = 0(A,B)

for all A ∈ ob(C) and h ∈ MorC (B, A), k ∈ MorC (A, C).

The above properties are easily checked (see Exercise 4.3.8 below) and the above-
mentioned unique morphism is called a zero morphism. As a result, a category C
with zero object has zero morphisms between all its objects.
In the sequel, we will assume that C has a zero object 0 and denote the zero
morphisms unambiguously by 0, too.
Let f ∈ MorC (A, B) for some A, B ∈ ob(C).

(i) A morphism i : K → A is a kernel of f if f i = 0 and for each D ∈ ob(C) and

g ∈ MorC (D, A) with fg = 0 there is a unique h ∈ MorC (D, K) making the
diagram below commutative

0
h
g

i f
K A B

0 (4.2.1)

Any kernel is a monomorphism and is, up to isomorphism, unique.

4.2 Exact Categories and Exact Functors 41

(ii) A morphism p : B → C is a cokernel of f if pf = 0 and for each D ∈ ob(C)

and g ∈ MorC (B, D) with gf = 0 there is a unique h ∈ MorC (C, D) making
the diagram below commutative

A B C
f p

g
h
0

D (4.2.2)

Any cokernel is an epimorphism and is, up to isomorphism, unique.

Since kernel and cokernel of a morphism f , if they exist, are unique, it suffices
to give their domain and codomain, respectively, a symbol; these are, respectively,
ker(f ) and cok(f ).
For instance, in R-Mod, the kernel of f ∈ HomR (R M, R N ) is the usual kernel
ker(f ) together with the canonical embedding ker(f ) → R M, and the cokernel of f
is the canonical projection π : R N → R N /im(f ). However, in R-mod, kernels
need not exist since a submodule of a finitely generated module may be not finitely
generated.
We are now in a position to introduce the concept of an exact sequence.

4.2.2 Exact Categories

Let C be a category with zero, kernels and cokernels.

f g
A sequence A B C in C is called exact if gf = 0 and in the commutative
diagram below the unique morphism h is an epimorphism

f g
A B C

h
ker(g) (4.2.3)
42 4 Noetherian Modules

A sequence is exact if each 3-point part of it is

exact. A short exact sequence in C is of the form

f g
0 A B C 0

with f a kernel of g and g a cokernel of f .

Note that the above definitions in R-Mod agree with our previous ones in
Chap. 3 since epimorphisms in R-Mod are surjective.
The category C is called exact if every morphism in C can be written as a
composition of an epimorphism and a monomorphism. For instance, the canonical
factorisation of module maps (Theorem 3.1.1) tells us that R-Mod is exact.

4.2.3 Exact Functors

Let C and D be exact categories. A covariant functor F : C → D is said to be left

exact, respectively right exact, if F(0) = 0 and for every short exact sequence

f g
0 A B C 0

in C the sequence

F(f ) F(g)
0 F(A) F(B) F(C)

is exact in D, respectively the sequence

F(f ) F(g)
F(A) F(B) F(C) 0

is exact in D. If F is both left and right exact we say that F is exact.

There is an analogous definition for contravariant functors with arrows reversed
(but the sides “left” and “right” remain unchanged!).

Example 4.2.3.1 The covariant Hom-functor HomR (R U , −) : R-Mod → AGr is

left exact for every R U ∈ R-Mod. To see this, we start with a short exact sequence

f g
0 RM RN RL 0
4.3 Exercises 43

in R-Mod and we need to show that (a) f∗ is a monomorphism, and (b) im(f∗ ) =
ker(g∗ ); then the sequence

f g
0 HomR ( R U , R M) HomR ( R U , R N ) HomR ( R U , R L)

will be exact.
If f h = 0 for some h ∈ HomR (R U , R M) then h = 0 as f is injective.
Therefore, f∗ is injective, thus a monomorphism. This proves (a).
In order to prove (b), take h ∈ im(f∗ ) ⊆ HomR (R U , R N) so that h = f k for
some k ∈ HomR (R U , R M). Then g∗ (h) = gh = gf k = 0 as im(f ) ⊆ ker(g). It
follows that h ∈ ker(g∗ ) and thus im(f∗ ) ⊆ ker(g∗ ). Now take h ∈ ker(g∗ ) so that
gh = 0. Then h(U ) ⊆ ker(g) = im(f ). As a result, for u ∈ U , there is m ∈ M
with h(u) = f (m) and, since f is injective, we can define k ∈ HomR (R U , R M) by
k(u) = m. It follows that h = f k ∈ im(f∗ ), which proves (b).

In a similar vein, the contravariant Hom-functor HomR (−, R U ) is left exact

but right exactness (of either functor) needs additional assumptions on R U , see
Exercise 6.4.8 in Chap. 6.

4.3 Exercises

Exercise 4.3.1 Let R M ∈ R-Mod for a unital ring R be Noetherian. Let f ∈

EndR (R M). Show that f is injective provided it is surjective.

Exercise 4.3.2 Prove that every left Noetherian unital ring R is von Neumann finite,
that is, for x, y ∈ R, xy = 1 implies that yx = 1.

Exercise 4.3.3 Let R be a commutative unital ring. Using Zorn’s lemma, show that
R is Noetherian if every prime ideal of R is finitely generated.

Exercise 4.3.4 Prove that, if R is a left Noetherian ring, then Mn (R), n ∈ N is a

left Noetherian ring.

Exercise 4.3.5 Provide an alternative proof of Corollary 4.1.2 by showing that

a module R M has all submodules finitely generated (i.e., is Noetherian) if R M
contains a Noetherian submodule R N such that R M/R N is Noetherian. (That is,
work only with condition (d) in Theorem 4.1. To this end, use Exercise 2.7.4 and an
appropriate isomorphism theorem.)
44 4 Noetherian Modules

Exercise 4.3.6 Let R M ∈ R-Mod for a unital ring R. Show that the following two
conditions on R M are equivalent.

(a) R M is finitely generated;

(b) for every family {R Mi | i ∈ I } ⊆ R-Mod and each epimorphism
f: i∈I R Mi → R M there exists a finite subset J ⊆ I such that the
composition

ιJ f
R Mj −→ R Mi −→ R M
j ∈J i∈I

is surjective (where ιJ denotes the canonical monomorphism).

Exercise 4.3.7 Let R M ∈ R-Mod for a unital ring R. Show that R M is finitely
generated if andonly if, for every family {R Ni | i ∈ I } of submodules R Ni of
M such that  i∈I R Ni  = R M, there exists a finite subset J ⊆ I such that
R
 j ∈J R Nj  = R M.

Exercise 4.3.8 Verify the properties of the zero morphism as stated in Sect. 4.2.1.

Exercise 4.3.9 Prove that the kernel and the cokernel of a morphism in a category
with zero are unique up to isomorphism, provided they exist.

Exercise 4.3.10 Let C be an exact category. Show that for every f ∈ mor(C) the
following identities hold:

ker(cok(ker(f ))) = ker(f ) and cok(ker(cok(f ))) = cok(f ).

Exercise 4.3.11 Let f, g ∈ mor(C) be “parallel arrows”, that is, both f and g
belong to MorC (A, B) for two objects A, B in C. An equalizer of (f, g) is a
morphism e ∈ MorC (C, A) such that f e = ge and for every h ∈ MorC (D, A)
with f h = gh there is a unique h ∈ MorC (D, C) such that eh = h. Show that in
R-Mod every equalizer is a kernel.

Exercise 4.3.12 Identify the kernel and the cokernel of a morphism f ∈ Ban1 .

Exercise 4.3.13 Let R M and R N be left R-modules. Let f : RM → RL and

g : R N → R L be module maps with the same codomain.

(i) Show that

RM ×L R N = {(x, y) ∈ R M × R N | f (x) = g(y)}

is a submodule of the product module R M × R N.

4.3 Exercises 45

(ii) Let π1 : R M × R N → R M and π2 : R M × R N → R N be the projections onto

the first and the second coordinate, respectively. Show that the diagram below
is commutative:
1
RM L RN RM

2 f

RN RL
g

(iii) Prove that R M ×L R N has the universal property of a pullback; that is, given
another commutative diagram of module maps of the form

1
RK RM

2 f

RN RL
g

for some R K ∈ R-Mod there is a unique module map h : R K → R M ×L R N

such that π1 ◦ h = σ1 and π2 ◦ h = σ2 .

Exercise 4.3.14 Let R M and R N be left R-modules. Let f : RL → RM and

g : R L → R N be module maps with the same domain.

(i) Show that

RT = {(f (x), −g(x)) | x ∈ R L}

is a submodule of the sum R M ⊕ R N.

46 4 Noetherian Modules

(ii) Let R M ⊕T R N denote the quotient module (R M ⊕ R N)/R T . Put α(y) =

(y, 0) + R T , y ∈ R M and β(z) = (0, z) + R T , z ∈ R N . Show that the diagram
below is commutative

f
RL RM

RN RM T RN

(iii) Prove that R M ⊕T R N has the universal property of a pushout; that is, given
another commutative diagram of module maps of the form

f
RL RM

RN RK

for some R K ∈ R-Mod there is a unique module map h : R M ⊕T R N → R K

such that h ◦ α = α and h ◦ β = β .
Artinian Modules
5

The dual condition to “ACC” is “DCC”, the descending chain condition which we
shall discuss in the present chapter. The resulting modules are termed Artinian after
Emil Artin (1898–1962).

Historical Note Born in 1898 in Vienna (Austria), Emil Artin became one of the
leading mathematicians of the twentieth century. He made major contributions
to algebraic number theory, notably class field theory, abstract algebra and braid
theory. In 1927 he solved Hilbert’s 17th problem.
Artin’s university studies at Vienna were interrupted when he was drafted in June
1918 but he continued them in Leipzig from 1919 onward where he was awarded his
PhD in 1921. After a 1-year postdoctoral position at Göttingen during which period
he worked closely with Emmy Noether and Helmut Hasse, Artin moved to Hamburg
where he advanced to the rank of Privatdozent, having completed his Habilitation in
1923. In 1925 he accepted a position as associate professor at Hamburg University
and was promoted to full professor in 1926.
Because of his wife’s Jewish father and his own distaste for the Hitler regime,
Artin lost his position in Hamburg and was forced to emigrate to the US where he
got a professorship at Notre Dame University, Indiana in 1937. In 1938 Artin and
his family moved again, to Indiana University in Bloomington. Apart from various
visiting positions at Stanford, Ann Arbor, Boulder as well as in Japan, Artin spent
the years between 1946 and 1957 at Princeton, which had become the mecca of
mathematics during those years. Finally, in 1958, he returned to Hamburg where he
would remain until he died of a heart attack in December 1962.
A detailed account of Emil Artin’s life and work, written by J. J. O’Connor and
E. F. Robinson can be found at
https://mathshistory.st-andrews.ac.uk/Biographies/Artin/

Interestingly enough there is no characterisation of Artinian modules in terms of

elements, the dual concept to “finitely generated” turns out to be formulated entirely

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 47

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_5
48 5 Artinian Modules

in categorical language so we need to introduce this notion first. We shall continue

our discussion of module-like categories in Sect. 5.4 below with a brief introduction
to abelian categories.

5.1 Finitely Cogenerated Modules

We first introduce the main concept of this chapter.

Definition 5.1.1 Let R be a unital ring. Let R M ∈ R-Mod. We say R M is Artinian

if it satisfies DCC, that is, every descending chain

RN1 ⊇ RN2 ⊇ . . . ⊇ RNk ⊇ . . .

of submodules of R M becomes stationary, i.e., R N k = R N m for all k ≥ m for some

m ∈ N. The ring R is called left Artinian if R R is Artinian.

There is of course an analogous notion of Artinian right modules and right

Artinian rings, which is independent from left Artinian. A unital ring which is both
left and right Artinian is called Artinian.
In order to obtain a similar characterisation of Artinian modules as in Theo-
rem 4.1, we need to introduce the following property. The reader may want to
compare this definition with Exercise 4.3.6.

Definition 5.1.2 Let R be a unital ring, and let R M ∈ R-Mod. We say R M

 every family {R Mi | i ∈ I } ⊆ R-Mod and each
is finitely cogenerated if, for
monomorphism g : R M → i∈I R M i , there exists a finite subset J ⊆ I such that
the composition
g pJ
RM −→ RMi −→ RMj
i∈I j ∈J

is injective (where pJ is the canonical projection).

Exercises 4.3.7 and 5.5.1 illustrate nicely the interplay between the concepts of
finitely generated and finitely cogenerated modules.
We now come to the promised characterisation of Artinian modules.

Theorem 5.1.3 Let R M ∈ R-Mod for a unital ring R. The following conditions
are equivalent.

(a) R M is Artinian.
(b) Each strictly descending chain R N 1 ⊃ R N 2 ⊃ . . . ⊃ R N k ⊃ . . . of submodules
R N i ≤ R M is finite.
5.1 Finitely Cogenerated Modules 49

(c) Every non-empty set of submodules of R M contains a minimal element.

(d) Every quotient module of R M is finitely cogenerated.

Proof The equivalence of the conditions (a) and (b) is evident.

(b) ⇒ (c) Let N be a non-empty set of submodules of R M without a minimal
element. Take R N 1 ∈ N; then there is R N 2 ∈ N with R N 1 ⊃ R N 2 . Let
R N 1 , . . . , R N k ∈ N be chosen such that R N 1 ⊃ R N 2 ⊃ . . . ⊃ R N k . Since R N k is
not minimal in N, there is R N k+1 ∈ N strictly contained in R N k . By induction, we
obtain an infinite strictly descending chain of submodules of R M. Thus (b) fails.
(c) ⇒ (d) Let {R N i | i ∈ I } be a family of submodules of R M. We put N =
{ j ∈J R Nj | J ⊆ I finite}, the set of all finite intersections of modules in {R N i |
i ∈ I }. Clearly N = ∅. Let R N min be a minimal element in N. For each R N ∈ N,
R N ∩ R N min ∈ N; hence R N ∩ R N min = R N min . It follows that R N min is the
smallest element in N. Since
  
RNi = R Nj ,
i∈I J ⊆I finite j ∈J

 
i∈I R N i = 0 implies that N= 0 and therefore R N min = 0. By Exercise 5.5.1
below, R M is finitely cogenerated.
Every homomorphic image of R M, and hence every quotient of R M, inherits
property (c). Thus, these quotient modules are finitely cogenerated too, proving (d).
(d) ⇒ (a) Let R N 1 ⊇ R N 2 ⊇  . . . ⊇ R N k ⊇ . . . be a descending chain of
submodules of R M.Put R N = k∈N R N k . By hypothesis, R M/R N is finitely
cogenerated. Since k∈N R N k /R N = 0, condition (d) together
 with Exercise 5.5.1
below imply that there is a finite subset J ⊆ N such that j ∈J R N j /R N = 0. Since
R N 1 /R N ⊇ R N 2 /R N ⊇ . . . ⊇ R N k /R N ⊇ . . . it follows that R N r /R N = 0 for
some r ∈ N, that is, R N r = R N . As a result, R N r+k = R N r for all k ∈ N wherefore
R M is Artinian. 


The proof of the next result is analogous to the one of Theorem 4.1.1 and is therefore
left to the reader, see Exercise 5.5.2.

Theorem 5.1.4 Let R be a unital ring. Let

f g
0 RM1 RM2 RM3 0

be a short exact sequence in R-Mod. Then R M 2 is Artinian if and only if both R M 1

and R M 3 are Artinian.

Unsurprisingly, similar consequences can be drawn from Theorem 5.1.4 as in the

Noetherian case. The proofs are identical to that situation.
50 5 Artinian Modules

Corollary 5.1.5 Let R be a unital ring. Let R N be a submodule of R M ∈ R-Mod.

Then R M is Artinian if and only if both R N and R M/R N are Artinian.

Corollary 5.1.6 Every finite direct sum of Artinian modules is Artinian.

We have the following characterisation of left Artinian rings.

Corollary 5.1.7 For every unital ring R, the following conditions are equivalent.

(a) R is left Artinian;

(b) Every finitely generated unital left R-module is finitely cogenerated.

Proof (b) ⇒ (a) As R is unital, R R is finitely generated. It follows that every

quotient module R R/R L, where L is a proper left ideal of R, is finitely generated as
well and, by assumption, finitely cogenerated. Therefore, by Theorem 5.1.3, R R is
Artinian.
(a) ⇒ (b) Let R M ∈ R-Mod. If R M is finitely generated, then R M is a
homomorphic image of a free finitely generated left R-module R N. As R N is
(isomorphic to) a finite direct sum of R R (Exercise 2.7.6), Corollary 5.1.6 entails
that R N is Artinian. Theorem 5.1.3 thus yields the statement. 


As a consequence, a unital ring R is left Artinian if and only if every module in

R-mod is Artinian, another parallel to the situation of Noetherian rings.
We shall now start to explore the differences between the theories of Noetherian
and of Artinian rings. For a start, it is very easy to give an example of a Noetherian
ring which is not Artinian: Z is, as a principal ideal domain, clearly a Noetherian
ring but it is not Artinian; for instance, the descending chain

Z ⊃ 2Z ⊃ 22 Z ⊃ . . . ⊃ 2k Z ⊃ . . .

is infinite.

5.2 Commutative Artinian Rings

In this section we want to study some special properties of Artinian rings that are
commutative.

Definition 5.2.1 Let R be a principal ideal domain. A module R M ∈ R-Mod is

said to be a torsion module if AnnR (m) = 0 for every m ∈ M.

Proposition 5.2.2 Every finitely generated torsion module over a principal ideal
domain is Artinian.
5.2 Commutative Artinian Rings 51

Proof Let R be a principal ideal domain, and let R M ∈ R-mod with generators
m1 , . . . , mk . Suppose that R M is a torsion module. Then, by Exercise 2.7.9,

RM = Rm1 + . . . + Rmk and Rmi ∼

= R/AnnR (mi ), 1 ≤ i ≤ k.
k
It follows that R M is a homomorphic image of i=1 R/AnnR (mi ); thus, by
Corollaries 5.1.6 and 5.1.5, it suffices to show that each summand R/AnnR (mi )
is Artinian.
In order to prove this claim, take a non-zero ideal I of R (this is where we use
AnnR (mi ) = 0 for each i). By hypothesis, I = aR for some a ∈ R. We will show
that the ideal lattice of R/I is finite; thus DCC on submodules of R/I certainly
must hold.
As R is a principal ideal domain, it is a unique factorisation domain. As a result,
we can write a in the form a = p1n1 p2n2 · · · pn with irreducible elements pj ∈ R
and n1 , . . . , n ∈ N. For J = bR, b ∈ R we have I ⊆ J if and only if b | a. The
uniqueness in the factorisation of a implies that there are only finitely many divisors
of a; hence only finitely many ideals in R containing I . Consequently, R/I has at
most finitely many ideals. 


The reader may want to compare the above result with the fact that every finitely
generated module over a principal ideal domain is Noetherian.
Commutative Artinian rings are in a way the simplest rings after fields. To give
evidence to that fact, we record a few more of their nice properties (some of which
do extend appropriately to non-commutative Artinian rings).

Definition 5.2.3 Let R be a commutative unital ring. Then


rad(R) = {I | I maximal ideal in R}
and

nil(R) = {P | P prime ideal in R}

are called the Jacobson radical and the nil radical of R, respectively.

By definition, both nil(R) and rad(R) are ideals of R and, since every maximal
ideal is a prime ideal—the quotient by it is a field, hence an integral domain—, we
have nil(R) ⊆ rad(R). In general, this inclusion will be strict.

Proposition 5.2.4 In an Artinian commutative ring R, every prime ideal is maxi-

mal. Hence nil(R) = rad(R).

Proof Let P be a prime ideal of R. Then S = R/P is an Artinian integral domain

(Corollary 5.1.5). Let x ∈ S \ {0}. Let us denote the ideal generated by a ∈ R
by (a). Since (x n ) ⊇ (x n+1 ) for all n ∈ N, by DCC there is some m ∈ N such that
52 5 Artinian Modules

(x m ) = (x m+1 ); that is, x m = x m+1 y for some y ∈ S. Since S is an integral domain,

we find that 1 = xy, in other words, x is invertible. It follows that every non-zero
element in S is invertible so that S is a field. As a result, P is a maximal ideal.  

Proposition 5.2.5 In an Artinian commutative ring R, the nil radical nil(R) is

nilpotent, that is, nil(R)k = 0 for some k ∈ N.

Proof Since nil(R)n ⊇ nil(R)n+1 for all n ∈ N, DCC gives nil(R)n = nil(R)k for
some k ∈ N and all n ≥ k. Put I = nil(R)k . Suppose I = 0; then the set J of all
ideals J in R with I J = 0 is non-empty. Since R is Artinian, there is a minimal
element J0 in J. Take x ∈ J0 with xI = 0. As (x) ⊆ J0 we conclude that (x) = J0 .
Since (xI )I = xI 2 = xI = 0, an analogous argument yields xI = (x) = J0 . It
follows that x = xy for some y ∈ I from which we obtain

x = xy = xy 2 = . . . = xy n for all n ∈ N.

Since y ∈ I = nil(R)k ⊆ nil(R) and every element in nil(R) is nilpotent

(Exercise 5.5.5 below), y = 0 for some ∈ N. This implies that x = xy = 0
contradicting the choice of x. We conclude that I = 0 as required. 


5.3 Artinian vs. Noetherian Modules

We already observed that the ring Z is Noetherian but not Artinian. Here is another
example of the same ilk.

Example 5.3.1 We shall use again the ring from Example 4.3. The ring

R = Zs, t | t 2 = ts = 0 = Z[s] ⊕ Z[s]t

is left Noetherian (but not right Noetherian). Moreover, R is neither left nor right
Artinian since the module Z[s] ∼ = R/Z[s]t is not Artinian; it contains (Z[s]s k )k∈N
as an infinite strictly descending chain of submodules.

In the opposite direction, there is the following surprising result for rings (it does
not extend to modules, though).

5.3.2 Hopkins–Levitzki Theorem Every left Artinian ring is left Noetherian.

We defer the proof until Chap. 7 but the reader can already try their hands at the
commutative case, see Exercise 5.5.8.
Of course, an analogous statement holds with “left” replaced by “right” in both
the assumption and the conclusion.
5.3 Artinian vs. Noetherian Modules 53

We conclude our discussion by bringing the theories of Artinian and Noetherian

modules close together once more. The following result should remind the reader of
some facts they learned in Linear Algebra.

Proposition 5.3.3 Let R be a unital ring, and let R M ∈ R-Mod. For every f ∈
EndR (R M), the following holds.

(i) If R M is Artinian then there is n ∈ N such that

ker(f n ) + im(f n ) = R M;

in particular, f is surjective if it is injective.

(ii) If R M is Noetherian then there is n ∈ N such that

ker(f n ) ∩ im(f n ) = 0;

in particular, f is injective if it is surjective.

(iii) If R M is both Artinian and Noetherian then there is n ∈ N such that

ker(f n ) ⊕ im(f n ) = R M.

In addition, the following conditions are equivalent:

(a) f is injective;
(b) f is surjective;
(c) f is bijective.

The third statement in the above proposition is also known as “Fitting’s Lemma”. In
all statements, f n stands for the n-th iterate of the mapping f .

Proof

(i) Suppose R M is Artinian. Then the descending chain of submodules

im(f ) ⊇ im(f 2 ) ⊇ . . . ⊇ im(f n ) ⊇ . . .

becomes stationary; that is, im(f n+k ) = im(f n ) for some n ∈ N and all k ∈ N.
Therefore, for given x ∈ R M, there is y ∈ R M such that f n (x) = f 2n (y);
consequently, x − f n (y) ∈ ker(f n ). It follows that x = x − f n (y) + f n (y) ∈
ker(f n ) + im(f n ), as claimed.
If ker(f ) = 0 then ker(f n ) = 0; thus im(f n ) = R M which implies
im(f ) = R M.
54 5 Artinian Modules

(ii) Suppose R M is Noetherian. Then the ascending chain of submodules

ker(f ) ⊆ ker(f 2 ) ⊆ . . . ⊆ ker(f n ) ⊆ . . .

becomes stationary; that is, ker(f n+k ) = ker(f n ) for some n ∈ N and all
k ∈ N. In particular, ker(f 2n ) = ker(f n ). Thus, for x ∈ ker(f n ) ∩ im(f n ),
say x = f n (y), y ∈ R M, we have f 2n (y) = f n (x) = 0. As y ∈ ker(f 2n ) =
ker(f n ), we obtain x = 0 as claimed.
If im(f ) = R M then im(f n ) = R M; thus ker(f n ) = 0 which implies that
ker(f ) = 0.
(iii) is a direct consequence of (i) and (ii). The proof is complete.



5.4 Abelian Categories

Let C be a category which is both additive and exact; see p. 23 and 42, respectively.
Then C is called abelian. Abelian categories were introduced by A. Grothendieck
in 1957 as an abstract setting which captured many features of module categories.
In this section we aim to illustrate this fact by discussing various results familiar
from module theory that can be obtained without the use of elements in the objects.

Proposition 5.4.1 Let f ∈ MorC (A, B), where C is an abelian category. Then f
is an isomorphism if and only if it is both a monomorphism and an epimorphism.

Proof The “only if”-part follows directly from the definitions; see Definition 1.14.
Suppose f is a monomorphism. If fg = 0 for some g ∈ MorC (D, A) then g = 0.
Therefore g can be uniquely factored through 0 −→ A, that is, ker(f ) = 0. Note
that cok(ker(f )) = cok(0) = 1A ; thus we have a commutative diagram

1A
0 A A

f h

cok(f ) B ker(cok(f ))

with a unique morphism h ∈ MorC (A, ker(cok(f ))). Thus f = ker(cok(f )).
5.4 Abelian Categories 55

Suppose in addition that f is an epimorphism. Then cok(f ) f = 0 so cok(f ) =

0. Hence ker(cok(f )) = 1B and the above diagram becomes

1A
0 A A

f h

0 B B
1B

The two properties f = ker(cok(f )) and f = cok(ker(f )), respectively now give
us the existence of unique morphisms k, k ∈ MorC (B, A), respectively making the
diagrams below commutative, compare (4.2.1) and (4.2.2):

that is, f k = 1B and k f = 1A . Thus f is an isomorphism as claimed. 



Corollary 5.4.2 A morphism f ∈ MorC (A, B), where C is an abelian category, is

an epimorphism if and only if ker(cok(f )) = B.

Proof In the proof of the above proposition we observed that f is an epimorphism

if and only if B = cok(ker(f )). Since ker(cok(f )) ∼ = cok(ker(f )), see Exer-
cise 5.5.10, we obtain the claim. 


A typical technique to obtain results in abelian categories that are analogous to

those in module categories, such as the isomorphism theorems for example (see also
Exercise 5.5.10 below), is the use of diagram lemmas. The difference to working in
a concrete category lies in the fact that, in a general abelian category, there are
no elements to chase around. Hence the arguments become somewhat lengthy; for
that reason we confine ourselves to the sample below, and leave the more elaborate
results, such as the 5 – Lemma, the 3 × 3 – Lemma or the Snake Lemma, to more
comprehensive expositions (see [30], e.g.).
56 5 Artinian Modules

5.4.3 Short Five Lemma Suppose we are given the following commutative dia-
gram in the abelian category C:

m e
0 A B C 0

f g h

0 A B C 0
m e

Suppose both rows are exact and the morphisms f and h are both monomorphisms
(epimorphisms). Then g is a monomorphism (an epimorphism).

Proof Let k = ker(g); then hek = e gk = 0. As h is a monomorphism this yields

ek = 0. Therefore k factors through m = ker(e), say as k = mk . Since 0 = gk =
gmk = m f k and m and f are monomorphisms we obtain k = 0 and thus k = 0,
that is, g is a monomorphism.
The second statement follows by duality, or explicitly as follows. Let = cok(g).
As m f = gm = 0 and f is assumed to be an epimorphism we get m = 0.
Hence factors through e = cok(m ), say = e . Since 0 = g = e g = he
and e and h are epimorphisms we obtain = 0, that is, g is an epimorphism. 


The relation between abelian categories and module categories is in fact rather
intimate. The Freyd–Mitchell theorem states that for every small abelian category
there is an full exact embedding in R-Mod for a suitable ring R. See, e.g., Sect. 4.14
in [30].
There exist various alternative but equivalent ways to define an abelian category,
some even derive the abelian group structure on the Hom-sets from other basic
axioms; see [17, Chap. 2] and [30, Chap. 4]. A very readable account is given in
Chap. 7 of [29].

5.5 Exercises

Exercise 5.5.1 Let R M ∈ R-Mod for a unital ring R. Show that R M is finitely
 if and only if, for every family {R Ni | i ∈ I } of submodules
cogenerated  R Ni of R M
such that i∈I R Ni = 0, there exists a finite subset J ⊆ I such that j ∈J R Nj = 0.

Exercise 5.5.2 Write out the details of the proof of Theorem 5.1.4.
5.5 Exercises 57

Exercise 5.5.3 Let R be an integral domain, and let R M ∈ R-Mod. The torsion
submodule of R M is defined as

tor(R M) = {m ∈ M | AnnR (m) = 0}.

Prove the following statements.

(i) tor(R M) ≤ R M;
(ii) tor(R M/tor(R M)) = 0;
(iii) for f ∈ HomR (R M, R N ), where R N ∈ R-Mod, we have f (tor(R M)) ⊆
tor(R N).

Exercise 5.5.4 In an Artinian commutative ring R, there are only finitely many
prime ideals.

Exercise 5.5.5 Let R be a commutative unital ring. Show that the definitions of the
nil radical of R given in Definition 5.2.3 and in Exercise 1.3.10 agree with each
other.

Exercise 5.5.6 Let K be a field and let V be a K-vector space. Show that the
following three conditions are equivalent.

(a) V is finite dimensional;

(b) V satisfies ACC;
(c) V satisfies DCC.

Exercise 5.5.7 Let R be a commutative unital ring and let M1 , . . . , M be maximal

ideals in R such that M1 M2 · · · M = 0. Show that R is Noetherian if and only if R
is Artinian.

Exercise 5.5.8 Show that an Artinian commutative ring R is Noetherian (compare

Theorem 5.3.2) by employing Propositions 5.2.4 and 5.2.5 together with Exer-
cises 5.5.4 and 5.5.7.

Exercise 5.5.9 Let C be an abelian category. Show that every morphism f ∈

mor(C) has a factorisation f = me, where m = ker(cok(f )) and e = cok(ker(f ))
(In particular, m is a monomorphism and e is an epimorphism. This is the analogue
of the canonical factorisation of module maps (Theorem 3.1.1).)

The next exercise is the analogue of the First Isomorphism Theorem in module
theory (Theorem 3.1.2) in abelian categories.
58 5 Artinian Modules

Exercise 5.5.10 Let C be an abelian category. For each f ∈ mor(C), there is a

canonically associated isomorphism h : cok(ker(f )) −→ ker(cok(f )) which arises
as follows. Starting with the diagram

i p
ker(f ) A cok(ker(f ))

p i
cok(f ) B ker(cok(f ))

for a given f ∈ MorC (A, B), use the definition of kernel and cokernel in
Sect. 4.2.1 in order to obtain unique morphisms g : A −→ ker(cok(f )) and
k : cok(ker(f )) −→ B, respectively with the properties f = i g and f = kp,
respectively. In the commutative diagram

i p
ker(f ) A cok(ker(f ))

g k
f h

p i
cok(f ) B ker(cok(f ))

g may be uniquely factored through cok(ker(f )) by h because 0 = f i = i gi, hence

0 = gi as i is a monomorphism. Likewise k factors through ker(cok(f )) via h .
Using i g = f = kp conclude that h = h and finally employ Proposition 5.4.1 to
prove that h is an isomorphism.

Exercise 5.5.11 Let R be a left Noetherian ring. Show that R-mod is an abelian
category.

Exercise 5.5.12 Repeat Exercise 4.3.11 for an abelian category C in place

of R-Mod.

Exercise 5.5.13 Fill in the details in the following statement: The additive category
Ban∞ is not abelian since the inclusion 1 → c0 is both a monomorphism and an
epimorphism but not an isomorphism. (For the definition of the spaces, see p. 68.)
Simple and Semisimple Modules
6

In this chapter, which is at the very heart of our exposition, we shall apply module
theory to study the structure of a certain class of rings; these will be termed
(classically) semisimple. To this end, we will, once again, introduce some special
classes of modules and investigate their properties.
From this point on, we shall only consider unital rings (but still state this
explicitly) and unital modules. These were introduced in the Exercises in Chap. 1
where it was also shown that this is no restriction of the generality.

6.1 Decomposition of Modules

We turn our attention to the modules that are of the utmost interest to us in this book.

Definition 6.1.1 Let R be a unital ring, and let R M ∈ R-Mod. Then

(i) RM is called simple (or, irreducible) if R M = 0 and R M contains no non-trivial

submodules; that is, 0 and R M are the only submodules of R M.
(ii) R M is called semisimple (or, completely reducible) if every submodule of R M
is a direct summand; that is, for each R N ≤ R M there is R N ≤ R M such that
RM = RN ⊕ RN .

Remark 6.1.2 Identifying R M ∈ R-Mod with the associated representation R →

End(M) (compare Sect. 1.1.4), simplicity of R M is the same as the property that
there are no non-trivial subgroups of (M, +) which are invariant under the R-action.
This interpretation gives rise to the terminology ‘irreducible’.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 59

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_6
60 6 Simple and Semisimple Modules

Examples 6.1.3

1. An abelian group G is simple as a group if and only if it is simple as a Z-module.

The maximal ideals of Z are of the form pZ for a prime number p. Hence, by
Exercise 6.4.1 below, G is simple if and only if G is isomorphic to Zp .
2. Let V be a non-zero vector space over the field K. Put R = EndK (V ). Then
V ∈ R-Mod via f · v = f (v), where f ∈ R, v ∈ V . We show that R V is
simple: take v ∈ V \ {0} and let w ∈ V . Extend {v} to a basis B of V and define
f ∈ EndK (V ) by f (v) = w and f (b) = 0 for each b ∈ B \ {v}. Then f · v = w;
consequently, Exercise 6.4.1 yields the result.

Our next result, though very easy to prove, is fundamental for the application of
simple modules in ring theory. It was first obtained by I. Schur in 1905 in a different
language (see Exercise 6.4.2 below).

6.1.4 Schur’s Lemma Let R be a unital ring, and let R M ∈ R-Mod be simple.
Then EndR (R M) is a division ring.

Proof Since ker(f ) ≤ R M for each f ∈ EndR (R M), either ker(f ) = 0 — in which
case f is injective — or ker(f ) = R M, that is, f = 0. Similarly, im(f ) ≤ R M
implies that im(f ) = R M — in which case f is surjective — or im(f ) = 0, that is
f = 0. It follows that every non-zero endomorphism of R M is bijective, and hence
EndR (R M) is a division ring. 


Remark 6.1.5 Suppose R M is a simple module over a commutative unital ring R.

Then EndR (R M) is in fact a field. By Schur’s Lemma, all we need to show is that
EndR (R M) is commutative. By Exercise 6.4.1, R M ∼ = R/I for a maximal ideal I
of R. Since R/I is a commutative unital ring (in fact, a field), R/I ∼
= EndR/I (R/I );
compare Exercise 2.7.1. Putting this together we find

EndR (R M) ∼
= EndR/I (R/I ) ∼
= R/I

is commutative as I = AnnR (M).

We shall now study the relations between simple and semisimple modules. It
is evident from the definition that every simple module is semisimple. To obtain
information in the other direction, we first record the following useful observation.

Proposition 6.1.6 Every submodule and every quotient of a semisimple module is

semisimple.
6.1 Decomposition of Modules 61

Proof Let R M ∈ R-Mod for a unital ring R be semisimple. Let R N ≤ R M.

Take R L ≤ R N ; by assumption, there is R L ≤ R M such that R L ⊕ R L = R M.
Therefore, setting R L = R N ∩ R L , we get

RL ⊕ RL = R L ⊕ (R N ∩ R L ) = R N ∩ (R L ⊕ R L ) = R N

by the modular law (Exercise 2.7.8). Hence R N is semisimple.

Let f be an R-module map defined on R M and let R L ≤ im(f ). Then R N =
f −1 (R L) ≤ R M and, by hypothesis, R N ⊕ R N = R M for some R N ≤ R M. Since
ker(f ) ⊆ R N , it follows that

im(f ) = f (R M) = f (R N) ⊕ f (R N ) = R L ⊕ f (R N )

and thus im(f ) is semisimple. The first isomorphism theorem (Theorem 3.1.2)
completes the argument. 


Lemma 6.1.7 Let R be a unital ring. Every non-zero semisimple module R M ∈

R-Mod contains a simple submodule.

Proof In case every submodule R N of R M is either 0 or R M, there is nothing to

prove. Thus we can take a non-zero submodule R N ≤ R M together with an element
m ∈ R M \ R N and, without loss of generality, we can assume that m generates R M,
cf. Proposition 6.1.6. Zorn’s Lemma provides us with a maximal submodule R L of
R M containing R N and not containing m. Indeed, the non-empty set N = {R K ≤
R M | R N ⊆ R K and m ∈ / R K} is inductively ordered so has a maximal element R L.
By hypothesis, R M = R L ⊕ R L for some non-zero R L ≤ R M. We claim that R L
is simple. Suppose R L is a non-zero submodule of R L . As R L is maximal in N,
R L ⊕ R L must contain m and hence, R L ⊕ R L = R M. As a result, R L ⊆ R L
wherefore R L = R L and R L is simple. 


With the above preparations at hand, we can now ‘completely reduce’ any
semisimple module into its simple constituents.

Proposition 6.1.8 The following conditions on R M ∈ R-Mod are equivalent:

(a) RM is semisimple;
(b) RM is the direct sum of simple submodules;
(c) R M is the sum of simple submodules.

Proof By convention, 0 = i∈∅ R M i ; 

thus we can without loss of generality
assume that R M = 0. Suppose R M = i∈I R M i where each R M i ≤ R M is
simple. Let R N ≤ R M. Let S be the set of all subsets J ⊆ I such that
 
(i) j ∈J R Mj is a direct sum; and (ii) RN ∩ j ∈J R Mj = 0.
62 6 Simple and Semisimple Modules

Clearly ∅ ∈ S so S = ∅. Since S is inductively ordered, Zorn’s lemma provides us

with a maximal element Jmax . By construction,

RM := R N + R Mj = RN ⊕ R Mj .
j ∈Jmax j ∈Jmax

We will show that R M i ⊆ R M for every i ∈ I which yields R M = R M . For each

i ∈ I,

RM ∩ RMi = RMi or RM ∩ RMi = 0

as each R M i is simple. The latter case implies R M + R M i = R N ⊕ j ∈Jmax R Mj ⊕

R M i and hence, by maximality of Jmax , i ∈ Jmax . The first case also entails that
R M i ⊆ R M and therefore R M = R M.
Choosing R N = 0 in the above argument gives the implication (c) ⇒ (b) and
choosing R N arbitrary yields (b) ⇒ (a).
(a) ⇒ (c) Let R N be the sum of all simple submodules of R M; then R N = 0, by
Lemma 6.1.7. Let R N ≤ R M be such that R N ⊕ R N = R M. By Proposition 6.1.6,
R N is semisimple, however it does not contain any non-zero simple submodule. By
Lemma 6.1.7, R N = 0 and therefore R N = R M. This proves (c). 


Proposition 6.1.9 Suppose that R M ∈ R-Mod is semisimple and that R M =

i∈I R M i is a decomposition of R M into simple submodules R M i . If R N ≤ R M is
simple then R N ∼ = R Mj for some j ∈ I .

Proof
 As R M =  i∈I R M i , we may, and will, consider R M as a submodule of
i∈I R M i . Let pj : i∈I R M i → R Mj be the canonical projection onto the j th
component. Since R N = 0, pj (R N ) = 0 for at least one j ∈ I . Therefore, f =
pj |R N ∈ HomR (R N , R Mj ) is a non-zero homomorphism. Since R Mj is simple,
im(f ) = R Mj ; since R N is simple, ker(f ) = 0. It follows that f is an isomorphism.



It can be shown that the cardinality of the set I in a decomposition such as in

Proposition 6.1.9 is unique; see Exercise 6.4.3 below.
A neat criterion to test a module for semisimplicity, which will turn out to be
useful in Chap. 7, is the following.

Proposition 6.1.10 Let R M ∈ R-Mod be finitely cogenerated. Then R M is

semisimple if and only if the intersection over all maximal submodules of R M is
zero.

Proof Suppose first that R M ∈ R-Mod is semisimple and that R M = i∈I R M i

is a decomposition of R M into simple submodules R M i . For each j ∈ I , set
R Nj =
∼
i=j R M i . Then R Nj is a maximal submodule of R M as R M/R Nj = R Mj
6.1 Decomposition of Modules 63


which is simple. We have j ∈I R Nj = 0, hence the intersection over all maximal
submodules of R M is zero.
Conversely, let {R N i | i ∈ I } be the collection of all maximal submodules
 j ∈ I , put R M i = R M/R N i which is a simple left R-module.
of R M. For each
Suppose that
 i∈I R N i = 0. By Exercise 5.5.1, there is a finite subset J ⊆ I
such that j ∈J R Nj = 0. Consequently, the R-module map R M −→ j ∈J R Mj
is an embedding of R M into a semisimple module, hence R M is semisimple
(Proposition 6.1.6). 


Our next immediate goal is to determine for which rings every (left) module
is semisimple. Clearly, the question for which ring every module is simple does
not make sense, since we can always form direct sums. The key to the answer to
the above question is the following concept for short exact sequences of modules.
(Compare with (3.3.2) on p. 32.)

f g
Definition 6.1.11 Let 0 RM 1 RM 2 RM 3 0 be a short exact
sequence in R-Mod. We say the sequence splits if im(f ) = ker(g) is a direct
summand in R M 2 .

Theorem 6.1.12 The following conditions are equivalent for a unital ring R.

(a) Every short exact sequence in R-Mod splits;

(b) Every R M ∈ R-Mod is semisimple;
(c) Every finitely generated R M ∈ R-Mod is semisimple;
(d) Every cyclic R M ∈ R-Mod is semisimple;
(e) R R is semisimple.

Proof (a) ⇒ (b) Let RN ≤ RMand consider the short exact sequence
0 RN RM R M /R N 0. . By hypothesis, this sequence splits
and hence R N is a direct summand in R M.
(b) ⇒ (c) ⇒ (d) ⇒ (e) are obviously valid.
(b) ⇒ (a) If each submodule of a module in R-Mod is a direct summand so in
particular the image 0 RM 1 R M 2 in a given short exact sequence as in
Definition 6.1.11.
(e) ⇒ (b) Let R M ∈ R-Mod. For m ∈ M, the cyclic module Rm is
semisimple since it is a homomorphic image of the semisimple module R R
 (m)
(Proposition 6.1.6). By Proposition 6.1.8, Rm is thus the sum j ∈Jm R Nj of

simple submodules R Nj(m) . As R M = m∈M Rm, it follows that R M is the sum of
simple submodules and therefore, by Proposition 6.1.8, is semisimple. 


We can now introduce the class of rings this book is centred around.
64 6 Simple and Semisimple Modules

Definition 6.1.13 A unital ring R satisfying any, and hence every, condition in
Theorem 6.1.12 is called (classically) semisimple or completely reducible.

Remarks 6.1.14

1. Strictly speaking we defined ‘left semisimple ring’ as we are working in R-Mod.

An analogous concept of a ‘right semisimple ring’ in Mod-R turns out to be an
equivalent notion (see Corollary 7.1.7) wherefore we drop the specification right
from the start.
2. Theorem 6.1.12 (e) states: R is semisimple if and only if R R is semisimple. The
submodules of R R are the left ideals of R and the simple submodules are the
minimal left ideals. We thus see that a semisimple ring can be reconstructed from
its ‘smallest building blocks’.
3. We emphasise “classically” in the above definition since there is another, more
modern version of “semisimple ring” which was introduced by N. Jacobson. This
concept, which we will briefly discuss in Chap. 7 and in some special cases in
Chap. 10, is also referred to as “semiprimitive” or “J-semisimple” and avoids
inherent finiteness conditions. Therefore it is more suited for applications in
infinite dimensions that are typical in Functional Analysis, for example.

6.2 Projective and Injective Modules

In this section we continue our study of the nice properties that semisimple rings
enjoy; this will lead us to the Artin–Wedderburn theorem in the next chapter. To this
end, we introduce two extremely important classes of modules.

Definition 6.2.1 Let R be a ring. The module R P ∈ R-Mod is said to be projective

if it satisfies the following condition:
Whenever two modules R M, R N ∈ R-Mod and an epimorphism f ∈
HomR (R M, R N ) as well as a homomorphism g0 ∈ HomR (R P , R N) are given,
there exists g ∈ HomR (R P , R M) such that fg = g0 .
(“Every homomorphism from R P can be lifted.”)

g
RM RP

f
g0

RN
6.2 Projective and Injective Modules 65

Definition 6.2.2 Let R be a ring. The module R I ∈ R-Mod is said to be injective

if it satisfies the following condition:
Whenever two modules R M, R N ∈ R-Mod and a monomorphism f ∈
HomR (R N , R M) as well as a homomorphism g0 ∈ HomR (R N, R I ) are given,
there exists g ∈ HomR (R M, R I ) such that gf = g0 .
(“Every homomorphism into R I can be extended.”)

g
RM RI

f
g0

RN

These two concepts from Homological Algebra allow us to formulate the

following characterisation of semisimple rings.

Theorem 6.2.3 The following conditions on a unital ring R are equivalent.

(a) R is semisimple;
(b) every left R-module is projective;
(c) every left R-module is injective.

Proof (a) ⇒ (b) Let R P ∈ R-Mod be part of the following diagram in which f is
an epimorphism

RP

g0

RM RN 0
f
66 6 Simple and Semisimple Modules

We can extend the exact sequence to a short exact sequence as follows

RP

g
g0

0 ker(f ) RM RN 0
f

By Theorem 6.1.12, this sequence splits; by Exercise 6.4.4, there is thus f ∈

HomR (R N , R M) such that ff = idN . Putting g = f g0 ∈ HomR (R P , R M)
we find that fg = ff g0 = g0 ; that is, R P is projective.
f
(b) ⇒ (a) Let 0 RK RM RN 0 be a short exact sequence
in R-Mod. By hypothesis, R N is projective; thus, for g0 = idN , we find f ∈
HomR (R N , R M) with ff = g0 = idN , that is, f is a section for f . It follows that
our sequence splits, and Theorem 6.1.12 yields that R is semisimple.
(a) ⇒ (c) Let R I ∈ R-Mod be part of the following diagram in which f is a
monomorphism

f
0 RN RM

g0

RI

We complete the exact sequence to a short exact sequence

f
0 RN RM R M/ im(f ) 0
f

g
g0

RI

By Theorem 6.1.12, this sequence splits; by Exercise 6.4.4, there is thus f ∈

HomR (R M, R N ) such that f f = idN . Putting g = g0 f ∈ HomR (R M, R I ) we
find that gf = g0 f f = g0 as desired; as a result, R I is injective.
6.2 Projective and Injective Modules 67

f
(c) ⇒ (a) Let 0 RK RM RN 0 be a short exact sequence
in R-Mod. By hypothesis, R K is injective; thus, for g0 = idK , we find f ∈
HomR (R M, R K) with f f = g0 = idK . Therefore the sequence splits, and
Theorem 6.1.12 entails that R is semisimple. 


A major step forward in the direction of the Artin–Wedderburn theorem is

provided by the next result revealing various finiteness properties of semisimple
modules.

Theorem 6.2.4 Let R be a unital ring. The following conditions on a semisimple

module R M ∈ R-Mod are equivalent.

(a) RM is finitely generated;

(b) RM is finitely cogenerated;
(c) R M is Noetherian;
(d) R M is Artinian;
(e) R M is a direct sum of finitely many simple submodules.

Proof The proof is organised in the following way

(c) (a)

(e)

(d) (b)

The implications (c) ⇒ (a) and (d) ⇒ (b) follow from Theorems 4.1 and
5.1.3, respectively. In order to show (a) ⇒ (e), suppose R M = i∈I R M i is
a decomposition of R M into a direct sum of simple submodulesR M i . As R M
is finitely generated, there is a finite subset J ⊆ I such that j ∈J R Mj =
j ∈J R Mj = R M, by Exercise 4.3.7. Since R M i ∩ R Mj = 0 for i = j , it follows
that J = I so that (e) holds.
In order to show (b) ⇒ (e) suppose R M is finitely cogenerated. Let R M =
i∈I R M i be a decomposition of R M into a direct sum of simple submodules R M i .
Fix j ∈ J and put R Nj =  i=j R M i . In this way, we obtain a family of submodules
of R M with the property j ∈I R N j = 0, as R Mj ⊆ R Nj . By Exercise 5.5.1, there
is a finite subset J ⊆ I such that j ∈J R Nj = 0. Let k ∈ I \ J ; then R M k ⊆ R Mj
for any j ∈ J by simplicity. Hence R M k ⊆ R Nj for all j ∈ J . It follows that
R M k = 0. As a result, I \ J = ∅, that is, I = J is finite and (e) holds.
68 6 Simple and Semisimple Modules

Finally suppose that R M = i∈I R M i for a finite set I . If all R M i are simple,
they are both Noetherian as well as Artinian and R M inherits these properties by
Corollaries 4.1.3 and 5.1.6, respectively. This shows (e) ⇒ (c) and (e) ⇒ (d). 


From this theorem, we gain a lot of insight into the structure of a semisimple ring
R as the module R R is clearly finitely generated.

Corollary 6.2.5 Every semisimple unital ring R is both left Noetherian as well as
left Artinian. Moreover, R = ni=1 Li for a finite set {L1 , . . . , Ln } of minimal left
ideals of R.

It is clear that there is a right-handed version for each of the last few results.
However, we aim for a symmetric version of the above corollary which will be
achieved in the next chapter.

6.3 Projective and Injective Objects

In a general category, objects which behave similarly to projective and injective

modules in R-Mod play an important role, especially in Homological Algebra.
They can be defined precisely as in Definition 6.2.1 and in Definition 6.2.2,
respectively, but in an abelian category there is the following equivalent description
in terms of the Hom-functors. Compare with Examples 3.2.2 and Sect. 4.2.3.

Definition 6.3.1 Let C be an abelian category. An object P ∈ ob(C) is called

projective if the functor Hom(P , −) : C → AGr is right exact. An object I ∈
ob(C) is called injective if the functor Hom(−, I ) : C → AGr is right exact. (We
followed here the usual convention to denote the morphism set MorC (A, B), A, B ∈
ob(C) by Hom(A, B) as it is an abelian group.)

Sometimes, to be injective does not require any additional property; e.g., in the
category of vector spaces, every object is injective. On the other hand, in AGr, the
injective objects are precisely the divisible abelian groups and the full subcategory
of finite abelian groups has no projective objects at all. We shall focus on injective
objects in the following, trying to illustrate what they might be good for. Evidently,
projectivity is the dual concept to injectivity.
In some categories the requirement on an injective object according to the above
definition is too restrictive: e.g., in Ban∞ , the category of all Banach spaces,
the categorical definition would imply that C is not injective. To see this, let 1
denote the Banach space of all absolutely summable sequences with the norm
(ξn )n∈N 1 = ∞ n=1 |ξn | and let c0 denote the Banach space of all null sequences
(sequences converging to 0) with the norm (ξn )n∈N ∞ = sup n |ξn |. Define a
∞
contractive linear functional g0 : 1 → C by g0 ((ξn )n∈N ) = n=1 ξn . Upon
1
embedding into c0 canonically we see that g0 cannot be extended to a bounded
linear functional g : c0 → C. Since C is complemented in every non-zero Banach
6.3 Projective and Injective Objects 69

space, it follows that 0 is the only injective object according to the categorical
definition. (Use Exercise 6.4.5 to see this.) An analogous argument shows that 0
is the only projective one, too.
The solution to this problem is to restrict the class of admissible monomorphisms.

Definition 6.3.2 Let M be a class of morphisms in the category C which consists

of monomorphisms in C and contains all isomorphisms. Assume further that M
is closed under composition. The object I ∈ ob(C) is said to be M-injective if it
satisfies the following condition:
Whenever two objects A, B ∈ ob(C) and a monomorphism f ∈ M, f : A → B
as well as a morphism g0 ∈ MorC (A, I ) are given, there exists g ∈ MorC (B, I )
such that gf = g0 .

g
B I

f
g0

The monomorphisms in M are often referred to as embeddings; e.g., the good

embeddings in the category of Banach spaces are the isometric linear mappings.
(Note, however, that with this class M of embeddings, M-injectivity in Ban∞ is
still different from M-injectivity in Ban1 : only the latter one gives the “extension
under preservation of the norm” property which is the preferred one in Functional
Analysis (Hahn–Banach theorem).)
Suppose M is a class of embeddings in the category C. We say C has enough
M-injectives if for every object A ∈ ob(C) there are an M-injective object I and an
embedding ι ∈ MorC (A, I ). For instance, Ban1 has enough M-injectives, where
M is the class of all linear isometries. This is a consequence of the Hahn–Banach
theorem which allows to embed every Banach space into ∞ () for a suitable index
set  (use Exercise 6.4.12 which also holds for non-abelian categories to show that
∞ () is M-injective). For a unital ring R, the category R-Mod has enough M-

injectives (where M is simply the class of all monomorphisms). This follows from
the fact that HomZ (R, G) is injective in R-Mod for every divisible abelian group.
For the details, see [29], Sect. 2.4 or [21], Sect. I.8.
70 6 Simple and Semisimple Modules

Let A ∈ ob(C) be an object in the abelian category C with a fixed class M of

embeddings. An injective resolution of A consists of a family {I n | n ∈ N0 } of M-
injective objects in C together with a family of morphisms {d (n) | n ∈ N0 }, where
d (n) ∈ MorC (I n , I n+1 ), as well as an M-morphism A → I 0 such that the sequence

d (0) d (1) d (2)

0 A I0 I1 I2 ...

is exact. It can be shown that, if C has enough M-injectives, then every object in C
has an injective resolution.
Let F : C → AGr be a left exact functor (compare Sect. 4.2.3), where C is an
abelian category with enough M-injectives. The right derived functors of F, Rn F,
n ∈ N0 , are defined as follows. For an object A in C, choose an injective resolution
as above and consider the complex

F(d (0) ) F(d (1) ) F(d (2) )

0 F(A) F(I 0 ) F(I 1 ) F(I 2 ) ...

and denote by H n (F(I n )) = ker F(d (n) )/ imF(d (n−1) ), n ≥ 1 the nth cohomology
group of the complex. Then Rn F(A) = H n (F(I n )) for n ≥ 1 and R0 F(A) = F (A).
Derived functors are at the heart of Homological Algebra; for more information,
see, e.g., [21] or [29].

6.4 Exercises

Exercise 6.4.1 Let R be a unital ring, and let RM ∈ R-Mod. Show that the
following conditions on R M are equivalent.

(a) RM is simple;
(b) RM is cyclic and every non-zero element m ∈ R M generates R M;
(c) ∼
R M = R/I for some maximal left ideal I of R.

Exercise 6.4.2 (The Original Formulation of Schur’s Lemma) Let G be a group

and let ρ : G → GLn (C) be a finite-dimensional irreducible representation of G.
Put V = Cn and let R = C[G]. Show that the only endomorphisms of R V are
the multiplications by elements of C. (You may use that the only finite-dimensional
division algebra over C is C itself.)

Exercise 6.4.3 Show that the index sets of any two decompositions of a non-
zero semisimple module into a direct sum of simple submodules have the same
cardinality.
6.4 Exercises 71

Exercise 6.4.4 Let R be a unital ring. Let

f g
0 RM1 RM2 RM3 0

be a short exact sequence in R-Mod. Show that the sequence splits if and only if
either of the following conditions holds.

(a) ∃ f ∈ HomR (R M2 , R M1 ) : f f = idM1 (that is, f is a section);

(b) ∃ g ∈ HomR (R M3 , R M2 ) : gg = idM3 (that is, g is a retraction).

With this notation, the following decomposition holds

R M2 = im(f ) ⊕ ker(f ) = ker(g) ⊕ im(g ).

Exercise 6.4.5 In generalising the previous exercise to an abelian category C, we

say that an object E ∈ ob(C) is a retract of F ∈ ob(C) if there exist morphisms
s ∈ Hom(E, F ) and r ∈ Hom(F, E) such that rs = 1E . In this case we call s a
section and r a retraction. An object E ∈ ob(C) is an absolute M-retract if every
ι : E → F in M for any F ∈ ob(C) is a section (where M is a class of embeddings
as in the text above).
Show that every M-injective object is an absolute M-retract. Every retract of an
M-injective object is M-injective. If C has enough M-injectives then every absolute
M-retract is M-injective.

Exercise 6.4.6 Apply the characterisation of semisimple modules to show that

every vector space over a field has a basis.

Exercise 6.4.7 Let R P ∈ R-Mod for a unital ring R. Show that the following
conditions on R P are equivalent.

(a) R P is projective;
(b) R P is (isomorphic to) a direct summand of a free left R-module;
(c) every epimorphism onto R P splits.

Exercise 6.4.8 The covariant Hom-functor HomR (R U , −) : R-Mod → AGr is

left exact for every R U ∈ R-Mod, see Example 4.2.3.1 in Chap. 4. Show that
HomR (R U , −) is right exact if and only if R U is projective.

Exercise 6.4.9 Show that a unital ring R which is simple as a left R-module is a
division ring.
72 6 Simple and Semisimple Modules

Exercise 6.4.10 Let R M ∈ R-Mod for a unital ring R and let n ∈ N. Given
f ∈ EndR (R M n ), let αk be the composition

i f pk
n n
0 RM RM RM RM 0

of the projection pk onto the kth factor with f with the injection i of the th
 Show that f → (αk ) defines a ring isomorphism between EndR (R M )
summand. n

and Mn EndR (R M) .

Exercise 6.4.11 Show that, for every unital ring R and each n ∈ N, the centre
Z(Mn (R)) of Mn (R) is isomorphic to Z(R).

Exercise 6.4.12 Suppose C is an (abelian) category which is bicomplete, that is,

arbitrary products and coproducts exist. Show that, if {Pα } is a family of projective
then α Pα is projective and that, if {Iα } is a family of injective objects
objects in C,
in C, then α Iα is injective.

Exercise 6.4.13 Let n ∈ N. Show that

with the canonical mappings and Zn = Z/nZ, is an injective resolution of Zn but

Zn itself is not injective (as a Z-module).
The Artin–Wedderburn Theorem
7

In this chapter, we shall obtain the full structure theorem for semisimple rings.
This result, due to J. H. M. Wedderburn (1907) for semisimple finite-dimensional
algebras and to E. Artin (1927) in the general case, enables us to determine
completely this class of rings from the more elementary class of division rings.
It is generally regarded as the first major result in the abstract structure theory of
rings. In Sect. 7.2 below, we will briefly discuss the semisimplicity of group rings
via Maschke’s Theorem 7.2.1. A detailed proof of the Hopkins–Levitzki theorem,
already stated in Theorem 5.3.2, is given in Sect. 7.3.

7.1 The Structure of Semisimple Rings

We start by recalling a well-known proposition from Ring Theory.

Proposition 7.1.1 Let R be a unital ring, and let n ∈ N. Every ideal J in Mn (R) is
of the form J = Mn (I ) for some ideal I in R. Thus, Mn (R) is simple if and only if
R is simple.

Proof Evidently, if I ⊆ R is an ideal, then Mn (I ) is an ideal of Mn (R). Conversely,

suppose J ⊆ Mn (R) is an ideal. Let eij , 1 ≤ i, j ≤ n denote the canonical matrix
units, that is, the entries of eij are all 0 but for the ith row and j th column where the
entry is 1. Define I ⊆ R as the set of all a ∈ R such that ae11 ∈ J . Clearly I is an
ideal in R. We aim to show that J = Mn (I ).
Let a = (aij ) ∈ Mn (R). For all i, j, k, , we have eij aek = aj k ei . Thus, if
a ∈ J , choosing i = = 1, we find that aj k e11 ∈ J and therefore aj k ∈ I for all
j, k. Consequently J ⊆ Mn (I ). Conversely, let a = (aij ) ∈ Mn (I ). By definition
aij e11 ∈ J for all i, j . It follows that aij eij = ei1 (aij e11 )e1j ∈ J and thus,
of I ,
a = i,j aij eij ∈ J . Hence Mn (I ) ⊆ J .
The final statement is now clear. 


© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 73

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_7
74 7 The Artin–Wedderburn Theorem

In the next result we collect a number of properties of matrix rings over division
rings.

Theorem 7.1.2 Let D be a division ring, and let R = Mn (D) for some n ∈ N.
Then

(i) R is simple, semisimple, left Artinian and left Noetherian;

(ii) up to isomorphism, R has a unique simple module R V ∈ R-Mod. On it, R
acts faithfully and R R ∼
= R V n;
∼
(iii) EndR (R V ) = D.

Proof Every division ring is simple; hence R is simple by Proposition 7.1.1. Let
V be the n-tuple column space D n , viewed as a right D-vector space. Then R =
Mn (D) acts on the left of V by matrix multiplication, so R V ∈ R-Mod is a faithful
unital left R-module. In fact, Mn (D) can be identified with EndD (VD ) in the usual
way by choosing a basis of VD . An argument analogous to the one in Examples 6.1.3
shows that R V is a simple R-module.
For each i ∈ {1, . . . , n}, let Li denote the left ideal of R consisting of those n×n-
matrices over D whose columns other than the ith are zero. Then R = L1 ⊕. . .⊕Ln
and, since R Li ∼= R V , we get R R ∼ = R V n . By Proposition 6.1.8, R R is semisimple
so the ring R is (left) semisimple. By Corollary 6.2.5, R is thus left Artinian as well
as left Noetherian, which completes the proof of (i).
The uniqueness of R V stated in (ii) now follows from Proposition 6.1.9.
In order to prove (iii), we define a mapping σ : D → EndR (R V ) by v · σ (d) =
v · d, v ∈ V , d ∈ D. Note that we write here endomorphisms on the right in order
to avoid the opposite ring D op ; see Remark 7.1.3 below. Clearly, σ is a unital ring
homomorphism and, since D acts faithfully on V , σ is injective. To show that σ is
⎛ ⎞ ⎛ ⎞
1 d
⎜0⎟ ⎜∗ ⎟
⎜ ⎟ ⎜ ⎟
surjective, let f ∈ EndR (R V ). Then ⎜ . ⎟ f = ⎜ . ⎟ for some d ∈ D and hence,
⎝ .. ⎠ ⎝ .. ⎠
0 ∗
⎛ ⎛ ⎞⎞ ⎛⎛ ⎞ ⎞
⎛ ⎞ ⎛ ⎞ 1 ⎛ ⎞ 1
a1 ⎜ 1 a 0 . . . 0 ⎜0⎟⎟ a 1 . . . ⎜⎜0⎟ ⎟
⎜ .. ⎟ ⎜⎜ . ⎟ ⎜ ⎟⎟ ⎜. ⎟ ⎜⎜ ⎟ ⎟
⎝ . ⎠ f = ⎜⎝ .. ⎠ ⎜ .. ⎟⎟ f = ⎝ .. ⎠ ⎜⎜ .. ⎟ f ⎟
an
⎝
an
0 ⎝ . ⎠⎠
an
0 ⎝⎝ . ⎠ ⎠
0 0
⎛ ⎞
⎛ ⎞ d ⎛ ⎞ ⎛ ⎞
a1 . . . ⎜ ⎟ a1 d a1
⎜ .. ⎟ ⎜ ∗⎟ ⎜ .. ⎟ ⎜ .. ⎟
=⎝. ⎠ ⎜ .. ⎟ = ⎝ . ⎠ = ⎝ . ⎠ σ (d)
an
0 ⎝.⎠
an d an
∗

so that f = σ (d). 

7.1 The Structure of Semisimple Rings 75

Remarks 7.1.3

1. By Schur’s Lemma (6.1.4), we know that EndR (R V ) is a division ring as R V is

simple. The point in the above theorem is that this division ring is the one we
start with.
2. The notational change in the proof of Theorem 7.1.2 to write endomorphisms
on the right allows us to avoid the opposite ring D op . It occurs naturally here
since right multiplication ρr : x → xr is a left R-module map on R and the
composition ρs ρr of ρr and ρs is ρrs and not ρsr so that the map r → ρr is not a
ring homomorphism.

Example 7.1.4 Let D1 , . . . , D be division rings and let n1 , . . . , n ∈ N. Then

R = Mn1 (D1 ) × . . . × Mn (D ) is a unital semisimple ring. This follows from
Theorem 7.1.2 together with the fact that every finite direct product of semisimple
rings is again semisimple.

Historical Note Joseph Henry Maclagan Wedderburn was born in 1882 in Scotland
as the 10th of 14 children. He entered the University of Edinburgh at the age of
16 and was elected Fellow of the Royal Society of Edinburgh when he was only
21 years old. After further studies at Leipzig, Berlin and Chicago he returned to
Edinburgh in 1905 and was awarded his DSc and his PhD in 1908. Wedderburn
served in the First World War between 1914 and 1918 after which he became
an Associate Professor at Princeton in 1921. Among his (three) PhD students at
Princeton was Nathan Jacobson. He made important advances in the theory of rings,
algebras and matrix theory; among these is his celebrated result that every finite
division ring is a field which also has important consequences for finite projective
geometry. His 1907 paper “On hypercomplex numbers" contains the structure
theorem for finite-dimensional semisimple algebras. See also Artin’s paper [5].
Wedderburn died in Princeton in October 1948.
https://www-history.mcs.st-andrews.ac.uk/Biographies/Wedderburn.html

We shall now see that the above Example 7.1.4 already exhausts all possibilities.

7.1.5 Artin–Wedderburn Theorem Every semisimple unital ring is isomorphic to

a finite direct product of matrix rings over division rings.

Proof Let R be a semisimple unital ring. By Corollary 6.2.5, there are finitely many
minimal left ideals L1 , . . . , Ln in R such that

R = L1 ⊕ . . . ⊕ Ln .
76 7 The Artin–Wedderburn Theorem

Some of the Li ’s may be isomorphic to each other as left R-modules; from each
isomorphism class we pick one representative and denote this simple submodule of
R R by R Mj (so that R Mj =
∼
 R M i for j = i). Suppose we have such isomorphism
classes. Then there exist n1 , . . . , n ∈ N such that

RR
∼ n n
= R M 11 ⊕ . . . ⊕ R M . (7.1.1)

The R M ni i are called the homogeneous components of R. By Schur’s Lemma (6.1.4),

EndR (R M i ) = Di is a division ring for each 1 ≤ i ≤ . Note that EndR (R M ni i ) ∼
=
Mni (EndR (R M i )) = Mni (Di ) for each 1 ≤ i ≤ by Exercise 6.4.10.
From identity (7.1.1) we obtain

 
EndR (R R) ∼
= EndR
ni
RMi
i=1

 nj

ni
= HomR RMi , R Mj
i=1 j =1

 
∼
= HomR
ni
RMi ,
nj
R Mj
i=1 j =1

 nj

ni
= HomR RMi , R Mj
i=1 j =1

 nj 
∼
= HomR ni
R M i , R Mj ,
i=1 j =1

where we used the universal properties of the direct sum and the direct product of
modules and the fact that these constructions coincides with each other if the index
set is finite, see Exercise 2.7.10.
We now apply a similar chain of arguments to compute the individual homomor-
phism groups.

 n 
 ni nj 
HomR R M ni i , R Mj j = HomR RMi , R Mj

ni nj
 
∼
= HomR R M i , R Mj
⎧ ni ni
⎨  
= HomR RMi , RMi if i = j
⎩
0 if i = j,
7.1 The Structure of Semisimple Rings 77

 
where we used that HomR R M i , R Mj = 0 for i = j since both modules are simple
and non-isomorphic. Putting these two computations together we find

 ni 
EndR (R R) ∼
ni
= HomR RMi , RMi
i=1

= EndR (R M ni i )
i=1

∼
= Mni (Di ).
i=1

Finally observe that R ∼

= EndR (R R) via r → ρr if we write the endomorphisms on
the right, compare Exercise 2.7.1. Consequently we arrive at

R∼
= Mn1 (D1 ) × . . . × Mn (D )

which was our aim. 



A representation as in the above result is unique in various ways, as we shall see

below in Theorem 7.1.9 and in Exercise 7.4.1.

Remark 7.1.6 If we want to avoid writing endomorphisms on the right in the above
proof, we obtain instead

R∼
= EndR (R R)op ∼
op
= Mni (Di )op = Mni (Di ),
i=1 i=1

where the last equality is given by the transpose mapping.

Corollary 7.1.7 A unital ring is left semisimple if and only if it is right semisimple.

Proof This follows from the fact that we have an analogous theory of semisimple
right modules and right semisimple rings leading to the same Artin–Wedderburn
Theorem which is a symmetric statement. 


Corollary 7.1.8 A commutative unital ring is semisimple if and only if it is a finite

direct product of fields.

Proof While the “if”-part is immediate, the “only if”-part follows from Theo-
rem 7.1.5 together with the fact that Mn (D) is commutative if and only if n = 1
and D is commutative. 

78 7 The Artin–Wedderburn Theorem

In addition, we have a very strong form of uniqueness in the decomposition in

the Artin–Wedderburn Theorem.

7.1.9 Uniqueness Theorem for Semisimple Rings Let R be a unital ring and
suppose
n m
R= Ri = Rj
i=1 j =1

are two decompositions with simple rings Ri , Rj for 1 ≤ i ≤ n, 1 ≤ j ≤ m. Then

n = m and for every i there is a unique j such that Ri = Rj .

Proof We will identify each Ri and Rj with a two-sided ideal in R in the canonical
way. Then Ri = RRi = Ri R and Rj = RRj = Rj R for all i, j . Therefore

m
Ri = Ri R = Ri Rj (1 ≤ i ≤ n).
j =1

Since each Ri Rj is thus an ideal of Ri , and Ri is simple, we have Ri Rj = 0 or

Ri Rj = Ri . As Ri = 0, there is some j such that Ri Rj = 0 and thus Ri Rj = Ri .
On the other hand, Ri Rj = Rj as Rj is simple so that Ri = Rj .
In the identification Rj → 0 × . . .× Rj × 0 . . .× 0, all Rj s are distinct from each
other, as are the Ri s. Therefore, for each Ri there can be only one Rj with Ri = Rj ;
this entails that n = m. 


Putting Theorems 7.1.5 and 7.1.9 together we arrive at a very satisfactory result
illustrating in a beautiful way what the structure theory of rings is about. This result
is at the very heart of our book.

Corollary 7.1.10 A unital ring R is semisimple if and only if it is the unique finite
direct product of simple Artinian rings R1 , . . . , Rn .

Note that we must insert “Artinian” in the sufficiency condition as not every simple
ring is semisimple; this will be illustrated in the example below.

Example 7.1.11 Let D be a division ring (for a concrete example, take D = Q). For
each n ∈ N, let Rn = M2n (D) which is a simple unital ring by Proposition 7.1.1.
We define unital ring monomorphisms Rn → Rn+1 by
!
A 0
M (D)  A −→
2n ∈ M2n+1 (D).
0 A
7.1 The Structure of Semisimple Rings 79

Identifying
∞ Rn with a unital subring of Rn+1 in this way, we put R = M2∞ (D) =
n=1 Rn . (This is an instance of a “direct limit” of a directed system of rings.)
Clearly, R is a unital ring.
In order to show that R is simple, let I be a non-zero ideal of R. Then there is
some Rn such that I ∩ Rn = 0. As I ∩ Rn is an ideal of Rn and Rn is simple, we
obtain I ∩ Rn = Rn , that is, Rn ⊆ I . Since 1R = 1Rn ∈ Rn it follows that I = R.
We next show that R is not left Artinian (and hence cannot be semisimple, by
Corollary 6.2.5). For n ∈ N, let en denote the matrix unit e11 in Rn . Then en+1 =
en+1 en ∈ Rn+1 . For instance,
!
10
e1 in R1 = M2 (D) is ,
00
⎛ ⎞
1000
⎜0 0 0 0⎟
e1 in R2 = M4 (D) is ⎜ ⎟
⎝0 0 1 0⎠ ,
0000
⎛ ⎞
1000
⎜0 0 0 0⎟
e2 in R2 = M4 (D) is ⎜ ⎟
⎝0 0 0 0⎠ ,
0000

so e2 = e2 e1 . In this way, we obtain a descending chain of left ideals

Re1 ⊇ Re2 ⊇ . . . ⊇ Ren ⊇ Ren+1 ⊇ . . .

Suppose en ∈ Ren+1 for some n. Then en = aen+1 where a ∈ Rj = M2j (D) for
some j > n. The (2n +1, 2n +1)-entry of en in Rn+1 is 1, however the (2n +1, 2n +
1)-entry of aen+1 is 0, which is impossible. As a result, the descending chain cannot
become stationary and R is not left Artinian.

As we shall see now, the Artinian property is the one that turns a simple ring into
a semisimple one.

Proposition 7.1.12 For every simple unital ring R, the following conditions are
equivalent.

(a) R is semisimple;
(b) R is left Artinian;
(c) R = Mn (D) for some division ring D and n ∈ N.

Proof The implications (c) ⇒ (a) and (c) ⇒ (b) are provided by Theorem 7.1.2.
(a) ⇒ (b) is Corollary 6.2.5 and (a) ⇒ (c) follows from Theorem 7.1.5.
80 7 The Artin–Wedderburn Theorem

(b) ⇒ (a) If R R is Artinian it contains a simple submodule by Theorem 5.1.3, see

also Exercise 7.4.2. Hence R contains a minimal left ideal. Define the socle soc(R)
of R as the sum of all minimal left ideals (compare Exercise 7.4.3). As soc(R)
is a non-zero ideal in the simple ring R, we obtain R = soc(R), that is, R R is a
sum of simple submodules. By Proposition 6.1.8, R R is semisimple so that R is a
semisimple ring. 


This result shows that a right Artinian simple ring must be left Artinian (and vice
versa) since a similar characterisation as in Proposition 7.1.12 is available. For this
reason, one simply speaks of ‘Artinian simple rings’.

Remark 7.1.13 For an approach to the Artin–Wedderburn theorem from the view-
point of abelian categories and Morita equivalence the reader is invited to consult
[30, Sect. 4.12].

7.2 Maschke’s Theorem

In this short section we shall discuss an important class of semisimple rings that is
related to group representations. In Sect. 1.1.5 we indicated the connections between
finite-dimensional representations of a group G and modules over the group ring
K[G], where K is a field. Indeed, given a finite group G and a field K, there is
a one-to-one correspondence between K-linear representations of G and finitely
generated left K[G]-modules. It turns out that the relation between the order of G
and the characteristic of K determines the structure of the group ring.

Theorem 7.2.1 (Maschke) Let K be a field with characteristic q (finite or zero).

Let G be a finite group such that q does not divide |G|. Then the group ring K[G]
is semisimple.

i
Proof Set R = K[G] and let the sequence 0 RN R M in R-Mod be
exact. As N and M are K-vector spaces, we can define a linear splitting p0 : M →
N easily: simply extend a basis of N to a basis of M and define p0 by p0 (i(b)) = b
for all basis elements in N and p0 = 0 otherwise. Then p0 i = idN . We intend to
“upgrade” p0 to a splitting map p ∈ R-Mod.
To this end, suppose that |G| = n and define p : M → N by

1
p(m) = g · p0 (g −1 m) (m ∈ M).
n
g∈G
7.3 The Hopkins–Levitzki Theorem 81

For all h ∈ G, we have

1 1
p(hm) = g · p0 (g −1 hm) = h(h−1 g) · p0 ((h−1 g)−1 m)
n n
g∈G g∈G

1 
=h· (h−1 g) · p0 ((h−1 g)−1 m)
n −1
h g∈G

= h · p(m)

as every element of G is uniquely expressed in the form h−1 g. Consequently, p ∈

HomR (R M, R N ). Since R N ≤ R M, g −1 m ∈ R N for every m ∈ R N ; hence, for
such m,

1 1 1
p(i(m)) = g · p0 (i(g −1 m)) = g(g −1 m) = m = m.
n n n
g∈G g∈G g∈G

Thus pi = idN and therefore R N is a direct summand of R M. By Theorem 6.1.12,

K[G] is a semisimple ring. 


On the other hand, it is not too hard to show that, if the characteristic of K does
divide |G|, then K[G] cannot be semisimple; see, e.g., [7, Proposition 4.1.5].
Since for every algebraically closed field K, the endomorphism ring
EndK[G] (R M) of a simple left K[G]-module R M is isomorphic to K
(Exercise 7.4.17), a combination of Theorems 7.1.5 and 7.2.1 yields the following
result.

Corollary 7.2.2 Let G be a finite group. Then the group ring C[G] is isomorphic
to a finite direct sum of matrix rings over C.

A comprehensive discussion of the semisimplicity of group rings can be found

in [24] and [31].

7.3 The Hopkins–Levitzki Theorem

We stated the Hopkins–Levitzki theorem (Theorem 5.3.2) already two chapters

ago but found that a proof would be better placed after a thorough discussion of
semisimple rings. Let us emphasise that ring will mean unital ring throughout.
82 7 The Artin–Wedderburn Theorem

We start with a generalisation of Definition 5.2.3 to the non-commutative setting.

Definition 7.3.1 For a ring R,


rad(R) = {L | L maximal left ideal in R}

is called the Jacobson radical of R.

Analogously, one could define a ‘right-handed’ version as the intersection of all

maximal right ideals. However, this is unnecessary as the next result shows.

Proposition 7.3.2 In every ring R, the following sets agree with each other:

(i) the intersection of all maximal left ideals;

(ii) the intersection of all maximal right ideals;
(iii) {y ∈ R | 1 − xy is invertible for all x ∈ R};
(iv) {y ∈ R | 1 − yx is invertible for all x ∈ R};
(v) {y ∈ R | 1 − xyz is invertible for all x, z ∈ R}.

In particular, the Jacobson radical rad(R) is an ideal of R.

Proof We start with an observation which is either attributed to Jacobson or

Kaplansky: for all x, y ∈ R,

1 − xy is invertible ⇐⇒ 1 − yx is invertible .

By symmetry, we only have to verify one implication, and the observation will
establish the equivalence of (iii) and (iv) in the proposition.
Suppose that (1 − xy)u = 1 = u(1 − yx) for some u ∈ R. Then

(1 − yx)(yux + 1) = yux − yxyux + 1 − yx

= yux − y(u − 1)x + 1 − yx = 1,
(yux + 1)(1 − yx) = yux + 1 − yuxyx − yx
= yux + 1 − y(u − 1)x − yx = 1

show that 1 − yx is invertible with inverse yux + 1. Clearly (v) implies (iii), so
assume (iii) holds. Then 1 − zxy is invertible for all z, x ∈ R whence, by the
observation, 1 − xyz is invertible, which was to show.
Suppose y ∈ R is not in the intersection of all maximal left ideals of R and let
L be a maximal left ideal such that y ∈ / L. Then L + Ry = R and there is x ∈ R
with 1 − xy ∈ L. It follows that 1 − xy cannot be (left) invertible as L = R which
shows that the set described in (iii) is contained in the intersection of all maximal
left ideals. On the other hand, if 1 − xy is not invertible for some x ∈ R then the left
7.3 The Hopkins–Levitzki Theorem 83

ideal R(1 − xy) is proper, thus contained in a maximal left ideal L (by a standard
application of Zorn’s lemma). It follows that y ∈ / L which establishes the reverse
inclusion.
The equality of the sets in (ii) and (iv) is shown analogously. 


Corollary 7.3.3 For every ring R, we have


rad(R) = {AnnR (M) | R M ∈ R-Mod simple}.

Proof Let R M ∈ R-Mod be simple and take y ∈ R. If ym = 0 for some m ∈ M

then Rym = M and thus, there is x ∈ R with (1 − xy)m = 0. As m = 0 it follows
that 1 − xy is not invertible so y ∈
/ rad(R). Conversely, let L be a maximal left ideal
of R. Then R M = R/L is a simple left R-module. If y ∈ R satisfies yM = 0 then
y ∈ L which proves “⊇” above. 


Definition 7.3.4 A ring R is called semiprimitive if rad(R) = 0, and it is called left

primitive if it has a faithful simple left module.

By Corollary 7.3.3, every left primitive ring is semiprimitive, and by Proposi-

tion 7.3.2, ‘semiprimitive’ is left-right symmetric. Analogously, one can define a
right primitive ring but it turns out that this, in general, is a concept different from
left primitive; cf. [24, Sect. 11].

Remark 7.3.5 Semiprimitive rings are also called Jacobson semisimple, or J-

semisimple for short. In Analysis, where one is mostly interested in infinite-
dimensional algebras, which hardly satisfy any finiteness conditions on ideals, it has
become customary to call semiprimitive algebras semisimple and the older version
classically semisimple. See, e.g., [1], Sect. 5.3.

Corollary 7.3.6 For every ring R, the quotient ring R/rad(R) is semiprimitive.

Proof Let us denote R/rad(R) by R̂ and its elements by x̂ etc. Note that rad(R)
is a proper ideal as 1 ∈ / rad(R). Take y ∈ rad(R̂); by Proposition 7.3.2, 1 − x̂ ŷ
is invertible in R̂ for all x̂ ∈ R̂. Let u ∈ R be such that û(1 − x̂ ŷ) = 1 in R̂.
Then u(1 − xy) = 1 + r for some r ∈ rad(R). As 1 + r is invertible in R, by
Proposition 7.3.2, u(1 − xy) and hence 1 − xy is left invertible for all x ∈ R. A
similar argument shows that 1 − xy is right invertible, thus invertible for all x ∈ R
which implies that y ∈ rad(R). As a result, ŷ = 0. 


By Corollary 7.3.3, every simple left R-module is a simple left R̂-module, where
R̂ = R/rad(R). Conversely, every simple left R̂-module is a simple left R-module
in a canonical way. (See Exercise 2.7.3.)
84 7 The Artin–Wedderburn Theorem

The reader may wonder why we introduced the Jacobson radical en route to a
proof of the Hopkins–Levitzki theorem. The reason will be revealed to them now.

Proposition 7.3.7 For every ring R, the following properties are equivalent.

(a) R is semisimple;
(b) R is left Artinian and semiprimitive.

Proof (a) ⇒ (b) Every semisimple ring is left Artinian, by Corollary 6.2.5, so we
only need to show that rad(R) = 0. Since rad(R) is a direct summand of R R, there
is a left ideal L ⊆ R such that L ⊕ rad(R) = R. Let e ∈ L, f ∈ rad(R) be such that
e + f = 1. Then e2 + f e = e and, as f e ∈ rad(R) ∩ L = 0, it follows that e2 = e.
Similarly, f 2 = f so that e and f are orthogonal idempotents in R. As e = 1 − f is
invertible, it follows that e = 1 and so f = 0. We conclude that rad(R) = Rf = 0
as claimed.
(b) ⇒ (a) We first observe the following: let L be a non-zero left ideal of R. As
R is left Artinian, L contains a minimal left ideal (use Theorem 5.1.3 (c)). Every
minimal left ideal N of R is a direct summand of R R. Indeed, since rad(R) = 0 and
N = 0 there exists a maximal left ideal K of R not containing N. Then N ∩ K = 0
and R R = N ⊕ K.
Now suppose that R is not semisimple. Take a minimal left ideal N1 in R and
write R R = N1 ⊕ K1 for a maximal left ideal K1 as above. As K1 = 0 there
exists a minimal left ideal N2 ⊆ K1 . As before, N2 is a direct summand of R R and
thus of K1 (use Exercise 2.7.8). We therefore can write K1 = N2 ⊕ K2 for a left
ideal K2  K1 . By induction, we obtain a strictly descending chain of left ideals
K1  K2  K3  . . . which contradicts the assumption that R is left Artinian.  

The fact that every semisimple ring is semiprimitive can also be deduced from
the Artin–Wedderburn theorem together with Exercise 7.4.7 since simple rings are
trivially semiprimitive. Of course, the argument above uses a less heavy tool.
We have now taken a substantial step towards the Hopkins–Levitzki theorem. If
R is left Artinian with rad(R) = 0 then, by the above proposition, R is semisimple
and hence left Noetherian, by Corollary 6.2.5. In particular, for any left Artinian ring
R, R/rad(R) is left Noetherian, by Corollaries 5.1.5 and 7.3.6. But, in the statement
of Theorem 5.3.2, there is no assumption on the Jacobson radical so what seems to
be an obstruction will have to be removed. We shall next see how an additional tool
enables us to achieve this goal.
The idea is to use a short exact sequence of modules like

0 rad(R) RR R/ rad(R) 0 (7.3.1)

together with Theorem 4.1.1 to obtain that R R is Noetherian from the same
properties of the outer modules. However, a priori, we do not know that rad(R)
is Noetherian, or even just finitely generated (though it will be once we have shown
7.3 The Hopkins–Levitzki Theorem 85

that R is left Noetherian). The remedy is to work with a suitable power of the radical
instead.
Throughout the remainder of this section we will denote rad(R) by J and
R/rad(R) by R̂.
For some n ∈ N, we replace (7.3.1) by

0 J n− 1 R R/J n −1 0 (7.3.2)

and use that, if R is left Artinian, then R̂ is semisimple. The choice of n will become
clear immediately.
Call an ideal I of R nil if every element in I is nilpotent and call I nilpotent if
I n = 0 for some n ∈ N.

Lemma 7.3.8 For every ring R, the Jacobson radical J contains each nil ideal of
R. If R is left Artinian then J is nilpotent.

Proof Let z ∈ R be nilpotent, say, zk = 0. Then 1 − z is invertible with inverse

1 + z + z2 + . . . + zk−1 . Let I be a nil ideal in R and take y ∈ I . Then, for all
x ∈ R, xy ∈ I is nilpotent so 1 − xy is invertible. By Proposition 7.3.2, y ∈ rad(R)
so I ⊆ rad(R).  
Suppose R is left Artinian. The descending chain J k k∈N becomes stationary,
say at n. Put I = J n and suppose that I = 0. As I 2 = I the set

{L ⊆ R | L is a left ideal and I L = 0}

is not empty, thus contains a minimal element, say L0 (Theorem 5.1.3). Take y ∈ L0
such that Iy = 0. Since I Iy = 0 and Iy ⊆ L0 it follows that Iy = L0 . Therefore
y = xy for some x ∈ I which implies that (1 − x)y = 0. Since x ∈ J n ⊆ J , 1 − x
is invertible so that y = 0, a contradiction. We conclude that J n = I = 0. 


The reader may want to compare the above proof with the argument in Propo-
sition 5.2.5; see also Exercise 7.4.8. We observe that, since every nilpotent ideal is
nil, the Jacobson radical of a left Artinian ring is the largest nil ideal. On the other
hand, it is the smallest ideal such that the corresponding quotient is semisimple, as
we shall observe now.

Proposition 7.3.9 In a left Artinian ring R, the Jacobson radical is the smallest
ideal of R whose corresponding quotient is semisimple.

Proof By Corollary 7.3.6 and Proposition 7.3.7, R/J is semisimple. Conversely, let
K ⊆ R be an ideal such that R/K is a semisimple left R-module, and let ρ : R →
R/K be the canonical epimorphism. As R/K is semisimple, there is a left ideal W
in R, containing K, such that ρ(J ) ⊕ ρ(W ) = R/K. Suppose that W = R. Then
there is a maximal left ideal L of R which contains W . By definition, J ⊆ L so
86 7 The Artin–Wedderburn Theorem

J + W ⊆ L. On the other hand, J + W = R which is a contradiction. As a result,

W = R and ρ(J ) = 0, in other words, J ⊆ K. 


Note also that, if K is an ideal of the left Artinian ring R and J ⊆ K, then R/K
is semisimple, as a homomorphic image of R/J .
We need another auxiliary result which can be obtained, for instance, using the
convenient characterisation of the Jacobson radical contained in Proposition 7.3.2.

Lemma 7.3.10 Let I be an ideal of the ring R such that I ⊆ J . Then rad(R/I ) =
J /I .

Proof Let π : R → R/I and σ : R/I → R/J be the canonical epimorphisms.

Take y ∈ rad(R); then 1 − xy is invertible in R for all x ∈ R and thus, 1 −
π(x)π(y) is invertible in R/I for all π(x). By Proposition 7.3.2, π(y) ∈ rad(R/I )
so rad(R)/I ⊆ rad(R/I ).
Take y ∈ R such that π(y) ∈ rad(R/I ). Then 1 − π(x)π(y) is invertible in R/I
and hence, 1 − σ (π(x))σ (π(y)) is invertible in R/J . We have seen in the proof of
Corollary 7.3.6 that this implies y ∈ J and thus, π(y) ∈ J /I . This establishes the
reverse inclusion rad(R/I ) ⊆ J /I . 


The above result will allow us to control the quotient R/J n−1 in (7.3.2). In order
to determine the behaviour of J n−1 as an R-module, we first need a definition.

Definition 7.3.11 Let I be an ideal of the ring R. We put rad(I ) = rad(R) ∩ I and
call it the Jacobson radical of I .

Clearly, this extends Definition 7.3.1 to non-unital rings. We collect some

properties of this radical. Recall that we put J = rad(R). The first observation
addresses the fact that a proper ideal has to be treated as an R-module, due to the
lack of an identity.

Lemma 7.3.12 For every ideal I of R, its radical is given by


rad(I ) = {L | L maximal submodule of R I }. (7.3.3)

Proof Let us abbreviate the set on the right-hand side of (7.3.3) by J . Note at first
that, by definition,

rad(I ) = J ∩ I = {L ∩ I | L maximal left ideal of R}

= {L ∩ I | L maximal left ideal of R and I  L},
7.3 The Hopkins–Levitzki Theorem 87

as I ⊆ L is tantamount to L ∩ I = I . Let L ⊆ R be a maximal left ideal such that

I  L. Then, L + I = R and thus, by Theorem 3.1.3,

R/L = (L + I )/L ∼
= I /(L ∩ I )

and since R/L is a simple left R-module, so is I /(L ∩ I ). By Proposition 2.2.1,

L ∩ I is a maximal submodule of I .
Now let L ⊆ I be a maximal submodule of R I . Then L ⊆ L for some maximal
left ideal L of R. Since L = L ∩ I ⊆ L ∩ I the maximality of L entails that, either
L = L ∩ I , or L ∩ I = I . Either case yields

{L ∩ I | L maximal left ideal of R and I  L} = J ,

which was to show.

If R I does not contain any maximal submodules then J = I . Moreover, every
maximal left ideal L of R has to contain I so I ⊆ J which implies that J ∩ I =
I = J in this case too. 


Combining this lemma with Proposition 6.1.10 we obtain the next result.

Corollary 7.3.13 An ideal I of a left Artinian ring R is semisimple (in R-Mod) if

and only if rad(I ) = 0.

Proposition 7.3.14 For every ideal I of a left Artinian ring R, we have rad(I ) = J I
and I /rad(I ) is semisimple.

Proof We prove the second statement first. From Theorem 3.1.3 we obtain

I /rad(I ) = I /(J ∩ I ) ∼
= (I + J )/J −→ R/J (7.3.4)

which together with Propositions 7.3.7 and 6.1.6 yields that I /rad(I ) is semisimple.
Clearly, J I ⊆ J ∩ I . By Exercise 7.4.10, rad(I /J I ) = rad(I )/J I so, in order to
establish the reverse inclusion, it suffices to show that rad(I /J I ) vanishes. Set M =
I /J I ; this is an Artinian left R-module, by Corollary 5.1.5. We can regard M as a
left R/J -module (Exercise 2.7.3 and Corollary 7.3.3); since R/J is a semisimple
ring (Corollary 7.3.6 and Proposition 7.3.7) it follows that M is a semisimple R/J -
module (Theorem 6.1.12) and hence semisimple as an R-module and an R/J I -
module (Exercise 2.7.3 again). By Corollary 7.3.13 above, rad(I /J I ) = 0 which
was to prove. 

88 7 The Artin–Wedderburn Theorem

Remark 7.3.15 Let I be an ideal in a left Artinian ring R. The embedding (as rings)
of I /rad(I ) in (7.3.4) enables us to consider I /rad(I ) as an ideal in R/J ; call this
Ī . Then

rad(I /rad(I )) = rad(Ī ) = Ī ∩ rad(R/J ) = 0,

by Corollary 7.3.6. Alternatively, this can be deduced from Exercise 7.4.10.

We are now in a position to provide a proof of the Hopkins–Levitzki theorem.

Proof of Theorem 5.3.2 Let R be a left Artinian ring and let n ∈ N be the least
integer k such that J k = 0 (Lemma 7.3.8). We prove that R is left Noetherian
by induction on n. Let n = 1; then J = 0 and the assertion follows from
Proposition 7.3.7 as every semisimple ring is left Noetherian (Corollary 6.2.5).
Now assume that n > 1 and, for every 1 ≤ k < n and every left Artinian
ring S with rad(S)k = 0, we have S is left Noetherian. We will apply the short
exact sequence (7.3.2) with S = R/J n−1 . By Corollary 5.1.5, S is left Artinian. By
 n−1
Lemma 7.3.10, rad(S) = rad(R)/J n−1 and hence, rad(S)n−1 = J /J n−1 = 0.
Thus, by induction hypothesis, S is left Noetherian.
It remains to show that J n−1 is left Noetherian and then to apply Theorem 4.1.1
to complete the proof. Clearly, J n−1 is left Artinian and, by Proposition 7.3.14,
rad(J n−1 ) = J J n−1 = 0 and J n−1 is semisimple (both as a left R/J - and a left
R-module). By Theorem 6.2.4, J n−1 is left Noetherian, which completes the claim.



Remark 7.3.16 The above proof is inspired by the arguments in Sect. 15 of [2]
where the theory of radicals of modules is discussed in detail.

Remark 7.3.17 The most commonly found proof of the Hopkins–Levitzki theorem
uses composition series and relies on the Jordan–Hölder theorem. Let R be a ring.
A composition series of a left R-module M is a descending chain of submodules

M = M0 ⊃ M1 ⊃ M2 ⊃ . . . ⊃ M = 0

such that each quotient Mi−1 /Mi is simple. In this case, is the length of the
composition series and the simple modules are called the composition factors. The
Jordan–Hölder theorem states that any two composition series of a module have the
same length and the composition factors are, up to permutation and isomorphism,
uniquely determined. This allows one to introduce the length of a module as the
length of a composition series, if such exists. It can be shown that a module has
a finite composition series if and only if it is both Artinian and Noetherian. The
7.4 Exercises 89

composition series that one uses in a proof of the Hopkins–Levitzki theorem derive
from the chain of inclusions

R ⊃ J ⊃ J2 ⊃ ... ⊃ Jn = 0

as each quotient J i /J i+1 is semisimple Artinian, so has a composition series

itself. For details, we refer the reader to Sects. 2.5 and 3.3 in [7], in particular
Theorem 3.3.5, and Chaps. 5 and 14 in [8].
The method using composition series yields versions of the Hopkins–Levitzki
theorem for modules too; see, e.g., Theorem 4.15 in [24].

7.4 Exercises

Exercise 7.4.1 (Uniqueness Theorem for Semisimple Modules) Let RM ∈

R-Mod for a unital ring R. Suppose that
n m
RM = R Mi = R Mj
i=1 j =1

are two direct sum decompositions of R M with simple summands R Mi ≤ R M and

R Mj ≤ R M. Then n = m and there is a permutation π of {1, . . . , n} such that
∼
R Mi = R Mπ(i) for each 1 ≤ i ≤ n.

Exercise 7.4.2 Show that every non-zero Artinian module contains a simple sub-
module.

Exercise 7.4.3 For a ring R, define its socle soc(R) as the sum of all minimal left
ideals of R and if R has no minimal left ideals, put soc(R) = 0. Show that soc(R)
is an ideal of R.

Exercise 7.4.4 For a ring R, an essential left ideal L of R is defined by the property
that L ∩ J = 0 for every non-zero left ideal J of R. Using Zorn’s Lemma show that,
for every left ideal I of R there is a left ideal I of R such that I ⊕ I is an essential
left ideal.

Exercise 7.4.5 Let R be a ring with non-zero socle soc(R). Using Zorn’s Lemma
show that, for each x ∈ R \ soc(R), there is an essential left ideal L of R containing
soc(R) such that x ∈ / L. Use this to prove that soc(R) is equal to the intersection of
all essential left ideals of R.

Exercise 7.4.6 Show that a unital ring R is semiprimitive if and only if it has a
faithful semisimple left module.
90 7 The Artin–Wedderburn Theorem

Exercise 7.4.7 Let R = R1 × R2 be the direct product of two rings R1 and R2 .

Show that rad(R) = rad(R1 ) × rad(R2 ).

Exercise 7.4.8 A proper ideal P of a ring R is said to be prime if, whenever I1 , I2

are ideals of R such that I1 I2 ⊆ P , we have I1 ⊆ P or I2 ⊆ P , and it is called left
primitive if it is the annihilator of a simple left R-module. Show that

(i) every maximal ideal is left primitive;

(ii) every left primitive ideal is prime;
(iii) if R is left Artinian then every prime ideal is maximal.

(Compare with Proposition 5.2.4 in Chap. 5.)

Exercise 7.4.9 A unital ring R is called semiprime if, for every ideal I of R, the
condition I 2 = 0 entails I = 0. Prove the following analogue of Proposition 7.3.7:
R is semisimple if and only if R is left Artinian and semiprime.

Exercise 7.4.10 Follow the argument in the proof of Lemma 7.3.10 to obtain the
following extension of this lemma: Let I be an ideal in a unital ring R and let K be
an ideal of R such that K ⊆ rad(I ). Then rad(I /K) = rad(I )/K.

Exercise 7.4.11 Let ρ : R → S be a surjective ring homomorphism between the

unital rings R and S. Show that ρ(rad(R)) ⊆ rad(S). Give an example to show that
equality may not always hold.

Exercise 7.4.12 Let R be a unital ring and let n ∈ N. Using Proposition 7.1.1, show
that rad(Mn (R)) = Mn (rad(R)).

Exercise 7.4.13 A unital ring R is called von Neumann regular if, for each r ∈ R,
there is s ∈ R such that rsr = r. Show that, if R M ∈ R-Mod is semisimple, then
EndR (R M) is von Neumann regular.

Exercise 7.4.14 Let R be a von Neumann regular ring. Show that every principal
ideal of R (that is, an ideal which is generated by one element) is generated by an
idempotent.

Exercise 7.4.15 Show that every von Neumann regular ring is semiprimitive. Let
C[0, 1] be the ring of all continuous real-valued functions on the unit interval [0, 1].
Prove that C[0, 1] is semiprimitive but not von Neumann regular.

Exercise 7.4.16 Let R be a reduced ring in which every prime ideal is maximal.
Show that R is von Neumann regular.

Exercise 7.4.17 An algebra over a field K is a ring A which at the same time is a
K-vector space such that λ(xy) = x(λy) for all λ ∈ K and x, y ∈ A. Suppose K is
7.4 Exercises 91

an algebraically closed field and that A is a finite-dimensional K-algebra. Let R M

be a simple left A-module. Show that EndA (R M) ∼ = K.

Exercise 7.4.18 Let (I, ≤) be a directed set, and let {R M i | i ∈ I } be a family of

modules over a ring R. Suppose that, for i ≤ j , there is a module homomorphism
fj i : R M i → R Mj such that fii = id|Mi and, for j ≤ k, fki = fkj ◦ fj i . Then
{(R M i , fj i ) | i, j ∈ I } is called a directed system of R-modules. Let R M =
i∈I R M i be the direct sum of this family and let R N be the submodule of R M
generated by {ιi (xi ) − ιj (fj i (xi )) | xi ∈ R M i , i ≤ j }, where ιi : R M i → R M,
i ∈ I is the canonical injection. Denote by lim −→ I R
M i the quotient module R M/R N .
This is called the direct limit of the directed system of modules.
Define εi : R M i → lim −→ I R
M i as the composition of ιi followed on by the
canonical quotient map R M → R M/R N . Show that {(lim −→ I R
M i , εi ) | i ∈ I } has
the following universal property:
It makes the right-hand triangle in the diagram below commutative and, whenever
{(R L, ηi ) | i ∈ I } consists of an R-module R L and module homomorphisms
ηi : R M i → R L, i ∈ I making the larger triangle commutative, then there is a
unique R-module map lim −→ I R
M i → R L making the entire diagram commutative.

RMi

RL lim I R M i
−→
f ji

R Mj

It should be obvious that the direct limit is uniquely determined up to isomorphism

by this universal property.
92 7 The Artin–Wedderburn Theorem

Exercise 7.4.19 (Nakayama’s Lemma) Let I be a left ideal of the (unital) ring R.
Show that the following three conditions are equivalent.

(i) I ⊆ rad(R);
(ii) for every M ∈ R-mod, I · M = M ⇒ M = 0;
(iii) for all R N , R M ∈ R-Mod with N ⊆ M such that R M/R N is finitely
generated, N + I · M = M ⇒ N = M.
Tensor Products of Modules
8

This chapter is devoted to a more sophisticated construction with modules, the so-
called tensor product. It is extremely useful, though fairly abstract. Its main benefit
lies in the fact that it allows us to convert bilinear mappings into homomorphisms of
abelian groups. The relations between tensor products and homomorphism groups
is fundamental and will lead us to the concept of adjoint functor in the later part of
the chapter.

8.1 Tensor Product of Modules

Here, we will be dealing with left and right modules over a unital ring R at the same
time.

Definition 8.1.1 Let MR ∈ Mod-R and R N ∈ R-Mod. Let G be an abelian

group. A mapping β : M × N → G is said to be R-balanced if, for all m, m1 , m2 ∈
M, n, n1 , n2 ∈ N and r ∈ R,

(i) β(m1 + m2 , n) = β(m1 , n) + β(m2 , n);

(ii) β(m, n1 + n2 ) = β(m, n1 ) + β(m, n2 );
(iii) β(m · r, n) = β(m, r · n).

We denote the set of all such mappings by B(M × N, G).

Example 8.1.2 Let R = R and put M = N = R3 (as R-vector spaces). We define

⎛ ⎞
x1    
β : R3 × R3 → M3 (R) via ⎝x2 ⎠ y1 y2 y3 = xi yj 1≤i,j ≤3 .
x3

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 93

M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_8
94 8 Tensor Products of Modules

This ‘vector product’ is an example of an R-balanced mapping into the abelian

group of 3 × 3-matrices over R.

Our approach to the tensor product is via its universal property.

Definition 8.1.3 A tensor product of the modules MR ∈ Mod-R and R N ∈

R-Mod over R is an abelian group T (M, N) together with an R-balanced mapping
τ : M × N → T (M, N) such that, for every abelian group G and each R-balanced
mapping β : M × N → G, there is a unique Z-module map β̂ : T (M, N) → G
satisfying β = β̂ ◦ τ ; that is, making the diagram below commutative

M N T (M, N)

Theorem 8.1.4 For all MR ∈ Mod-R and R N ∈ R-Mod over a unital ring R
there is a tensor product which is unique up to isomorphism.

Notation As a consequence of Theorem 8.1.4, we shall denote this unique tensor

product by M ⊗R N and the associated mapping τ by (x, y) → x ⊗ y.

Proof We will prove the uniqueness first. Suppose that both (T , τ ) and (S, σ )
satisfy the universal property of (8.1.3). By applying it to σ and to τ , respectively
we obtain the following two commutative diagrams

M N T M N S

S T

We claim that the group homomorphisms σ̂ and τ̂ are inverses of each other so that
T and S are isomorphic. We have

σ̂ τ̂ σ = σ̂ τ = σ and τ̂ σ̂ τ = τ̂ σ = τ.

As a result, we obtain the following commutative diagrams

M N S M N T

S T
8.1 Tensor Product of Modules 95

The uniqueness hypothesis in (8.1.3) implies that σ̂ τ̂ = idS and τ̂ σ̂ = idT

which was to prove.
We next show the existence of a tensor product. Let F be the free Z-module with
basis M × N and consider M × N as a submodule  of F . Recall that this means,
each element of F is a unique Z-linear combination ki=1 ni (xi , yi ) with ni ∈ Z,
xi ∈ M, yi ∈ N and k ∈ N; compare Sect. 2.5. Let K be the Z-submodule of F
generated by elements of the form

(x1 + x2 , y) − (x1 , y) − (x2 , y),

(x, y1 + y2 ) − (x, y1 ) − (x, y2 ),
(x · r, y) − (x, r · y),

where x, x1 , x2 ∈ M, y, y1 , y2 ∈ N and r ∈ R. Put T = F /K and π = πK , the

canonical epimorphism. We set

τ : M × N → T, τ (x, y) = π(x, y)

and verify the conditions in (8.1.3).

Let x, x1 , x2 ∈ M, y ∈ N, r ∈ R; then

τ (x1 + x2 , y) = π(x1 + x2 , y) = (x1 + x2 , y) + K

= (x1 , y) + K + (x2 , y) + K = τ (x1 , y) + τ (x2 , y)

by the definition of K. Similarly, τ is additive in the second variable. Moreover,

τ (x · r, y) = π(x · r, y) = (x · r, y) + K = (x, r · y) + K = τ (x, r · y).

This shows that τ is R-balanced.

Now let β : M × N → G be an R-balanced mapping into an abelian group G.
As F is free, there is a unique extension of β to a Z-module map β̃ : F → G and
we put β̂ ◦ π = β̃. This is a Z-module map provided it is well defined. To this end,
take x, x1 , x2 ∈ M, y ∈ N, r ∈ R and observe that
 
β̃ (x1 + x2 , y) − (x1 , y) − (x2 , y) = β(x1 + x2 , y) − β(x1 , y) − β(x2 , y)
= β(x1 + x2 − x1 − x2 , y) = 0

and similarly for the second variable. Moreover,

 
β̃ (x · r, y) − (x, r · y) = β(x · r, y) − β(x, r · y) = 0.

This shows that K = ker π ⊆ ker β̃ and so π(f ) = 0 implies β̃(f ) = 0 for each
f ∈ F . As a result, β̂ is well defined.
Finally, β̂ ◦ τ = β̂ ◦ π|M×N = β yields the universal property. 

96 8 Tensor Products of Modules

Remarks 8.1.5

1. For any MR ∈ Mod-R and R N ∈ R-Mod, every element in M ⊗R N as

constructed in (8.1.4) can be written as


k
ni (xi , yi ) + K,
i=1

k ni (xi , yi ) + K = (ni xi , yi ) + K and n

but i xi ∈ MR so we can simply write
k
i=1 (xi , yi ) + K which we will denote by i=1 xi ⊗ yi .
2. If the ring R is commutative, every left R-module can be considered as a right
R-module in a canonical way and vice versa; see Exercise 8.4.1. In this case,
M ⊗R N carries a natural module structure. It follows that, if {xi | i ∈ I } and
{yj | j ∈ J } are sets of generators of MR and R N , respectively then {xi ⊗ yj |
(i, j ) ∈ I × J } is a set of generators of M ⊗R N. In particular, the tensor product
of two finitely generated modules over a commutative ring is finitely generated.

The tensor product is rather well behaved and compatible with other construc-
tions with modules. As an example, we mention the result below.

Proposition 8.1.6 Let MR , MR ∈ Mod-R and R N, R L ∈ R-Mod. Then

(i) (M ⊕ M ) ⊗R N ∼
= M ⊗R N ⊕ M ⊗R N;
(ii) M ⊗R N ∼
= N ⊗R M provided R is commutative;
(iii) (M ⊗R N) ⊗R L ∼
= M ⊗R (N ⊗R L) provided R is commutative.

Proof We leave the verification of (i) and (iii) to the reader; see the Exercises. Note
that, if R is commutative, the tensor product carries a natural R-module structure
and therefore, it makes sense to iterate the construction as in (iii).
For illustration, we prove assertion (ii). Suppose R is commutative so that we can
consider the modules M and N as modules on the left and on the right. We have the
following commutative diagram

M N N M

1 2

2
M R N N R M
1

in which the ‘flip’ ι : (x, y) → (y, x) is clearly a bilinear isomorphism.

8.1 Tensor Product of Modules 97

The additive maps τ̂1 , τ̂2 are given by the universal property of the tensor product
and satisfy τ̂2 ◦ τ1 = τ2 ◦ ι and τ̂1 ◦ τ2 = τ1 ◦ ι−1 . Therefore

τ2 ◦ ι = τ̂2 ◦ τ1 = τ̂2 ◦ τ̂1 ◦ τ2 ◦ ι

and

τ1 ◦ ι−1 = τ̂1 ◦ τ2 = τ̂1 ◦ τ̂2 ◦ τ1 ◦ ι−1

which gives τ̂2 ◦ τ̂1 = id|N⊗R M and τ̂1 ◦ τ̂2 = id|M⊗R N , respectively, where, again,
we applied the universal property. Consequently, M ⊗R N ∼ = N ⊗R M as abelian
groups.
Since, for all x ∈ M, y ∈ N,

τ̂2 (r · x ⊗ y) = τ̂2 ◦ τ1 (r · x, y) = τ2 ◦ ι(r · x, y) = τ2 (y, r · x)

= y ⊗ r · x = r · (y ⊗ x)
= r · τ2 (y, x) = r · τ2 ◦ ι(x, y) = r · τ̂2 ◦ τ1 (x, y)
= r · τ̂2 (x ⊗ y)

and similarly for τ1 , we see that this isomorphism is indeed an isomorphism of R-

modules. 


Unless M is an R-bimodule, we cannot endow M ⊗R N with a left module

structure. There is the following important special case, however.

Proposition 8.1.7 Let R be a unital ring, and let R N ∈ R-Mod. Then R ⊗R N ∼

=
R N as left R-modules.

Proof We consider R as standard right R-module RR . Define β : R × N → N by

β(a, x) = a · x, a ∈ R, x ∈ N. Then β is R-balanced:

β(a1 + a2 , x) = (a1 + a2 ) · x = a1 · x + a2 · x = β(a1 , x) + β(a2, x);

β(a, x + y) = a · (x + y) = a · x + a · y = β(a, x) + β(a, y);
β(ar, x) = (ar) · x = a · (r · x) = β(a, r · x)

for all a, a1 , a2 , r ∈ R and x, y ∈ N.

Let β̂ : R ⊗R N → N be the unique group homomorphism satisfying β̂(a ⊗x) =
a · x given by (8.1.4). In this case, R ⊗R N is also a left R-module in a canonical
way and β̂ is an R-module map:

r · β̂(a ⊗ x) = r · (a · x) = (ra) · x = β̂(ra ⊗ x)

which extends to sums via additivity.

98 8 Tensor Products of Modules

Let x ∈ N; then x = 1 · x = β̂(1 ⊗ x), thus β̂ is surjective. Put

γ : N → R ⊗R N, x → 1 ⊗ x.

Then, for each i ai ⊗ xi ∈ R ⊗R N,
     
γ β̂ ai ⊗ xi = γ ai · xi = 1 ⊗ ai · xi = ai ⊗ xi .
i i i i

Therefore γ β̂ = idR⊗R N and β̂ is injective. The statement follows. 



Our next aim is to investigate the relation between the tensor product and
homomorphism groups. To this end, we first record the following special situation.

Proposition 8.1.8 Let R be a unital ring, and let RN ∈ R-Mod. Then

HomR (R, R N ) ∼
= R N as left R-modules.

Proof Let f ∈ HomR (R, R N ); then, for r ∈ R, r · f defined by (r · f )(a) =

f (ar), a ∈ R is in HomR (R, R N ) so HomR (R, R N) is a left R-module. Define
ϕ : HomR (R, R N) → R N by ϕ(f ) = f (1), f ∈ HomR (R, R N); we have

ϕ(f1 + f2 ) = (f1 + f2 )(1) = f1 (1) + f2 (1) = ϕ(f1 ) + ϕ(f2 )

and

ϕ(r · f ) = (r · f )(1) = f (r) = r · f (1) = r · ϕ(f )

for all r ∈ R and f, f1 , f2 ∈ HomR (R, R N ). Thus ϕ is an R-module map. Take

f ∈ ker ϕ; then f (r) = r · f (1) = 0 for every r ∈ R so f = 0. Hence ϕ is injective.
Take y ∈ R N . Define f ∈ HomR (R, R N) by f (a) = a · y. As y = f (1) = ϕ(f ),
ϕ is surjective and altogether an isomorphism. 


Combining the last two propositions we have, as abelian groups,

HomR (R ⊗R M, R N ) ∼
= HomR (R M, R N ) ∼
= HomR (R M, HomR (R, R N ))

for all R M, R N ∈ R-Mod. We shall now see that a similar isomorphism can be
established in general once we allow ourselves to work with bimodules.

Theorem 8.1.9 Let S and R be unital rings, let R M ∈ R-Mod, S N ∈ S-Mod

and S UR ∈ S-Mod-R. Then

HomS (S U ⊗R M, S N ) ∼
= HomR (R M, HomS (S U , S N)).
8.1 Tensor Product of Modules 99

Proof In order to define a group isomorphism

ϕ : HomS (S U ⊗R M, S N) → HomR (R M, HomS (S U , S N)),

take f ∈ HomS (S U ⊗R M, S N ). Fix x ∈ M and put fx : U → N, fx (u) =

f (u ⊗ x) for each u ∈ U . In this way, we get an S-module map fx :

fx (u1 + u2 ) = f ((u1 + u2 ) ⊗ x)
= f (u1 ⊗ x + u2 ⊗ x) = f (u1 ⊗ x) + f (u2 ⊗ x)
= fx (u1 ) + fx (u2 ),
fx (s · u) = f (s · u ⊗ x) = s · f (u ⊗ x)
= s · fx (u)

for all u, u1 , u2 ∈ U and x ∈ M. Hence fx ∈ HomS (S U , S N) and we define

ϕ(f )(x) = fx , x ∈ M. For x, y ∈ M, we have

ϕ(f )(x + y) = fx+y = fx + fy = ϕ(f )(x) + ϕ(f )(y)

since fx+y (u) = fx (u) + fy (u) for all u ∈ U . Moreover,

fr·x (u) = f (u ⊗ r · x) = f (u · r ⊗ x) = fx (u · r) = (r · fx )(u) (u ∈ U )

where the last equality comes from the definition of r · fx . Consequently, for x ∈ M
and r ∈ R, we have

ϕ(f )(r · x) = fr·x = r · fx = r · ϕ(f )

and thus ϕ(f ) is an R-module map.

Next we show that ϕ is a group homomorphism. Take x ∈ M, f, g ∈
HomS (S U ⊗R M, S N ). For all u ∈ U ,

(f + g)x (u) = (f + g)(u ⊗ x) = f (u ⊗ x) + g(u ⊗ x) = fx (u) + gx (u)

and thus

ϕ(f + g)(x) = (f + g)x = fx + gx = ϕ(f )(x) + ϕ(g)(x) = (ϕ(f ) + ϕ(g))(x).

Take f ∈ ker ϕ. Then fx (u) = f (u ⊗ x) = 0 for all x ∈ M and therefore f = 0.

We now define a right inverse of ϕ to show that ϕ is surjective, which will
complete the proof. Let h ∈ HomR (R M, HomS (S U , S N)). Put

βh : U × M → N, βh (u, x) = h(x)(u)
100 8 Tensor Products of Modules

whenever u ∈ U and x ∈ M. It is evident that βh is additive in the first and in the

second variable separately. In addition, for u ∈ U , x ∈ M and r ∈ R,

βh (u · r, x) = h(x)(u · r) = (r · h(x))(u) = h(r · x)(u) = βh (u, r · x)

so that βh is R-balanced. The universal property of the tensor product yields a

unique group homomorphism ψ(h) : S U ⊗R M → N with the property ψ(h)(u ⊗
x) = βh (u, x) for all u ∈ U , x ∈ M. From
 
ϕ(ψ(h))(x) (u) = ψ(h)x (u) = ψ(h)(u ⊗ x) = h(x)(u) (u ∈ U )

we obtain ψ(h)x = h(x) for all x ∈ M and therefore ϕ(ψ(h)) = h. As a result,

ϕ ◦ ψ = id which was to prove. 


In categorical language the above theorem states that the functors U ⊗R – and
HomS (U, –) are adjoint functors; compare Sect. 8.3 below.
We next introduce the tensor product of module maps.

Proposition 8.1.10 Let R be a unital ring. Let MR , MR ∈ Mod-R and

R N, R N ∈ R-Mod. For every pair of module maps f ∈ HomR (MR , MR ),
g ∈ HomR (R N , R N ) there is a unique Z-module map

f ⊗ g : M ⊗R N −→ M ⊗R N

such that (f ⊗ g)(x ⊗ y) = f (x) ⊗ g(y) for all x ∈ M, y ∈ N.

Proof Define β : M × N → M ⊗R N by β(x, y) = f (x) ⊗ g(y), x ∈ M, y ∈ N.

Then

β(x, y1 + y2 ) = f (x) ⊗ g(y1 + y2 ) = f (x) ⊗ (g(y1 ) + g(y2 ))

= f (x) ⊗ g(y1 ) + f (x) ⊗ g(y2 ) = β(x, y1 ) + β(x, y2 )

for all x ∈ M, y1 , y2 ∈ N. Similarly, we show that β is additive in the first variable.

Furthermore, for x ∈ M, y ∈ N and r ∈ R,

β(x · r, y) = f (x · r) ⊗ g(y) = f (x) · r ⊗ g(y) = f (x) ⊗ r · g(y)

= f (x) ⊗ g(r · y) = β(x, r · y).

Hence β is R-balanced and, by the universal property of the tensor product, we

obtain a unique extension to a Z-module map M ⊗R N → M ⊗R N ; we denote
this map by f ⊗ g. 

8.2 Tensor Product of Algebras 101

8.2 Tensor Product of Algebras

In this section, K will denote a field and A, B, C will be unital K-algebras. That
is, A is a K-vector space and carries an associative multiplication (x, y) → xy
such that λ(xy) = x(λy) for all x, y ∈ A and the two distributivity laws hold.
Furthermore, there is a multiplicative identity which we will denote by 1.
Our aim is to endow the module tensor product A ⊗K B with an algebra structure
and study the properties of this algebra. To this end, we first amend the definition of
a K-balanced mapping in a natural way. A mapping β : A × B → C will be called
K-bilinear if, for all a, a1 , a2 ∈ A, b, b1 , b2 ∈ B and λ ∈ K,

(i) β(a1 + a2 , b) = β(a1 , b) + β(a2, b);

(ii) β(a, b1 + b2 ) = β(a, b1) + β(a, b2);
(iii) β(λa, b) = β(a, λb) = λβ(a, b).

Note that the only difference to the original definition of an R-balanced mapping
(p. 93) is in the last identity of the last line above which is possible as C is a vector
space too.
A slight modification of the construction of the module tensor product in
Theorem 8.1.4 yields a K-vector space A ⊗K B which has the universal property
with respect to K-bilinear mappings as defined above, that is, results in a linear
mapping β̂ from A ⊗K B into a K-vector space E.

A B A K B

There is a canonical way of turning this vector space tensor product into a K-
algebra.

Definition 8.2.1 Let A and B be unital algebras over the field K. The tensor
product algebra A ⊗K B is the vector space tensor product endowed with the unique
associative product satisfying (a ⊗ b)(c ⊗ d) = ac ⊗ bd for all a, c ∈ A and
b, d ∈ B.

It should by now be evident how to define the product alluded to above: we first
have to linearise the K-bilinear mapping
 
(A ×B)×(A ×B) −→ (A ×B)⊗K (A ×B), (a, b), (c, d) − → (a, b)⊗(c, d),
102 8 Tensor Products of Modules

where A × B carries the canonical algebra structure. In a second step we linearise

the bilinear mapping

(A × B) ⊗K (A × B) −→ A ⊗K B, (a, b) ⊗ (c, d) −→ ac ⊗ bd

in order to obtain the desired multiplication.

Remark 8.2.2 Just as in Remark 8.1.5 we observe that, whenever {ai | i ∈ I } is a

basis of A and {bj | j ∈ J } is a basis of B, then {ai ⊗bj | (i, j ) ∈ I ×J } is a basis of
the vector space A ⊗K B. Therefore, if dimK A = n < ∞ and dimK B = m < ∞,
the dimension of A ⊗K B is finite, in fact, it is equal to nm.

The mappings a → a ⊗ 1 and b → 1 ⊗ b are injective algebra homomorphisms

from A into A ⊗K B and from B into A ⊗K B, respectively, and their images,
denoted by A ⊗ 1 and 1 ⊗ B, respectively, commute with each other and generate
A ⊗K B.

8.3 Adjoint Functors

Let R be a unital ring. For fixed MR ∈ Mod-R, M ⊗R – is a covariant functor from

R-Mod into AGr, the category of abelian groups (use Exercises 8.4.11 and 8.4.12).
Likewise, for fixed R N ∈ R-Mod, –⊗R N is a covariant functor from Mod-R into
AGr. These functors are additive, by Proposition 8.1.6, and right exact. (To show
this requires some work; we refer to [8, Chap. 15] for details.) However they are in
general not left exact (compare Exercise 8.4.13).

Definition 8.3.1 A left R-module R N ∈ R-Mod is called flat if for every

monomorphism f ∈ HomR (MR , MR ) in Mod-R the induced Z-module map
f ⊗ id : M ⊗R N → M ⊗R N is a monomorphism in AGr.

Proposition 8.3.2 For every unital ring R, the left module R R is flat.

Proof A right-handed version of Proposition 8.1.7 states that MR ∼= MR ⊗ R for

every right R-module MR , and the analogous proof shows that γM : x → x ⊗ 1 is an
isomorphism. Given a monomorphism f ∈ HomR (MR , MR ) in Mod-R we thus
−1
have f ⊗ id = γM ◦ f ◦ γM is a monomorphism. 

 
Generalising Proposition 8.1.6 one can show that ⊕i∈I Mi ⊗R N ∼ = ⊕i∈I Mi ⊗R
N for every family {Mi | i ∈ I } in Mod-R
 and
 each module R N ∈ R-Mod. This
is done of course by showing that ⊕i∈I Mi ⊗R N is a coproduct in AGr of the
family {Mi ⊗R N | i ∈ I }. Together with Proposition 8.3.2 this implies immediately
that every free left R-module is flat. Using this fact together with Proposition 8.1.6
and Exercise 6.4.7 gives us the following corollary.
8.3 Adjoint Functors 103

Corollary 8.3.3 Every projective left module over a unital ring is flat.

Recall the concept of a von Neumann regular ring R: for each r ∈ R there
is s ∈ R such that rsr = r (Exercise 7.4.13). It turns out that a unital ring R
is von Neumann regular if and only if every left R-module is flat; see, e.g., [8,
Theorem 15.22]. This requires a more detailed study of the finitely generated left
ideals in a von Neumann regular ring.
By definition, a module R N ∈ R-Mod is flat if and only if the functor – ⊗R N
is exact (compare Sect. 4.2.3). We shall now discuss the “Fundamental theorem of
tensor products” (Theorem 8.1.9) from the point of view of adjoint functors.

Definition 8.3.4 Let C and D be categories and suppose F : C → D and G : D →

C are covariant functors. Suppose there exists a natural equivalence

η = ηAB : MorD (F(A), B) −→ MorC (A, G(B)), A ∈ ob(C), B ∈ ob(D)

of functors C op × D → S. Then F is called left adjoint to G and G is called right

adjoint to F. Moreover, η is called the adjugant.

The naturality of η in the above definition can be made explicit as follows. For
all f ∈ MorC (A , A), g ∈ MorD (B, B ) and h ∈ MorD (F(A), B), we have

η(g ◦ h ◦ F(f )) = G(g) ◦ η(h) ◦ f.

Let us look at some examples of adjoint functors we already encountered.

Examples 8.3.5

1. Let F : S → R-Mod be the ‘free functor’ (compare Sect. 2.5) and let
G : R-Mod → S be the forgetful functor. Then F is left adjoint to G and G
is right adjoint to F by Theorem 2.5.3. This example extends to any concrete
category which has free objects in the sense of Sect. 2.6.1.
2. Theorem 8.1.9 tells us (in essence) that, whenever R, S are unital rings and S UR ∈
S-Mod-R, then the functor S U ⊗R – is left adjoint to the functor HomS (S U , –).
3. In a similar vein, if R and S are unital rings and ρ : R → S is a unital ring
homomorphism then, considering S as a right R-module (see Exercise 1.3.3), the
‘extension-of-scalars’ functor S ⊗R – : R-Mod → S-Mod is left adjoint to the
‘restriction-of-scalars’ functor S-Mod → R-Mod.
4. Let R be a unital ring and let G be a (multiplicative) group. The construction of
the group ring (see Examples 1.2 (ix)) provides us with a functor R[–] from the
category of groups into the category of unital rings. It is left adjoint to the functor
in the opposite direction which associates to a unital ring R its group of units R ∗ .
5. We return to Example 3.2.4 from our present point of view. For a commu-
tative unital C* -algebra A, its Gelfand space (A) is homeomorphic to the
space of all multiplicative linear functionals on A equipped with the weak *-
104 8 Tensor Products of Modules

topology; thus the functor can also be viewed as the representable functor
MorAC∗ (−, C) : AC1∗ −→ Comp. In the other direction, the functor C
1
associates with a compact Hausdorff space X the morphism set MorComp (X, C).
Therefore we obtain the adjunction

MorComp (MorAC∗ (A, C), Y ) ∼

= MorAC∗ (MorComp (Y, C), A),
1 1

where we have to take care of the fact the both functors are contravariant.

Adjoint functors are ubiquitous in Mathematics; for a detailed study see, for
instance, Sect. II.7 in [21]. They have pleasant properties; e.g., left adjoint functors
preserve colimits while right adjoint functors preserve limits. As an illustration
of the arguments, we prove a special case of the latter statement. A pullback in
a general category, if it exists, is defined by the universal property as stated in
Exercise 4.3.13.

Proposition 8.3.6 Suppose that G : D → C is a covariant functor which has a

left adjoint (so it is itself a right adjoint functor). Then G preserves products and
pullbacks.

Proof Let {Bi | i ∈ I } be a family of objects in D and let (B, pi )i∈I denote their
product (assuming it exists). To show that (G(B), G(pi ))i∈I is the product of the
family {G(Bi ) | i ∈ I } in C, let fi : A → G(Bi ), i ∈ I be morphisms in C. Let
η : F → G be an adjugant and let ξ be its inverse. For each i, ξ(fi ) : F(A) → Bi so
that there exists a unique morphism g : F(A) → B in D such that pi ◦ g = fi for
all i ∈ I . It follows that

G(pi ) ◦ η(g) = η(pi ◦ g) = fi (i ∈ I ),

and η(g) is the unique morphism with this property as every morphism f : A →
G(B) is of the form f = η(g ) for some g in mor(D).
To prove the second assertion,

f G(f )
Y A G(Y ) G(A)

g g G(g) G(g )

B X G(B) G(X )
f G(f )

which we need to show is a pullback. Suppose h : Z → G(A) and k : Z → G(B)

satisfy G(g ) ◦ h = G(f ) ◦ k. Upon applying ξ we obtain g ◦ ξ(h) = f ◦ ξ(k)
and thus there exists a unique morphism ρ : F(Z) → Y such that f ◦ ρ = ξ(h) and
8.4 Exercises 105

g ◦ ρ = ξ(k). Now applying η yields G(f ) ◦ η(ρ) = h and G(g) ◦ η(ρ) = k and, as
for products, η(ρ) is the unique morphism satisfying these identities. 


8.4 Exercises

Exercise 8.4.1 Let R be a commutative ring and let R M ∈ R-Mod. Show that M
becomes a right R-module in a natural way by defining x · r = r · x, x ∈ M, r ∈ R.
As a result, when considering modules over commutative rings, one usually does
not specify “left” or “right”.

Exercise 8.4.2 Prove the two outstanding statements in Proposition 8.1.6.

Exercise 8.4.3 Show that, for every finite abelian group G, Q ⊗Z G = 0.

Exercise 8.4.4 Let K be a field and let G, H be groups. Prove that K[G] ⊗K
K[H ] ∼
= K[G × H ] as K-algebras.

Exercise 8.4.5 Let K be a field and let G be a group. Take M, N ∈ Mod-K[G]

and consider M and N as K-vector spaces in order to form their tensor product
M ⊗K N. Show that (m ⊗ n)g → (mg) ⊗ (ng) defines a right K[G]-module
structure on M ⊗K N.

Exercise 8.4.6 Let K be a field and let A be a unital K-algebra. Show that, for all
n ∈ N, Mn (A) ∼
= Mn (K) ⊗K A.

Exercise 8.4.7 Let V and W be vector spaces over the field K. Their vector space
tensor product V ⊗ W is defined as the module tensor product V ⊗K W . Show that,
for every z ∈ V ⊗W , there exist n ∈ N, linearly independent vectors
 v1 , . . . , vn ∈ V
and linearly independent vectors w1 , . . . , wn ∈ W such that z = ni=1 vi ⊗ wi .

Exercise 8.4.8 For a K-vector space V , let L(V ) denote the K-algebra of all linear
mappings from V to itself. Suppose V and W are K-vector spaces. Show that, for
each S ∈ L(V ), T ∈ L(W ), the function γS,T defined by V × W → V ⊗ W ,
γS,T (v, w) = Sv ⊗ T w for all v ∈ V , w ∈ W yields a linear mapping S,T : V ⊗
W → V ⊗ W such that S,T (v ⊗ w) = Sv ⊗ T w and from this, a bilinear mapping
 
γ : L(V ) × L(W ) −→ L(V ⊗ W ), γ (S, T ) = S,T S ∈ L(V ), T ∈ L(W )

is obtained. Use the universal property of the algebra tensor product to obtain a
unique homomorphism

ϕ : L(V ) ⊗ L(W ) −→ L(V ⊗ W )

106 8 Tensor Products of Modules

such that ϕ(S ⊗ T ) = S,T for all S ∈ L(V ), T ∈ L(W ). Use the previous exercise
to show that ϕ is injective, and refer to Remark 8.1.5 to prove that ϕ is surjective
provided both V and W are finite dimensional.

Exercise 8.4.9 Let A and B be finite-dimensional semisimple algebras over an

algebraically closed field K. Use the isomorphism from the previous exercise
together with the Artin–Wedderburn theorem to prove that A ⊗K B is semisimple.

Exercise 8.4.10 With the assumptions and notation as in Proposition 8.1.10 show
that f ⊗ g is surjective if both f and g are surjective. Does an analogous statement
hold for injectivity?

Exercise 8.4.11 Let MR ∈ Mod-R, RN ∈ R-Mod. Show that idM ⊗ idN =

idM⊗R N .

Exercise 8.4.12 Suppose MR , MR , MR ∈ Mod-R and R N, R N , R N ∈

R-Mod. Let f ∈ HomR (MR , MR ), f ∈ HomR (MR , MR ) and let g ∈
HomR (R N , R N ), g ∈ HomR (R N , R N ). Show that

(f ◦ f ) ⊗ (g ◦ g) = (f ⊗ g ) ◦ (f ⊗ g).

Exercise 8.4.13 Writing Z2 for Z/2Z consider the exact sequence of Z-modules

f g
0 2 0

where f (k) = 2k. Tensoring with Z2 yields

f id g id
0 2 2 2 2 0.

Show that f ⊗ id = 0 and thus cannot be a monomorphism since Z ⊗ Z2 ∼

= Z2 by
Proposition 8.1.7.

Exercise 8.4.14 Given the short exact sequence in R-Mod

f g
0 R N1 RN2 RN3 0

with both R N1 and R N 3 flat, show that the module R N 2 is flat.

Exercise 8.4.15 Let R be a unital ring. Show that, if {Mi | i ∈ I } is a family of

modules in Mod-R then i∈I Mi is flat if and only if every Mi is flat.
8.4 Exercises 107

Exercise 8.4.16 Suppose the functor G : D → C has a left adjoint. Show that G
preserves equalizers and hence kernels. (Compare Exercise 4.3.11.)

Exercise 8.4.17 A full subcategory C of a category D is called reflective if the

inclusion functor C → D has a left adjoint, called the reflector. Let Metric
be the category of metric spaces (as objects) and uniformly continuous mappings
(as morphisms). Show that the category CMetric of complete metric spaces is a
reflective subcategory and the reflector is the “completion of a metric space”.
Exchange Modules and Exchange Rings
9

Our penultimate chapter is devoted to an extension of the concept of semisimple

module and, as a consequence, the notion of a semisimple ring. The exchange
property for modules was introduced by Crawley and Jónsson in 1964 and the
concept of an exchange ring belongs to Warfield (1972). Our main sources for this
chapter are [13] and [16], which contain references to the original articles and a lot
of additional information.
Throughout this chapter, all rings will be unital, all modules will be unital left
modules, and we shall drop the explicit reference to the ring; that is, instead of R M
we simply write M for a unital left module over the ring R.

Definition 9.1 Let R be a unital ring. The module M ∈ R-Mod has the exchange
property if, for every module A ∈ R-Mod and every decomposition

A=M ⊕N = Ai (9.1.1)
i∈I

with N ∈ R-Mod, Ai ∈ R-Mod and M ∼

= M, there exist submodules Ai ≤ Ai
such that

A=M ⊕ Ai . (9.1.2)
i∈I

If the above condition holds for every finite index set I , M is said to have the finite
exchange property.

The exchange property has a forerunner in the following “Steinitz’ Exchange

Lemma for semisimple modules”:

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 109
M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_9
110 9 Exchange Modules and Exchange Rings

Let A ∈ R-Mod be semisimple and decomposed as a direct sum A = i∈I Ai

of simple submodules Ai ≤ A. For each submodule M ≤ A, there is a subset J ⊆ I
such that A = M ⊕ j ∈J Aj .
This follows immediately from the arguments in the proof of Proposition 6.1.8.

9.1 Basic Properties of Exchange Modules

We first observe that, the submodules Ai ≤ Ai in Definition 9.1 above are in fact
direct summands of the Ai . This follows by applying Exercise 9.4.1 to the chain of
inclusions
 
Ai ⊆ Ai ⊆ Ai ⊕ M ⊕ Ai
j =i

 
which yields Ai = Ai ⊕ Ci with Ci = Ai ∩ M ⊕ j =i Ai for each i ∈ I .
For finitely generated modules, the finite exchange property already implies the
full exchange property.

Proposition 9.1.1 Let M ∈ R-mod be a finitely generated module. If M has the

finite exchange property then M has the exchange property.

Proof Suppose that A = M ⊕ N = ∼

i∈I Ai ; then, as M = M, M is finitely
generated and therefore, M ⊆ j ∈J Aj for a finite subset J of I . Applying
Exercise 9.4.1 we find that
 
Aj = M ⊕ N ∩ Aj .
j ∈J j ∈J

Since M has the finite exchange property there exist submodules Aj ≤ Aj , j ∈ J

such that

Aj = M ⊕ Aj
j ∈J j ∈J

and consequently,

A= M ⊕N = Aj ⊕ Ai = M ⊕ Aj ⊕ Ai ,
j ∈J i∈I \J j ∈J i∈I \J

which proves the assertion. 



As it turns out it is in fact sufficient for a module M to have the finite exchange
property for two-element decompositions; see Theorem 9.1.5 below. Working
towards this statement, we establish the following results.
9.1 Basic Properties of Exchange Modules 111

Proposition 9.1.2 Let {Ai | i ∈ I } be a family in R-Mod and let Bi ≤ Ai for all
i ∈ I . Suppose A, M, N and L in R-Mod are such that

A=M ⊕N ⊕L= Ai ⊕ L,
i∈I

A/L = (M + L)/L ⊕ (Bi + L)/L.

i∈I

Then

A=M⊕ Bi ⊕ L.
i∈I


Proof The assumption on A/L provides us with A = M + i∈I Bi + L. In order
 is a direct sum, take m ∈ M, x∈ L and finitely many bi ∈ Bi such
to show this
that m + i∈I bi + x = 0. Then m + L + i∈I (bi + L) = 0 in A/L wherefore,
by assumption, m ∈ L and each bi ∈ L. Since the direct sum decomposition of A
yields m = 0 and bi = 0 (as Bi ∩ L ⊆ Ai ∩ L), it follows that x = 0 which was to
prove. 


We say that a module M ∈ R-Mod has the c-exchange property if the defining
property in Definition 9.1 holds for all index sets I of cardinality at most c.

Corollary 9.1.3 Let {Ai | i ∈ I } be a family in R-Mod and suppose A, M, N and

L in R-Mod are such that

A =M ⊕N ⊕L = Ai ⊕ L.
i∈I

If M has the c-exchange property then there exist submodules Bi ≤ Ai , i ∈ I such

that

A=M⊕ Bi ⊕ L.
i∈I

Proof Apply Proposition 9.1.2 to A/L ∼

= M ⊕L and the c-exchange property of M.



This result will now be used to show that the exchange property is well behaved
under direct sums.
n
Theorem 9.1.4 Let M = i=1 Mi be a finite direct sum of modules Mi ∈
R-Mod. Then M has the c-exchange property if and only if each Mi , 1 ≤ i ≤ n
has the c-exchange property.
112 9 Exchange Modules and Exchange Rings

Remark This result does not extend to infinite direct sums; an example can be found
in [13, Example 12.15].

Proof Clearly it suffices to deal with the case n = 2; thus assume that M = X ⊕ Y
for some X, Y ∈ R-Mod. Suppose that A = M ⊕ N = i∈I Ai , where I has
cardinality at most c. If Y has the c-exchange property then, for some Bi ≤ Ai ,

A=X⊕Y ⊕N =Y ⊕ Bi .
i∈I

If X has the c-exchange property too, then, by Corollary 9.1.3, A = X⊕Y ⊕ i∈I Ci
for some submodules Ci ≤ Bi . It follows that M has the c-exchange property.
Conversely suppose that M = X ⊕ Y has the c-exchange property. If

A=X ⊕N = Ai
i∈I

with X ∼
= X and |I | ≤ c then, setting B = A ⊕ Y , we have

B =M ⊕N =Y ⊕ Ai ,
i∈I

where M = X ⊕ Y ∼
= M. Fix j ∈ I . Since

B = M ⊕ N = Y ⊕ Aj ⊕ Ai
i=j

and M has the c-exchange property, there exist submodules C ≤ Y ⊕ Aj and Bi ≤

Ai , i = j , with the property

B =M ⊕C⊕ Bi .
i=j

Applying Exercise 9.4.1 to the inclusions Y ⊆ Y ⊕ C ⊆ Y ⊕ Aj gives Y ⊕ C =

Y ⊕ Bj , where Bj = (Y ⊕ C) ∩ Aj . Hence, M ⊕ C = (X ⊕ Y ) ⊕ C = X ⊕ Y ⊕ Bj .
When substituting this into the above identity we find

B = X ⊕ Y ⊕ Bj ⊕ Bi = X ⊕ Y ⊕ Bi .
i=j i∈I

Since X ⊕ i∈I Bi ⊆ A, the modular law (Exercise 2.7.8) entails

 
A ∩ Y + (X ⊕ Bi ) = (A ∩ Y ) + (X ⊕ Bi )
i∈I i∈I
9.1 Basic Properties of Exchange Modules 113

and thus, A = X ⊕ i∈I Bi so that X has the c-exchange property. A similar

argument yields Y has the c-exchange property too, so the proof is complete. 


Our next result may be somewhat surprising.

Theorem 9.1.5 A module which has the 2-exchange property has the finite
exchange property.

Proof Suppose M has the 2-exchange property; we shall show that M has the n-
exchange property for every n ∈ N. Let n > 2 and assume that M has the (n − 1)-
exchange property. Let N, A1 , . . . , An ∈ R-Mod be such that M ⊕ N = ni=1 Ai .
Setting A = ni=1 Ai and B = ni=2 Ai we have A = A1 ⊕ B. By the 2-exchange
property,

A = M ⊕ A1 ⊕ B ,

where

A1 = A1 ⊕ A1 , B = B ⊕ B and B = B ∩ (M ⊕ A1 ).

By the induction hypothesis, since M ∼

= A1 ⊕ B , we infer from Theorem 9.1.4 that
B has the (n − 1)-exchange property so, from
n
B =B ⊕B = Ai ,
i=2

we obtain B = B ⊕ ni=2 Ai for some submodules Ai ≤ Ai , 2 ≤ i ≤ n. As

B = B ∩ (M ⊕ A1 ) and thus B ⊆ M ⊕ A1 ⊆ B ⊕ (B ⊕ A1 ), we can apply
Exercise 9.4.1 to find

M ⊕ A1 = B ⊕ B where B = (M ⊕ A1 ) ∩ (B ⊕ A1 ).

As a result,

A = M ⊕ A1 ⊕ B = B ⊕ B ⊕ B
n
= B ⊕B = Ai ⊕ B ⊕ B
i=2
n
=M⊕ Ai
i=1

entailing that M has the n-exchange property. 


114 9 Exchange Modules and Exchange Rings

Corollary 9.1.6 Let M ∈ R-mod. If M has the 2-exchange property then it has
the exchange property.

Proof This follows immediately from Proposition 9.1.1 and Theorem 9.1.5. 


It may be time to see a module which fails the (2-)exchange property.

Example 9.1.7 The module Z fails the 2-exchange property. Note at first that Z is
indecomposable as a module: if Z ∼ = X ⊕ Y with X, Y ∈ Z-Mod then X or Y must
be trivial. Now start with the identity

Z ⊕ Z = Z(1, 0) ⊕ Z(0, 1) = Z(7, 3) ⊕ Z(5, 2)

and suppose that Z had the 2-exchange property. Then there are submodules X ≤
Z(7, 3), Y ≤ Z(5, 2) such that Z ⊕ Z = Z(1, 0) ⊕ X ⊕ Y . It follows that Z ∼
= X ⊕Y
whence X = 0 or Y = 0. Therefore

Z ⊕ Z = Z(1, 0) ⊕ Z(7, 3) or Z ⊕ Z = Z(1, 0) ⊕ Z(5, 2)

which is impossible as neither {(1, 0), (7, 3)} nor {(1, 0), (5, 2)} form a Z-basis of
Z ⊕ Z.

In fact, indecomposable modules which have the exchange property can be

characterised as those with local endomorphism ring [16, Theorem 2.8].

9.2 Exchange Rings

Definition 9.2.1 A unital ring R is said to be an exchange ring if, for any pair of
elements a and b in R with a + b = 1, there exist idempotents e ∈ Ra and f ∈ Rb
such that e + f = 1.

In this section, we will show that R is an exchange ring if and only if the standard
left module R R has the exchange property and study some of the properties of
this type of ring. It appears that we have defined a property that depends on the
choice of the standard left module R R over its right-handed analogue RR ; however
Proposition 9.2.3 below shows that this is not the case.
We prepare this by the following lemma.

Lemma 9.2.2 The following conditions are equivalent for an element x in a unital
ring R.

(a) ∃ e = e2 ∈ R : e − x ∈ R(x − x 2 );
(b) ∃ e = e2 ∈ Rx, c ∈ R : (1 − e) − c(1 − x) ∈ rad(R);
9.2 Exchange Rings 115

(c) ∃ e = e2 ∈ Rx : R = Re + R(1 − x);

(d) ∃ e = e2 ∈ Rx : 1 − e ∈ R(1 − x).

Proof (a) ⇒ (b) Suppose e − x = r(x − x 2 ) where e is an idempotent in R and

r ∈ R. Then e = r(1 − x)x + x ∈ Rx and

1 − e − (1 − rx)(1 − x) = 0 ∈ rad(R).

(b) ⇒ (c) If 1 − (e + c(1 − x)) ∈ rad(R) for some c ∈ R then e + c(1 − x) is

invertible in R and therefore, R = Re + R(1 − x).
(c) ⇒ (d) Let e ∈ Rx be an idempotent such that, for some s, t ∈ R, 1 =
te + s(1 − x). Put f = e + (1 − e)te; then f is an idempotent in Rx and

1 − f = 1 − e − (1 − e)te = (1 − e)s(1 − x) ∈ R(1 − x).

(d) ⇒ (a) If 1 − e ∈ R(1 − x) for some idempotent e ∈ Rx then e − x =

e(1 − x) − (1 − e)x ∈ R(x − x 2 ). 


Proposition 9.2.3 A unital ring R is an exchange ring if any, and hence every, of
the conditions in Lemma 9.2.2 holds for all x ∈ R. Moreover, this is equivalent to
the condition that, for any pair a and b in R with a + b = 1, there exist idempotents
e ∈ aR and f ∈ bR such that e + f = 1.

Proof Since it is evident that condition (d) in Lemma 9.2.2 is equivalent to the
condition that, for any pair of elements a and b in R with a + b = 1, there exist
idempotents e ∈ Ra and f ∈ Rb such that e + f = 1, the first assertion in the
proposition holds.
Now assume that the condition in the second assertion of the proposition is
satisfied; we show that this implies that R is an exchange ring, and the converse
will be true by symmetry. Take a, b ∈ R with a + b = 1 and pick e = e2 ∈ aR,
f = f 2 ∈ bR such that e + f = 1. Choose r ∈ Re, s ∈ Rf with the property
e = ar and f = bs. It follows that r = re = rar and rbs = rebs = ref = 0 and
similarly, s = sf s and sar = 0. Put

r = 1 − sb + rb and s = 1 − ra + sa.

Then r s = 0 = s r and, as ar + bs = e + f = 1, we find

ar = a(1 − sb) + arb = a(1 − sb) + (1 − bs)b = a(1 − bs) + b(1 − sb) = 1 − sb.
116 9 Exchange Modules and Exchange Rings

This entails r ar = r (1 − sb) = r , and a similar argument yields bs = 1 − ra

and s bs = s . Put e = r a ∈ Ra and f = s b ∈ Rb to obtain idempotents which
add up to 1:

e + f = r a + s B = (1 − sb + rb)a + (1 − ra + sa)b
= a − sba + rba + b − rab + sab
= a + b + s(ab − ba) + r(ba − ab) = a + b = 1

since a and b = 1 − a commute. This was to prove. 



Corollary 9.2.4 Every homomorphic image of an exchange ring is an exchange

ring.

Proof This follows immediately from the definition or any of the conditions listed
in Lemma 9.2.2. 


Let L be a subgroup of (R, +). One says that idempotents lift modulo L if,
whenever x ∈ R satisfies x − x 2 ∈ L, there is an idempotent e ∈ R such that
e − x ∈ L. Proposition 9.2.3 entails the following characterisation of exchange
rings.

Corollary 9.2.5 A unital ring R is an exchange ring if and only if idempotents lift
modulo every left ideal of R.

We will now make contact with the exchange property, which will justify the
choice of the terminology ‘exchange ring’.

Theorem 9.2.6 Let R be a unital ring. Then M ∈ R-Mod has the finite exchange
property if and only if EndR (M) is an exchange ring. In particular, R is an
exchange ring if and only if R R has the exchange property.

Proof By Proposition 9.2.3, R is an exchange ring if and only if its opposite ring
R op is an exchange ring. Since R op ∼
= EndR (R R), see Exercise 2.7.1, the statement
on R follows immediately from the main statement in the theorem together with
Corollary 9.1.6.
Put S = EndR (M) and suppose S is an exchange ring. In order to deduce that
M has the finite exchange property, it suffices to show that M has the 2-exchange
property (Theorem 9.1.5). Assume, thus, that

A = M ⊕ N = A1 ⊕ A2 in R-Mod. (9.2.1)

Denote by p ∈ S the projection onto M and by qi ∈ S the projections onto Ai , i =

1, 2, respectively. These are idempotents satisfying p = pq1 p+pq2 p. Since pSp ∼
=
EndR (M) as rings, by hypothesis, pSp is an exchange ring. We can therefore apply
9.2 Exchange Rings 117

Lemma 9.2.2 to pSp to choose orthogonal idempotents si ∈ pSp(pqi p), i = 1, 2

with s1 +s2 = p (use condition (a) in Lemma 9.2.2 and note that pq2 p = p−pq1 p).
We can write si = ai qi p for some ai ∈ pSp with si ai = ai , i = 1, 2. Define ti ∈ S
by ti = qi ai qi , i = 1, 2. Then ti tj = 0 for i = j and

ti − ti2 = qi ai qi − qi ai qi qi ai qi = qi ai pqi − qi ai pqi ai pqi = qi (si − si2 ) = 0

for i = 1, 2. Hence, t1 , t2 are orthogonal idempotents with ti A ⊆ Ai , i = 1, 2. As a

result, Ai = ti A ⊕ Ai with Ai = Ai ∩ ker ti , i = 1, 2, and we obtain

A = (t1 A ⊕ t2 A) ⊕ A1 ⊕ A2 .

The argument will be finished once we show that A = M ⊕ A1 ⊕ A2 .

Note first that ti p = qi ai qi p = qi si and thus,

ai ti p = ai qi si = ai qi ai qi p = ai qi pai qi p = si2 = si

for i = 1, 2. Suppose x ∈ M ∩ (A1 ⊕ A2 ). Then t1 x = 0 = t2 x and therefore,

x = px = s1 x + s2 x = a1 t1 px + a2 t2 px = a1 t1 x + a2 t2 x = 0.

It follows that M ∩ (A1 ⊕ A2 ) = 0.

Next take x ∈ A and write it (uniquely) as x = x1 + x2 + y with xi ∈ ti A,
i = 1, 2, and y ∈ A1 ⊕ A2 . Observe that

ti aj tj = ti paj qj tj = qi si sj aj qj tj = δij tj

for i, j = 1, 2. It follows that xi − ai xi ∈ Ai for each i and consequently,

x − a1 x1 − a2 x2 = x1 − a1 x1 + x2 − a2 x2 + y ∈ A1 ⊕ A2 .

Since ai xi ∈ M for each i this shows that A = M ⊕ A1 ⊕ A2 proving that M has

the 2-exchange property.
In order to prove the “only if”-part suppose that M has the finite exchange
property. Let a, b ∈ S = EndR (M) be such that a + b = 1. Write A := M ⊕ M =
N1 ⊕ N2 , where Ni is the image of the ith canonical projection from A onto M,
i = 1, 2. Put D = {(x, x) | x ∈ M} and M = {ax, −bx) | x ∈ M}. Then M ∼ =M
and A = M ⊕ D = N1 ⊕ N2 . By hypothesis, there exist Ni ≤ Ni , i = 1, 2 with
the property A = M ⊕ N1 ⊕ N2 . Consequently, for each x ∈ M, there is a unique
decomposition

(x, x) = (ay, −by) + (x1 , 0) + (0, x2 ) (9.2.2)

118 9 Exchange Modules and Exchange Rings

with y ∈ M and (x1 , 0) ∈ N1 , (0, x2 ) ∈ N2 . We define a , b ∈ S by a x = x2 and

b x = x1 . Then

(aa x, aa x) = (aa x, −ba x) + (0, 0) + (0, a x),

(bb x, bb x) = (−ab x, bb x) + (b x, 0) + (0, 0)

and thus, a aa = a and b bb = b so that e = aa ∈ aS and f = bb ∈ bS are

idempotents. From (9.2.2) we obtain x = ay + b x = −by + a x and therefore,
y = (a + b)y = (a − b )x. Substituting back we find

x = a(a − b )x + b x = aa x + (1 − a)b x = aa x + bb x

which implies that e + f = 1. This proves that EndR (M) is an exchange ring, in
view of Proposition 9.2.3. 


Remark The concept of an exchange ring was originally introduced by Warfield but
the equivalent formulation via Definition 9.2.1 is due to Nicholson; the proof of
Theorem 9.2.6 presented here is the original one in [28].

Corollary 9.2.7 A unital ring R is an exchange ring if and only if every finitely
generated projective module in R-Mod has the exchange property.

Proof Only the “only if”-part needs proof. By Theorem 9.2.6, R R has the exchange
property and therefore, every finitely generated free R-module has the exchange
property, by Theorem 9.1.4. As every finitely generated projective module is a direct
summand of a finitely generated free module, the same theorem yields the claim.



Another characterisation of exchange rings involves the Jacobson radical.

Theorem 9.2.8 A unital ring R is an exchange ring if and only if the quotient ring
R/rad(R) is an exchange ring and idempotents lift modulo rad(R).

Proof The “only if”-statement follows from Corollaries 9.2.4 and 9.2.5. Now
suppose that R = R/rad(R) is an exchange ring and let x ∈ R. Write x for
x + rad(R). There are an idempotent a ∈ R x and an element c ∈ R such that
1 − a = c (1 − x ), by condition (b) in Lemma 9.2.2. Without loss of generality we
can assume that a ∈ Rx. If idempotents lift modulo rad(R), there is f = f 2 ∈ R
such that f = a . Set u = 1 − f + a, which is a unit in R. It follows that
e = u−1 f u = u−1 f a is an idempotent in Rx. Moreover e = f = a and thus
(1−e)−c(1−x) ∈ rad(R) which proves that R is an exchange ring by Lemma 9.2.2.


9.3 Commutative Exchange Rings 119

Here is a list of some examples of exchange rings.

Examples 9.2.9

1. Every von Neumann regular ring R is an exchange ring (cf. Exercise 7.4.13).
Let x ∈ R and take y ∈ R with xyx = x. Then f = yx is an idempotent
in Rx. Put e = f + (1 − f )xf to obtain an idempotent e ∈ Rx such that
1 − e = (1 − f )(1 − xf ) ∈ R(1 − x). The statement follows from Lemma 9.2.2.
2. Call a unital ring R clean if every element in R is the sum of a unit and an
idempotent. Every clean ring is an exchange ring. For, take x ∈ R and write it as
x = u + f where u is a unit and f is an idempotent. Set e = u−1 (1 − f )u. Then

u(e − x) = (1 − f )u − u(u + f ) = u − f u − uf − u2 = x − x 2

and thus, e − x ∈ R(x − x 2 ) so that Lemma 9.2.2 and Proposition 9.2.3 yield the
claim.

Remark 9.2.10 Exchange rings have rather interesting properties. For instance, it
was shown by Ara et al. [4] that a unital C* -algebra is an exchange ring if and
only if it has real rank zero, a rather important structural property useful for the
classification of C* -algebras. Ara [3] also introduced the concept of non-unital
exchange rings proving, amongst others, that a (possibly non-unital) ring R is an
exchange ring if and only if it contains an ideal I such that both I and R/I are
exchange rings and idempotents lift modulo I . For these, and further comments, see
[13, 11.43].

9.3 Commutative Exchange Rings

In this section we shall study some special properties of commutative exchange

rings. We have already observed (Exercise 9.4.5) that a commutative (unital) ring is
a exchange ring if and only if it is clean. We will now establish a connection with
various kinds of dimension for a commutative ring.
A good proportion of the interest in commutative rings comes from their
appearance in Analysis, especially as rings of continuous functions. We will assume
the reader has a basic knowledge in (point-set) Topology, such as provided by,
e.g., [32]. For a compact Hausdorff space X, we denote by CR (X) the space of
continuous real-valued functions on X, which is a commutative unital ring under
pointwise operations. (Compare also Example 3.2.2)
The Lebesgue covering dimension of X is defined via refinements of open
coverings of X; see [19], for example. However, we will only need the case
when X is zero-dimensional which is equivalent to the condition that X is totally
disconnected, that is, singleton sets are the only non-empty connected subsets
of X. This in turn is equivalent to the existence of a basis for the topology of
X consisting of clopen (= closed and open) subsets of X. We will use this latter
120 9 Exchange Modules and Exchange Rings

characterisation since it entails that the ring CR (X) is abundant with idempotents:
for each clopen subset U ⊆ X, its characteristic function χU belongs to CR (X) and
is an idempotent.
The following is a standard characterisation of zero-dimensional compact Haus-
dorff spaces; see, e.g., [19, Theorem 16.17].

Proposition 9.3.1 The following conditions on a compact Hausdorff space X are

equivalent.

(a) X is zero-dimensional;
(b) for every pair of disjoint closed subsets A and B of X, there exist disjoint open
subsets U and V of X such that X = U ∪ V , A ⊆ U and B ⊆ V ;
(c) for every closed subset A of X and every open subset U of X containing A,
there exists a clopen subset V of X such that A ⊆ V ⊆ U .

The exchange property for CR (X) turns out to be the equivalent of zero-
dimensionality of X.

Proposition 9.3.2 A compact Hausdorff space X is zero-dimensional if and only if

the ring CR (X) is an exchange ring.

Proof Suppose at first that X is zero-dimensional. Let a, b ∈ CR (X) satisfy a +b =

1 and let A = a −1 (0) and B = b −1 (0). Then A and B are disjoint closed subsets
of X so, by Proposition 9.3.1, there exist U, V ⊆ X clopen and disjoint such that
X = U ∪V , A ⊆ U and B ⊆ V . Set e = χV and f = χU , which are idempotents in
CR (X) with sum 1. Define c ∈ CR (X) by c = a1 e and d ∈ CR (X) by d = b1 f ; more

1
if t ∈ V
precisely, c(t) = a(t ) and similarly for d. Then e = ca and f = db
0 if t ∈ U
wherefore, by Definition 9.2.1, CR (X) is an exchange ring.
Now suppose that CR (X) is an exchange ring. Let A and B be two disjoint
closed subsets of X. By Urysohn’s Lemma (see, e.g., Theorem 1.5.6 together with
Theorem 1.6.6 in [32]) there is a ∈ CR (X) such that a(X) ⊆ [0, 1], A ⊆ a −1 (0) and
B ⊆ a −1 (1). Let e ∈ CR (X) be an idempotent such that e = ca and 1−e = d(1−a)
for some c, d ∈ CR (X). Let V ⊆ X be such that e = χV . Then V is clopen and,
with U = X \ V , A ⊆ U and B ⊆ V . By Proposition 9.3.1, X is zero-dimensional.



Next we introduce two algebraic versions of dimension for a general commu-

tative unital ring that appeared in the work of Canfell ([11] and [12]). For the
remainder of this section, R will always denote a commutative unital ring and for
an element a ∈ R, the ideal generated by a will be written as (a).

Definition 9.3.3 A finite set (a1 ), . . . , (an ) of principal ideals of R is said to be

uniquely generated if, whenever (ai ) = (bi ), 1 ≤ i ≤ n for a set {b1 , . . . , bn } ⊆ R,
9.3 Commutative Exchange Rings 121

there exist elements ui ∈ R such that ai = bi ui for all 1 ≤ i ≤ n and R =

(u1 ) + · · · + (un ). The first Canfell dimension of R, denoted by dim1 R, is the least
natural number n such that every set of n + 1 principal ideals is uniquely generated.

Definition 9.3.4 The order of a finite subset {a1, . . . , an } ⊆ R which generates

R is the largest m ∈ {1, . . . , n} with the property that m + 1 members of
{a1 , . . . , an } have non-zero product. A refinement of {a1 , . . . , an } is another finite
subset {b1 , . . . , bk } ⊆ R such that, for each 1 ≤ j ≤ k, there is 1 ≤ i ≤ n with
bj ∈ (ai ). The second Canfell dimension of R, denoted by dim2 R, is the least d ∈ N
such that every generating subset of R has a refinement of order at most d.

Both Canfell dimensions of CR (X) agree with the covering dimension of the
compact Hausdorff space X, which was shown in [11] and [12], respectively.
Consequently, by Proposition 9.3.2, CR (X) is an exchange ring if and only if
dim1 CR (X) = 0 which is equivalent to dim2 CR (X) = 0. These results partly extend
to arbitrary commutative rings.

Theorem 9.3.5 Let R be a commutative unital ring. If R is an exchange ring then

dim1 R = 0.

Proof Let a, b ∈ R be such that (a) = (b). Then there are s, t ∈ R with a = bs and
b = at so that b(1 − st) = 0. Since R is an exchange ring, there is an idempotent
e ∈ R such that e ∈ (1 − st)R and 1 − e ∈ stR. For some v, u ∈ R we thus have
e = (1 − st)v and 1 − e = stu. Replacing v by ve and u by u(1 − e) if necessary,
we can assume that v = ve and u = u(1 − e). Note that be = b(1 − st)v = 0 so
that b = b(1 − e).
Let k = s(1 − e) + (1 − st)e. Then k is invertible with inverse tu + v:
  
k(tu + v) = s(1 − e) + (1 − st)e tu(1 − e) + ve
= stu(1 − e) + (1 − st)ve = 1 − e + e = 1.

In addition,

bk = b(s(1 − e) + (1 − st)e) = b(1 − e)s + b(1 − st)e = bs = a

wherefore the first Canfell dimension of R must be zero. 



Remark 9.3.6 Every integral domain has first Canfell dimension zero. This follows
immediately from a = bs and b = at, hence b(1 − st) = 0 so that st = 1 (unless
b = 0, the trivial case). However, not every integral domain is an exchange ring;
e.g., Z is not (Example 9.1.7).

The relation to the second Canfell dimension is even more pleasing.

122 9 Exchange Modules and Exchange Rings

Theorem 9.3.7 Let R be a commutative unital ring. Then R is an exchange ring if

and only if dim2 R = 0.

Proof Let us first observe that dim2 R = 0 means that, whenever a1 , . . . , an ∈ R

with (a1 ) + · · · + (an ) = R are given, there exist b1 , . . . , bm ∈ R with (b1 ) + · · · +
(bm ) = R and, for each 1 ≤ i ≤ m, there is 1 ≤ k ≤ n with bi ∈ (ak ) with the
property that bi bj = 0 for all i, j , i = j .
Now suppose  that R is an exchange ring. Let a1 , . . . , an ∈ R and r1 , . . . , rn ∈ R
be such that ni=1 ai ri = 1. By Exercise
 9.4.6, there exist orthogonal idempotents
ei ∈ Rai ri , 1 ≤ i ≤ n such that ni=1 ei = 1. Since ei ∈ (ai ) for all i it follows
that dim2 R = 0.
Suppose dim2 R = 0. Take a ∈ R; as R = aR + (1 − a)R, by hypothesis, there
exist a1 , . . . , a ∈ aR and a +1 , . . . , am ∈ (1 − a)R such that ai aj = 0 for i = j
and

a1 r1 + · · · + am rm = 1

for some ri ∈ R, 1 ≤ i ≤ m. Letting e = a1 r1 + · · · + a r we have e ∈ aR and

1 − e ∈ (1 − a)R. Since

e(1 − e) = (a1 r1 + · · · + a r )(a +1 r +1 + · · · + am rm ) = 0,

as all the ai s are mutually orthogonal, we find e = e2 and R is an exchange ring. 



In the remainder of this section we will introduce and investigate a variety of

concepts for commutative unital rings that all turn out to be closely related to the
notion of an exchange ring. Our source for this is a paper by McGovern [27], which
also contains the references to the original papers where these concepts were first
put forward.

Definition 9.3.8 Let R be a commutative unital ring. Then

(i) R is a Gelfand ring if, for every pair of elements a and b in R with a + b = 1,
there exist r, s ∈ R such that (1 + ar)(1 + bs) = 0.
(ii) R is a pm-ring if every prime ideal of R is contained in a unique maximal ideal.
(iii) R is a tb-ring if, for every pair of distinct maximal ideals of R, there is an
idempotent belonging to exactly one of them.

There is a canonical topological space associated to every commutative unital

ring.
9.3 Commutative Exchange Rings 123

Definition 9.3.9 For a commutative unital ring R, let Spec(R) denote the set of all
prime ideals of R endowed with the following topology: a basis for the open subsets
is given by sets of the form

U (a) = {P ∈ Spec(R) | a ∈
/ P },

a ∈ R. This is the prime ideal space or spectrum of R and the topology is referred
to as the hull-kernel topology or Zariski topology.

The prime ideal property implies that U (ab) = U (a) ∩ U (b) for all a, b ∈ R.
It is customary to denote the complement of U (a) by V (a), that is, V (a) =
{P ∈ Spec(R) | a ∈ P }. In general, Spec(R) has poor separation properties;
its closed points are the maximal ideals in R. The subset Max(R) of all maximal
ideals is a subspace of Spec(R) when equipped with the subspace topology, and is
called the maximal spectrum of R. The basic subsets in Max(R) are thus U (a) =
U (a) ∩ Max(R), a ∈ R, and their complements will likewise be denoted by V (a).
Nevertheless both Spec(R) and Max(R) are compact spaces. More details on the
prime spectrum can be found in [6] or [10, Chap. II].

Notation Let E(R) denote the following subset of Max(R):

E(R) = {U (e) | e = e2 ∈ R}.

The next result provides us with the relations between the various types of rings
defined above.

Theorem 9.3.10 The following conditions on a commutative unital ring R are

equivalent.

(a) R is an exchange ring;

(b) R is a Gelfand ring and Max(R) is zero-dimensional;
(c) R is a pm-ring and Max(R) is zero-dimensional;
(d) R is a tb-ring;
(e) E(R) is a basis for the Zariski topology on Max(R);
(f) for each a ∈ R, there is an idempotent e ∈ R such that V (a) ⊆ U (e) and
V (1 − a) ⊆ U (1 − e).

The proof of Theorem 9.3.10 will be split up into a series of propositions which
emphasise more clearly the interconnections between the different conditions. We
start by examining pm -rings; clearly the acronym stems from the assumption that,
for each prime ideal, there is a unique maximal ideal containing it.
124 9 Exchange Modules and Exchange Rings

Proposition 9.3.11 A commutative unital ring R is a pm-ring if and only if, for
every pair of distinct maximal ideals M and M of R, there exist a ∈
/ M and b ∈
/M
such that ab = 0. In this case, Max(R) is a Hausdorff space.

Proof Let P be a prime ideal in R and let M be a maximal ideal containing P . If

the condition is satisfied, for every other maximal ideal M = M, take a ∈
/ M and
b ∈/ M such that ab = 0. As ab ∈ P it follows that b ∈ P and hence P  M .
Therefore R is a pm -ring.
Conversely, take two different maximal ideals M, M of R and put

S = {ab | a ∈
/ M, b ∈
/ M }.

If the semigroup S does not contain contain 0, a standard argument using Zorn’s
Lemma yields the existence of a prime ideal P with P ∩ S = ∅. Hence P ⊆ M ∩ M
and R is not a pm -ring.
In order to show that Max(R) is Hausdorff whenever R is a pm -ring let M, M ∈
Max(R), M = M . Take a ∈ / M and b ∈/ M such that ab = 0. Then M ∈ U (a),
M ∈ U (b) and U (a) ∩ U (b) = U (ab) = ∅ which proves the claim. 


Let X be a compact Hausdorff space; then CR (X) is a pm -ring, see [19,

Theorem 2.11 together with Remark 6.6 (c)] or [19, Theorem 7.15]. Since
Max(CR (X)) is homeomorphic to X (e.g., [6, Exercise 26 in Chap. I] or [19,
Chap. 7]), dim Max(CR (X)) = 0 is equivalent to CR (X) being an exchange ring
(Proposition 9.3.2). Consequently, the equivalence of conditions (a) and (c) in
Theorem 9.3.10 is a direct generalisation of Proposition 9.3.2. Working towards this
equivalence, we note the following.

Proposition 9.3.12 Let R be a commutative unital ring. Then

R exchange ring "⇒ R tb-ring "⇒ R pm-ring. (9.3.3)

Proof Suppose first that R is an exchange ring. Let M, M ∈ Max(R) with M =

M . Take a ∈ M \ M and observe that (a) + M = R. Thus, for some x ∈ M,
r ∈ R, 1 = ar + x. By hypothesis, there is an idempotent e ∈ Rx ⊆ M such that
1 − e ∈ R(1 − x) = Rar ⊆ M . It follows that e ∈ / M and R is a tb -ring.
Suppose now that R is a tb -ring. Let M, M ∈ Max(R) with M = M and let
e ∈ M \ M be an idempotent. Since 1 − e ∈ / M and e(1 − e) = 0 ∈ M it follows
that 1 − e ∈ M entailing that R is a pm -ring, by Proposition 9.3.11. 


Corollary 9.3.13 Let R be a commutative exchange ring. Then the space Max(R)
is a compact Hausdorff space.
9.3 Commutative Exchange Rings 125

The terminology “tb -ring” stands for “topologically boolean ring” and is moti-
vated by the fact that a compact zero-dimensional Hausdorff space is also called a
boolean space and the result below.

Proposition 9.3.14 Let R be a tb-ring. Then Max(R) is zero-dimensional.

Proof Note first that Max(R) is a compact Hausdorff space, by Propositions 9.3.11
and 9.3.12. It therefore suffices to show that the topology on Max(R) has a basis
consisting of clopen subsets. In fact, these will be the elements of E(R). By
Exercise 9.4.9, each U (e), e an idempotent in R, is clopen and, whenever a closed
subset K ⊆ Max(R) is covered by a family {U (ei ) | i ∈ I } of subsets from E(R),
by compactness, a finite subfamily suffices to cover K and, by Exercise 9.4.9, their
union belongs to E(R), so K is in fact contained in one subset of the form U (e).
Let K ⊆ Max(R) be closed and let M ∈ Max(R) be in the complement of K.
For each N ∈ K, there is an idempotent eN ∈ / N which belongs to M. Then N ∈
U (eN ), M ∈ U (1 −eN ) and U (eN )∩U (1 −eN ) = ∅. Since K ⊆ N∈K U (eN ),
the above argument yields an idempotent e ∈ R such that K ⊆ U (e) and M ∈ /
U (e). It follows that E(R) is a basis for the Zariski topology on Max(R). 


Remark 9.3.15 The argument in the above proof in fact establishes the implication
(d) ⇒ (e) in Theorem 9.3.10. Let us now assume that condition (e) holds. This
entails that E(R) is the collection of all clopen subsets of Max(R). Take a ∈ R;
then V (a) is a closed subset of Max(R) and hence contained in some U (e), e = e2
by the above argument. Likewise V (1 − a) ⊆ U (1 − e) and hence condition (f) in
Theorem 9.3.10 holds.
Assuming (f) we will now show that R must be an exchange ring. Take a ∈ R. By
hypothesis, there is an idempotent e ∈ R such that V (a) ⊆ U (e) and V (1 − a) ⊆
U (1 − e). Let M be a maximal ideal of R. Suppose a ∈ M; then e ∈ / M and thus
e−a ∈ / M. Suppose a ∈ / M; if e − a ∈ M then a + M = e + M = 1 + M since
R/M is a field and e + M is a non-zero idempotent. It follows that 1 − a ∈ M so
1−e ∈ / M wherefore e ∈ M, which is impossible. As a result, e−a is invertible in R
and therefore, a = e − a + e can be written as a sum of a unit and an idempotent, in
other words, R is a clean ring. It follows from Example 9.2.9.2 that R is an exchange
ring.

In summary we have established so far the following chain of inclusions

(a) ⇒ (d) ⇒ (e) ⇒ (f) ⇒ (a)

of the conditions in Theorem 9.3.10, by Proposition 9.3.12 and Remark 9.3.15, as

well as (d) ⇒ (c) by Propositions 9.3.12 and 9.3.14. We shall now see that the
properties ‘Gelfand ring’ and ‘pm -ring’ are always equivalent for a commutative
unital ring so that conditions (b) and (c) are equivalent as well.
126 9 Exchange Modules and Exchange Rings

Proposition 9.3.16 A commutative unital ring R is a Gelfand ring if and only if it

is a pm-ring.

Proof Suppose R is a Gelfand ring. Let M and M be two distinct maximal ideals
in R. For any m ∈ M \ M , M + (m) = R and thus, m + rm = 1 for some m ∈ M
and r ∈ R. By assumption, there are s, t ∈ R such that (1 + sm )(1 + trm) = 0.
Since 1 +sm ∈ / M and 1 +trm ∈ / M Proposition 9.3.11 entails that R is a pm -ring.
Now take a, b ∈ R with a + b = 1 and suppose that R fails to be a Gelfand ring.
Then the semigroup

S = {(1 + ar)(1 + bs) | r, s ∈ R}

does not contain 0. The usual argument invoking Zorn’s Lemma yields the existence
of a prime ideal P with P ∩ S = ∅. Since P + (a) = R (otherwise 1 = x − ar
with x ∈ P , r ∈ R would force x = 1 + ar ∈ S which is impossible) there is
a maximal ideal M ⊂ R containing P + (a). Similarly, there is a maximal ideal
M ⊂ R containing P + (b). As M = M (for a + b = 1), Proposition 9.3.11
implies that R is not a pm -ring. 


To complete the proof of Theorem 9.3.10, we need to show that one of the
conditions (b) or (c) in the theorem imply that R is an exchange ring. The strategy
to achieve this is to construct two suitable orthogonal ideals of R whose sum is R.
These will arise as intersections of maximal ideals, and since the intersection of
maximal ideals is not a maximal but only a prime ideal, we will have to work within
the prime spectrum for the first time.

Proposition 9.3.17 Let R be a pm-ring whose maximal spectrum is zero-

dimensional. Then R is an exchange ring.

Proof We aim to show that E(R) is a basis for the Zariski topology of Max(R); as
seen in Remark 9.3.15 above, this implies that R is an exchange ring. First assume
that the nil radical nil(R) is zero. Let K ⊆ Max(R) be a clopen subset. Set

K = {P ∈ Spec(R) | P ⊆ M for some M ∈ K}.

As K is the preimage of K under the canonical mapping Spec(R) → Max(R) which

associates to each prime ideal the
 unique maximal ideal containing it, K is clopen
(see Exercise 9.4.14). Put J1 = {P | P ∈ K} and J2 = {Q | Q ∈ Spec(R)\K}.
As J1 ∩ J2 = nil(R) = 0, J1 and J2 are orthogonal ideals in R.
Since Spec(R) is compact and K is clopen, there exist a1 , . . . , an ∈ R and
b1 , . . . , bm ∈ R such that

"
n "
m
K= U (ai ) and Spec(R) \ K = U (bj ).
i=1 j =1
9.3 Commutative Exchange Rings 127

Equivalently,


m 
n
K= V (bj ) and Spec(R) \ K = V (ai ).
j =1 i=1

As a result, bj ∈ P for every P ∈ K, 1 ≤ j ≤ m and ai ∈ Q for every Q ∈

Spec(R) \ K, 1 ≤ i ≤ n. It follows that {a1 , . . . , an , b1 , . . . , bm } ⊆ J1 + J2 . Since

"
n "
m
Spec(R) = U (ai ) ∪ U (bj )
i=1 j =1

there is no prime ideal of R which contains J1 + J2 ; in other words, J1 + J2 = R.

Take e1 ∈ J1 and e2 ∈ J2 such that e1 + e2 = 1. Then e1 = e1 (e1 + e2 ) = e12 so
e1 is an idempotent wherefore e2 = 1 − e1 is an idempotent as well. Take x ∈ J1 .
Then x = x(e1 + e2 ) = xe1 and hence J1 = Re1 . Similarly J2 = Re2 = R(1 − e1 ).
It follows that K = V (e1 ) and thus, K = V (e1 ), which was to show.
In the general case, that is, if nil(R) is not necessarily zero, we apply the above
argument to R/nil(R). Since the spectra of R and R/nil(R) are homeomorphic (e.g.,
[6, Exercise 21 in Chap. I] or [10, Remark on p. 101]), the assumptions apply, and
since idempotents lift from R/nil(R), see the result below, we obtain the claim.  

Lemma 9.3.18 Let R be a commutative unital ring. Then idempotents lift modulo
the nil radical nil(R).

Proof Let a ∈ R be such that a −a 2 ∈ nil(R). Then, for some n ∈ N, (a −a 2)n = 0.

The binomial formula yields
! !
n n−1 2 n n−2 4
an − a a + a a − · · · + a 2n = 0
1 2

whence a n − a n+1 b = 0 for some b ∈ R. Put e = (ab)n ; then a n bn = a n+1 bbn =

a n+1 bn+1 = . . . = a n+n bn+n = a 2n b2n . Therefore, e = e2 and

a + nil(R) = a n + nil(R) = a n+1 b + nil(R) = (a n+1 + nil(R))(b + nil(R))

= (a + nil(R))(b + nil(R)) = ab + nil(R)

and hence

a + nil(R) = a n + nil(R) = (a + nil(R))n = (ab + nil(R))n

= (ab)n + nil(R) = e + nil(R)

so that the idempotent a + nil(R) in R/nil(R) lifts to the idempotent e in R. 


128 9 Exchange Modules and Exchange Rings

9.4 Exercises

Exercise 9.4.1 Let M1 , M2 and N be modules satisfying M1 ⊆ N ⊆ M1 ⊕ M2 .

Show that N = M1 ⊕ (N ∩ M2 ).

Exercise 9.4.2 Use Theorem 9.1.4 to prove that, for modules M, N, L, K and T ∈
R-Mod, if M = L ⊕ K = N ⊕ T and T has the finite exchange property and
N ≤ L, then K has the finite exchange property.

Exercise 9.4.3 Let R be an exchange ring and let e ∈ R be an idempotent. Show

that the subring eRe of R is an exchange ring. (The identity of eRe is e.)

Exercise 9.4.4 A unital ring R is said to be semiregular if R/rad(R) is von

Neumann regular and idempotents lift modulo rad(R). Show that every semiregular
ring is an exchange ring.

Exercise 9.4.5 Let R be a commutative unital ring. Show that, if R is an exchange

ring, then R is clean.

Exercise 9.4.6 Let R be an exchange n ring. Show by induction that, for every finite
 i=1 ai = 1, there exist orthogonal idempotents
family a1 , . . . , an in R such that
ei ∈ Rai , 1 ≤ i ≤ n such that ni=1 ei = 1.

Exercise 9.4.7 Use Theorems 9.1.4 and 9.2.6 to show the following. Let e be an
idempotent in a unital ring R. Then R is an exchange ring if and only if eRe and
(1 − e)R(1 − e) are exchange rings.

Exercise 9.4.8 Show directly from the definition that every commutative exchange
ring is a Gelfand ring.

Exercise 9.4.9 For a commutative unital ring R, let E(R) = {U (e) | e = e2 ∈ R};
compare p. 123. Show that E(R) is closed under finite intersections and unions as
well as complements.

Exercise 9.4.10 Let R be a commutative unital ring. Show directly from the
definitions that, if R is a tb -ring, then R is an exchange ring.

Exercise 9.4.11 Let R be a commutative unital ring. Show directly from the
definitions that, if R is a clean ring, then R is a tb -ring.

Exercise 9.4.12 Use Proposition 5.2.4 to show that every Artinian commutative
ring is a pm -ring.
9.4 Exercises 129

Exercise 9.4.13 A commutative unital ring R is called local if it contains a unique

maximal ideal. Show that every such ring is a pm -ring.

Exercise 9.4.14 Suppose R is a pm -ring. Show that the canonical mapping

Spec(R) → Max(R) which associates to each prime ideal the unique maximal
ideal containing it is continuous.
Semiprimitivity of Group Rings
10

The starting point for this chapter is the discussion around Maschke’s theorem,
Theorem 7.2.1, which provides us with conditions under which the group ring K[G]
is semisimple, provided G is a finite group. For any field K, the elements of G form
a basis of the K-vector space K[G] and if the ring K[G] is semisimple, then it is
necessarily Artinian, hence finite dimensional (Corollary 6.2.5 and Exercise 5.5.6).
As a result, we cannot expect K[G] to be semisimple for an infinite group G.
The ‘next best’ property one can expect for infinite groups therefore is semiprimi-
tivity, see also Proposition 7.3.7. An important and still open problem in ring theory,
which was popularised by Kaplansky, asks for characterisations of those groups for
which K[G] is semiprimitive. (This was known as the “semisimplicity problem”
before the terminology changed.) The larger part of this chapter will be devoted to
a study of some of the known results. In our approach, we will use techniques from
functional analysis (developed in detail in Sect. 10.2) to obtain a positive answer
when K = C, the field of complex numbers. This is a nice illustration of how
different areas of Mathematics can work together.
Throughout this chapter, K will denote a field and groups will be written
multiplicatively.

10.1 Basic Properties

We begin by establishing some basic properties of group rings. For the definition of
the operations, see (ix) in Examples 1.2.

Proposition 10.1.1 Let G be a finitely generated torsion-free abelian group. Then,

for some n ∈ N, K[G] is contained isomorphically between the polynomial ring
K[x1, . . . , xn ] and the rational field K(x1 , . . . , xn ). In particular, K[G] is an
integral domain.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 131
M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3_10
132 10 Semiprimitivity of Group Rings

Proof The fundamental theorem of abelian groups yields a decomposition G =

g1  × . . . × gn  as a product of infinite cyclic groups, for some natural number n.
The canonical homomorphism K[x1 , . . . , xn ] −→ K[G] given by xi → gi ,
1 ≤ i ≤ n is an embedding since every element of G is uniquely of the form
g1m1 · · · gnmn . Under this embedding, for each α ∈ K[G], there is m ∈ N such that
(x1 , . . . , xn )m α ∈ K[x1, . . . , xn ] and hence, K[G] is an integral domain. As such,
it is canonically contained in the field of fractions K(x1 , . . . , xn ). 


As a consequence we obtain the following result.

Corollary 10.1.2 Let G be an infinite cyclic group. Then K[G] is a principal ideal
domain.

Proof By the above proposition, we already know that K[G] is an integral domain.
Let I be a non-zero ideal of K[G]. Let g ∈ G be a generator. Then every element
of K[G] is of the form g k f (g) for some k ∈ Z and a polynomial f (g) ∈ K[G].
Suppose g k f (g) ∈ I ; then f (g) = g −k g k f (g) ∈ I . Take a non-zero polynomial
t (g) ∈ I of minimal degree and divide f (g) by t (g). By the division algorithm
and the minimality assumption we conclude that t (g) divides f (g) and thus t (g)
generates the ideal I . Consequently, K[G] is a principal ideal domain. 


Suppose H is a subgroup of the group G; then, clearly, K[H ] is a unital subring

of K[G]. We have the following relation between the Jacobson radicals of these two
rings.

Proposition 10.1.3 Let H be a subgroup of G. Then

rad(K[G]) ∩ K[H ] ⊆ rad(K[H ]).

We prepare the proof of this proposition by the following lemma.

Lemma 10.1.4 Let R be a unital subring of the unital ring S. Suppose, as left R-
modules, R is a direct summand of S. Then rad(S) ∩ R ⊆ rad(R).

Proof Suppose that R S = R R⊕R M for some left R-module R M. Take α ∈ rad(S)∩
R and fix β ∈ R. Since there is s ∈ S such that (1 − αβ)s = 1 = s(1 − αβ) and s
can be written (uniquely) as s = r + m with r ∈ R, m ∈ M, we have

1 = (1 − αβ)r + (1 − αβ)m.

As α belongs to R and R is a unital subring, this entails that 1 = (1 − αβ)r and

(1 − αβ)m = 0. Upon multiplying

1 = r(1 − αβ) + m(1 − αβ)

10.2 Some Analytic Structure on C[G] 133

on the right by r we obtain

r = r(1 − αβ)r + m(1 − αβ)r = r + m

and hence m = 0. As a result, 1 − αβ is invertible in R and so α ∈ rad(R), by

Proposition 7.3.2. 


Proof of Proposition 10.1.3 The set M = {α ∈ K[G] | ag = 0 for all g ∈ H } is a

left K[H ]-submodule of K[G] and K[G] = K[H ] ⊕ R M. By Lemma 10.1.4, the
assertion follows. 


We turn our attention to fields with finite characteristic. Suppose K is a field with
characteristic p > 0 and let A be an algebra over K. If A is commutative, the power
map α → α p is a homomorphism, as is easily checked. In the general case, we have
the following generalisation. We shall denote by [A, A] the subspace of A spanned
by all commutators [α, β] = αβ − βα.

Lemma 10.1.5 Let K be a field with characteristic p > 0, and for given n ∈ N,
set q = pn . Let A be an algebra over K and let α1 , . . . , αm ∈ A. Then, for some
β ∈ [A, A], we have
q q
(α1 + · · · + αm )q = α1 + · · · + αm + β. (10.1.1)

q q
Proof We have (α1 + · · · + αm )q = α1 + · · · + αm + β, where β is the sum of
all words αi1 · · · αiq with at least two distinct indices. Let Cq be the cyclic group
of order q. Suppose ω1 = αi1 · · · αiq and ω2 = αij · · · αiq αi1 · · · αij−1 are cyclic
permutations of each other. Then,

ω1 − ω2 = (αi1 · · · αij−1 )(αij · · · αiq ) − (αij · · · αiq )(αi1 · · · αij−1 ) ∈ [A, A].

Hence, modulo [A, A], all cyclic permutations of a word are equal and the number
of formally distinct permutations of a word which occur in β as above is the size of
a non-trivial orbit in Cq , therefore divisible by p. This proves the assertion. 


This result will become handy in Sect. 10.3 below.

10.2 Some Analytic Structure on C[G]

Let G be an arbitrary group. When G is infinite, C[G] is an infinite-dimensional

vector space over the complex numbers and it is expedient to introduce some
analytic structure on it. Firstly, we see that it is an inner product space in a canonical
way.
134 10 Semiprimitivity of Group Rings

 
Definition 10.2.1 Let α = g∈G ag g, β= g∈G bg g belong to C[G]. We put

(α | β) = ag b̄g
g∈G

(where b̄ denotes the complex conjugate of b ∈ C) and call this the canonical inner
product on C[G].

It will be shown in Exercise 10.4.1 that the above indeed defines an inner product
on C[G] and that α2 = (α | α)1/2 therefore is a norm on C[G].  However, the
associated metric space is in general not complete. In other words, C[G], (· | ·) is
in general not a Hilbert space.
Beside the structure of an inner product space, C[G] also has the structure of a
normed algebra.
 
Definition 10.2.2 Let α = g∈G ag g ∈ C[G]. We put α1 = g∈G |ag |, where
|a| of course denotes the absolute value of the complex number a.

Proposition 10.2.3 The above definition of α1 turns C[G] into a normed space.
Moreover, for all α, β ∈ C[G], we have αβ1 ≤ α1 β1 and C[G] is a complex
normed algebra with identity 1.

Proof It is straightforward to check that, for all α, β ∈ C[G], α + β1 ≤ α1 +

β1 , that cα1 = |c| α1 for every c ∈ C and that α1 = 0 implies that α = 0.
Thus,  · 1 defines a norm on C[G].
From the triangle inequality and the fact that αg1 = α1 we obtain
# #  
# #
αβ1 = # αbg g # ≤ αbg g1 = α1 |bg | = α1 β1 .
1
g∈G g∈G g∈G

 
As a result, C[G],  · 1 is a normed algebra. Clearly, 1 is a multiplicative identity.



As all good things come in threes, we will introduce another norm on C[G].
Let H = 2 (G) be the Hilbert space obtained from C[G], (· | ·) , compare
Exercise 10.4.1. The algebra C[G] acts on H in a canonical way: for clarity
 let us
denote the elements in H by ξ so that ξ ∈ C[G] can be written as ξ = g∈G xg g
with xg ∈ C. For such ξ and α ∈ C[G] we have αξ ∈ C[G] ⊆ H , that is, C[G]
acts as left multiplication operator. The above inequalities imply that this action is
continuous, so can be extended to all of H . Moreover, since αξ 2 ≤ α2 ξ 2 for
all ξ ∈ H , the action is injective (α1 = 0 ⇒ α = 0) and it makes sense to define
our third norm as
$ %
αop = sup αξ 2 | ξ ∈ H, ξ 2 = 1 .
10.2 Some Analytic Structure on C[G] 135

Because C has a distinguished conjugation z → z̄, we can define an involution, that

is, an anti-linear automorphism α → α ∗ on C[G] in the following way:
 
for α = ag g we put α ∗ = ā g g −1 .
g∈G g∈G

Proposition 10.2.4 For all α, β, γ ∈ C[G], we have (α + β)∗ = α ∗ + β ∗ , (αβ)∗ =

β ∗ α ∗ and (α ∗ )∗ = α. Furthermore,

(αγ ∗ | β) = (α | βγ ) = (β ∗ α | γ ). (10.2)

Proof The first three identities are very easy 

to check. Next we recall the definition
of the trace on C[G] from Exercise 1.3.9: tr g∈G a g g = a1 and, therefore (α |
β) = tr αβ ∗ . As a consequence, for all α, β, γ ∈ C[G],

(αγ ∗ | β) = tr αγ ∗ β ∗ = tr α(βγ )∗ = (α | βγ )
= tr (βγ )∗ α = tr γ ∗ β ∗ α = (β ∗ α | γ ),

where of course we used the trace property. 



In the language of bounded linear operators on the Hilbert space H , α ∗ is the

adjoint of α: (ξ | αη) = (α ∗ ξ | η), by (10.2), compare Exercise 10.4.4.
The “operator norm” has the additional nice property that αop = α ∗ op
and that α ∗ αop = α2op . With the involution α → α ∗ , the completion of
 
C[G],  · op becomes what is known as a C*-algebra. See also Example 3.2.4
and Exercises 10.4.4 and 10.4.5.
The C* -algebra which is obtained in this way is the reduced group C*-algebra
of G, denoted by Cr∗ (G). A special property of this kind of C* -algebra is that it
carries a faithful trace. The trace tr on C[G] satisfies

|tr α|2 ≤ α22 = (α1 | α1) = (α ∗ α1 | 1) ≤ α ∗ αop = α2op

so that tr is a bounded linear functional and thus can be extended from the
dense subspace C[G] to the whole algebra Cr∗ (G). We shall denote this extended
functional by τ .

Lemma 10.2.5 The functional τ is a normalised faithful trace on Cr∗ (G), that is,
τ (1) = 1, τ (ab) = τ (ba) for all a, b ∈ Cr∗ (G) and τ (a ∗ a) ≥ 0 with τ (a ∗ a) = 0 if
and only if a = 0.

Proof By definition, τ (1) = tr 1 = 1 and since tr is a bounded linear functional on

the dense subspace C[G] and the multiplication in Cr∗ (G) is continuous, it follows
that τ (ab) = τ (ba) for all a, b ∈ Cr∗ (G). Let a ∈ Cr∗ (G). Let ε > 0 and take
136 10 Semiprimitivity of Group Rings

α ∈ C[G] such that a − αop < ε. Then

a ∗ a − α ∗ αop ≤ a ∗ − α ∗ op αop + a ∗ op a − αop

< ε(ε + aop ) + aop ε = ε(ε + 2aop ) = ε .

Since tr α ∗ α = g∈G |ag |
2 ≥ 0, for ε > 0 such that tr α ∗ α − ε ≥ 0, we obtain

τ (a ∗ a) = tr α ∗ α − τ (α ∗ α − a ∗ a) ≥ tr α ∗ α − a ∗ a − α ∗ αop > tr α ∗ α − ε ≥ 0.

In the case where τ (a ∗ a) = 0 the above inequality yields tr α ∗ α = 0, that is, α = 0,

and as a result, a must be zero too. Therefore the trace τ is faithful. 


The next lemma is a useful general result in C* -algebras and can be found in
many of the standard textbooks. We provide a proof for the convenience of the
reader.

Lemma 10.2.6 Let A be a unital C*-algebra. Every idempotent in A is equivalent

to a projection in A. That is, if e ∈ A satisfies e = e2 then there is p ∈ A with
p = p2 = p∗ such that, for some x, y ∈ A, we have e = xy and p = yx.

Proof We first observe that every element of the form 1 + b∗ b, b ∈ A is invertible

in A. One of the many ways to see this is to consider the unital C* -subalgebra of A
generated by 1 and b∗ b. Since b∗ b is selfadjoint, this C* -subalgebra is commutative
and therefore isomorphic to C(X) for a compact Hausdorff space X; compare
Examples 3.2.2 and 3.2.4. As a function on X, 1 + b∗ b is of the form 1 + f̄ f and
this function is certainly invertible as it has no zeros. Consequently, the (isomorphic
image of the) inverse serves as the inverse of 1 + b∗ b in A.
We now set a = 1 + (e − e∗ )∗ (e − e∗ ) for our given idempotent e. We compute

ea = e + e(e∗ e − e − e∗ + ee∗ ) = ee∗ e = e + (e∗ e − e − e∗ + ee∗ )e = ae,

that is, e and a commute and hence e and a −1 commute. We put p = ee∗ a −1 and
observe that p∗ = a −1 ee∗ = p (since a is selfadjoint, so is its inverse). Furthermore,

p2 = p∗ p = a −1 ee∗ ee∗ a −1 = a −1 aee∗a −1 = p

so that p is a projection. Finally, we have ep = p and pe = p∗ e = a −1 ee∗ e =

a −1 ae = e so that e and p are indeed equivalent. 


The above observations enable us to draw a consequence on the range of the

trace on idempotents in Cr∗ (G). Let e ∈ Cr∗ (G) be an idempotent and let p be a
projection equivalent to e. Then τ (e) = τ (p) = τ (p∗ p) ≥ 0 and equal to 0 only if
e = 0 (Lemma 10.2.5). Applying the same reasoning to 1 − e we find that τ (e) = 1
10.3 The Semiprimitivity Problem 137

if and only if e = 1. This can be used to determine an algebraic property of the

group ring C[G].
Recall from Exercise 4.3.2 that a unital ring R is von Neumann finite, if, for
x, y ∈ R, xy = 1 implies that yx = 1.

Theorem 10.2.7 For every group G, the group ring C[G] is von Neumann finite.

Proof Let x, y ∈ C[G] satisfy xy = 1; then e = yx is an idempotent different

from 0. Since tr e = tr xy = 1 if follows that e = 1, i.e., yx = 1. 


This theorem in fact holds for all groups rings K[G] where K has characteristic
zero. A result by Kaplansky from 1969, it can be derived from the above special
case, or by other means, see, e.g., [31, Corollary 2.1.9].

10.3 The Semiprimitivity Problem

A primitive ring is one that has a faithful simple module; according to the convention
in this book, we focus on the left-handed version (compare Definition 7.3.4). As
primitive rings are somewhat scarce, one wants to study more general rings by
looking at all their simple modules (equivalently, irreducible representations) but for
this to work, the canonical obstruction, the Jacobson radical, has to vanish. These
are the semiprimitve rings (Definition 7.3.4).
When is the group ring K[G] semiprimitive? The answer to this question will
depend on properties of the field K and the group G and possibly on the interaction
of the two. We will discuss this question for infinite groups in this section.
The first case we investigate is the group ring with complex numbers as
coefficients, this will lead us to one of the first results on the above question,
originally obtained by Rickart in 1950. We present a modified version of his
original proof which uses functional analytic techniques which we already prepared
ourselves for in the previous section.

Theorem 10.3.1 For every group G, the group ring C[G] is semiprimitive.

Proof When G is finite, C[G] is semisimple by Maschke’s theorem (Theo-

rem 7.2.1), hence semiprimitive. For the case when G is infinite, we will use the
Banach algebra 1 (G) discussed in Sect. 10.2 and Exercise 10.4.2. The strategy of
 For αn ∈ rad(C[G]), we will construct an
the proof is as follows. entire function F
such that F (1) = ∞ n=0 tr α ; this will entail that limn→∞ tr α = 0. On the other
n

hand, we will show that the only α ∈ rad(C[G]) that allows for the construction of
the above-mentioned F is α = 0, which will complete the argument that C[G] is
semiprimitive.
Let us prove the last assertion first. Take β ∈ rad(C[G]) \ {0} and put α =
ββ ∗ ∗
β2 2 , where we use the notation of Sect. 10.2. Then α = α ∈ rad(C[G]) since the
138 10 Semiprimitivity of Group Rings

Jacobson radical is an ideal of C[G]. It follows that

tr ββ ∗ β22
tr α = = = 1.
β22 β22
m
Suppose tr α 2 ≥ 1 for some m ∈ N0 . Then

= tr α 2 (α 2 )∗ = α 2 22 ≥ |tr α 2 |2 ≥ 1;
m+1 m m m m
tr α 2
m
thus, by induction, tr α 2 ≥ 1 for all m ∈ N0 and therefore the sequence (tr α n )n∈N
alluded to above cannot tend to zero.
We shall now go about the construction of the function F . Since, for every z ∈ C,
1 − zα is invertible in C[G] (Proposition 7.3.2), we can define φ : C → C[G] by
φ(z) = (1 − zα)−1 . Suppose α = 0 and that |z| < α 1
. In this case the series
∞ −1
1

n=0 (zα) converges with limit (1 − zα)

n (the proof is the same as for the usual
geometric series in Real Analysis, see Exercise 10.4.3).
We define F : C → C by F (z) = tr φ(z). On the open disk {z ∈ C | |z| <
∞
1
α1 }, F is given by the power series F (z) = n n
n=0 z tr α , since the trace is linear
and continuous. We show that F is entire; then the identity theorem entails that F
is given by this power series on the entire complex plane, in particular, F (1) =
 ∞ n
n=0 tr α which was to prove.
In order to show that F is differentiable at every z ∈ C, note first that φ(z) and
φ(w) commute for all z, w ∈ C. Therefore,

φ(z) − φ(w) = (1 − zα)−1 − (1 − wα)−1

= ((1 − wα) − (1 − zα))(1 − zα)−1 (1 − wα)−1
= (z − w)α φ(z)φ(w).

Taking traces yields, for z = w,

F (z) − F (w)
= tr (α φ(z)φ(w)),
z−w

thus, by continuity of φ and the trace,

F (z) − F (w)
lim = tr (α φ(z)2 ).
w→z z−w

As a result, the derivative of F at z exists and is equal to tr (α φ(z)2 ). 



There do exist purely algebraic proofs of Theorem 10.3.1 but it is instructive to see
how Analysis can be used to obtain an algebraic result.
10.3 The Semiprimitivity Problem 139

From Lemma 7.3.8 we know that rad(K[G]) contains every nil ideal of K[G].
This is a starting point to obtain another case where K[G] is semiprimitive.

Lemma 10.3.2 In every group ring, each algebraic element in the radical is
nilpotent.

Proof Let α ∈ rad(K[G]) and suppose it satisfies a polynomial equation over K

which we write as

α n (1 + c1 α + c2 α 2 + · · · + cr α r ) = 0

for suitable r, n ∈ N0 and ci ∈ K. Set γ = c1 α + c2 α 2 + · · · + cr α r ∈ rad(K[G]);

then 1 + γ is invertible and thus α n = 0. 


Lemma 10.3.3 Let G be a group and let K be a field with the property that |K| >
dimK K[G] (as a strict inequality of possibly infinite cardinal numbers). Then every
element in rad(K[G]) is algebraic.

Proof Let α ∈ rad(K[G]); since 1 − cα is invertible for every c ∈ K (Proposi-

tion 7.3.2), and dimK K[G] is strictly less than the cardinality of K, there have to
be finitely many ai , ci ∈ K \ {0}, 1 ≤ i ≤ n, for some n ∈ N, such that

a1 (1 − c1 1α)−1 + a2 (1 − c2 α)−1 + · · · + an (1 − cn α)−1 = 0.

Upon multiplying the above identity by (1 −c1 α) · · · (1 −cn α) we obtain a non-zero

polynomial over K of which α is a root. 


From the two results above we see that, for very large coefficient fields, every
element in the radical is nilpotent; hence the radical itself is a nil ideal.

Theorem 10.3.4 Let K be an uncountable field and let G be a group. Then

rad(K[G]) is a nil ideal.

Proof Let α = g∈G ag g ∈ rad(K[G]) and let H be a finitely generated subgroup
of G which contains {g ∈ G | ag = 0}. Then dimK K[H ] is countable so strictly
smaller than |K|. By Lemma 10.3.3 together with Lemma 10.3.2, rad(K[H ]) is
a nil ideal. As rad(K[G]) ∩ K[H ] ⊆ K[H ] (Proposition 10.1.3), it follows that
α ∈ rad(K[G]) ∩ K[H ] is nilpotent, so altogether K[G] is a nil ideal. 

140 10 Semiprimitivity of Group Rings

Thus the remaining task is to determine when K[G] has no non-zero nil ideals.

Theorem10.3.5 Let K be a field with characteristic p > 0 and let G be a group.

For α = g∈G ag g ∈ K[G], let

i
p-supp α = {g ∈ G | ag = 0 and g p = 1 for some i ∈ N}

be the p-support of α. Suppose α is nilpotent. Then

  
tr α + ag = 0.
g∈p-supp α

Therefore, either tr α = 0 or p-supp α = ∅.

n
Proof Since α is nilpotent, there is n0 ∈ N such that α p = 0 for all n ≥ n0 .
n
Moreover, as p-supp α is finite, we can choose n such that g p = 1 for all g ∈
p-supp α. Put q = pn . Applying Lemma 10.1.5 we obtain
 q
0 = αq = ag g q + β
g∈G

& '
for some β ∈ K[G], K[G] . Since the trace vanishes on commutators this entails
that
 q   q
0 = tr α q = ag = ag .
g q =1 g q =1

the choice ofn, g = 1 if and only if g ∈ p-supp α or g = 1. Thus, tr α +

By q

g∈p-supp α ag = 0 as claimed. 


Let p be a prime. A group G is called a p -group if, for all g ∈ G, g = 1 and all
i
i ∈ N, g p = 1. We now combine the last two theorems to obtain an answer to the
semiprimitivity problem for certain fields and certain groups.

Theorem 10.3.6 Let K be an uncountable field with characteristic p > 0 and let
G be a p -group. Then K[G] is semiprimitive.

Proof By Theorem 10.3.4, it suffices to show

 that the only nilpotent element in
rad(K[G]) is the zero element. Let α = g∈G ag g ∈ rad(K[G]) be nilpotent.
−1
Then ag = tr (αg ) = 0, by Theorem 10.3.5, since G is a p -group. This proves
the claim. 

10.4 Exercises 141

Remark 10.3.7 The above result, due to Amitsur (1957), also holds for fields with
characteristic 0 and arbitrary groups. To this end, one has to extend the previous two
theorems to this setting and add quite a bit of additional terminology and techniques.
Since these would take us here too far afield, we refer to [31, Chap. 7] for the details.

A wealth of information on the semiprimitivity problem is contained in Passman’s

book [31] as well as in [24]. On the other hand, the following fundamental question
remains open:
Is the group ring Q[G] semiprimitive for every group G?

Remark 10.3.8 The three famous ‘Kaplansky conjectures’ on the structure of group
rings are as follows. Let K be a field and G be a torsion-free group. Then

(U) K[G] has no non-trivial units, that is, every element in K[G]× is of the form
ag with a ∈ K ∗ and g ∈ G.
(Z) K[G] has no non-trivial zero divisors, that is, every zero divisor is equal to 0.
(I) K[G] has no non-trivial idempotents, that is, every idempotent in K[G] is either
equal to 1 or to 0.

The relation between them is

(U) "⇒ (Z) "⇒ (I).

The second implication is easy: every idempotent e ∈ K[G] \ {0, 1} yields non-
trivial zero divisors via e(1 − e) = e − e2 = 0. The first is a bit harder and
uses Connell’s result ([31, Theorem 4.2.10]) that K[G] is a prime ring under
the given assumptions. They were open for many decades and proven in a large
number of special cases. Only in 2021, conjecture (U) was settled in the negative by
Gardam [18].
Since these questions are related to other important problems, for example, the
Baum–Connes conjecture in operator algebras, there is still quite some activity in
the study of group rings.

10.4 Exercises

Exercise 10.4.1 Show that the mapping (α, β) → (α | β) as defined in

Definition 10.2.1 gives C[G] the structure of an inner product space and therefore
α2 = (α | α)1/2 yields a norm on C[G]. Show further that the associated metric
(α, β) → α −β2 turns C[G] into a metric space which in general (i.e., for infinite
groups) is not complete. Its completion to a Hilbert space is usually denoted by
2 (G). (For the terminology, see any basic book on Functional Analysis such as

[1, 14] or [26].)

142 10 Semiprimitivity of Group Rings

Exercise 10.4.2 Prove that the normed algebra C[G] as described in Proposi-
tion 10.2.3 is in general not complete. Its completion to a Banach algebra is
generally denoted by 1 (G) and called an “L1 -group algebra”. (You find the
definition of 1 (G) in any of the above-cited books.)

Exercise 10.4.3 LetA be anunital Banach algebra. Show that, for every b ∈ A with
b < 1, the series ∞n=0 b converges and has limit (1−b) −1 . (This series is called

the Neumann series after Carl Neumann. See, e.g., [1, Sect. 4.4].)

Exercise 10.4.4 Let H be a complex Hilbert space. A linear mapping T : H →

H is called bounded if, for some constant μ ≥ 0 and all ξ ∈ H , the inequality
T ξ 2 ≤ μ ξ 2 holds (we continue to denote the norm on H by  · 2 ). In this
case, one can define the operator norm of T as
$ %
T op = sup T ξ 2 | ξ ∈ H, ξ 2 = 1 .

Show that this indeed defines a norm on the complex vector space B(H ) of all
bounded linear mappings on H . Show also that, if S, T ∈ B(H ), then ST op ≤
Sop T op and that T op = T ∗ op and T ∗ T op = T 2op , where the adjoint
T ∗ of T is given by the formula (T ξ | η) = (ξ | T ∗ η) for all ξ, η ∈ H . Also prove
that B(H ) with this norm is complete. (If you need help with this exercise, look into
[1, Sect. 2.14], [26, Chap. 3, Sect. 9] or [32, Sect. 3.2].)

Exercise 10.4.5 A C*-algebra is a subalgebra of some B(H ) (as in the previous

exercise) which contains with every operator also its adjoint and which is closed
under convergence with respect to the operator norm. Suppose {Ai | i ∈ I }is a
family of C* -algebras contained in (a fixed) B(H ). Show that the intersection Ai
is a C* -algebra in B(H ). Consequently, for every non-empty subset S of B(H )
there is a smallest C* -algebra in B(H ) containing S. (It is called the C*-algebra
generated by S.)

Exercise 10.4.6 Let K be a subfield of C which is closed under complex con-

jugation and let G be a group. Use the involution ∗ on K[G] (compare Proposi-
tion 10.2.4) to show that K[G] does not contain any non-zero nil ideal.

Exercise 10.4.7 For a group G, let F(G) denote the set of all its finitely generated
subgroups. Let α ∈ K[G], for some field K. Show that α ∈ rad(K[G]) if and only
if α ∈ rad(K[H ]) for all H ∈ F(G).

Exercise 10.4.8 Use Proposition 10.1.3 together with the previous exercise to show
that, for a group G, K[G] is semiprimitive if K[H ] is semiprimitive for all H ∈
F(G), and that rad(K[G]) is a nil ideal if rad(K[H ]) is nil for all H ∈ F(G).

Exercise 10.4.9 Let R be a non-zero ring and let G be an infinite group. Define the
augmentation map  : R[G] → R by (r) = r for all r ∈ R and (g) = 1 for all
10.4 Exercises 143

g ∈ G. Assuming that R[G] is semisimple, ker  will be a direct summand which

can be written as Re for some idempotent e ∈ R. Show that the idempotent 1 − e
has to involve every group element which is impossible as G is infinite. Conclude
that R[G] can never be semisimple.
Bibliography

1. Allan, G.R.: Introduction to Banach Spaces and Algebras. Oxford Graduate Texts in
Mathematics, vol. 20. Oxford University Press, Oxford (2011)
2. Anderson, F.W., Fuller, K.R.: Rings and Categories of Modules. Graduate Texts in Mathe-
matics, vol. 13. Springer-Verlag, New York (1992)
3. Ara, P.: Extensions of exchange rings. J. Algebra 197, 409–423 (1997)
4. Ara, P., Goodearl, K.R., O’Meara, K.C., Pardo, E.: Separative cancellation for projective
modules over exchange rings. Isr. J. Math. 105, 105–137 (1998)
5. Artin, E.: The influence of J. H. M. Wedderburn on the development of modern algebra. Bull.
Am. Math. Soc. 56, 65–72 (1950)
6. Atiyah, M.F., Macdonald, I.G.: Introduction to Commutative Algebra. Addison-Wesley,
London (1969)
7. Beachy, J.A.: Introductory lectures on rings and modules. London Mathematical Society
Student Texts, vol. 47. Cambridge University Press, Cambridge (1999)
8. Blyth, T.S.: Module Theory: An Approach to Linear Algebra. Clarendon Press, Oxford (1990)
9. Blyth, T.S., Robinson, E.F.: Basic Linear Algebra. Springer Undergraduate Mathematics
Series. Springer-Verlag, London (1998)
10. Bourbaki, N.: Elements of Mathematics. Commutative Algebra. Springer-Verlag, Berlin
(1989)
11. Canfell, M.J.: Uniqueness of generators of principal ideals in rings of continuous functions.
Proc. Am. Math. Soc. 26, 571–573 (1970)
12. Canfell, M.J.: An algebraic characterization of dimension. Proc. Am. Math. Soc. 32, 619–620
(1972)
13. Clark, J., Lomp, C., Vanaja, N., Wisbauer, R.: Lifting Modules; Supplements and Projectivity
in Module Theory. Frontiers in Mathematics. Birkhäuser Verlag, Basel (2006)
14. Conway, J.B.: A Course in Functional Analysis. Springer-Verlag, New York (1985)
15. Eilenberg, S., MacLane, S.: General theory of natural equivalences. Trans. Am. Math. Soc.
58, 231–294 (1945)
16. Facchini, A.: Module Theory: Endomorphism Rings and Direct Sum Decompositions in
Some Classes of Modules. Progress in Mathematics, vol. 167. Birkhäuser Verlag, Basel
(1998)
17. Freyd, P.J.: Abelian Categories. An Introduction to the Theory of Functors. Harper & Row,
New York (1964)
18. Gardam, G.: A counterexample to the unit conjecture for group rings. Ann. Math. 194, 967–
979 (2021)
19. Gillman, L., Jerison, M.: Rings of continuous functions. Graduate Texts in Mathematics, vol.
43. Springer-Verlag, New York (1976)
20. Golan, J.S.: The Linear Algebra a Beginning Graduate Student Ought to Know. Springer-
Verlag, Dordrecht (2007)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 145
M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3
146 Bibliography

21. Hilton, P.J., Stammbach, U.: A Course in Homological Algebra. Graduate Texts in Mathe-
matics, vol. 4. Springer-Verlag, New York (1971)
22. Jacobson, N.: Lectures in Abstract Algebra, vol. II. D. Van Nostrand, Toronto (1953)
23. Knapp, A.W.: Basic Algebra. Birkhäuser, Boston (2006)
24. Lam, T.Y.: A first course in noncommutative rings. Graduate Texts in Mathematics, vol. 131.
Springer-Verlag, New York (1999)
25. MacLane, S.: Categories for the Working Mathematician. Graduate Texts in Mathematics,
2nd. edn., vol. 5. Springer-Verlag, New York (1978)
26. Mathieu, M.: Funktionalanalysis. Ein Arbeitsbuch. Spektrum Akademischer Verlag,
Heidelberg-Berlin (1998)
27. McGovern, W.Wm.: Neat rings. J. Pure Appl. Algebra 205, 243–265 (2006)
28. Nicholson, W.K.: Lifting idempotents and exchange rings. Trans. Am. Math. Soc. 229, 269–
278 (1977)
29. Osborne, M.S.: Basic Homological Algebra. Graduate Texts in Mathematics„ vol. 196.
Springer-Verlag, New York (2000)
30. Pareigis, B.: Categories and Functors. Academic Press, New York/London (1970)
31. Passman, D.S.: The Algebraic Structure of Group Rings. John Wiley & Sons, New York
(1977)
32. Pedersen, G.K.: Analysis Now. Graduate Texts in Mathematics, vol. 118. Springer-Verlag,
New York (1995)
33. Taylor, M.E.: Linear Algebra. Pure and Applied Undergraduate Texts, vol. 45. American
Mathematical Society, Rhode Island (2020)
34. Wallace, D.A.R.: Groups, Rings and Fields. Springer Undergraduate Mathematics Series.
Springer-Verlag, London (1998)
Index of Symbols

Symbols ob(C), 9
B(H ), 142 MorC (A, B), 10
C(X), 28 1 (G), 141
K-algebra, 101 1, 68
L(V ), 105 2 (G), 141

Mn (R), 2 AGr, 10
R[x], 2 Ban1 , 10
Tn (R), 2 Ban∞ , 10
Z(R), 23 Cop , 12
c0 , 68 Comp, 10
p -group, 140 CMetric, 107
R[G], 3 C, 9
R R, 6 Gr, 10
HomR (MR , NR ), 5 HTop, 10
HomR-S (R MS , R NS ), 5 Lat, 10
End(G), 2 Metric, 107
GLn (K), 7 Mod-R, 4, 10
Max(R), 123 Ring1 , 10
Spec(R), 123 Ring, 10
AC1∗ , 31 S, 10
R M i , 18 Top, 10
i∈I soc(R), 89
codom(f ), 10 A ⊗K B, 101
cok(f ), 41 B(M × N, G), 93
dim1 R, 121 CR (X), 119
dim2 R, 121 Cr∗ (G), 135
dom(f ), 10 (A), 31
C, 2 E(R), 123
H, 2 M ⊗R N, 94
Q, 2 R × , 13
R, 2 R-Mod-S, 4, 10
Z, 2 R-Mod, 4, 10, 11
HomR (R M, R N), 5 x ⊗ y, 94
im(f ), 12 p-support, 140
ker(f ), 12, 41 AnnR (S), 16
p-supp
 α, 140
R S, 17
R M i , 17 πN , 16
i∈I
mor(C), 9 tor(R M), 57
RX, 19

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 147
M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3
Index

A D
Algebra Descending chain condition (DCC), 48
normed, 134 Dimension
over a field, 91 first Canfell dimension, 121
Annihilator, 16 second Canfell dimension, 121
Artin, Emil, 47 zero-dimensional space, 119
Artin–Wedderburn Theorem, 75 Direct limit
Ascending chain condition (ACC), 35 of modules, 91
Direct product
of modules, 17
B Direct sum
Bimodule, 4 of modules, 18
Dorroh superring, 13

C
C*-algebra, 135, 142 E
commutative, 31 Endomorphism
reduced group, 135 of a group, 2
Category, 9 Epimorphism, 12, 15
abelian, 54 canonical epimorphism, 16
additive, 23 Equalizer
arrow in a, 9 of two morphisms, 44
codomain of a morphism, 10 Evaluation map, 30
concrete, 10, 27 Exact sequence, 41
domain of a morphism, 10 of modules, 32
dual category, 12 short exact sequence, 32, 42
exact, 42 split, 63
final object, 23
full subcategory, 11
initial object, 23 F
morphism in a, 9 Free object
object in a, 9 in a category, 21
reflective subcategory, 107 Functor, 27
small category, 10, 11 additive, 33
subcategory, 11 constant, 27
zero object, 23 contravariant, 27
Cohomology group, 70 covariant, 27
Composition series, 88 exact, 42
Coproduct faithful, 27
in a category, 22 forgetful, 27
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 149
M. Mathieu, Classically Semisimple Rings,
https://doi.org/10.1007/978-3-031-14209-3
150 Index

full, 27 basis of a module, 19

Hom-functor, 28 completely reducible, 59
identical, 27 cyclic, 17
inclusion, 28 exchange property, 109
left adjoint, 103 faithful, 17
left exact, 42 finite exchange property, 109
representable, 29 finitely cogenerated, 48
right adjoint, 103 finitely generated, 11, 17
right derived, 70 flat, 102
right exact, 42 free module on a set, 19
generated submodule, 17
generator of a submodule, 17
H injective, 64
Hahn–Banach theorem, 69 irreducible, 59
Hilbert’s Basis Theorem, 37 left, 3
Hilbert space, 141 Noetherian, 36
Homomorphism projective, 64
module homomorphism, 4 quotient module, 16
canonical factorisation, 25 right, 4
Hopkins–Levitzki Theorem, 52 semisimple, 59
simple, 59
standard R-module, 6
I submodule, 4
Ideal torsion module, 50
left primitive, 90 torsion submodule, 57
nil, 85 unital, 3, 13
nilpotent, 85 Monomorphism, 12, 15
prime, 13, 90 Morphism, 9
Injective resolution, 70 cokernel of a morphism, 41
Inner product space, 141 kernel of a morphism, 40
Involution zero morphism, 40
in a C* -algebra, 135
Isomorphism, 12, 15
N
Nakayama’s Lemma, 92
J Natural equivalence, 31
Jordan–Hölder theorem, 88 Natural isomorphism, 29
Natural transformation, 29
Noether, Emmy, 36
L
Linearly independent set, 19
O
Object
M injective, 68
Mapping M-injective, 69
bi-additive, 3 projective, 68
K-bilinear, 101 representing, 29
R-balanced, 93
Maschke’s theorem, 80
Modular law, 24 P
Module, 3 Polynomials, 2
c-exchange property, 111 Prime ideal space, 123
Z-module, 8 Product
Artinian, 48 in a category, 21
Index 151

Pullback semiregular, 128

in R-Mod, 44 semisimple, 64, 83
Pushout left semisimple, 77
in R-Mod, 45 right semisimple, 77
simple, 2
tb -ring, 122, 125
Q topologically boolean, 125
Quaternions, 2 unital, 2
von Neumann finite, 43, 137
von Neumann regular, 90
R
Radical
Jacobson radical, 51, 82, 86 S
nil radical, 13, 51 Schur’s Lemma, 60
Representation, 7 Short Five Lemma, 55
of a group, 7 Socle
left regular representation, 7 of a ring, 80, 89
of a ring, 7 Spectrum
Ring, 1 of a commutative ring, 123
Artinian, 48 maximal spectrum, 123
left Artinian, 48
right Artinian, 48
clean, 119 T
commutative, 2 Tensor product, 94
completely reducible, 64 of algebras, 101
exchange, 114 of modules, 94
Gelfand, 122 of vector spaces, 105
left primitive, 83 Topology
local, 129 hull-kernel topology, 123
minimal unitisation, 13 Zariski topology, 123
Noetherian, 37 Trace
left Noetherian, 36 on C* -algebra, 135
right Noetherian, 36 on group ring, 13
pm -ring, 122, 123 normalised faithful, 135
reduced, 5, 90
right primitive, 83
semiprime, 90 W
semiprimitive, 83 Wedderburn, J. H.M., 75