Rings, Fields and

Groups
An Introduction to Abstract Algebra
R B J T Allenby
Senior Lecturer, School of Mathematics, University of Leeds

Edward Arnold
A division of I Iodder & Stotugliton
I10N1)ON NEW YORK ME AI'CKLANI)
For Janet,
Elizabeth and
Rachel

© 1991 R. B. J. T. Allenby
First published in Great Britain 1983
Reprinted with corrections 1985 and 1986
Reprinted 1988. 1989
Second edition 1991

Distributed in the USA by Routledge, Chapman and Hall. inc..

29 West 35th Street. New York, NY 10001

Britich Library Cataloguing Publication Data

Allenby. R. B. J. T.
Rings, fields and groups.
1. Rings (Algebra)
1. Title
512'.4 QC251
ISBN 0-7131-3476-3

All rights reserved. No part of this publication may be reproduced

or transmitted in any form or by any means. electronically or
mechanically. including photocopying, recording or any information
storage or retrieval system, without either prior permission in
writing from the publisher or a licence permitting restricted copying.
In the United Kingdom such licenccs arc issued by the Copyright
Licensing Agency: 90 Tottenham Court Road, London WIP 9HE
Typeset in 10/l2pt Times by MS Filmsetting Limited, Frome, Somerset
Printed in Great Britain for Edward Arnold,
a division of Hodder and Stoughton Limited, Mill Road,
Dunton Green, Sevenoaks, Kent TNI3 2YA
by St Edmundsbury Press Ltd. Bury St Edmunds, Suffolk
and bound by Hartnolls Ltd, Bodmin, Cornwall
Preface to the first edition

Theoverall aim of this book is to present a fairly leisurely introduction to

some of the results, methods and ideas which are increasingly to be found in
first and second year abstract algebra courses in British universities and
polytechnics and in equivalent courses elsewhere.
There are several ways in which an author might present such an introduc-
tion. One is the (to some) aesthetically most pleasing take-it-or-leave-it purely
axiomatic approach in which the reader is given a list of the appropriate
definitions and is then led through proofs of those theorems universally agreed
by those already in the know to be basic to the subject. The guiding spirit
behind such an approach would almost certainly force the author to begin
with a long and rather dry axiomatic development of the set-theoretical
language needed. Although the purity of this particular approach would
probably be preferred by a majority of practising pure mathematicians who
already have a degree of familiarity with the material, the present author has
found that it is appreciated by only a handful in every class of fifty or so
beginners. Many students become restless at such an approach as they find
difficulty in connecting the discussion with ideas with which they are already
familiar. The sudden change from the world of concrete examples usually
found in school mathematics to the abstract setting seemingly remote from the
real world is one which can lead some students away from abstract algebra in
particular and pure mathematics in general, a state of affairs which naturally
saddens the author who much enjoys sharing with his beginning students the
pleasure (even excitement!) to be obtained from following through some of the
clever ideas and neat arguments to be found there.
A second approach, possibly more attractive to the beginner, would be to
present a detailed historical account of the develolpment of algebra from, say,
1500 to the present day. To the student who is aware of some of the upheavals
in school mathematics courses in recent years, the inclusion of an account of
some of those theories which were once vigorously developed and were
expected to become important but have generally failed to find favour or
application might prove especially interesting! However, such an approach
would leave, in a volume of reasonable size, little room for a really detailed
account of any of the theories reviewed.
An intermediate course, and the one taken here, is to try to get fairly quickly
into the spirit of abstract algebra, whilst at the same time interjecting
occasional remarks and comments, either because of their historical content
iv Preface

or because, quite simply, the author thinks they are fascinating (or both!). In
particular, several complete sections are included more as light reading than as
essential material. Amongst these I include Sections 3.5, 3.9, 4.4, 5.2, 5.12 and
6.7. The present approach, therefore, is a mixture of the formal and the
informal. The author has certainly found this mixture acceptable in courses he
has given on both sides of the Atlantic. On the one hand we take an informal
approach to set theory because our concern is with the algebra; on the other
hand we do not want to throw all formal working to the wind, present day
algebra being, as it is, an axiomatic discipline. Indeed, one of the chief aims of
the book is to develop the reader's critical faculties and we believe it preferable
to begin this in Chapter 1 where, because of the student's (intuitive) familiarity
with much of its content, the critical approach seems all the more prominent.
(See also the Problems posed at the end of the Prologue.) Here formal
definitions are given when some readers might feel that informal ones would
suffice. (See, for instance, the development of the idea of polynomial from
Sections 1.6 to 1.8). Such readers should not need encouraging to read,
carefully, the reasons put forward for preferring the more formal approach.
We also use this chapter to try and answer, by example, a question often asked
by beginners, namely 'How much proof should I give?' The answer clearly
depends upon the knowledge and maturity of the people trying to correspond.
To help the reader through his first encounter with several of the proofs the
author has been more expansive than he would normally be in communicating
with a colleague. Those extra portions, which can be omitted as confidence
grows and which to some extent are the answers to the questions the reader
should be asking himself as he works through the text, have been put in square
brackets.
We begin with a prologue in which we attempt to answer some of the
questions students seem afraid to ask: What is abstract algebra? How did it
develop? What use is it? The historical account of the development of algebra
will include many words not familiar to the beginner, but we feel that in a new
land it is preferable to possess a map, even one in a foreign language, than no
map at all. In placing this material before rather than after the main body of
the text we hope to whet the reader's appetite and heighten his sense of
excitement with a description of the discoveries and inventions made by some
of the mathematical giants of the past and that this excitement will fire him
sufficiently to read this book avidly even when (as it probably will) the going
gets a bit difficult.
The numbering of Chapter 0 indicates that we view it as a preliminary to the
text proper. Chapters 1, 2, 3 and 4 concentrate on algebraic systems known as
rings and fields (though these names are not formally introduced until Chapter
3) the concept of group not being mentioned until Chapter 5. A majority of
texts on abstract algebra offer a study of the theory of groups before a study of
rings, the reason often given being that groups, having only one binary
operation, are simpler to begin with than are rings and fields which have two.
The author (a group theorist!) feels that there is a rather strong case for
 Preface v

reversing this order; the fact is that natural concrete examples of rings and
fields (the integers, polynomials, the rational, real and complex numbers) are
much better known to the beginner than are the equivalent concrete examples
of groups (mainly symmetries of 2- and 3-dimensional figures). (This author
will just not accept the complaint 'But the integers form a group under
addition': so they do but that is not the natural way to look at them. Indeed
the author rebels strongly against the argument which, briefly, runs: 'The
integers under addition form a group. Therefore we must study group
theory.') What in the author's opinion really clinches the argument for
studying rings and fields before groups are the several exciting applications
that can quite quickly be made to easily stated yet non-trivial problems in the
theory of numbers and of geometrical constructions. (See especially Sections
3.8 and 4.6.)
The placing of ring theory before group theory will, it is true, give rise to a
little more repetition of corresponding elementary concepts than might have
been the case with the more usual presentation. The author does not, however,
feel any need to apologise for that! (On the other hand Chapters 5 and 6 make
little essential use of Chapters 3 and 4, so they can be studied directly after
Chapter 2.)
Throughout the text problems, numbered only for ease of reference, have
been inserted as they have occurred to the author. Some are reasonably easy,
some solved later in the text and some are quite hard. I leave you to find out
which! The purpose of these problems is (i) to set you thinking and then
discussing them with a colleague or teacher; (ii) to get you into the habit of
posing questions of this kind to yourself. Active participation is always much
more exciting (and instructional) than passive reading!
In this book the numbering of theorems, lemmas, etc. in any one chapter is
consecutive, thus: Theorem 5.5.4, Example 5.5.5, Notation 5.5.6, Definition
5.5.7. When referring to a theorem, lemma, etc. given elsewhere in the text
usually only its number is given (e.g. 3.8.2). Reference to an exercise is
however given in full (e.g. exercise 3.2.14) except when the exercise referred to
is at the end of the section concerned. Thus, within Section 3.2, exercise 3.2.14
would be referred to as 'exercise 14'.
In producing this text I have received help from several people, especially
from the secretaries in the School of Mathematics in the University of Leeds.
In particular I should like to thank Mrs M M Turner, Mrs P Jowett, Mrs A
Landford and Mrs M R Williams. Several colleagues in Leeds and elsewhere
have offered helpful comments and gentle criticism on parts of the manuscript.
Here I especially with to thank Drs J C McConnell, E W Wallace and J R
Ravetz. For supplying me with photographs I thank the keeper of the David
Eugene Smith collection at Columbia University, New York, and especially
Prof Dr Konrad Jacobs of the University of Erlangen—Nürnberg who kindly
donated the pictures of Emmy Noether and Richard Dedekind.
Leeds RBJTA
1982
Contents
Preface to the first edition

Preface to the second edition vi

How to read this book

Prologue

0 Elementary set theory and methods of proof 1

0.1 Introduction
0.2 Sets
0.3 Newsets from old 3
0.4 Some methods of proof 5

I Numbers and polynomials 10

1.1 Introduction 10
1.2 The basic axioms. Mathematical induction 10
1.3 Divisibility, irreducibles and primes in 1 20
Biography and portrait of Hubert 25
1.4 GCDs 26
1.5 The unique factorisation theorem (two proofs) 33
1.6 Polynomials—what are they? 35
1.7 The basic axioms 37
1.8 The 'new' notation 39
1.9 Divisibility, irreducibles and primes in Q[x] 42
1.10 The division algorithm 49
1.11 Roots and the remainder theorem 51

2 Binary relations and binary operations 57

2.1 Introduction 57
2.2 Congruence mod n. Binary relations 57
2.3 Equivalence relations and partitions 61
2.4 63
Biography and portrait of Gauss 68
2.5 Some deeper number-theoretic results concerning
congruences 69
viii Contents

2.6 Functions
2.7 Binary operations

3 Introduction to rings
3.1 Introduction
3.2 The abstract definition of a ring
Biography and portrait of Hamilton
3.3 Ring properties deducible from the axioms
3.4 Subrings, subfields and ideals
Biography and portrait of Noether
Biography and portrait of Fermat
3.5 Fermat's conjecture (FC)
3.6 Divisibility in rings
3.7 Euclidean rings, unique factorisation domains and principal
ideal domains
3.8 Three number-theoretic applications
Biography and portrait of Dedekind
3.9 Unique factorisation reestablished. Prime and maximal ideals
3.10 Isomorphism. Fields of fractions. Prime subfields
3.11 U[x] where U is a UFD
3.12 Ordered domains. The uniqueness of Z

4 Factor rings and fields

4.1 Introduction
4.2 Return to roots. Ring homomorphisms. Kronecker's theorem
4.3 The isomorphism theorems
4.4 Constructions of R from Q and of C from R
Biography and portrait of Cauchy
4.5 Finite fields
Biography and portrait of Moore
4.6 Constructions with compass and straightedge
4.7 Symmetric polynomials
4.8 The fundamental theorem of algebra

5 Basic group theory

5.1 Introduction
5.2 Beginnings
Biography and portrait of Lagrange
5.3 Axioms and examples
5.4 Deductions from the axioms
5.5 The symmetric and the alternating groups
5.6 Subgroups. order of an element
5.7 Cosets of subgroups. Lagrange's theorem
 Contents ix

5.8 Cyclic groups 213

5.9 Isomorphism. Group tables 216
Biography and portrait of Cayley 220
5.10 Homomorphisms. Normal subgroups 223
5.11 Factor groups. The first isomorphism theorem 229
5.12 Space groups and plane symmetry groups 233

6 Structure theorems of group theory 242

6.1 Introduction 242
6.2 Normaliser. Centraliser. Sylow's theorems 242
6.3 Direct products 250
6.4 Finite abelian groups 254
6.5 Soluble groups. Composition series 258
6.6 Some simple groups 268

7 A brief excursion into Galois theory 274

7.1 Introduction 274
Biography and portrait of Galois 275
7.2 Radical Towers and Splitting Fields 276
7.3 Examples 280
7.4 Some Galois groups: their orders and fixed fields 283
7.5 Separability and Normality 287
7.6 Subfields and subgroups 291
7.7 The groups Gal(R/F) and Gal(S1/F) 297
7.8 The groups - 300
7.9 A Necessary condition for the solubility of a polynomia
equation by radicals 302
Biography and portrait of Abel 303
7.10 A Sufficient condition for the solubility of a polynomial
equation by radicals 305
7.11 Non-soluble polynomials: grow your own! 308
7.12 Galois Theory—old and new 311

Partial solutions to the exercises 315

Bibliography 364

Notation 371

Index 373
Preface to the second edition
Some while ago I decided to inflict, on the mathematical community, yet
another algebra book. One reason was my belief that few texts, if any, then
available both proved how exciting abstract algebra is and showed the reader
'behind the scenes' of the subject. Indeed it is not as easy to do this on paper as
in the lecture theatre where one can laugh with or shout at the audience to
convince them what a good time they are having (and there is no literary
equivalent of enlivening the lecture by losing the chalk or falling off the
podium). Nevertheless, I found myself writing as if I were lecturing, occa-
sionally wanting to offer tit-bits of information or what I hoped would be
helpful asides. So it was most gratifying to learn that this style was acceptable
to many readers—students and teachers alike.
In asking me to prepare a second edition, the publishers kindly offered me
extra pages which, after some consultation, I decided to fill with (i) a brief
account of Galois Theory and (ii) hints/outline solutions to many of the 800 or
so exercises. (I hope, from the teacher's [students'] point of view, that I haven't
overdone [underdone] the latter.) Of course I have also taken the opportunity
to consider if my original text said what I had intended. Where it didn't I've
incorporated changes. The necessity of amending what Churchill might have
called 'terminological inexactitudes' was pointed out to me by many friends
and colleagues. In particular I should like to thank: Seymour Bachmuch (U.C.
Santa Barbara), Allen Bell and Ken Goodearl (University of Utah), Al Hales
(U.C.L.A.), Bob Gregorac (Iowa State University), John Meldrum (Univer-
sity of Edinburgh) and John Silvester (King's College, London) as well as
some of my colleagues at Leeds. (Special thanks are due to Bell, Goodearl and
Hales for sending me, unrequested, full solutions to the exercises they set in
class. If they were thereby hinting at something, I'm afraid I've failed to grasp
the message!) Incidentally I am not averse to receiving comments directly from
students and I thank all who have written, including one from the University
of Sana'a.
I have also updated the bibliography, including some books and papers
which have appeared since the first edition of RFG went to press. However I
(continue to) invite the reader to experience the pleasure of seeking out others
I haven't found room to mention.
Finally I must thank (most effusively!) my colleague John McConnell who
willingly surrendered the notes of his own successful lectures on Galois Theory
and then (wilfully?) misunderstood my invitation to 'look over Chapter 7',
taking it as a challenge to make me keep my account close to the way Galois
Theory should be presented! Without his help the new chapter might well have
contained more 'terminological inexactitudes' than did the whole of the
original text.
Leeds RBJTA
1991
How to read this book

I hope it won't be considered too presumptuous if I offer some advice to the

reader who finds himself, perhaps for the first time, left to work on his own.
To most of us, reading a mathematics book is a lot harder than reading a
novel. A certain amount of application is required. Begin by getting up out
of that armchair and, with pencil and paper, find yourself a nice clear desk and
a comfortable but upright seat.
Perhaps the first thing to realise is that learning is not a linear business. I
have known students who feel obliged to know pages 1—65 almost by heart
before they dare turn to page 66. This is silly. If your conscience allows you
to say that you really are stuck at some point don't be afraid to pass on to
the next paragraph. Things can often become clearer with hindsight.
If, however, you find yourself carrying three or four such problems, it seems
clear that there is nothing for it but to go back and attack the first difficulties
again. Incidentally, one way to motivate yourself if things do get sticky is to
imagine that you have to explain the subject to the class the next day. (Think
of all the unpleasant questions your classmates might ask you and make sure
you can answer them!)
Definitions are sometimes hard to grasp on first reading. One method I've
found helpful is to learn the definition by heart. Once you have control over
the words you can concentrate all your efforts on their meaning.
Perhaps the best approach to reading a proof for the first time is positively
to disbelieve each assertion made. This commits you to examining each word
and can be quite tiring. On the other hand the exhilaration following the
successful completion of such a task is well worth the effort. Moreover, having
mastered the proof as it were 'locally', it should now not be too difficult to
master it 'globally' by identifying the key points that make the proof work.
Explain the method—not the details—to yourself.
Regarding exercises: before attempting to answer a question do make sure
you know the meaning of all the words in it! It may seem obvious but some
appear to forget that if you don't understand what a question asks you have
little chance of answering it! If a problem (or a proof) seems too difficult as
it stands one can try looking at a special or a simpler case. This may point
the way in regard to the original problem. In any event, if you can't manage
a simpler case you are surely not going to succeed with the more difficult one.
The message here is: Don't be afraid to examine lots of concrete examples.
Drawing pictures, as I do in this text, can also be helpful.
xii How to read this book

One more (obvious) point. Don't expect the various parts of solutions to
problems to occur to you in the 'right' order. (I can assure you that this book
was not written straight through from beginning to end!) Having got a rough
draft of a solution, now write it out neatly and in a logically developed manner
so that you will be able to read and understand your solution in 6 months time.
Finally may I say that I hope that (most days) you enjoy reading this book
as much as I (most days) enjoyed writing it.
Prologue

In this prologue we discuss some of the questions which few beginners seem
to have the courage to ask and yet to which they would surely like some kind
of answer The questions considered are (i) what is algebra? (ii) what is its
history? and (iii) what is it good for? We urge the reader to dip frequently
into the historical outline below. The discoveries mentioned there of some
of the world's best mathematicians should whet the appetite for, and also
place in some kind of perspective, the mathematics covered in Chapters 1
through 6.

What is algebra?
The word algebra derives from the word al-jabr which appears in the title of
a book written in the 9th Century by the Persian mathematician Mohammed
Al-Khowarizmi (from whose name comes the word algorithm). This book,
in a Latin translation, had great influence in Europe. Its concern with problems
equivalent to those of solving polynomial equations, especially those of degree
2, led to the word algebra eventually becoming synonymous with the science
of equations. This state of affairs persisted into the 19th Century, Serret, in
1849, observing that 'Algebra is, properly speaking, the analysis of equations.'
One possible definition of algebra [1321* is that it is the study of operations,
of rules of computation. This is slightly unfair since, as we observe in Chapter
2, not all operations are interesting. In any case the word algebra is nowadays
often prefixed, to indicate different stages of development, by adjectives such
as classical, modern and abstract. Whilst there does not appear to be any
universal agreement on the precise meaning to be attached to these prefixes
(compare the definitions of modern algebra in [80, p. 669] and [92, p. 702)]
we can fairly safely say that classical algebra is the present synonym for the
theory of equations, a theory in which are manipulated symbols which invari-
ably represent numbers, be they complex, real, or rational. The term modern
algebra can then be used to describe that subsequent algebra, some of it
arising from more detailed investigations within the classical theory, in which
the symbols manipulated are no longer restricted to representing numerical
quantities. (Cauchy, in 1815, had defined multiplication of permutations,
whilst Gauss, in 1801, had combined pairs of binary quadratic forms and
also integers 'mod p'.) Finally, we give the name abstract algebra to that
*
Brackets such as these refer to the bibliography.
xiv Prologue

generalisation of modern algebra in which the main object is the study of

algebraic systems defined solely by postulates or axioms (usually chosen not
arbitrarily but with several concrete instances in mind), no particular meaning
being attributed to the symbols being manipulated.

Abstraction and axiomatisation

The introduction of the concepts of abstraction and axiomatisation is attribu-
table to the Greek mathematician/philosophers of the period 600—300 BC.
Thus, neither is a recent invention.
The method of abstraction has several interconnected uses. By stripping
concrete objects of their less essential features, they become less involved
and hence more amenable to mathematical treatment. Also one increases the
chance of revealing similarities between superficially distinct objects, so that
a theorem from one area of mathematics may suggest an analogue in another
area. Further, several concrete theories can be studied simultaneously under
the umbrella of one general theory of which the concrete theories are special
cases. Finally, one penetrates to the real reasons for the success (or failure!)
of a theory. In this way a deeper understanding should result.
As a simple illustration of these remarks we note that the Babylonians
posed many problems [80, p. 34] of the form: 'Find the side of a square if
the area less the side is 870.' The verbal solution proceeds: 'Take half of 1,
that is multiply by that is 1. Add this to 870 to get 870L This is the
square of 29k. Add to to get 30, the side of the square.' Of course this
same procedure is applicable to other problems of a like nature. But how
much deeper in content and how much easier to understand is the observation
that infinitely many such problems can be dealt with all at once, as it were,
by replacing numerical coefficients by literal coefficients of no initial specific
numerical value. In short: if ax2+bx+c =0 then x =(—b±'1b2—4ac)/2a.
(In the above problem a = 1, b = —1, c = —870.)
The axiomatic method which rejects proofs based on intuition involves
deducing by logical argument alone, from initial statements assumed without
question (the axioms), other statements (the theorems). The most widely
quoted example of the use of this method is Euclid's Elements (c. 300 BC) in
which the author supposedly sought to give a consistent foundation to
geometry. (For a different point of view see [121].) It is argued in [125] that
Euclid's fifth axiom (The whole is greater than the part) was included
specifically to eliminate from mathematics the paradoxes (c. 450 BC) of the
infinite introduced by Zeno ([85, p. 35—7]). Furthermore, the 'crises' brought
about by the discovery that -'12 is not a rational number clearly called for
rigorous investigation: after all, the intuitively 'true' had been proved false!
(See problem 3 below.) Thus the axiomatic method was called upon to help
confirm the basic consistency of mathematics.
For the next 2000 years the abstract/axiomatic approach was with a few
exceptions replaced by a more concrete intuitive approach. However, in the
headlong rush to develon the newborn calculus of Leibniz and Newton.
 Prologue xv

absurdities arising from the free use of intuitive geometrical arguments (see
problem 1) led to the so-called second great crisis and a call for analysis to
be made more rigorous; in other words based on arithmetic (whose foundations
were obviously (!) secure).
The publication by Lobachevsky in 1829 of a consistent (non-Euclidean)
geometry, in which Euclid's parallel postulate is denied, should perhaps have
turned mathematicians' attention back to a study of axioms, especially as it
had long been appreciated that Euclid's use of the axiomatic method was, to
say the least, inconsistent. ([121] contains stronger views.) However, this work
apparently attracted little attention for several years.
In due course the 'arithmetisation' of analysis got under way: Dedekind
insisted that 'what is provable should be proved', observing that even the
equality x '/3 = had not yet (1858) been satisfactorily estab-
lished.
Eventually the desire continually to express concepts in terms of yet more
'fundamental' ones led to Peano setting down in 1889 his symbolic and
axiomatic description of the set of integers (in terms of the undefined concepts:
set, belongs to, zero, number, successor of). Furthermore the whole numbers
themselves were shown to be definable entirely in terms of Cantor's new
notion of set (Section 0.1). Unfortunately Cantor's definition was too
wide-ranging: intuition had failed once again at the most basic level (see
Russell's Paradox in Section 0.2) and the third great foundational crisis was
at hand. One result was Zermelo's attempt (1908) to build a formal set theory
on an axiomatic basis.
One of the facets of 19th Century algebra, the ever-increasing number of
concrete structures of distinct outward form but with similar underlying
properties, encouraged their cataloguing and comparison by abstracting com-
mon features. Indeed Weber's book [40] of 1893 talks not just of (groups of)
permutations, matrices, etc., but of (groups of) 'things' (Din gen, in German).
Basing his proposals on the more commonly used properties shared by
permutations, matrices, etc., he postulated that his 'things' be subject to similar
rules (i.e. axioms). It would then follow (as Boole had said in 1847) that 'the
validity .. does not depend on the interpretation of the symbols .... Every
.

system of interpretations ... is equally admissible.' Note that whilst the

axiomatic method permitted, in theory, the assuming of arbitrarily chosen
sets of axioms, those adopted were not chosen at random, the aim being to
reflect properties of concrete systems already deemed important.
Thus the abstract/axiomatic method has a role to play in classification and
reorganisation; in making special results more intelligible by identifying com-
mon features. The method thus supplies greater transparency and insight and
leads to a unified approach offering progress along a wide front. It also helps
us avoid making intuitively obvious but unfounded assertions and hence
'proving' false theorems.
A final remark. Since Euclid's time the concept of axiom has changed
somewhat. To Euclid axioms were unshakeable truths (e.g. every two distinct
points uniquely determine a line). Today we interpret the word axiom
xvi Prologue

differently. We do not ask whether or not an axiom is 'true'—just as we don't

ask* if the rules of chess are 'true'. (One can of course question whether or
not the axioms are appropriate if one is trying to model a concrete example.)
The Greek word axiorna originally meant 'request'; the reader is requested
to accept the axioms unquestioningly as the rules of the game. All he can
(and should) question is whether or not the asserted conclusions follow
logically from the axioms.

Historical development of algebra

As stated above, the algebra now called classical concerned itself with (poly-
nomial) equations, in particular with attempts to supply formulae for the roots
—b
of equations of degrees 3, 4, 5, etc. The solutions x of the
= 2a
general quadratic equation ax2 + bx + c =0 were known to the Babylonians
via the process of completing the square (although only positive solutions
were found acceptable).
A long search ensued for similar formulae in terms of radicals (that is,
formulae involving just +, —, ÷, and the coefficients of the given
equation) for the roots of equations of higher degree, but none appeared until
the 16th Century when a formula for the cubic was found by the Italians
Scipione del Ferro, Niccolo Fontana (more commonly known as Tartaglia,
'the stammerer', because of a speech impediment brought about by injury in
childhood) and Jerome Cardano. There is no need here to go into the rivalries
of the various mathematicians involved in the search for the cubic formula—
however, it makes for fascinating reading ([80], [84], [85], [89]).
Not long afterwards the solution of the general quartic (or biquadratic)
equation was found by the Italian Lodovico Ferrari and in 1545 both the cubic
and quartic formulas were published in Cardano's Ars Magna. After this
double success hope must have been high that a solution in the case of the
general quintic would soon be forthcoming. New ways of solving the quartic
equation were discovered but still the quintic equation resisted attack.
In 1770/71 Lagrange set about analysing the various methods then known
for dealing with the general equations of degrees 2, 3, 4 and he found that
they all depended on the same general principle (see Section 5.2). In particular
Lagrange showed cause for finding the number of formally distinct 'values'
taken by a function, for example xy + zt, on permuting the symbols x, y, z, t,
in all possible ways. Although Lagrange seemed to cherish hopes that his
work would show the way to the solution of the general quintic, the results
obtained indicated a distinct possibility that there might be no corresponding
formula in the case of equations of degree greater than 4, and a supposed
proof of this was given by Ruffini in 1799. It was apparently difficult to see
if Ruffini's proof was complete and, despite a further attempt by him at the
*[78, p. 231.
 Prologue xvii

problem, in 1813, the credit for supplying the first generally accepted proof
of impossibility goes to Abel (pronounced 'Arbel') in 1824. Thus it was proven
that no universal radical formula for obtaining all the roots of every quintic
was available. Yet it was undeniable that such formulae were available for
certain special quintic equations (for instance the five roots of x5 — 1 are each
radically expressible; see exercise 4.6.6). In 1832 Evariste Galois described,
by associating with each equation a finite group, exactly which equations were
treatable. This result is a mere corollary to a much more general theory, still a
subject of research, called Galois Theory. (See Chapter 7). Galois, it is usually
said,* coined the word group at this time and introduced the concept of
normal subgroup. He was also the first to investigate fields with finitely many,
pfl, elements where p is a prime and (See Section 4.5.)
At the turn of the century other ideas, later to be seen as part of algebra,
were coming from the pen of Carl Friedrich Gauss, one of the greatest
mathematicians who ever lived. Before he was 19, Gauss had constructed,
by straightedge and compass, a regular 17-gon, the first 'new' constructible
regular polygon for 2000 years, and in 1799 he gave the first satisfactoryt
proof (where Newton, Euler and Lagrange had failed) of the fundamental
theorem of algebra (4.8.1). Gauss' most influential contribution is probably
his Disquisitiones Arithmeticae (1801), a work in which appear his 17-gon
(see Section 4.6), his introduction of the notation of congruence (see Section
2.2), his proof that all the roots of f' — 1 are expressible in radicals, and a
proof of the quadratic reciprocity law. (For an integer a and a prime r not
dividing a, define to be 1 or —1 according as the congruence
x or not. The reciprocity law states that, for distinct
odd primes, (!')Q')=(_ See, for example [42], Chapter 8.)
Despite the fact that at the turn of the century mathematicians were happily
(in most casest) using complex numbers, there was some residual disquiet
relating to them. First, although Wessel (1797) and Argand (1806) had tried
to make complex numbers a little more respectable by showing how to
interpret them, their addition and multiplication geometrically, there
remained, possibly because of doubts concerning the intuitive use of
geometrical arguments, the desire to put them ona firmer basis. Furthermore
there was still the problem 'What exactly is (see Section 4.4). Suffice
it here to say that this problem was finally dispensed with in 1833, by banishing
(!): Hamilton replaced the objectionable a + ib by the ordered pair
(a, b) of real numbers (see Section 4.4), thus duplicating an earlier (unpub-
lished) work of Gauss. Second, unease was caused by the continued manipula-
tion of letter symbols as if they were positive integers in situations in which
*
Kiernan [108] thinks it might have been Galois' teacher L Richard.
t Taking into account the somewhat less rigorous demands of the period.
t There were still some who thought negative whole numbers absurd!
xviii Prologue

they clearly were not. For instance, the equality (a —b)(c —d) =
ac + bd — ad — bc, acceptable to everyone whenever a, b, c, d were whole
numbers with b less than a and d less than c, appeared to remain valid when
particular irrational or complex numbers were substituted for the letters. In
addition, due to the efforts of Woodhouse, Babbage, Herschel and Peacock
from 1803 onwards, Leibniz' notation and methods in the calculus had
gradually replaced Newton's in England and as a consequence much of
continental analysis had become available to the English. In particular the
calculus of operations became an English preserve [109]. In this calculus one
combines functions as if one is dealing with algebraic quantities. Thus in the
differential equation + + 2y =0 one separates the 'operator'

—r+3 —+2 from the operand y and even factorises it as

dx dx dx /dx
(Indeed, later (1843), Boole applied the method of partial fractions writing
1 1 1__.)
as So there was obviously raised the question
d d — d
—y+3—+2
dx dx
—+1
dx
—+2
dx
as to the validity of doing this sort of thing. It is in fact the case that whereas
not all the early investigators even tried to validate their reasoning, several,
including Cauchy, Servois and Boole, certainly did. And it was in this connec-
tion that Servois, in 1815, introduced the notions of functions which are
'distributive' and 'commutative', terms still used today (see Section 1.2).
It was in this atmosphere that Peacock, his friends Babbage and Herschel
having worked in the calculus of operations, introduced (1830, 1833, 1842)
his two concepts of algebra: arithmetic algebra and symbolic algebra. Arith-
metic algebra, declares Peacock, concerns the use of symbols representing
positive integers so that, for instance, the expression a —b is meaningless if
a <b. Symbolic algebra is the same as arithmetic algebra except that its
operations are universally applicable: a —b has meaning for all a, b. The rules
for operating in symbolic algebra are to be just those of arithmetic algebra.
A consequence of this is Peacock's Principle of Permanence of Equivalent
Forms which essentially says 'Whatever the occasion, a = b a'. It was
against this background, namely that it seemed preposterous to suppose that
a b could ever be other than equal to b a in any consistent algebraic setting,
that Hamilton looked for more than 10 years for an extension of complex
numbers suitable for application to the physics of 3-dimensional space. At
last, in 1843, he broke through the psychological barrier by founding a
consistent set of 4-dimensional 'hypernumbers' in which all the usual laws of
arithmetic hold with the exception that a b # a in general. He had invented
the quaternions. (The reader who fails to see how any system of entities in
which the laws a b = b a and a (b . c) =(a . b) . c both fail can have any
practical value has only to recall the familiar 3-dimensional vector calculus,
 Prologue xix

denoting vector multiplication. Subtraction on the set of integers is another

example!)
At this very time, in Germany, H Grassmann was generalising complex
numbers still further by inventing an unlimited number of arithmetics of
n-tuples. Unfortunately his exposition was not easy to read and it was
Hamilton's quaternions which gained the more attention. Hamilton had great
hopes for quaternions but they were not quite what the physicists wanted. It
is true that the great mathematical physicist Maxwell referred to them when
making use (separately) of the 'scalar' and 'vector' parts of a quaternion, but
the vector calculus of Gibbs and Heaviside (in the 1880s) seemed to suit the
physicists better. For a time a battle raged between the supporters in each
camp. Gibbs—Heaviside eventually won as regards applicability but the quater-
nions have the honour of being the first to demonstrate the existence of
consistent number systems not satisfying the commutative law of multipli-
cation.
During the 1840s, the quaternions soon inspired the manufacture of other
consistent number systems which violated the most obvious laws of arithmetic.
In 1845 Cayley described his (8-dimensional) octonions, still called the Cayley
numbers. Here not only is a b = b a not universally valid, the associative
law a (b c) = (a b) c is also broken.* In 1853 Hamilton invented
biquaternions (quaternions with complex coefficients which turn out to be
2 x 2 complex matrices in disguise). Here a b = 0 is possible even when a 0
and b 0. That is, the biquaternions possess divisors of zero. The 'smaller'
systems of hypernumbers introduced were put into some kind of order by the
American B Peirce in a paper published posthumously in 1881. Peirce had
set himself the task of methodically classifying, and looking for applications
of, the n-dimensional systems for all n 7.
Did Hamilton have to go to 4 dimensions to find his algebra? It appears
that he did search for 3-dimensional examples but could find none without
divisors of zero.t Indeed, Hamilton believed that all 3-dimensional systems
had to have divisors of zero and considered the fact that the quaternions had
no divisors of zero to be one of its chief merits. In 1861 Weierstrass proved
(essentially) that the only finite dimensional extensions of the real numbers
in which all the usual laws of arithmetic hold are the real numbers themselves
and the complex numbers. In 1878 Frobenius showed that relinquishing the
commutative law of multiplication adds only the quarternions to the list and,
using algebraic topology, Bott, Milnor and Kervaire showed, in 1957,
that relinquishing in addition the associative law adds only the Cayley numbers.
For more on this see the article by C W Curtis in [2].
Returning to 1847, Book invented another sort of algebra, since called
Boolean algebra, in order to put logic on a symbolic mathematical basis. After
his death his wife Mary wrote that the idea of symbolising logic had occurred
*
Here the remaining laws of arithmetic still hold.
tSee [37, p. 1061.
xx Prologue

to him at the age of 17 (Leibniz had had similar but less developed ideas as
early as 1666), but several subsequent writers (see [109, p. 235]) have indicated
that Boole's work on the calculus of operations in the early 1840s must have
at least influenced his approach if not actually initiated it. In this book we
have insufficient space to be able to do justice to Boole's ideas by indicating
applications to logic, probability and computer design. Fortunately there are
several introductory books on the subject; 1741 and [75] are just two of them.
Looking at the 19th Century development of number theory, we return to
Gauss and his reciprocity law. He was able to extend his law to cubic and
biquadratic residues but, to state his results elegantly, he found it helpful to
introduce numbers of the form a + pb and a + ib respectively where a, b are
integers, p is a complex cube root of unity and, of course, i is the usual square
root of —1. In this work he needed to know that these numbers factorised
uniquely into primes (3.7.13) just as do the ordinary integers. Kummer
endeavoured to study higher residues and considered, for the purpose, num-
bers of the form a0 + a + 2( where the a1 are integers and
p a prime. These numbers are also relevant for attempts to solve
Fermat's Conjecture (see Section 3.5) and indeed the FC would be solved if
only the uniqueness of factorisation theorem valid for numbers of the form
a +pb and a + ib extended to them. Unfortunately, as Kummer knew,* p =23
provided the first instance (of infinitely many) of the failure of unique factorisa-
tion. To try and get round the problem Kummer introduced extra 'ideal'
numbers (Section 3.9) to help him regain uniqueness of factorisation in many
cases. His analysis showed that the FC is indeed true for all prime exponents
p C 100, except for 'irregular' primes p = 37, 59, 67, which cases he dealt
with later.
Starting in 1871, Dedekind extended Kummer's ideas further. Any complex
number which is a root of an equation a0+a1x = 0 where the a1
. .

are integers will, said Dedekind, be called an algebraic number. (See exercise
3.2.14.) Those for which, in addition, = 1, will be termed algebraic integers
. . .

is one such (!) since it is a root of x2 +7x+15=0). Dedekind

.

2
introduced the term number field (Zahlkörper, in German) to denote a
collection of complex numbers satisfying the field axioms to be found in
Definition 3.2.2(10). One can prove that the algebraic numbers form a field,
but not so the algebraic integers: given (algebraic) integers a, (3 we find that
a/$ need not be an integer. The concept introduced here, then, needs a
name. It was called number ring (Zahlring, by Hilbert, in 1897). Our present
concept of field grew out of Dedekind's work and also that of Kronecker, the
basic notions being present already in the work of Abel and Galois. Dedekind
looked at collections of numbers gathered into a completed whole, this concept
being essential in his construction (Section 4.4) of the real numbers on the
basis of the rational numbers. Kronecker would have none of this. He insisted
* There is some well known folk-lore relating 10 this. See [46, p. 801 and the references mentioned
there.
 Prologue xxi

that every entity asserted to exist in mathematics must be shown to exist by

a finite set of instructions whose applications yield the entity. Even the
statement that 'obviously any polynomial in x with integer coefficients must
be expressible (by using a 'degree' argument) as a product of polynomials
which cannot be further decomposed' was of little value to him unless it be
accompanied by a method which in each instance would supply the indecom-
posable factors. He supplied such a method (see exercise 1.11.11).
Since Dedekind's algebraic numbers extend Kummer's concepts, his alge-
braic integers also lack uniqueness of factorisation. He reinterpreted
Kummer's concept of ideal number in terms of collections of already existing
numbers, called these collections 'ideals' (see Sections 3.4 and 3.9), and
showed how every ideal could be expressed uniquely as a product of prime
ideals (see Section 3.9).
In 1893 H Weber gave an account of Galois' theory which is 'applicable
to every case ... from function theory ... to number theory' using the field
concept 'without reference to any numerical interpretation'. Weber can thus
be regarded as the founder of abstract field theory.
One further item which should be mentioned here since it has strong
connections with present day ring theory is the subject of algebraic invariants.
Algebraic quantities remaining essentially unchanged under a change in coor-
dinates are important in coordinatised geometry since they correspond to
intrinsic geometric properties. In number theory too there was interest in
representing whole numbers by, amongst other things, binary quadratic forms
f(x, y) = ax2 + 2bxy + cy2 and in quantities such as — ac which remain

unchanged when, inf(x, y), x, y are replaced by x=ocx'+fiy', y=yx'+öy'

1. Motivated by an 1841 paper
where a, fJ, y, S are integers and aS —/37 =
of Boole, Cayley, who was interested at the time in algebraic aspects of
projective geometry, began seeking invariants of homogeneous forms of
degree ii in two and more variables. Cayley attracted his friend Sylvester into
studying invariants (the term 'invariant' is due to Sylvester) and these two did
so much research on the subject that they were named the Invariant Twins.
Cayley's work soon encouraged many mathematicians to invariant theory,
described by Sylvester as 'the essence of modern algebra', and so many
invariants were found that, to bring some order to the subject, minimal systems
of invariants were sought in terms of which all others could be expressed.
Gordan (1868) proved that to each binary form f(x, y) there is such a finite
system. The proof is hard. The whole subject was brought to a sudden climax
when Hilbert, in 1888, showed that for a form of any degree in any number
of variables a finite 'basis' always exists—and this without giving any indication
of how to find such a set in any particular instance! What Kronecker said
doesn't seem to have been recorded;* Gordan called it 'Theology, not math-
ematics'. Hilbert's theorem in its ring-theoretic form is stated in Section 3.4.
Though it is sometimes said that Hilbert's theorem killed invariant theory,
this is not entirely correct. Invariant theory continued—albeit at a reduced
result, proved in 1888, appeared in print in 1890. Kronecker died in 1891.
xxii Prologue

level of intensity—and in recent years activity has begun to pick up again.

(For more on the history see [102].)
A subject of which invariant theory formed a considerable part was that
of algebraic geometry. Algebraic geometry which, very loosely speaking, is
concerned with curves and surfaces in n-dimensional space which are defined
by algebraic equations, is a meeting ground for several mathematical disci-
plines including geometry, complex analysis, topology and number theory as
well as algebra. Around 1900 many deep results were obtained especially by
the Italian geometers although the validity of some of their methods was not
always apparent. The subject was set upon firm foundations around 1930 by
Emmy Noether and van der Waerden using an abstract algebraic approach.
Along with algebraic number theory, algebraic geometry can claim to be one
of the main motivating factors behind an autonomous theory of current
interest, that of commutative rings.
What about the development of group theory from Galois' time up to 1900?
First, one should note that it was not until about 1846, when Liouville
published two of his papers, that Galois' work became better known. As
suggested in [108, p. 94], it was possibly in connection with the announcement
of publication, made as early as 1843, that in the period 1844—6 A L Cauchy
was especially active in developing the theory of permutations. Adding to his
work of 30 years before he proved Galois' assertion that if a finite group G
contains pin elements, p being a prime, then G contains a subgroup of order
p. One should note however that Cauchy did not use the word group but
talked rather of 'a system of conjugate substitutions'. Although Galois had
introduced the word group, he had used it inconsistently. In 1854 Cayley
explicitly defined the term 'group'. He thought only in terms of finite groups
but, since he specifically insisted on the associative law being satisfied his ideas
were similar to those of today. In this paper Cayley's concern was with systems
of elements satisfying the equation x" = 1 and for both n = 4 and n 6 he
showed that there is essentially just one other system besides the set of
complex nth roots of unity.
Up to about 1867 only finite groups were considered. Furthermore the
term group was not generally employed and the main interest was in finding
the number of formally distinct values a function of n variables takes when
the variables are permuted. In 1867 Jordan, motivated by earlier studies of
crystal structure by the physicist Bravais, considered groups with infinitely
many elements—in particular groups of movements. In 1870 Jordan produced
his Traité des Substitution's et des equations algébriques. This work organised
the known theory of permutation groups and its relationship with Galois
Theory. It also introduced many new results and the concept of homomorph-
ism (5.10.1) as well as providing the atmosphere for the eventual finding by
Fedorov and Schönflies around 1890 of the 230 crystallographic space groups
(see Section 5.12).
In 1872 the Norwegian L Sylow extended the Galois/Cauchy result men-
tioned above by replacing p by in hypothesis and conclusion (see 6.2.8
and 6.2.12).
 Prologue xxiii

In the same year Felix Klein, in his famous inaugural address at the
University of Erlangen, stated his aim of using group theory to bring a unity
to the various classical geometries that had been found since the announcement
of the first non-Euclidean one by Lobachevsky in 1829. Thus geometries
would be classified by groups of transformations which left certain geometrical
aspects invariant. (The concept of invariance was definitely the 'in' subject at
the time and it was later to provide a central idea in the theory of relativity.)
Since rotations and translations of the plane can be arbitrarily small the
notion arises of an (infinite) continuous group. In 1874 and 1883 Sophus Lie
(pronounced 'Lee', Norwegian) used the idea to attempt a classification and
simplification of the solutions to certain differential equations. In studying his
continuous transformation groups (groups whose elements depend upon a
system of continuously varying parameters satisfying certain differentiability
conditions) Lie was led naturally to study some non-commutative, non-
associative algebras subsequently named after him: Lie algebras. (In a Lie
algebra multiplication and addition satisfy a b = —b a and (a . b) . c +
(b c) a + (c . a) b = 0.) Lie groups and Lie algebras form a major com-
ponent of the present day theoretical physicists' armoury.
We note, in passing, that initially there was no universal agreement about
what exactly constituted a group. For example, whereas Cayley, in 1854,
specifically demanded that the associative law should be satisfied, Lie and
Klein, in their earlier work, did not feel obliged explicitly to mention the
requirement; in all cases of interest to them the condition was automatically
satisfied! As the group concept became yet more prominent, it became increas-
ingly desirable to standardise terminology. In 1882 H Weber gave a set of
postulates for abstract groups of finite order. These postulates are essentially
those in use today.
Two other directions taken in the 19th Century by the theory of groups
should perhaps be mentioned. One is Dyck's concentration on systems of
generators for a group and on relations satisfied by these generators. These
concepts came to be of prime importance with the introduction of non-abelian
groups into topology, specifically via the fundamental group of a topological
space. The second is the introduction of group representation theory in which
groups are represented (via homomorphisms) by groups of matrices with
complex number entities. Matrices have the advantage that they can be added
together and multiplied by scalar quantities; further the concepts of deter-
minant and trace are available to aid computation. This theory, developed
by Frobenius, Molien, Schur and Burnside (see [104]) is of vital importance
today ([104], [36, Chapter 12], [62]) in the theory of finite groups and also in
representing certain groups which arise naturally from symmetry consider-
ations in chemistry and physics.
Possibly inspired by Hilbert's (1899) full axiomatisation of Euclidean
geometry, the 20th Century began with many attempts to find independent
sets of axioms for fields and for groups, the main worker being E V Huntington
around 1902—5. In 1905 J H M Wedderburn proved that every finite division
ring is a field, a result which provides the only known proof that in a finite
xxiv Prologue

projective plane Desargue's theorem implies that of Pappus [19]. In 1907

Wedderburn proved one of the fundamental theorems of modern non-
commutative ring theory. This theorem, proved for algebras over an arbitrary
field, was extended by Artin in 1927 to more general rings. It describes, just
as does the Fundamental Theorem for Finite Abelian Groups (6.4.4), the
exact structure, in easy terms, of a wide class of rings (the so-called semisimple
ones).
In 1908 K Hensel introduced for number-theoretic purposes a new type of
number—the p-adic numbers. Inspired by Hensel's construction, E Steinitz,
in a 140 page paper in 1910, undertook a classification of all fields, several
apparently unrelated sorts of which had now come into existence. He proved
that all fields fall into two categories: those whose unique minimal (so-called
prime) subfield is essentially the same as the rational numbers and the others
where it is essentially the (finite) Galois field 74, (see Section 3.10). Further,
every field can be obtained from its prime subfield by successively 'adjoining'
elements.
1914 saw the first axiomatic declaration of exactly what constitutes a ring.
(It is less general than the current one.) This axiomatising process was carried
further in the 1920s and 1930s by Emmy Noether, the greatest of all women
algebraists. Following her doctoral thesis which, supervised by Gordan, closed
with a complete list of 331 covariant forms for a given ternary quintic, her
strong inclination to unify, organise and generalise via axiomatisation took
over. In 1921 she investigated differential operators in quantum mechanics
by abstracting their essential properties, taking these properties as axioms
and building all consequences thereupon. This approach she adopted in all
her subsequent work thereby introducing a revolutionary style of attack on
problems of algebra. In the period 1920—6 she brought within her general
theory of ideals several theories which until that time had been seen as
independent. Her methods also greatly simplified many of the earlier proofs,
a number of them being very important in algebraic geometry. The rings she
and her school investigated, in one sense dual to those of Artin mentioned
above, are now called, naturally enough, Noetherian rings. Penetrating
theorems about these rings were first obtained as recently as 1960 by A W
Goldie.
Amongst the major achievements of the century in group theory one must
mention the modular representation theory of groups by matrices over finite
fields as pioneered by Richard Brauer and the subsequent use of this theory
in investigations into finite simple groups (Section 6.6). In the theory of
infinite groups various conjectures of Burnside have been established by the
Russian mathematicians Golod, Kostrikhin, Novikov and Adjan. We refer
the reader to [29] for precise statements.
The group-theoretic result of Golod (1964) is a consequence of joint work
with Shafarevich (also 1964) in which was solved affirmatively an old problem
of field theory closely related to the work described here in Sections 3.6, 3.7
and 3.9 (see [52, p. 125] for brief details).
 Prologue xxv

Some uses of algebra

It is impossible for the author to know exactly what the reader will accept as
being a 'proper' application of abstract algebra. Applications within pure
mathematics other than those found in the text are far too numerous and
diverse to mention here. Even excluding these the following represents only
a small sample.
First, algebra offers its basic notations and concepts as a most convenient
means for expressing, mathematically, certain concrete ideas. For instance,
professional mathematicians often prefer the phrase 'ring of integers' to
'collection of all integers', even when no use is to be made of the ring structure:
the algebraic terminology gives a more complete, a more faithful, picture. (In
this text we make a similar use of set theory. That is, we use its more elementary
concepts and notations as a language in which to express, succinctly, our
algebraic notions.) Of course there is no obligation to use the language of
algebra in this way any more than there is to use the symbol x in solving
problems involving quadratic equations: as mentioned earlier, the Babylonians
readily solved such problems writing down everything in longhand. It's just
that the new terminologies and notations offer extra insight and clarity of
expression and enable more powerful methods to be developed.
Regarding the more specialised applications which seem to use the algebra
(of the type presented here)* in an essential way, the theory of groups, being
(in a rather wide sense) the mathematical formulation of symmetry, is naturally
widely employed in physics and chemistry—for instance in applications to
crystallography, spectroscopy, general relativity, molecular vibrations,
molecular orbitals, solid state physics and especially in the modern theory of
elementary particles. (In February 1964 the Omega minus particle, which
group theory had previously predicted should exist, was first identified.)
Ring theory is not quite as old a subject as group theory and direct 'practical'
applications seem to be somewhat limited in number This ceases to be the
case when one considers rings which have extra structure, as for instance,
when the ring is an algebra or a field. In particular algebras of matrices and
of polynomials occur frequently: and in quantum mechanics use is made of
the algebra of all polynomials in two (non-commuting) letters x, y between
which the relationship xy = yx + 1 is assumed to hold. (Researchers in ring
theory are still investigating this ring!) In other applications rings (algebras)
also arise from groups. Both finite and infinite multiplicative groups give rise
to so-called group algebras via attempts to 'represent' the group elements by
matrices, and in studying a Lie group an investigation of the associated Lie
algebra is invaluable.
The theory of fields naturally underlies all appearances of the fields of
rational, real and complex numbers—but also finite fields (as well as finite
groups, polynomial rings and power series rings) are put to practical use in
*
Unfortunately there is no space in this book for that branch of algebra, namely linear algebra
Which has found the most 'practical' applications.
xxvi Prologue

the construction of efficient (from the point of view of cost!) error detecting
and correcting codes in the area of data communications. Finite fields are
also of importance in statistics via their association with (sets of orthogonal)
Latin squares.
Finally, the algebra introduced by Boole to model logic mathematically has
found application to the design of computers and telephone switching circuits,
again via the very real problem of reducing construction costs.
The following problems are not algebraic in content. They are placed here
mainly for your enjoyment and for subsequent discussion with friends and
teachers. However, it is intended that they should extract from the reader
that kind of critical attitude with which he should read this book from Chapter
1 onwards.

Problems
1 Sketch the graph of y = I. Clearly this function is continuous everywhere
and fails to be differentiable only at the origin (i.e. x =0 is the only point at
which there is no tangent). Invent a function which is continuous for all x
and yet not differentiable whenever x is a whole number. Can there exist a
continuous function which is not differentiable for any x? (Try sketching such
a function and then use your intuition. Finally, ask your teacher.)
2 Euclid defined a point as 'That which has no part'. Criticise this definition.
3 Discuss: Given any two straight line segments, say , it is
obvious that there must exist some (perhaps very small) unit of length in
terms of which the lengths of the above lines are viz and ii units respectively,
m and n being whole numbers.
4 Draw a circle C. Call the interior of C 'the plane', each point inside C 'a
point' and each chord of C (except for its end points) 'a straight line'. Defining
two 'straight lines' to be 'parallel' if they do not meet in a 'point' (i.e. inside
C) show that: Given a 'straight line' L and a 'point' P, not on L, in the 'plane',
it is possible to draw through P infinitely many 'straight lines' which are
'parallel' to L.
0
Elementary set theory and
methods of proof
01 Introduction
In 1895, at the beginning of his work Beiträge zur BegrQndung der transfiniten
Mengenlehre, Georg Cantor* made the following definition:
By a set we understand any collection M of definite, distinct objects
in of our perception or of our thought (which will be called the
elements of M) into a whole.
Thus examples of sets are: the set 7 of all whole numbers, here called the
integers; the sets 0, R and C of all rational, all real and all complex numbers
respectively; the set M comprising all moons of Mars; and even the set H of
all ten-legged octopodes which visited Archangel last 1 April.
Cantor's need for such a definition had arisen around 1872 from his
investigations concerning the possible uniqueness of representation of func-
tions by trigonometric series. In due course it became apparent that all of
mathematics could be made to rest upon a set-theoretic base. In particular
Cantor and Richard Dedekind, in his Stetigkeit und irrationale Zahlen (1872),
showed how the somewhat intangible irrational numbers (that is, those ele-
ments of R which are not in 0) could, using the set concept, be made
respectable in terms of 0 (see Section 4.4) and Gottlob Frege (1884) demon-
strated how the natural numbers 0, 1, 2, 3,... (on which 7 and ultimately 0
can be based — see exercise 4.4.17 and 3.10.5(iii)) could be defined in
set-theoretic terms. In addition, the concept of function can also be defined
set-theoretically (see Section 2.6).
It is therefore not surprising that the notations, terminology and simpler
notions of set theory now form an essential part of the language in which
contemporary mathematical discussions are conducted. The next two sections
introduce the simple set-theoretic ideas useful in this book.

0.2 Sets
We shall consider the words set, collection, aggregate as synonymous. The
elements of a set we shall sometimes call its members. If A is a set and if an
object a is an element of A we write a A. One usually reads the symbolism
a eA as 'a belongs to A'. If a is not a member of A we write and say
*
Oeorg Cantor (3 March 1845 —6 January 1918).
2 Elementary set theory and methods of proof

'adoes not belong to A'. Thus 3€!, ir+ieC,

Phobos M, Isaac Newton H.
Sets can be described by listing their members between pairs of curly
brackets (also called braces). For instance, M may be written alternatively as
M = {Phobos, Deimos}. It is of course impossible to describe sets containing
infinitely many elements this way. On such occasions we might be tempted
to write {x: P(x)} where P is a property characterising those and only those
elements of the set in question, P(a) indicating that the object a has property
P and hence lies in the set.* Thus the set t of all positive integers may be
written = {x: x 1 and x > O}. (This symbolism is read as 'z plus is the set
of all x such that x belongs to z and x is greater than 0'.) However, the
notation ={1, 2,3,4,. .}, where the three dots indicate the vague
.

expression 'and so on', is sometimes used. Small finite sets can be exhibited
in several ways. We have, for instance, M = {x: x is Phobos or x is Deimos}
or again M = {x: x is a Martian moon}.
If A and B are sets and if each element of A belongs to the set B we say
that A is a subset of B or that A is contained mt B, and we write A c B (or
B 2A, the latter being read also as 'B contains A'). In particular A cA for
each set A and if A cB and B cC then A cC. Given A cB, if we know
(and if we care!) that B contains elements not in A then we will write A c B
(or B nA) or even A (B A is then
called a proper subset of B. Thus we write 7 c 0, 7 c Q or 7 0 according
to the emphasis required. If A is not a subset of B we write A B (B A).
Note that A B when and only when A contains at least one element which
is not in B. We say sets A and B are equal, and write A = B, when and only
when they contain precisely the same members. Thus A = B when and only
when both A c B and B c A hold simultaneously. It follows that a useful way
to establish the equality of two sets A and B is to prove both A c B and B c A.
Warning: Try not to confuse and c. Somewhat roughly stated: is used
in relating a set to its elements; c is used in relating a set to its subsets. To
illustrate this, suppose A = {1, {l}, 2, {3, 4}, 7}. Thus A is a set with five
elements, namely 1,{1}, 2,{3,4}andl. Hence 1€A,
{2}cA, {3,4}€A, leA, (Inci-
dentally, sets like {2} which contain exactly one element are called singletons.)
NowconsiderthesetsF={x: x €landx2<0}, G ={ }andH asmentioned
earlier. It appears that each contains no members at all. A set with no members
*
It is perhaps surprising that so simple a definition can give rise to contradictions. In fact even
Cantor realised this, although the best known example is due to the British mathematician/phil-
osopher Bertrand Russell who took P(x) to be the condition x x. Clearly I I and Z Z so that
R ={x: is surely non-empty. One can ask: 'Does R belong to R?' If so, then R satisfies
the condition (x x) for being an element of R. Hence R R. If not, then R fails the test x x and
we deduce R e R. Thus R a R when and only when R R. Various axiomatisations of set theory
(in particular Zermelo's; see Prologue, p. xv) have been proposed to exclude the appearance of
such paradoxes. The sets we shall consider will all be acceptable in Zermelo's scheme.
t Some mathematicians prefer to say 'included in' to avoid possible confusion with subsequent
use of 'A contains a' to describe a a A This dual use of 'contains' is bad but common. Cf. above
'Warning'.
 New sets from old 3

is called an empty set. Despite the differences in definition of F, G and H we

can show that there is only one empty set. To prove this, and more, let 0 be
an empty set and let A be any set. Then 0 cA, for otherwise 0 and
so 0 would contain an element not to be found in A. But this is silly since
0 has no elements. Now suppose 01 is another empty set. Since 0 is empty
0 01. Since 01 is empty 01 c 0. These two inequalities, taken together,
imply 0 = 01. Thus we can talk of the (unique) empty set.

Problem 1 What is wrong with the following argument showing 0 A?

To prove B c A we must show that every element of B is also contained
in A. Since 0 has no elements we cannot show 0 cA. Hence 0

Problem 2 In connection with the first footnote on p. 2, can you think of a set
A for which AeA?

03 New sets from old

New sets can be made from old in several ways, one of which we saw above
in defining in terms of 1.
Let A and B be sets. Then the sets

AnB—{x:x€A andxeB}
and
AUB—{x:xEA orxEB (orboth)}*
are called, respectively, the intersection and the union of A and B. Thus if
A ={3, 1, 4} and B ={ir, 4, Oliver Cromwell, 1} then A nB {1, 4} whilst
A uB ={3, 1,4, ir, Oliver Cromwell}.
It follows immediately that for sets A and B we have
AnA=A, AnB=BnA, An0=0
AuA=A, AuB=BuA, AuO=A
Intersection and union of sets can be thought of in terms of the shaded
regions in Fig. 0.1; such figures are called Vennt diagrams.
If we introduce a third set C and shade the region common to A u B and
C (Fig. 0.2) it appears that the sets (AuB)nC and (AnC)u(BnC) are
equal. This can be checked (exercise 5(b)) by an argument not depending
upon pictures. [Pictures can be deceptive. What value has this pictorial proof
if, say, CnBz= 0 or ifAcB? See [124].]
There is no difficulty in extending the definition of union and intersection
to larger finite or even infinite collections of sets. For example, if for each
nE we define
t 1 1

n n
*
In mathematics the word 'or' is taken to include the possibility of both. This is not always the
case in ordinary conversation, for example 'would you like tea or coffee (but not both)?'
John Venn (4 August 1834—4 April 1923).
4 Elementary set theory and methods of proof

AnB

AuB

Fig. 0.1

(AuB)nC = (ArC) u(BnC)

Fig. 0.2

then the set of elements common to all the is denoted by In fact

fl Sn {O}, the set containing the one real number 0. Here we have 'indexed'
the various S with the elements of the set We can also use the elements
of lV (see exercise 6) or indeed the elements of any set (see 3.4.5(F)) as an
indexing set.
Another way of producing a new set from two old ones A and B is to
define their difference A\B = {x: x E A and x B}. In particular the set of all
non-zero real numbers is then denoted by* R\{O}.
The next definition is suggested by the way coordinates are introduced
into the real plane. Points in the plane are made to correspond to pairs of
*
Also denoted by R*.
 Some methods of proof 5

real numbers and vice versa in such a way that if points P and Q are given
coordinates and (b1, b2) respectively then P and 0 coincide when
and only when a1 b1 and a2 = b2. In particular (1, 2) and (2, 1) correspond
to distinct points. Since the only distinction between (1, 2) and (2, 1) is the
order in which the numbers 1 and 2 are written down we refer to such pairs
of numbers as ordered pairs.
Much more generally we make

Definition 0.3.1 Let A and B be sets. Then A X B denotes the set

{(a, b): a E A and b B} of ordered pairs. A x B is called the Cartesian product
of A and B. The word 'ordered' implies that elements (a, b), (c, d) in A x B
are defined to be equal when and only when both a = c and b = d.

Notes 0.3.2
(i) The concept of ordered pair can be described in a purely set-theoretic
manner. See exercise 8.
(ii) In a like manner one can define the set A x B x C of ordered triples
(a, b, c) where a A, b B and c C. One can even form the Cartesian
product of a collection of infinitely many sets (see exercise 6.3.20).

Example 0.3.3 If A ={1, H}, B ={1, e, h} then A xB ={(1, 1), (1, e), (1, h),
(11,1), (H,e), (H,h)}. Note that AXB
and BXA each have 6 (=2.3=3.2)
elements but that A x B B x A. [Why not?]

One final word on set production:

Definition 0.3.4 Let A be any set. By we denote the set of all subsets
of A. 2P(A) is called the power set of A.

Example 0.3.5 If A ={a,5,y} then

= {ø, {a}, {5}, {y}, {a, B}, {B, y}, {y, a}, {a, B, y}}

Note that A has 3 elements whilst has

0.4 Some methods of proof

From the beginning of Chapter 1 onwards, the majority of this book is taken
up with assertions described as theorems (or lemmas or corollaries) followed
by explanations purporting Ui be 'proofs'. Just as we have adopted an intuitive
approach to the concept of set so we shall allow our intuition to guide us in
the matter of whether or not an explanation is logically acceptable: to formalise
the notion of acceptability would take us too far afield, into symbolic logic.
(The reader keen to see how this can be achieved is referred, for instance,
to [57], [59].) On the other hand there are a few techniques, frequently used
6 Elementary set theory and methods of proof

in constructing proofs, to which it is perhaps useful to draw attention as they

have been known to give students the occasional difficulty. We do this via
some concrete examples, simple enough to allow the logical principle involved
to be easily seen.
Consider, for instance, how you would prove the following assertion:
(i) If x is an odd integer then x2 is an odd integer.
You would probably say: 'Assuming x is odd, we may write x = 2n +1 for
some suitable neZ. Then x2=(2n +1)2=4n2+4n +1, which is clearly odd.'
Here we go directly from the hypothesis that x is odd to the conclusion
that x2 is odd. Such a proof is called a direct proof.
For a second example consider the following assertion about the integers
x and y:
(ii) (1).
A direct proof is possible: From we deduce that (x —y)(x
It follows that x —y 0 and hence that x y.
However, the proof most of you would probably have given is:
If x = y then x2 = y2 (2), a contradiction.
Hence the result.
This approach, using the method of indirect proof, conceals a number of
logical points. Two we wish to draw attention to are most easily discussed in
symbolic terms. Let A and B stand for 'x2 # y2' and 'x y' respectively.
Further let ——A and —B denote their negations or denials. Thus —A is 'x2 =
and —B is 'x = y'. Then (1) is the assertion: If A then B. We write this briefly
as A 4' B and read it as 'A implies B'. On the other hand (2), apparently
logically equivalent to (1), is the assertion —B 4) —A. In fact it is shown, in
courses on elementary logic, that the assertions A 4' B and —B 4' —A are
equivalent—as is the pair B 4' A and A 4) B. What are clearly not (in
general) equivalent are the assertions A 4' B and B 4' A. [In terms of our
concrete example this latter assertion reads x y 4' x2 y2, which is blatantly
false, whereas (1) is (blatantly) true.] The assertion B 4' A is called the
converse of the assertion A 4' B.
To summarise: to prove that an asserted conclusion B follows from a
hypothesis A one can proceed directly or prove, equivalently, that —A is
deducible from hypothesis —B. This kind of indirect proof is called proof by
contraposition, —B 4' —A being the contrapositive to A 4) B.
Another frequently adopted method of indirect proof is that of reductio ad
absurdurn or proof by contradiction. Here the negation of the hoped for
conclusion is shown to lead to a contradiction (or absurdity). Possibly the first
ever use of this method was in a proof of:

Theorem 0.4.1 If x eR and x2=2 then In other words, —12 is not a

rational number.

Proof We suppose —12 e 0 so that we may write x = rn/n where rn, n c 7.

Clearly we may assume rn, n have no common divisor >1 in 7; such a common
 Some methods of proof 7

divisOr can be eliminated before proceeding. Then from x = rn/n we deduce

that rn2/n2 2, that is rn2 = 2n2. It follows that rn2, and hence* rn, is even.
Writing m = 21 we find 412 = 2n2; that is 212 = But then n2, and hence*
n, is even. Thus our assumption, namely that = x = rn/n e Q, has led to
the contradiction (absurdity) that numbers rn, n with no common divisor >1
are both even. Consequently Q.
Now consider the following two assertions.
(i) Every positive integer is the sum of four (non-negative) integer squares.
(ii) Every positive integer is the sum of eight non-negative integer cubes.
It is not very difficult to check that (i) is certainly true for each positive
c 100. For example 75 = 82+32+12+12; 86=92+22+12+02. You should
note however that even if you check (i) for all positive integers up to, say,
1050000, (i) would still not be established for every positive integer n; indeed,
for all you then know, (i) might fail for n = io5000°+ 1. No, to prove (i) you
must devise a proof which covers all positive integers n.
In fact (i) is true; it was first proved by Lagrange in 1770. On the other
hand (ii) is false: not all positive integers can be expressed as asserted. For
instance 239 cannot so be expressed. 239 is thus a coun:erexarnple to
assertion (ii). Actually there is only one other positive integer which is not
expressible as a sum of eight cubes. Can you find it? It is less than 50. The
point is that even though only two out of infinitely many positive integers fail
to satisfy (ii) the existence of even one such 'nasty' integer is sufficient to
destroy (ii)'s claim to be a valid assertion. In particular, finding the other
'nasty' number changes nothing—the challenge was set just for fun.
There are other principles of reasoning we could mention. For instance in
showing that 2.5.1 follows from 2.5.1' we split the proof into two cases (one
when p divides n and one when it does not) and establish each case separately.
This kind of proof is called proof by cases. Its use in the instance quoted is
so straightforward that no special attention needed to be drawn to it. We
refer the reader interested in a detailed account of different methods of proof
to [59, pp. 30—46].
We close with one or two remarks on terminology and notation. We have
already observed that the assertion 'If A thee B' is written A 4 B and read
'A implies B'. We also say 'A is a sufficient condition for B' (since A is
enough, all by itself, to allow us to conclude B) or that 'B is a necessary
condition for A' (since B necessarily follows from A—whether you like it or
not). Alternatively, mathematicians say 'B if A' or 'A only B'. If we know
A 4 and only if B', an assertion we write briefly
as A B or A if B. Technically all definitions should be in 'if' form. If, for
instance, in 1.3.1 we had defined a to be a divisor of b if b = ac this would
have left open the question of whether or not we are to call a a divisor of b
if no such c existed. Thus, use of 'itT' indicates that a will be called a divisor
of b when and on/v when the required c exists. (Cf. Definition 1.3.3!)
*
Exercise Ii.
8 Elementary set theory and methods of proof

In symbolic logic much use is made of the signs 3 (there exists) and V (for
all), although we don't often employ these here. As an example, note that
one of the properties of the equality relationship on 1 used in the proof of
1.2.1(i) may be stated succinctly as (Vx)(Vy)(Vz)(x = y xz = yz), whilst
axiom A3 in Section 1.2 includes the statement that (3x)(Vy)(x + y = y).
Finally note that the negation of (3x)(P(x)), that is —4(3x)(P(x))), is
and that, similarly,

Exercises
1 Which of the following assertions are true?
(a) kz; (b)(ir+i)2eC\R; (c)* (d) (e—ir)2eR\R4;
(e) ir —e eR
2 Let A = {ø, {ø}, 1, {1, ø}, 7}. Which of the following are true?
(i) ØEA; (ii) {ø}eA; (iii) {1}eA; (iv) (v) 7cA;
(vi) (vii) (viii) (ix) {ø,{ø},{1, ø}}cA;
(x) {{ø}}cA.
3
AnB, BnC, AnBnC, AuB, (AuB)nC, (AnC)u(BnC). Draw a
Venn diagram involving A, B and C.
4 LetA={(x,y):xeR,yeR,x2+y2=1},B={(x,y):xeR,yeR,y2=4x},
C = {(x, y): x eR, y eR, y2 = x3}. Find the intersections and unions asked for
in exercise 3.
5 (a) ForsetsA andB showthatAuB=B iffAcB andthatAnB=A
iffAgB.
(b) For sets A, B and C prove that (i) (AuB)nCc_(AnC)u(BnC);
(ii) (AnC)u(B nC)c_(A¼JB)nC. Deduce that (iii) (A uB)nC=
(A n C) u (B n C). (Cf. exercise 3.)

Rn{x:xEiR and ——<xEn}.FindflTr, URN,

,ER* n=1 n=1

7 Let U be a set with subsets A and B. Define N = U\A etc. Show

(i) (ii) (iii) (AUBY=AcnBc;
(iv) (A nB)c =Ac ((iii) and (iv) are called de Morgan's laws.t)
8 Define [a, b] to be the set {{a}, {a, b}}. Show that [a, b] = [c, d] if a = c
and b = d. (This explains note 0.3.2(i).)
9 (i) Showthat ifA and B arenon-emptythen AXBBXA if AB.
(ii) where CcA and DcB?
Let ScA xB. Need S be of the form CxD,
*
denotes the set of positive real numbers.
t Augustus de Morgan (?June 1806 — 18 March 1871).
 Some methods of proof 9

io How many elements has PP(A) if: (i) A has 10 elements? (ii) A = 0?
11 Let m €1. Prove that: if in2 is even then in is even.
12 Give another example of assertions A, B where A 4' B is true but B 4' A
is false.

13 Prove that e 1+
111
+ + is
in
irrational as follows. Suppose e = —.

Then
/ 11 1
But t=—+---—
1
+...
1! 2. n!i n+1 (n+1)(n+2)
—2--- + = —, a contradiction. What kind of proof have we used
n+l (n+1) n
here?
14 What kind of proof did we use in proving that 0 c A for each set A?
15 Letx,y andzel.Showthatx2+y2+z2cannotbeoftheform8k+7
when: (i) exactly one of x, y, z is odd; and (ii) all of x, y, z are odd. Deduce
that no integer of the form 8k + 7 is expressible as a sum of three integer
squares. What method of proof have we used here?
16 Write in words the following assertion, where x, y, z C
(Vx)(Vy)(3z)(xy =z2). Is the assertion true? Write symbolically: For all x
and y in 1 for which x cy, there exists z such that x + z = y.
17 Write in words the following assertions, in which x, ye!:
(i) (Vx)(By)(y >x); (ii) (By)(Vx)(y >x). Deduce that one may not
necessarily be able to interchange V and B without the risk of changing the
meaning.
I
Numbers and polynomials

1.1 Introduction
As implied in the Prologue, one of the central concepts of modern algebra
is that of 'ring', two of the most fundamental examples being the ring 7
of all integers (Sections 1.2 to 1.5) and the ring Q[xJ of all polynomials in the
'indeterminate' x with coefficients in the field Q of rational numbers (Sections
1.6 to 1.11). The terms ring and field will be defined formally in Chapter 3.
The main objectives of this chapter are as follows.
(1) To present some of the simpler properties of the sets 7 and O[x] in such
a manner as to emphasise their similarities (and their differences!). Thus we
shall have to hand important concrete examples and theorems which will help
motivate, and can act as test-cases in, our later development.
(2) To introduce some terminology and notation in common use in the
following chapters. One advantage of doing this at this stage is that the reader
will probably feel able to devote a little extra effort to learning this terminology
as the statements of the theorems themselves will take little remembering—
many of them should be fairly familiar already.
Finally and by far the most important is:
(3) To introduce into this familiar setting a few notes of caution. Here we
hope to develop the reader's critical faculties by showing, especially in relation
to O[x], that not everything is quite as straightforward as might be expected
(see in particular Section 1.6). We hope the reader will examine proofs of
theorems with one question continually in mind, namely 'Why can he (the
author) say that'?
As promised earlier, many of the proofs in this chapter are written in
expansive style with square brackets indicating those portions of proofs that
could, without great loss, be omitted. The threefold purpose of these brackets
is detailed in the preface.

1.2 The basic axioms. Mathematical induction

We start, then, by looking at the set 7 of all integers. As noted in Chapter
o a definition of integer can be given in set-theoretic terms (see, for example
[58]). The algebraist is, however, little interested in what the integers are; he
 The basic axioms. Mathematical induction 11

is mainly interested in the fact that these integers, whatever they may be, are
added and multiplied together, two at a time, according to the following
axioms (where we use the symbol to denote multiplication).
For every three integers a, b, c (distinct or not) we have:
Al a+b=—b+a Ml
A2 (a+b)+c =a+(b+c) M2 b) c=a . (b
A3 There exists in 7 a unique M3 There exists in 7 a unique
integer, namely 0, such that integer, namely 1, such that
0+a=a+0=a
A4 To each a in 7 there exists a
unique integer, namely —a,
such that
a +(—a)=(—a)+a =0
D
a non-empty subset N such that
(i) each element of 7 belongs to exactly one of the sets N, {0}, —N
where —N denotes the set {—x: x E N},
(ii) for alla, b eN we have a +b eN and a beN.
I If U is a subset of N such that I U and such that a + I U whenever
a U then U = N.
Remarks
(i) The axiom Al is called the commutative law of addition in 7L: Ml is the
commutative law of multiplication. A2 and M2 are the associative laws of
addition and multiplication respectively. A3 and M3 announce the existence
of additive and multiplicative identity (or neutral) elements. A4 asserts each
integer has an additive inverse. D lists the distributive laws.
(ii) The reader will probably have recognised N as having properties usually
ascribed to the set of positive integers and I as being the principle of math-
ematical induction.
(iii) Despite the fact that the algebraist is not interested in the nature of the
integers themselves he certainly gets joy out of the fact, to be proved later
(see Section 3.12), that there is essentially only one system of objects satisfying
the above axioms Al through to I. (The idea of two algebraic systems being
'essentially the same' will first be defined formally in 3.10.1. Before that, part
(ii) of the Remark in Section 1.8 might prove helpful.)
(iv) It is notable that for the set 7 there is no multiplicative analogue of
A4. However, as is well known, the sets* 0, R, C do all satisfy a near analogue
of A4, viz:
M4 To each non-zero a in 0 (or or C) there exists in 0 (or R or C) a
unique number, namely a1, (also written such that a a' a= 1.

*
We temporarily use 0, R and C intuitively for illustrative purposes only. We shall construct
them formally in Sections 3.10 and 4.4.
12 Numbers and polynomials

Problem I Assuming, for the moment, the truth of the assertion of unique-
ness made in Remark (iii) above, one deduces that, amongst the axioms listed
for 7, there must be at least one which cannot be satisfied by Q. Can you
identify which axiom(s) from Al through to I are not satisfied if one attempts
to apply them to Q instead of 7?

The reader can no doubt think of other properties usually ascribed to 7

which we have so far failed to mention. For instance, we are all familiar with
Z: If a, b e 7 and if a b =0 then a =0 or b =0 (or both; see the footnote
on p. 3).
C: If a, b,c el, if c a =c b and if c then a =b.
M: For all a, be 7 it is always the case that (—a) . (—b) = a b.
These properties might be called the zero-divisor law, the cancellation law
and the mysterious law (see [5, p. 5]) respectively. Why do we not add them
to the list of axioms given above? The answer is that we can, without using
intuition, speculation or hearsay concerning 7, prove that Z, C and M are
logical consequences of the axioms Al through to I. Now whilst algebraists
do not regard it as their prime duty to reduce all such sets of axioms to a
minimum size, it is part of an algebraist's function to investigate consequences
of axioms such as those just referred to. In this manner the algebraist can
hope to discover which features of a given system are essential and which
only incidental.
We shall perform a number of consequence-seeking calculations of this
type in a more abstract setting from Chapter 3 onwards. Let us content
ourselves for the moment with finding out exactly which of the axioms given
earlier are required to establish property M. First, then, we prove, giving all
the details,

Lemma 1.2.1*
(i)

For all c e 7 we have —(—c) = c.

Comment If these results seem rather too trivial to bother about, let's see
if we can make them more impressive (and less 'obvious'?) by stating their
conclusions in words:
(i) The product of the additive identity with any integer always yields the
additive identity.
(ii) The additive inverse of a product [that is —(a . b)] is equal to the product
of the additive inverse of the first [that is —a] with the second [namely b].

*
A lemma is a result which helps in the proof of a future theorem (cf. the German word Hilfsatz)
but is not deemed to be of sufficient importance by itself to warrant the title Theorem' (German
Satz).
 The basic axioms. Mathematical induction 13

(iii) The additive inverse of the additive inverse of a given integer is equal
to the given integer.

proof
(i) [From the property of 0 as stated in A3 we have, on setting a = 0,]
o+00 (A3)
[Multiplying each side by c we have]
Hence
c c (using the property x =y =yz of equality)
[Now using the second axiom in D on the left-hand side of this equation]
Consequently
(axiomD)
[By A4 the element —(0 c) certainly exists: we add it to each side. Reflecting
the fact that on the left-hand side it is being added to the element 0 c + 0 c
we write]
and so
(0 c +0 c)) =0 +(—(0 c))
(using the property x = y 4' x + z = y + z of equality)
[Now using the associative law on the left-hand side]
It follows that
0 c +(0 . C +(—(0 c))) = 0 c +(—(0 .
. c)) (A2)
[And replacing 0 c +(—(0 c)) by 0 on using A4 once on each side]
O'c+O=O (A4twice)
[Finally using A3 on the left-hand side]

0c=0 (A3)
as required.
(ii) Given a we obtain successively
a+(—a) =0 (A4)
(a + b =0 b (property of equality)
a =0 (Donlhs;(i)onrhs)
But ab+(—(a•b)) =0 (A4)
(—a) b = —(a . b) [since both are additive inverses for a b
and] by A4 [there is a unique such
inverse].
14 Numbers and polynomials

(iii) Given c e 7 we have

c+(—c)=O and (—c)+c=O (A4)
.. (—c) +c = 0 and c + (—c) = 0 (Switching the equations over—or
just using Al)
Hence c is an additive inverse for —c (A4)
But this inverse is unique (A4)

[Now the unique additive inverse of any element x is denoted by —x. Hence
the unique additive inverse of —c is denoted by —(—c). But c is this additive
inverse.]
Hence —(—c)=c.

Theoreml.2.2

Proof [Using 1.2.1(u) on the elements a and —b we get]

(—a) • (—b) = —(a • (—b)) (i) (1.2.1(u))
But a (—b) = —(a b) (ii) (Proof identical to 1.2.1(u))
Combining (i) and (ii)
(—a) . (—b) = a b (using 1.2.1(iii))

Remarks
(i) If the reader feels we have gone to a lot of trouble to establish a result
which is 'obvious' we ask him on what grounds he bases his belief in this
result? Is it merely 'experience'? Or has he had it 'on good authority' that it
is true? What we have done is to show that the semi-mystical assertion that
times minus is plus' is deducible as a consequence of other ('more
obvious') axioms of arithmetic.
In later chapters we shall prove several results of the above type where the
symbols used will not necessarily stand for integers. Then, assertions of the
above type will certainly be far from 'obvious', since we will lack the appropri-
ate 'experience'.
(ii) We were trying to find out exactly what we needed to assume in order
to prove 1.2.2. Including the (necessary) Lemma 1.2.1 it appears that our
proof of 1.2.2 depends on various properties of the = symbol together with
several applications of A4, A3 and D together with just one application of
A2. Notice that Al, Ml, M2 and M3 were not called upon.

Note 1.2.3 One other thing 1.2.2 does for us is to prove that 7 can contain
only one subset N with the properties listed in axiom P. (If M is a subset of
 The basic axioms. Mathematical induction 15

z satisfying P then either 1 or —1 belongs to M. It follows from 1.2.2 that

1 = 12 = (_1)2 and then from axiom P(ii) that, in any case, 1 eM. Similarly
1 eN and hence leMnNc_N. If, now, a EMnN then a +1eMnN since
both M and N satisfy axiom P(ii). But then M n N = N since 1 satisfies axiom
I. This means that N cM in 7L. Clearly N M is impossible [why?].)

The reader who is beginning to doubt the author's sanity should note that
there is something needing proof here. The innocuous looking set S =
{a + b'12: a, b_c Z} can support two distinct such Ns, namely the subset N1
of all a + b which are_positive real numbers in the usual sense and the
set N2{c +d'12 C

To deduce the familiar properties of ordering amongst the integers we make:

Definition 1.2.4 The (unique) subset N of 7 described in axioms P and I

will be called the set of positive integers; the subset —N the set of negative
integers. If a, b EL we say that a is less than b and write a <b if! b—a eN.
If we only know (or care) that b—aeNu{O} we write (b—a is the
shorthand notation for the more accurate b + ( — a).) When convenient we
shall also write b > a (b a) as an alternative to a < b (a b).

P(i) then says that each integer is either positive or zero or negative (and
never has two of these properties simultaneously).
P(ii) says that the sum and the product of positive integers are both positive.
I says that any subset of positive integers which contains 1 and which
contains a + I whenever it contains a is precisely the set of all positive
integers.
One can use axiom P and 1.2.4 to establish all the familiar properties of
the < sign. We treat just a couple, place some in the exercises and leave the
rest for the reader to prove for himself or look up in, for example, [32].

Theorem 1.2.5 If a, b,c €7 are such that a <b and b Cc then a Cc.

Proof [To say a <b is,] according to 1.2.4, [to say that] b—a eN. [To say
b Cc is,] similarly, [to say that] c — b N. [But] from P(ii) we [can then] deduce
[that b—a+c—b€N, that is*,] that c—aeN. Thus [using 1.2.4 again this
simply says] a Cc, as required.

Theorem 1.2.6 If a, b, ccl are such that ac b and 0 < c then <bc
Proof (Briefly) We are given that b — a and c are both in N. Hence, by axiom
P(ii), (b—a)ceN. It follows* that bc—ace N, that is ac < bc.
From now on we shall often write ac (etc.) in place of a.c. (etc.)
These are a couple of points which need care. See exercises 1, 2 and 3 following.
16 Numbers and polynomials

Now seems a fairly appropriate time to make

Definition 1.2.7 Let a eZ. We define (called modulus* of a) by

Ial=O if a=O
IaI=a if 0<a (i.e.ifaeN)
Ial=—a if a<O (i.e.iffae—N).
Thus

Example 1.2.8 I—171 = 17, 1311 = 31.

We first use this concept in Theorem 1.4.5.

We concentrate for a while on I, the principle of induction.t Let us suppose
that to each positive integer n there corresponds a statement which we denote
by S(n). For example, S(n) might be Let us
23 n n
further suppose that: (i) We can show S(1) to be a true statement; and that
(ii) for each positive integer k we can prove, under the assumption that S(k)
is a true statement, that S(k + 1) is also a true statement. Then I tells us that
for every positive integer n the statement S(n) is true.
For: Let U denote the subset of N comprising all those positive integers
n for which the statement S(n) is true. That is, U ={n: n eN and S(n) is
true}. Then I c U [since we are supposing that we can prove 5(1) to be true]
and, whenever k e U [that is, whenever 5(k) is true] then k + 1 U [since
we are supposing that we can then prove S(k + 1) true]. By principle I we see
that U = N. Thus for each n C N we have n C U; that is 5(n) is true, as required.
After talking about proofs by induction we should also mention the tech-
nique of definition by induction. Consider for example the famous Fibonaccit
sequence u1, U2, u3,... where (i) uj = 1, u2= 1 and (ii) for each integer n
is defined not directly but relative to previously defined terms of the
sequence by the formula + One is tempted to say that the
Fibonacci sequence is defined by (i) and (ii) but there are some logical subtleties
associated with this temptation. We shall not discuss them here as they can
be overcome. The reader who is sufficiently intrigued by this warning is
referred to a very readable article [106] by L Henkin.
We now look at two variants of I. Consider the following assertions:
Let VbeasubsetofN suchthat 1€ V andsuchthata + 1€ Vwhenever

*
Also called absolute value.
The first acceptable statement of the principle is often credited to Blaise Pascal (19 June
1623 — 19 August 1662) after whom Pascal's triangle is named.
t Leonardo Fibonacci lived in the period c. 1170—1240.
 The basic axioms. Mathematical induction 17

W: Every non-empty set of positive integers contains a least member.

That is, if T N and T 0 then T contains an element t such that
<z for any other z e T.

Theorem 1.2.9 The statements 1, W are equivalent. That is, each implies
the other two.

Remarks
(i) On the face of it, it looks as if we have six results to prove, namely:
14W, W4'I; W In fact we only need to prove
I I which we can write in abbreviated form as I 4.
W I. (Compare this with the deduction of a = b = c given that a b, b c
and c a, a, b, c E /.)
(ii) W is also expressed by saying 'The positive integers, taken in their natural
order, are well-ordered'. W is called the well-ordering principle—hence the
letter W! W is manifestly untrue when applied to 0 and R. [Why? And what
about C?]

We shall use W a lot in this chapter in particular and call upon in 1.10.1.

Note 1.2.10 Despite the remarks above on economy of effort we shall prove
here only the equivalence of I and W. We leave the reader to prove that

I W. [Here we wish to show that every non-empty

set of positive integers has a least member so...] assume [to the contrary]
that there is a non-empty set T, say, of positive integers such that T has no
least member. Then I T [since 1 is the smallest positive integer. Which
axiom(s) prove this for us? See exercise 19]. Define the subset V of N by
V_—{x:xeN and Xc for all teT}. Then 1eV. Let keV. Then k+leV:
otherwise k c k + 1 for some s e T so that [since k * T and since no integer
lies between k, and k + 1 (why not?)] s = k + 1 is the least integer in T [contra-
dicting the assumption that T has no such least integer]. Thus, by principle I,
V = N whence T =0. This manifest contradiction (T #0T =0) tells us
that our one assumption is not tenable. Thus no non-empty set T exists and
the required result is thereby proved.
W 1. Let U be a subset of N such that 1 E U and a + 1 e U whenever
a cU. Suppose U Then the set T ={x: x eN and U} is non-empty
and so [since we are assuming W holds we can conclude that] T has a least
element t, say. Thus t — 1T and so t — e U [is t — I a positive integer?].
1

Then [by the given property of U] t — + I e U. That is t e U. But t e T [we

1

chose it as the least element of T] and so U n T 0, a contradiction. Thus

[once again] our one supposition [this time that U N] must be invalid, so
that U = N, as required.
18 Numbers and polynomials

Exercises
1 Let a, b, c, d e 1. Show that using axiom A2 alone one can prove
((a + b) + c) + d = (a + (b + c)) + d = a + ((b + c) + d) = a + (b + (c + d)) =
(a +b)+(c +d).
[This shows that the 5 different ways one might set about working out the
sum of the four integers a, b, c, d all yield the same answer which can therefore
be unambiguously denoted by a + b + c + d. More generally one can show
(2.7.7) that the sum of n integers a1, a2,. . . , can be denoted unambiguously
by a1 Similar remarks apply to products.]
2 Use the same sort of reasoning, together with careful use of axiom Al, to
show that b +(—a)+c +(—b) —c +(—a), as asserted in 1.2.5.
3 Prove, with the same degree of detail, that —(a+(—b))=b+(—a), and
that (a —b) c =a c—b 'c. [For this second part think what a—b means.]
4 Show that the second axiom D is a consequence of the first axiom D and
Ml.
5 Prove that property C is a consequence of property Z (and which axioms?).
6 Using the style adopted in the comment following 1.2.1 state the result
of 1.2.2 in words.
7 Let C[0, I] be the set of all real valued continuous functions* defined on
the interval [0, 1]. For 1' g E C[0, 1] define, for each x E [0, 1], (f+g)(x) =
f(x) +g(x) and (f. g)(x) =f(x)g(x). Show that with these definitions of +
and C[0, 1] satisfies Al, A2, A3, A4, Ml, M2, M3, D but not Z. Thus Z
is not a logical consequence of Al through to D alone.
8 Prove, in full detail, that if a, b, c E / are such that a <b then a + c <b + c.
9 Prove that if ae/ and if a$0 then a. Deduce that 0<1.
10 Prove directly from A3 and A4 that —0=0. Show also (Lemma 1.2. 1(u)
might be helpful) that if a $0 and b $0 then ab $0. (That is, prove property
Z from axioms Al through to I.)
11 Showthat(i)ifa1>biandifa2>b2thenaia2+b1b2>a1b2+bia2;and
that (ii) if a1 >b1 >0 and if a2>b2>0 then a1a2>b1b2>0.
12 Prove that, for all a, b 7, (i) jab I Ia . lb and
I
(ii) a + b I Ia I + lb I.

[In problems 13 and 14 you are meant to get your hands dirty by experiment-
ing. Just try any example that comes into your head—if it turns out to be no
good, throw it away and start again! Why not try {3n +1: n E 7} for a
start?]
13 Can you find a non-empty set of numbers which fails to satisfy I but
satisfies all of the remaining axioms from Al through to I?
*
A formal definition of function is given in 2.6.1. You are asked to proceed informally here.
 The basic axioms. Mathematical induction 19

14 What about the same problem with A3, A4, M3, P(i) false (and the rest
true)?
15 Let K denote the set {E, 0}. On this set define addition (+) and multipli-
cation (') by
E+E=0+0=E; E+00+E=0
F E=E 0=0 E=E; 0 * 0=0
Which of the axioms Al, A2, A3, A4, Ml, M2, M3, M4, D, Z are satisfied
in this case? [Have you any ideas why we chose letters F, 0 rather than, say,
a, 1,?]
16 Prove by induction (using principle I) that, for all positive integers n,
n(n+l)
(i) l+2+3+ 2

(ii) '+(2n—l)=n2;
12+22 + = n(n + 1)(2n + 1);
(iii)

(iv) '+n3=(l+2+' .
17 Can you prove by induction that, for all positive integers n, 1 + + +

n ii

18 Can you prove by induction that, for all positive integers ii, 1+ +++

[17 shouldn't cause you any trouble if you know something about the harmonic
series, but a proof of 18, by induction on n, might cause you a few headaches.]
19 Prove, from principle I, that: for each positive integer n, 1 n. Deduce
that there is no integer c such that 0cc <1 and that, for each positive integer
k, there is no integer it such that k <it <k + 1.
20 Suppose a, b e 1 with a >0 and ab = 1. Show that a = 1. [This assertion
is called upon in Section 1.3. Hint: Since 0<a and 0<1 we have 0<b. Hence
a, 1 b (by exercise 19) and hence 1 a ab = 1.]
21

22 What is wrong with the following 'proof' that all triangles in the plane
are pairwise congruent? Let S(k) be the statement 'In every set of k triangles
in the plane all the triangles are congruent to one another.' Clearly S(l) is a
20 Numbers and polynomials

true statement since every triangle is congruent to itself! Now suppose S(n)
is a true statement if n=k and let T1,..., Tk+ be any k+ 1 triangles in the
1

plane. By the induction hypothesis T1,..., Tk and also T2 Tk + 1 are sets of

mutually congruent triangles. Putting these two sets back together we see that
all the T1 are congruent.
23 For the Fibonacci sequence prove: for each k t (i) 145k is a multiple
2 2 2
of
n(n—1)(n2—5n+18)
24 Evaluate for n = 1, 2, 3, 4, 5. Make a conjecture
24
involving 2Pt_1. Try to prove this conjecture by induction.

1.3 Divisibility, irreducibles and primes in Z

We move now to the central concept in this chapter by making

Definition 1.3.1Let a, b e 7. We say that a divides b (or that a is a divisor

of b) and we write a lb if there exists c 7 such that b = ac. If a does not
divide b we write a 4'b.

Remark If we wish to emphasise that we are demanding c 7 we might say

a divides b in 7. We shall have to be careful over this point in Section 3.8.
(See also the paragraph following 1.9.15.) As trivial examples we offer

Examples 1.3.2 31—12, 12k 18, 010 [is this one correct? I might be teasing
to see if you are awake!], 510, 0t2. Note that 12118 in 0.

Definition 1.3.3 If a, b, d €7 are such that dia and dlb then d is called a
common divisor of a and b.

The following result is easy but important.

Lemma 1.3.4 If dla and dlb and if s, t 7 then dlsa + it.

Proof Since dla there exists u €7 such that du = a. Since dlb there exists
v €1 such that dv =b. Then sa +tb =sdu +tdv —d(su -3-tv) so that dlsa +tb,
as required.

The next definition, which we formulate from a desire to get at the funda-
mental building blocks as far as multiplication in 7 is concerned, is intentionally
unconventional, introducing, as it does, a familiar concept in an unfamiliar
way. The reader will recall that one of the author's main aims in this chapter
is to encourage a critical attitude on the part of the reader to statements made
in this book. The main reason for adopting this definition will reveal itself in
Sections 3.6 and 3.7.
 Divisibility, irreducibles and primes in 1 21

Definition 1.3.5
(i) If u e Z is such that ul 1 (in 1) then u is called a unit.
(ii) If a C 1 is neither 0 nor a unit we say that a is irreducible if, whenever
a is expressed as a product, a be with b, c €1, it follows that either b
or c is a unit. [Both can't be units. Why not?]
(iii) If a C Z is neither 0 nor a unit we say that a is prime if, whenever a
divides a product, that is, a lbc where b, c C 7 it follows that a lb or ale
(or both).
(iv) If a, be 7 are such that a = bu, where ii is a unit, then a and b are associates.

Examples 1.3.6 1, —1 are the only units in 7 (exercise 1.2.20); 3, —7 are

irreducibles; 3, —7 are also primes; 8, —8 is a pair of associates.

Remarks The reader's first comments on the above definition might well
include:
(1) What is the point of making 1.3.5(i) when the only two integers satisfying
the property are 1 and —1?
(2) I thought the concept of primeness was defined by (ii) and that (iii)
describes a well-known property of prime numbers.
(3) What is the point of introducing the two definitions, namely (ii) and (iii),
when they express precisely the same concept?
(4) I can see why you exclude a =0 from (ii) and (iii) but why debar a from
being 1 and —1?
(5) I find from (iii) that —3 is a prime. Surely you don't allow negative numbers
to be primes?

Points (1), (2) and (4) will be commented on later (see Remarks (i) after
1.9.3, Theorem 1.4.10 and the Remark following 1.5.1 respectively). For the
moment we dismiss (5) with the answer 'Yes we do'! Regarding (3) we offer
the reader the

Challenge: Are you absolutely sure these concepts are the same? If you are
50 sure, a proof shouldn't be hard to come by. So, go to it before reading on!

We now reveal that the concepts of irreducibility and primeness do coincide

in 7. It is, however, precisely because of the fact that these concepts do not
Coincide in every kind of number system we shall meet (see exercise 6) that
the celebrated conjecture of Fermat remains unproven to this day. (See Section
3.5.)
Returning to 7 we see that to prove the above assertion of coincidence we
ifiust show: (i) that every irreducible element of 7 is necessarily a prime
element; and (ii) that every prime element of 7 is necessarily irreducible. The
latter we can do immediately; the former will take a little longer (Theorem
1.4.10).
22 Numbers and polynomials

Theorem 1.3.7 Every prime element in 1 is necessarily an irreducible one.

The strategy of this particular proof is to head directly from hypothesis to

conclusion as follows: Let a be any prime element of 1. We wish to prove
that a is irreducible. This will (essentially) be achieved if we can show that,
whenever we write a as a product, that is a = bc with b, c E /, then one of
b, c is a unit.

Proof Let a be a prime in 7. Then [by 1.3.5(iii)] a 0 and a is not a unit,

[and it only remains to establish the property described in 1.3.5(u)]. [Thus]
suppose a = bc where b, c E 7. Then certainly a bc. [But a is given to be prime
and so, from 1.3.5(iii),] we deduce alb or alc (or both). WLOG we suppose
a lb. Then [by definition] there exists in 7 an element s, say, such that as = b.
It follows that asc = bc. But bc = a and so asc = a = a 1. Since a Owe may
deduce from property C (Section 1.2) that sc = 1. Thus [since cs = sc = 1] c
is a unit—as required.

Remarks
(i) Note that under the assumption that a lb we have shown that c is a unit.
If you think that we should now give another proof of the same length to
show that the possibility that a Ic leads to the conclusion that b is a unit, you
have cheated yourself in that you have accepted the statement we
can suppose a Ib' above as a valid one when you don't even understand what
it says! If you've been caught out here please do not be so careless again.
(ii) The word 'irreducible' is a good one for elements with the property
listed in 1.3.5(u) since according to that definition an irreducible element is
one which cannot be represented as a product of two properly 'smallef
elements. Thus irreducible elements are seen to be the fundamental building
blocks with respect to multiplication for the system 7.

Let us suppose for the moment that we have shown that in 7 irreducibles
and primes are the same thing. A natural question, to which you probably
know the answer (but not a proof?) is

Question 1.3.8 Are there infinitely many primes (i.e. irreducibles)?

To answer this we use:

Lemma 1.3.9 Let a be an integer such that I <a. Then a can be expressed
as a product of finitely many positive irreducibles (i.e. primes).
*
Without I.oss Of Generality. Roughly speaking this means: We need show you the proof only
in one particular case. All other cases can be dealt with in an identical manner with only the
most trivial modifications.
 Divisibility, irreducibles and primes in 1 23

Remarks
(i) We extend the usual meaning of product to include the case of single
numbers standing alone. Thus whilst 2 3 and 3 2 are the two ways of writing
6 as a product of (positive) irreducibles the expression 7 is regarded as the
required product decomposition for the integer 7.
(ii) It follows immediately from 1.3.9 that if a and if a <—1 then a is
expressible as a product of primes, one of which is negative.

proof of Lemma 1.3.9 [We use the principle W.] Let S be the collection of
all those integers (greater than 1), if any, which are not expressible in the
desired form. [If S = 0 the required result is immediate—there are no 'nasty'
integers.] If S 0 then [S is a non-empty set of positive integers and]
principle W asserts that S has a least member. Let this member be called m
[m for minimum?]. Then m cannot itself be irreducible [since such an m
would have a product decomposition of the required kind, namely m itself].
Thus m can [definition of irreducible] be expressed as a product, say m =
mlm2, where 1<m1<m and 1<m2<m. But [since m is the least element
of S we have] m1 and m2 can be expressed as
products of (positive) irreducibles, m1 = k 1k2. . k,, m2 = .say. It follows
. . .

that m = m1m2 = k1. . . kj1. . . 1. [which shows that m is expressible as a

product of irreducibles after all]. Thus m 5, contradicting the assumption that
m E S. Hence [this assumption is wrong and] S = 0.

Remark Whilst this proof uses the 'contradiction method' it gets to this
contradiction by supposing the existence of a counterexample, hence, by
principle W, a smallest counterexample and thence the contradiction. Accord-
ingly this type of proof might be called proof by minimum counterexample.

We can now answer Question 1.3.8, giving a proof essentially due, to Euclid.*

Theorem 1.3.10 There are infinitely many primes.

The proof is yet another 'by contradiction'. Recall that we are temporarily
assuming that irreducibles and primes are one and the same thing in 7.
Proof Suppose there is only a finite number, n, say, of positive primes. Let
them be listed in increasing order as Pt, P2,. . , p,, [so that Pi = 2, P2 = 3, etc.].
.

Now form the integer = (P1P2. . . J?n)+l. By 1.3.9, can be written as a

product of finitely many primes [i.e. irreducibles] = tit2 . . . tr, say, where
each t1 must be one of the primes from the complete list P2,. . Suppose .

tlPm. Then . .p,, and Hence [by 1.3.4, with s=—1 and
1= 1,we see that] That is, pmII. [But this is absurd
*
Not much seems to be known about Euclid except that he taught mathematics in Alexandria
c. 300 BC. He was supposedly a modest and kindly man who is alleged to have given money to
a student who asked the use of studying geometry 'since he must make gain of what he learns'.
24 Numbers and polynomials
since by definition a prime cannot be a unit.] This absurdity completes the
proof.
As an entertaining diversion and one on which you can again get your
hands dirty we consider a small point which naturally arises here, namely: Is
it in fact the case that each of the numbers is itself a prime? After all with
Pt = 2, P2 = 3, p3 = 5, p4 = 7,..., etc. we get, successively, N1 3, N2 = 7,
N3 = 31, N4 = 211, N5 = 2311, all of which are primes. We make a

Conjecture* If we set = (P1P2PS. .

. p4+1 then, for each positive integer
n, is itself a prime.

Can you amend the proof of 1.3.10 to obtain a proof of this conjecture—or
can you supply a single counterexample which will kill off the conjecture?

Lemma 1.3.9 raises, and the remarks following it answer, an obvious

question, namely: Is the way of expressing any integer (other than —1, 0 and
1) as a product of irreducibles unique? Clearly not, since we may write, for
example, not only 6 = 2 but also 6 = (—3) = (—2). Rather
more to the point are questions like 'Are 132 71 . 103 and. 53 . 73
equal?' Disregarding variations of ordering and sign as in the decompositions
of 6 as given above one can establish a uniqueness-type theorem (1.5.1 below).
We first prove 1.5.1 by employing the standard proof involving definite use
of the concepts of irreducibility and primeness. Then, for comparison, enter-
tainment and for later examination, we offer a more novel approach which
apparently uses only irreducibility.

Exercises
I Show that if a lb and b a in 1 then a, b are associates.
2 Show that if a = bu where u is a unit in 1 then b = av where v is a unit
in 1.
3 Use the method of induction to prove that if a is a prime in 1 and if

4 and 31y.
[Hint: If 3 then x can be written in the form 3t + 1 or it — 1. Similarly
for y.]
S Show that if x, y, z €7 and then SIx or Sly or 51z.
6 Let H be the set of all positive integers of the form 4k + I where k €7.
Thus H ={1, 5, 9, 13,. .}. Call an H number h H-irreducible iff h 1 and h
.

*
Some work on this conjecture by an American undergraduate student can be found in volume
26, p. 567, of the Mathematics of Computation journal. See also volume 34, p. 303.
 Divisibility, irreducibles and primes in 1 25

cannot be expressed as a product of two smaller H numbers. Call an H

number p H-prime if p 1 and whenever with a, b E H we always have
or p1b in H. Write down the first ten H-irreducihles. Which, if any, of
these are H-primes? Are all H-primes necessarily H-irreducible? Factorise
441 in two essentially distinct ways as products of H-irreducibles. [This
example is usually credited to David Hilbert, below: hence the letter H.
We shall meet other similar examples in Section 3.6.]
[In Questions 7 and 8 assume that primes and irreducibles are the same things.]
7 Show that there are infinitely many positive primes of the form 4k -1-3 in
I [Hint: Suppose only finitely many,p1,p2 say. Set .. .p,—l.
Show that must be divisible by at least one prime of the form 4k +3 and
obtain a contradiction.]
8 Show that there are infinitely many positive primes of the form 6k 5 in
Z. Where does this method break down when applied to numbers of the form
8k +7? So there aren't infinitely many 8k +7 primes, right?

David Hubert (23 January /862— /4 February /943)

Hilbert was born at Kbnigsberg (now Kaliningrad), the son and the
grandson of judges. In 1885 he obtained his doctorate with a thesis
On the theory of invariants. After becoming a professor at
Konigsberg in 1892 he obtained the chair at Gättingen in 1895, a
POsition he held until his retirement in 1930.
Hilbert's mathematical interests ranged widely, encompassing the
theory of invariants, algebraic number theory, foundations of
geometry analysis and relativity theory. His outstanding longer
works include the 370 page Zahlbericht (1895—7), in which he
rewrote much of algebraic number theory, and his axiomatic
approach to Euclid's geometry (1899). At the International Congress
of Mathematicians in 1900 Hilbert presented his famous list of 23
problems to which he believed mathematicians should address
themselves. Several of these still remain unsolved.
26 Numbers and polynomials

After his work in geometry perhaps Hubert's greatest wish was to

prove the consistency of arithmetic and thereby to resolve the
'foundations-crisis' that attracted philosophers such as Bertrand
Russell. Some rejected his proposed method of procedure and in
1931 Kurt Gödel dashed all hopes by proving that in a consistent
system formalising the natural numbers there is a theorem A such
that neither A nor not-A can be proved within the system.
Around 1903 Hilbert, to help deal with a problem on integral
equations, introduced that infinite dimensional extension of
Euclidean space now called Hilbert Space.

1.4 GCDs
To get to the standard proof of 1.5.1 will take a while. We begin with

Definition 1.4.1 Let a, b €1. An integer c €1 is termed a greatest common

divisor (gcd) or highest common factor (hcf) of a and b ill
(i) cia and cib and
(ii) if dia anddlbthendic.
Remark We talk of a gcd rather than the gcd of a and b since 1.4.1 says
nothing about uniqueness (nor even about existence!). In anticipation of
proving existence of a positive gcd we introduce

Notation 1.4.2 The positive gcd of a, b (not both of which are zero) is
denoted by (a, b).

Thus

Examples 1.4.3
(i) (60, 24)= 12; (17, —42)= 1.
(ii) For a, b E/
(not both zero) we have (a, b) = (b, a) = (Ia lb I). Thus we
only ever need consider pairs of non-negative integers (see exercise 3 fol-
lowing).

Problem 2 It seems rather obvious that any pair a, b of integers (not both
zero) must possess a unique positive gcd. Surely one simply takes from the
set of all positive common divisors d of a and b the largest? It should be
easy to prove that d is the required gcd. You may not even think there is a
problem at all since the words 'greatest' and 'largest' are synonymous in the
English language. Notice, however, that for us 'largest' is to be interpreted
in the sense of ordering whereas 'greatest' (as used in 'greatest common
divisor') is used in the sense of division. Bearing in mind these remarks you
might try to prove here and now that the numerically largest common divisor
d of a and b, as described above, is indeed their (positive) gcd. (Warning:
 GCDs 27

your proof had better not apply to the H-numbers introduced above (see
exercise 15 below) nor to some of the number systems introduced in Section
3.6 (see exercise 3.6.12) where this assertion on gcds is false!)

Incidentally there is no problem in proving the uniqueness of the positive

gcd—given that a gcd actually exists at all. The main difficulty seems to be
with establishing existence.
A theorem which is very important for us and which proves not only that
any two integers a, b (not both zero) do have a gcd, but much more besides
is

Theorem 1.4.4 Any two integers a, b (not both zero) have a unique positive
gcd. Further, if this gcd is denoted by c, we can find s, t E I such that c = sa + it.

To help prove this theorem we shall need to call upon the following result,
which is known as the Division Algorithm.

Theorem 1.4.5 (The Division Algorithm) Let a, b e I with b 0. Then there

exist unique m, r El such that a = nib +r where r
[The letters m, r are chosen as they are the initial letters of the words
multiple and remainder respectively. A careful choice of notation can often
be very helpful.]

Examples 1.4.6
(0 With a=17 and b=—5 we find in=—3 and r=2 so that 17=
(—3)(—5)+2. Note that in El, rEl and
(ii) With a =—19 and b =—12 we find in =2 and r=5 so that —19=
2(—12) + 5.

Our intuition and experience tells us that the division algorithm is clearly
true. After all we have been working examples like this since age seven or
so. However, let us note that the verification of 1.4.5 in as many as 20 million
cases by no means establishes the truth of the theorem for all possible cases
(cf. the remarks in Section 0.4 on sums of four squares). Furthermore, intuition
is not always a good guide, as problem 3 in the Prologue demonstrates.
One difficulty the beginner might find on being asked to prove 1.4.5 is that
of deciding exactly what to write down. It seems difficult to explain so obvious
an assertion in simpler terms. A proof using a may occur to the reader:
Choose in €1 such that in is the greatest integer not exceedingThen
trivially ci so that a —nib cb (assuming for the moment that 0<b).
Setting a — nib = r we see that a = mb + r where 0 r <b. This proof suffers
from a slight defect in that one of our later aims is to show how to construct
Q on the basis of I so that officially 0 does not yet exist. Another defect is
28 Numbers and polynomials

raisedby the question 'How do you know that the stated m exists?' (Exercise
3.12.10 gives an example not so unlike 0 in which no such m can be found.)
We base a proof on principle W.
Assuming for the moment that 0 a and 0< b we give

Proof of Theorem 1.4.5 LetS be the set of all non-negative integers belonging
to {a — rnb: m e Z}. Then S is not empty since a E S. By the obvious extension
of principle W from 1 to 7 u{0}, S has a least member, which we shall
denote by r. Thus r = a —m1b for some m1 €7. We claim: r <b. For otherwise
0<bcr. It then follows that a—(m1+1)b=r—b is an element of S smaller
than the smallest element r of S. This absurdity leads us to conclude that
r <b. Thus mi and r are such that a = m + r where 0 r <b.
If a <0 we can, by the above, find m, r such that —a = mb + r where 0 c r <b.
Then a=(—m)b—r where —b<—rEO. But then a=(—m—1)b+(b—r).
Clear!y —m — 1 7 and 0< b — r <b if 0< r. (If r = 0 there is nothing to prove
since then a = (—m)b +r where immediately.)
The proof of 1.4.5, in the case where b <0, and of the assertions of
uniqueness we leave to exercises 8 and 9.
We can at last give the

Proof of Theorem 1.4.4 Let S denote the set of all positive integers of the
form ma +nb where m, n are free to range over the whole of 7. Then clearly
at least one of a, —a, b, —b belongs to S. [Why?] Thus S is not empty. Invoking
principle W we see that S contains a least member—let us call it c. We claim:
c is the required (unique) positive gcd of a and b.
In proving that c is a common divisor of a and b we in fact prove rather
more, namely: if w €5 then ciw.
Indeed, let w S and use the division algorithm to write w = kc + r where
k, r 6 7 and 0 r <c. Noting that if r = 0 we have nothing left to prove, we
may assume that 0 < r. Since w and c can be expressed in the forms ua ± vb
and m0a + n0b respectively, where u, v, m0, n0 7 we see that r =
(u — km0)a + ft — Thus r [being greater than zero and of the correct
form] belongs to S. But this contradicts the choice of c [as the least member
of 5; this contradiction shows that the assumption 0< r just cannot hold].
Thus r = 0 and hence ciw, as required. It follows readily that cia and clb.
Finally, c is a gcd for a and b. For, if dia and dib then dlmoa±nob=c
[Lemma 1.3.4].
The uniqueness of c is left to the exercises.

Remarks
(i) The author would like to think that the reader got some enjoyment out
of reading over that proof. It is a rather tidy kind of argument with no loose
ends.
(ii) Whilst being a very agreeable proof of the universal existence of gcds
the above proof is not very helpful in determining (a, b) for any particular
 GCDs 29

pair of integers a and b. To do this we call upon the procedure known as the
Euclidean Algorithm*. In order to clarify the procedure we work the 'general
case' first and only then give a couple of concrete examples.

Suppose, then, that a, b €7 are given and that, to avoid trivialities, neither
a nor b is zero. We use the division algorithm repeatedly as follows:
At step (1) we find tn1, r1 €7 such that a = in1b +r1 where <Ib!
At step (2) we find m2, r2 E 7 such that b = m2r1 + r2 where 0 r2 < r1
At step (3) we find in3, r3 7 such that r1 = rn3r2 + r3 where 0 r3 < r2

and in general
Atstep 0') we find in,, r,€7L such that r,_2—=m,r, 1+r, where
Since b > r1 > . and all the r1 are non-negative we must eventually
reach a first integer I for which r1 = 0. Thus the lth step in the above then reads:
At step (1) we find m,,r7e7 such thatr1 2=in1r,_1+0
This last equality implies rj_11r1 2 and as a consequence r1 = (r1_2, r1 ').
Now it is not difficult to see that (a, b) = (b, r1) (exercise 13) and that
similarly (b, r1) (r1, r2) = . . (r, 2, r1_1) = r1_1. Consequently we see that
.

(a, b) = r1 which can (except in the case where bla, that is where r1 0) be
1

described as 'the last non-zero remainder in the above process'.t

We offer two concrete examples where the gcds are perhaps not entirely
obvious in advance!

Examples 1.4.7
(i) Find (10 113,21 671).
We apply the above procedure to the pair 21 671, 10 113. Successively we
get

21671=
10113= 1445+1443
1445= 1443+2
1443 2-143 the last non-zero remainder.
2= 1+0
Ihus (10 113, 21 671) = 1. (Note that in fact neither of the given numbers is
Itself prime, since 10 113 = 3 3371 and 21 671 = 1667.
(ii) Find (30 031, —16 579).
*
Obtained by Euclid in Book 7 of the Elements. Euclid's work was by no means concerned
Solely with geometry.
G Lamé, whom we meet again in Section 3.5, proved that the number of steps required to
find the gcd is at most 5 times the number of digits in the smaller of the two given numbers.
See [56, p. 43].
30 Numbers and polynomials

We apply the procedure to the pair 30031, 16579. Successively we get

30031 = 116579+13452
16 579= 452+ 3127
3127+944
944+295
944=3 295+
295=5 59+0
Thus (30 031, —16 579) = 59, a result achieved without factorising either
number! (Incidentally if the number 30031 has no special significance for
you, I think I can safely claim that you have not properly settled the conjecture
following 1.3.10.)
1.4.4 asserts that there exist s,tel such that 1=10 113 s±21 671 t. To
find these we simply read the steps of the Euclidean Algorithm in 1.4.7(i)
backwards, as follows.
1 -=1443—721 2

=1443-721(1445 —1443)=722(1443)—721(1445)
=722(10113—6 1445)—721(1445)=722(10113)—5053(1445)
=722(10 113)—5053(21 113)

= 10828(10 113)—5053(21 671)

as required.
Thus we may choose s = 10 828 and t = —5053. Of course these coefficients
are by no means uniquely determined since we can also write
1 =(10828—21 671 k)(10 U3)—(5053—10 113k)(21 671)
where k is any integer.

On occasions it might prove convenient to compute the gcd and s and

simultaneously by putting the two procedures described above into one table,
as shown in Table 1.1. We leave the reader to work out exactly how the
various rows in the table are arrived at.

Remark The above working of the Euclidean Algorithm in the general case
is easily seen to provide another proof of the existence of and the formula
for the positive gcd of any two integers (not both zero). However, I am sure
everyone would agree that the proof of 1.4.4 is much, much sweeter, the
proof based on the Euclidean Algorithm being somewhat ungainly. Thus we
have the interesting situation of having two proofs concerning gcds. The first
offers a beautiful proof of existence but is fairly useless for calculations. The
 GCOs 31

Table 1.1

sa+tb s t

21671 0 1

10113 1 0
1445 —2 1

1443 13 —6
2 —15 7
1 10828 —5053

second is perfect for specific calculations and also gives a proof of existence.
On the other hand this proof seems to have little aesthetic merit.
Finally in this section we prove (at last!) that every irreducible element in
1 is necessarily a prime element so that the concepts of primeness and
irreducibility coincide in 7. We need a definition and a trivial consequence.

Definition 1.4.8 Two integers a, b are said to be relatively prime (or coprime)
iff(a,b)= 1.
Combining this definition with 1.4.4 we obtain immediately

Theorem 1.4.9 Let a, b 8 7. Then a and b are relatively prime if there exist,
in 7, integers s and t such that sa + tb = 1.

At last we have

Theorem 1.4.10 Let a e 7 be an irreducible element. Then a is a prime

element.

Proof Suppose b, c e 7 and a bc. [We want to show that a b or a c (or both).]
If a tb. It follows that (a, b) = 1. [Since a is
irreducible its only divisors are a, —a, 1, —1, and since atb the only divisors
a can have in common with those of b are 1 and —1.] Thus by there
exist s, t 7 such that sa + tb = 1. But then sac + tbc = c. Now a sac [you can
see that it does] and a'tbc [because. why?]. Thus aic, as required.
. .

[In summary: we have proved that either aib or, failing that, alc. Thus by
1.3.5 (iii), a is a prime element in 7.]

Remark As already noted in the system H of exercise 1.3.6, irreducible

elements are not necessarily prime ones. Of course H is intended only as an
easy illustration of what might go wrong—and go wrong it does in some of
the important generalisations of 7 that arise in Section 3.6.

Exercises (Do not use uniqueness of factorisation in 7 in these exercises.)

1
Show that the pair 0,0 has infinitely many common divisors but only one
common divisor.
32 Numbers and polynomials

2 Show that if c and d are gcds of a, b then c and d are associates. Deduce
that (a, h) is unique (as implied by the Remark and by 1.4.2 on p. 26).
3 Show that for all a, b (not both zero) we have
(a,b)=(—a,b)=(a,—b)=(—a,—b)=(b,a)
4 Showthatif(a,c)=(b,c)=lthen(ab,c)=1.[Usel.4.9.]

5 Show that if cia +b and if (a, b)= 1 then (a, c)=(b, c)= 1.
6 Prove or disprove: For a,b,cE7L (with (a,(b,c))=((a,b),c). [If
the equality holds we could unambiguously denote either gcd by (a, b, c).]
State and prove the analogue of 1.4.4 for three integers a, b, c.
7 Show that for each nE7L (5n+2,12n+5)=1. [Hint: c=(a,b)4'
cil2a —Sb.]
8 Show that once a, b are given, the values m, r in 1.4.5 are uniquely
determined. [Hint: Suppose a = rn1b +r1 +r2.]
9 Complete the proof of 1.4.5 by proving it in the case of b <0.
10 Let S be a non-empty set of integers such that a E 5, b ES a — b E S.
Prove that either S comprises 0 alone or else S contains a smallest positive
member c and S comprises precisely all the integral multiples of c.
11 Show that when we write c =sa +11,, where c = (a, b), then (s, t) = 1.
12 Given that sa + tb = 1 prove or give a counterexample to each of the
following assertions
(i) (sa, tb) = 1; (ii) (sb, ta) = 1; (iii) (st, ab) = 1.

13 Show that if a = nib +r, as in the division algorithm, then (a, b) = (b, r).
14 Find (527, 901) and write it in the form 527s+901 t. Find that pair of
integers s, t for which s is as small as possible positive. Prove there is no
solution s, it for which 25 <s <50.
15 Let H be the set of numbers defined in exercise 1.3.6. Show that 21 is
numerically the largest common H-divisor of 441 and 693 but show that there
is no gcd in the sense of 1.4.1.

16 Show that if — 1 is a prime (where k is a positive integer) then (i) k

is a prime and (ii) 2k-1(2k — 1) is a perfect number. [An integer greater than
1 is called perfect if is equal to the sum of all its positive divisors, including
it

1 but excluding itself.]

Find four perfect numbers. [Euclid proved (ii). Euler (see p.69) proved
every even perfect number is of that form. No odd perfect number is known.
The largest known is 2216090(2216091 — 1).]
17 Let p be a prime. For each i such that 0< i <p show that the binomial
coefficient is divisible by p. Use this to prove, by induction on n, that for
each n e we have p In — n. (This is Fermat's Little Theorem.)
 The unique factorisation theorem (two proofs) 33

is Show that for all where u1, u2, u3,... is the

Fibonacci sequence (Section 1.2).

1.5 The unique factorisation theorem (two proofs)

Even though we have just proved the equivalence of the concepts of primeness
and of irreducibility in / we ask the reader to note that in the following
theorem the concept of irreducibility is associated with the existence of a
decomposition of the asserted kind whereas the property of primeness is used
to establish the uniqueness of this decomposition.

Theorem i.5.V (The Unique Factorisation Theorem for /; also called the
Fundamental Theorem of Arithmetic) Let a be a non-zero element of 7.
Then either a is a unit or a can be expressed as a product of a unit and
finitely many positive primes. Further, if a = uptp2. . . Pr = vqtq2. . . q, where
u, v are units and Pt, . . . . . , are positive primes then u = v, r = s
and the and the q1 can be paired off in such a manner that paired primes
are equal.

Proof Half of the theorem has been proved already: Lemma 1.3.9 and
Remark (ii) following it show that every integer greater than I (respectively,
less than —1) can be expressed as a product of (respectively, —1 times a
product of) finitely many positive irreducibles (which we now know to be
primes).
Now suppose there exists an integer a with decompositions as above but
in which the and q1 do not pair off. If a is a negative integer then lal is a
positive integer with the same nasty property, and the set S of all positive
nasty as is non-empty. Thus S contains a smallest member. WLOG let it be
lal. Now . .and clearly ptIq'q2. .By exercise
1.3.3 we deduce [because Pt is prime] that for some i. But is irreducible
and so Pi = 1 or —1 or qi or —qi. Since Pt is not a unit and since Pi and qi are
both positive we are forced to conclude that p i = q. Thus the above equality
reducestop2p3.. . Pr =qiq2.. . . q5.[Whyarewe allowed tocancel?]
. .

But this integer is clearly smaller than a and so it does not lie in S. That is,
unique factorisation does apply to this smaller integer and we can deduce
that the remainingp2,. , Pr and qi,. , q1 -1, qi + 1, . , pair off (in particular
. . . . . .
r 1 = s — 1 so that r = s) in the manner described in the statement of the
theorem. Since Pt and q1 have already been paired off in the appropriate
manner (i.e. they are equal!) we find that is not a nasty integer after all.
That is, S is empty, and this proves the theorem.

Remark If we admitted 1 and —l as primes we would lose the uniqueness

We ye just obtained since, for instance, 7 = 1 7 = 1 1 7. Hence the restriction
01' primes and irreducibles being non-units.
*
Euclid: Book 9.
34 Numbers and polynomials

Before closing this section we offer a delightful second proof of this last
theorem in which (the reader is invited to verify) no use is made of the concept
of primeness. The reader is also invited to ponder which of the two proofs
gives him the greater personal satisfaction. (The author's preference should
be obvious!) This time we shall work throughout with decompositions of
integers into products of positive irreducibles.

Second proof of Theorem 1.5.1 Once again we quote 1.3.9 (which makes no
use of the primeness property) to establish the existence of a decomposition
(into irreducibles) in each case. To establish the uniqueness of this decom-
position we proceed as follows.
If the assertion relating to uniqueness is incorrect then there exists a smallest
positive integer c, say, such that c is expressible in two essentially distinct
ways. Suppose indeed that c =P1P2. . Pr =qlq2.
. . . where the p1 and q1 are
(positive) irreducibles. Since [by the minimality of cJ no can be equal to
any q1, we may assume [WLOG] that Pi <q1. Set d . . so that d Cc.
Now c—d=(q1—p1)q2. . w,q2. . where w1w2... is the
decomposition of qi —PI into irreducibles. (If q1 = 1 we get c —d =
q2q3. . . q5.) Now piIc and pild. Hence piIc —d. This means that c —d has a
decomposition [into irreducibles] of the form c — d = P1VIV2 . . t',,, where
.

V1V2. . . Vm is a decomposition of (c — d)/p1 intQ irreducibles. Thus c — d has

two essentially distinct decompositions into products of irreducibles. For: Pi
occurs in one decomposition but not in the other since (i) Pi is not equal to
any of q2, q3,. . . , and (ii) cannot be'equal to any w1, W2,. .., (or else
.. w1 =q1 —Pi whence pijqi, which is impossible—why?).
Since 0Cc — d Cc it appears that our assumption that there exists a smallest
counterexample is untenable and so the theorem is proved.

Remark In Section 3.6 we shall discuss number systems in which the concepts
of prime and irreducible definitely do not coincide and in exercise 3.7.16 we
shall ask the reader to criticise the application of the above proof (since it
leads to a manifest contradiction!) to these systems. See also exercise 2 below
and exercise 3.11.1.

Exercises
I Is 71 103 = 53 73? [Unhelpful hint: Both integers end in
a 7.]
2 Consider once again the set H of exercise 1.3.6. Clearly H satisfies
principle W. Every element of H has at least one factorisation into
H-irreducibles and 441, for example, has the two distinct decompositions
9 49 and 21 21 into H-irreducibles. Determine where the above second
proof of 1.5.1 fails for the set H.
3 Provethatifd=(a,b)then(a/d,b/d)=1.
 Polynomials—what are they? 35

4 Prove that if dila, d2Ia and (d1, d2) = 1 then did2Ia.

5

6
Prove that if ajbc and if (a, b) = 1 then aic.
Prove that if C2 = ab where (a, b) = 1 and where a, b
are both perfect squares.
t then a and b

1.6 Polynomials—what are they?

We now try to duplicate for polynomials some of the basic definitions and
theorems set down earlier when investigating /. We shall find out that, despite
the obvious differences in appearance between the elements of 1 and those
of Ok], the underlying structures of these two systems are quite similar.
Before beginning this attempted duplication we ought perhaps to remind
ourselves exactly what polynomials are. This is easy. A polynomial in x with,
let us say, rational number coefficients is simply an expression of the form
a0+a1x+'
for some m €1 and some €0. Further we define addition and multiplication
of polynomials in the usual way. That is we define addition by
in
. .+b,,x n )
=(ao+bo)+(ai+bi)x +...
and multiplication by

=a0b0+(a0bi+aib0)x +•
Before continuing we might just stop to clarify a few apparently trivial points.

(i) Are we to consider 2 +x2 as being of 'the same form' as 2 +Ox + 1x2 +0x3
or not?
(ii) What actually isx?
(iii) In the above definition of addition it seems that the + sign which is being
defined (that is, the one between the brackets) is being defined in terms
of itself! (Witness the same + signs inside the brackets.) Am I right? Is
this fair?
It would appear then that not everything in the garden is quite as rosy as
'night have been suspected. Certainly one would like the answer to (i) to be
and in fact that's easily arranged. And (iii) causes no real problem. One
Just changes the sign between the brackets to $ and proceeds a little more
carefully. Point (ii) however is a bit more problematical. To say that x is
an 'indeterminate' is no good. (What on earth is an indeterminate?) No
'nore can x be 'something which stands for a number' since one could
Cqually well imagine it standing for a matrix or even the differential operator
36 Numbers and polynomials

D (meaning d/dx), where D,

the
of of a
the a + bi'. Questioner: What is 'i'? Responder:
It is But what is that? Is it a number? Well. . .yes. It can't be a real
number since its square is negative. Of course. So on what grounds do you
assert that, for instance, 2 . i = i . 2?. . No reply!
.

We shall deal with these questions in Section 4.4.]

There's an easy way to avoid embarrassing questions about x (just as there
is about i). Dispense with x! Notice that the symbols x, x2 x s,... are really
nothing more than place markers. This is especially noticeable (now you know
to look for it!) when multiplying two polynomials in the usual manner. Taking
into account the problem raised in (i) above we make the formal

Definition 1.6.1
(i) A polynomial with coefficients in 0 is an infinite* sequence (a0, a1, a2,...)
in which the a, all belong to 0 and in which all the a, are equal to 0 from some
point onwards. (More formally: all the a, belong to 0 and there exists N eZ
such that 0 N and = 0 for all n such that N n. Of course N is allowed to
vary from one sequence to another.)
The set of all polynomials with coefficients in 0 will, for reasons which will
emerge in Section 1.8, be denoted, as usual, by 0[x].
(ii) The polynomial (b0, b1, b2,...) and the above polynomial
(a0, a1, a2,...) are said to be equal if a, = b, for all non-negative integers i.
(iii) Addition is defined by

where, for each non-negative integer i, c, = a, + b,.

(iv) Multiplication is defined by
(a0,a1,a2,. . .)O(bo,b1, b2,. . .)(d0,d1,d2,. . .)

where, for each non-negative integer i, d, = a0b, + a 1b,_1 . . + a,!,o. That is,
we putt d,
= k=O

Remarks
(i) The symbols $, 0 (the first of which is often calledt 'hot cross plus'!) are
used in preference to + and to distinguish them from the signs by which we
*
Wouldn't finite sequences be better? Think about it!
4
t means 'sum'. Thus Wk is shorthand for w1 + w2 + w3 + w4.

I am grateful to Dr R J S C?ossley of the University of York for informing me of the alternative

names 'plus bun' and 'dot bun'—appropriate terms since, as he says, ®, C are examples of
'bunnery' operations! (See Section 2.1.)
 The basic axioms 37

are already combining elements of 0. The new symbols are sufficiently

unfamiliar to remind us of their defining role but sufficiently similar to + and
to help us remember their meaning with a minimum of effort.
(ii) It is probably clear to the reader that, although we are trying to work
formally and to assume nothing other than that given, we do not choose our
definitions of ® and ED in a perfectly random fashion, but rather make them
reflect what we want to happen. Remember the aim here is to get rid of the
problems concerning x, not to get rid of the concept of polynomial itself!
Exercises 2 and 3 will convince doubters that our definitions of $ and ED are
the right ones!
(iii) According to the above definitions, the sum and product of two given
polynomials are again polynomials. Clearly each sum and product is a sequence
so the crux of the matter is to show that, in each case, it is a sequence with
only finitely many non-zero terms (exercise 6 below).

Exercises (Recall: a polynomial is an infinite sequence)

1 Determine the polynomial z, say, which is such that, for all polynomials
u, we have u $z = z $u = u. (z is clearly uniquely determined and is called
the zero polynomial.)
2 Let u=(l,3,O,—5,2,O,...) and v=(O,2,O,—1,O,3,O,...) where all
the entries not shown are to be assumed equal to 0. Find (a) u $ v (b) u ED v
(c)u®u (d)vEDu.
3 Informally using the more familiar notation suppose f = 2x — + 3x5 and
g=1+3x—5x3+2x4.Find(a)f+g (b)f'g (c)g2
4 Did your find 3(b) easier to do than 2(d)? If so, can you explain why?
S Find u, v each having at least two non-zero entries such that u C v =
(1, 3, 1, —1, 0,. . .). [Recall: all we want is the answer. You may use any
method you wish to get to that answer.]
6 Prove that the sum ($) and product (C) of two polynomials is again a
polynomial.

1.7 The basic axioms

We would now like to investigate which of the axioms Al,. , M4, etc. listed
. .

in Section 1.2 hold for the set Q[x]. We shall check a few and leave the rest
for the reader. For instance:
For A2 (in great detail): Here we are given A = (a0, a1,.. . , . .), B =
.

(b0, b1, .. , ba,...) and C = (co, c1, .

. , ca,...) where, for safety, we include
. .

the 'general' entries c,,. NowA$B = (a0+b0, a1 +b1,.. . ,

Hence (A$J3)$C =({a0-t-b0}-c-c0,{a1±b1}+c1,.. ,{a,, . .). On .

38 Numbers and polynomials

= (bo+co, b1 + c1,. . . , .). HenceA ®(B ®C) =

(ao+{bo+co}, a1 +{b1 +ci},. . , . . . .). The question is: Is the
polynomial (A ®B)®C equal to the polynomial A ®C)?
The answer is: Yes, provided that for each non-negative integer n we have
= And this indeed we do since, in Q, addition is
associative.
Note, then, that the fact that 9 is associative depends only upon the actual
definition of e and the fact that + is associative on 0.
For M3 (more briefly): Let I denote the polynomial (1, 0, 0, .). Then . .

[by definition] A 01 = (d0, d1, d2 d1,...) where, for each i, d1 =

did I think of trying I? Answer: I asked myself 'What polynomial g(x) (in
the old-fashioned sense) is such that, for all polynomials f(x), we always have
f(x)g(x) =f(x)?' The answer, g(x) = was obvious, and I is the sequence
1

version of the polynomial 1 + Ox + Ox2 The point being made here is.

that, whilst one is forbidden to use intuition to prove theorems, one can use
any means, and intuition is a good means, to feel one's way to an answer or
proof. Once an outline has emerged it is often not too difficult to fill in the
formalities.]
Note that I exists only by kind permission of 0, I e 0. That I has the
properties asked of it is due entirely to the fact that in 0 we know that. . . what9
We can 'order' the members of 0[x] so that P(i) and P(ii) are satisfied,
although one could not expect 0[x] to satisfy all the axioms Al through to
I. [Why not?] We shall leave this to exercise 4.
For the moment we concentrate on showing that property Z holds in 0[x].
ForZ:LetA=(a(hal,...,a,,O,...),B=(bO,bl,...,bf,O,...)wherea,,b,
are the last non-zero terms in A, B respectively. Setting A 0 B
(do, d1, . . . ,

d, = a0b, f + +a, b,. ' a,!,, -'-a, * h, + +a, .,h,

[= 0 ± 0 +-- 0 -ta,b-- 0 4--- 0 =a,h,I

Thus d.,, 0 [since a, b, are given to be non-zero in 0]. In particular A 0 B

is not the zero polynomial (0, 0, 0,. .). .

Note that 0[x] satisfies Z essentially by kind permission of the fact that 0
itself does.

Note 1.7.1 Although we don't need it just now we note that if we look at
the terms for s + t <i, in the above product, we see that each is equal to
0. Thus d,, is the last non-zero term in the above product.

Finally for M4: With A, B as above, we have seen that d5÷, 0 if a, and b,
are respectively the last non-zero terms of A and B. Thus if a1 0 for any i
 The 'new' notation 39

such that 0< i it is impossible to choose B such that A C B = L Indeed we

note that

Theorem 1.7.2 Given the polynomial A, there exists a polynomial B such

that A C B I if a0 0 in 0 and =0 in 0 for all i such that 0< i.

We leave the remaining details to the reader.

Exercises
1 Prove that for all u, v E 0[x] we have u v v u. That is, prove that e
satisfies axiom Al.

2 Prove that $, C together satisfy axiom D.

3 Prove that C satisfies axiom M2.
4 Find in 0{x] a subset N so that axioms P(i) and P(ii) are both satisfied.
[If you have difficulty, why not team up with a colleague and argue over it?]
You can't find two such subsets can you? (Cf. exercise 3.12.2.)
5 We define the set of all power series over 0 to be the set of all
infinite sequences as in 0[x] except that the restriction that only finitely many
of the a1 be non-zero is dropped. Show that the element (a0, a1, a2,...) has
a multiplicative inverse in if a0 0.

1.8 The 'new' notation

Definition 1.6.1 is all very well in that it gets rid of any nasty problems
concerning x, by getting rid of x! Further, exercises 1.6.2 and 1.6.3 indicate
that these sequence-type polynomials do appear to behave as we wanted them
to. (This is scarcely surprising since we defined $ and C so that they would!)
But exercises 1.6.2 and 1.6.3 and perhaps, more especially, exercise 1.6.5
emphasise that sequence-type polynomials are a little difficult to handle. Why
is this? The answer is simple. Unfamiliarity. We can work using the x-form
faster than we can the sequence form simply because been using the
former for years. Let's see how we can recover the familiar x-form from the
sequence-type definition.
To begin with, consider the subset of 0[x] comprising those polynomials
(i.e. sequences) whose every term, excepting possibly the first, is zero. Thus
we are looking at polynomials such as 0,.. ,) and (0, 0,. .). For two such
.

polynomials (a, 0,...) and (b, 0,...) we note that, according to 1.6.1, their
sum and product are respectively (a + b, 0,.. .) and (a b, 0,. .). .

We see, then, that the brackets, commas, zeros and dots are just so much
dead wood in that, if we write (a, 0,0,...) more briefly as a, then the
above addition and multiplication can be written much more briefly as a 3b =
a +b and a
40 Numbers and polynomials

We achieve even more economy by giving the name 'x' to the polynomial
(0, 1, 0,. .). For then we observe that
.

(O,a1,O,...)=(ai,O,...)®(O,1,O,...)
= a1 S x in our new notation

= a2 S x C x in our new notation

and in generalt
inournewnotation.
n xs here

Finally we observe that

(a0,a1,a2 .)=(a0,O,.. .)®(O,ai,O,.. .)$

®a19x $. . GxOxQ • Gx

and so, if we replace the $ sign by +, if we simply drop the C sign and if we
then agree to write x C x '0 x as x we establish a notational way of
writing polynomials that we might just find a little easier to work with! That
is, we are now proposing to write (a0, a1,. , 0,...) more intelligibly as
. .

ao+ã1x +a2x2+'' It appears that the bars over the as are

also superfluous since (on using the new notations + rather than ® etc.)
a+b—a+b and aE=ab. So we drop the bars too and find that with
the conventions described above we have shown that every polynomial can
be expressed in the form a0+a1x +a2x2+' . Note that the ways of
adding and multiplying polynomials in this new notation are precisely those
you have always used.

Remarks
(i) The reader who feels both exhausted and cheated at this outcome (if any
such reader there be) will perhaps comment: 'What a waste of time. All that
just to get back to where we started.' In fact we haven't quite got back to
where we started. We have shown that polynomials can be thought of in the
way we have always thought of them, secure in the knowledge that uncomfort-
able questions about x can be circumvented. -
(ii) In recognising that the sets Q acted upon by + and ' and 3 acted upon
by $ and C are essentially the same, we are meeting for the first time a
concept of prime importance in algebra, namely that of isomorphism. This
concept can perhaps be more readily appreciated by looking at Fig. 1.1 which
also clearly indicates that whilst the elements of 0 are totally distinct from
* The triple product is unambiguous sinct 0 is associative (exercise 1.7.3). Similar remarks apply
to the other multiple products. See exercise 1.2.1, Theorem 2.7.7 and exercise 1 below.
± Note the neat sidestepping of a proper proof by induction. See exercise 1 below.
 The 'new' notation 41

those of Q[x), nonetheless Q[x] contains the subsystem C which to all intents
and purposes is the 'same' as 0 so that, if it proves conNenient, 0 may be
identified with it. Such identification of the elements of 0 with those of 0 is
in exactly the same spirit as the identification of the elements of 1 with certain
elements of 0 (see Fig. 1.2), an identification to which you've probably never
previously given much thought. (For more on this see Sections 3.10 and 5.9.)

0,

Fig. 1.1

Fig. 1.2

(iii) At this point we should like to draw the reader's attention to a second
aspect of polynomial algebra which is sometimes badly dealt with. The question
arises: What are we to make of the equality x2—3x +2 = 0? Since the left-
and hence the right-hand sides of this equality are polynomials they are simply
shorthand for (2, —3, 1, 0,.. .) and (0, 0,. . .) respectively. Hence 1.6.1(u) tells
us to deduce that 2 = 0, —3 =0 and 1 = 0 in 0! We leave the reader to think
about this until we reach Section 1.11.

Note however that the assertion x2+2 = (x — 1)(x —2)+3x is certainly one
which is meaningful—and also happens to be true! These remarks indicate
that, if nothing else, we ought to proceed with circumspection.
42 Numbers and polynomials

Exercises
1 Since by exercise 1.7.3 polynomial multiplication is associative, we see (cf.
exercise 1.2.1 and Theorem 2.7.9) that the definition of x' as ihe product
x S x S. S x with n occurrences of x' is unambiguous. In particular we
can write as x" x. Using this equality prove, by induction, that, for each
non-negative integer t, xt is the polynomial (0, 0,. . ., 1, 0,. . .) with all entries
equal to 0 except for the lone 1 in the (t + 1)st place.
2 Using the set Q[x] rather than the set 0 to form sequence-type poly-
nomials, define the concept of 'polynomial in two letters x and y with
coefficients in 0'. Prove from your definition that xy yx. [The set of all such
polynomials we naturally denote by (0[x])[y] — more briefly, 0[x, y]j

1.9 Divisibility, irreducibles and primes in Q(x]

The fact that both / and 0[x] satisfy, with respect to addition and multiplica-
tion, most of the same axioms (both failing to pass M4) leads us to ask if
other concepts defined for 7 have analogues in 0[x]. (Of course, even if they
do, it will not necessarily mean that these concepts will have the same
importance for 0[x] as for 7.)
Following Section 1.3 we begin with

Definition 1.9.1 (cf. 1.3.1) Let f, g 0[x]. We say that f divides g (or that
f is a divisor of g) and we write if there exists h e 0[x] such that g =fh.
If! does not divide g we write ft g.

Examples 1.9.2
(i)
(ii) x +2tx +2x —4x + 12 (try it by long division!)
Definition 1.9.3 (cf. 1.3.5)
(i) If u e 0[x] is such that u 1 then u is called a unit.
(ii) If f e 0[x] is neither the zero polynomial 0 nor a unit we say that f is
irreducible if, whenever f is expressed as a product, f = gh with g, h C
it follows that either g or h is a unit. A non-zero non-unit polynomial f will
be called reducible if it is not irreducible.
(iii) If feQ[x] is neither 0 nor a unit we say that f is prime if, whenever f
divides a product, that is, where g, he Q[x], it follows that or (or
both).
(iv) If f, g e 0[x] are such that f = gu, where u is a unit, then f and g are
associates.

Remarks
(i) We emphasise that the requirement ull in 1.9.3(i) demands that we find
an element v in 0[x] such that uv = 1. Thus according to 1.7.2 the units in
 Divisibility, irreducibles and primes in QExI 43

Q[x] are precisely the non-zero constant polynomials—and there are infinitely
many of them. It follows that each non-zero element of O[x] has infinitely
many associates.
(ii) Instead of Q{x] we could equally well have considered Z[x], C[x],
each defined in the obvious way. Note that in Z[x] the only units are the
polynomials 1 and —1 and that whilst 2 +2
in Z[x].
(iii) Whether or not a polynomial is irreducible depends upon which set of
polynomials it is considered as belonging to. For example, the polynomial
2x2+2 is irreducible in Q[x] and R[x], but reducible in Z[x] and in C[x].
Can you see why in each case?

Notes 1.9.4
(i) In Section 1.5 we proved that each non-zero non-unit element of /
factorises into a product of irreducibles in an essentially unique way. We
naturally enquire whether or not the corresponding result holds in Q[x].
Analysis of 1.5.1 shows that it does if O[x] satisfies analogues of 1.3.9, 1.4.10,
exercise 1.3.3 and property C of Section 1.2. Now 1.4.10 calls upon 1.4.4
which itself uses 1.4.5 which. The energetic reader is invited to supply
. . !

the details. To give it maximum prominence the statement concerning factor-

isation in Q[x] is placed at the end of this section (1.9.18). However, instead
of checking this chain of results for every concrete example where we seek
to prove uniqueness of factorisation, it would be better if we could isolate
the essentials behind these theorems leaving us in each specific case only these
essentials to verify. This we shall do in Section 3.7.
(ii) Because (the analogue of) property C can be proved in Q[x], the analogue
of 1.3.7 is easily seen to hold there. Together the analogues of 1.4.10 and
1.3.7 show that, in Q[x], an element is prime if and only if it is irreducible.
Thus appears a very minor reason for our introducing the prime versus
irreducible battle in 1.3.5 and 1.9.3: primes (i.e. irreducibles) in / are usually
called primes! Primes (i.e. irreducibles) in Q[x] are usually called irreducibles!
Thus any difference implied by use of different words in / and Q[x] is illusory.
(iii) One final point. The proofs of 1.3.9, 1.4.4 and in particular the Division
Algorithm 1.4.5 make use of the fact that property W holds in 7. Since, despite
exercise 1.7.4, Q[x] does not seem to satisfy W (try it!), we are fortunate that
the appropriate analogue for Q[x] of 1.4.5 depends upon 7, rather than upon
Q[x], satisfying W. We obtain this analogue in 1.10.1.
Somewhat more perplexing however is the following. In proving in 7 and
in Q[x] that all irreducibles are primes we make essential use of the appropriate
division algorithm. However it is easily seen (exercise 1.10.2) that there can
be no analogue of 1.10.1 in 7[x]. This in itself does not immediately imply
that there are irreducibles in Z[x] which are not primes. Of course if there
are, the analogue for Z[x] of 1.5.1 could well fail since the proof as given
certainly wouldn't then go over to Z[x]. And yet surely factorisation in
44 Numbers and polynomials

is unique? What do you think? We pose

Problem 3 Are all irreducibles in Z[x] also primes?

Get your hands dirty by doing a bit of practical experimentation. You might
even stumble upon a theorem or two in your researches.

We now ask the obvious question: Which polynomials in Q[x] are irreduc-
ible? The best reply we can give is: There is a test (see exercise 1.11.11) due
to Kronecker* which will always tell in a finite number of steps (which may
easily be large enough to require a computer) whether or not a given element
of Q[x] is irreducible. Otherwise there is a criterion (1.9.16) which is easy to
apply in practice and which describes infinitely many (but not all) polynomials
irreducible over Q[x]. Other tests are given by the remark following 1.11.7
and 4.2.10(u).
To establish 1.9.16 we make some definitions and proofs relating to Z[x].
For ease of recognition we denote polynomials in Z[x] by capital letters.

Definition 1.9.5 A polynomial F—z0+z1x +. . EZ[x] is called

primitive itT the positive gcd in 2 of all the z1 is equal to 1.

Example 1.9.6
. . .. 72x3 +2x —42x +8 is not.
lOx2 + 15x +6 is primitive, 2

Notation 1.9.7 Extending the notation of exercise 1.4.6 we denote the

positive gcd of z0, z1,. .. , z,, (not all zero) by (zo, z1,.. ,zj. .

Such a gcd always exists and is unique. A word for word copy of the proof
of 1.4.4 shows this gcd is equal to the smallest positive integer in the set
.

A passage from Q[x] to Z[x] is provided by

Lemma 1.9.8 If f Q[x] then there exists in Z[x] an associate F = af of f

(where a c C is chosen to be positive) such that F is primitive.

Proof I'll leave you to outline this. The following example shows you the way.

Example 1.9.9 If f=d—9x+'9x2—14x3——ih[77—1155x+330x2—140x3]

The next result is due to Gauss (see p.68).

Theoreml.9.1O +ymxm
Z[xJ are both primitive then so is FG.

Proof Let FG is not primitive then

Co, C,n±n have a common divisor (other than I and —1) in 2 and hence
*
Leopold Kronecker (7 December 1823—29 December 1891).
 Divisibility, irreducibles and primes in Q[x] 45

some common prime divisor p, say. Now p doesn't divide all the nor all
the z1. [Why not?] Let s and t be the least suffices for which p .f' Ys and p ,f' Zt.
Consider* = y0z5±1+y1z5+1_1+'' +YS±IZO. Nowp iF'y5z1 [since
p '1'Ys' p z1 and p is a prime]. On the other hand [by choice of s, it] p does
divide every other term in the above equality including itself. [This
manifest contradiction shows that the assumption that PG is not primitive is
untenable.] Thus FG is primitive, as required.
For the next theorem we need two definitions. First

Definition 1.9.11 Let

.

The gcd (zo, z1,. . , z,.) is called the content of F.

.

Example 1.9.12 The content of —18x2+45x —30 is (—18,45, —30) =3.

Second

Definition 1.9.13 Let f = a0 + a 1x + where 0. Then n is called

the degree off We write deg f = n. (We do not attribute a degree to the zero
polynomial 1=0.)

We can now prove the important

Theorem 1.9.14 (Gauss) If FE Z[x] and if we can write F = gh where

g, h E Q[x] then we can write F = GH where G, H e Z[x], deg G = deg g and
degH =degh.

Remark The real meat in this theorem lies in the equalities of the degrees
of G and g and of H and Ii. Without these equalities the theorem is true but
absolutely trivial.

Proof
(i) We first suppose that F is primitive and write g = G and h = H where

are such that (a,b)=1=(c,d), where G,H€Z[x] and'are

Primitive and where degg =deg G and degh =degH.
Thus F=gh Hence bdF=acGH in Z[x]. Now F and GH are
both primitive polynomials [by assumption and by 1.9.10 respectively] and
so the content of bdF is bd [why?] whilst the content of acGH is ac. [Since
the polynomials bdF and acGH are equal their contents must be.] Thus
ac = bd and F = GH follows [from the equality above] as required.
*
if r > n we of course take z, to be 0.
46 Numbers and polynomials

(ii) If now F is not primitive, write F = zF1 where z is the content of F and
where F1 is primitive. Then, from F = gh we get F1 = h. Hence, by (i),
F1 =01111 and consequently F = (zG1)H1 where 01, zG1 and 111 E Z[x].
As an immediate corollary we obtain a result you might have stumbled on
in your researches into Problem 3.

Corollary 1.9.15 If F E Z[x] and if one can't write F as a product of poly-

nomials, each of smaller degree and with integer coefficients, then one still
can't write F as a product of polynomials each of smaller degree even if one
is prepared to use polynomials with rational coefficients. Equivalently: If
F E Z[x] is reducible in Q[x] then it is already reducible in Z[x].

We now give the solution to Problem 3. We take care in stating exactly

where each division is taking place. (Recall that 213 is false in 7 but true in Q.)

Solution to Problem 3 Let F€ 7[x] be irreducible (in 7[x]). Then F is prime

(in Z[x]).

Proof Suppose FIGH where 0, HE 71[x]. [We must show that FIG in Z[x]
or FIH in Z[x].] Now F is irreducible in Z[x]. Thus, if deg F 1, F is primitive
and is also irreducible in Q[x] [by 1.9.15]. Anticipating the proof (see 3.7.2
(ii) and 3.7.8) that in Q[x] irreducibles are necessarily primes we see that F
is prime in Q[x]. Hence [by the definition of primeness] either FIG in Q[x]
or FIH in Q[x]. Suppose WLOG the former. Then there exists s EQ[x] such
that Fs = 0. Writing s = 5, where a, b et, (a, b) = 1 and SE Z[x] is primi-
tive, we obtain aFS bG. Equating contents [noting that FS is primitive] we
see that a = b (content of 0). Thus
in Z[x].
. t whence s c 7L[x]. This proves FIG

If deg F = 0 then F is irreducible, hence prime, in 7 and hence prime as

an element of 7[x]. [Three assertions have just been made. Do your under-
stand them? Do you believe them? Can you prove them? Cf. exercise 5.]

We now come to the first of the simpler tests for irreducibility over Q[x].
This one is generally ascribed to Eisenstein.* As implied by 1.9.8 we can do
all our working in 7[x].

Theorem 1.9.16 (Eisenstein's Test) Let F = a0+a1x + +a,,x" E 7[x]. If

there exists a prime p in 7 such that
(i) pIao,pIai,. .

*
F G M Eisenstein (16 April 1823 — ii October 1852), a pupil of Gauss. Gauss placed Eisenstein
in the top three mathematicians of all time.
 Divisibility, irreducibles and primes in Ok] 47

(ii) p24'a0,
then F is irreducible in O[x].

proof Supposing F is reducible in Q[x], we may assume [which theorem?]

that FGH where .+bfeZ[X] and H=
sX 7[v] and where r<nand s<n. We have
a0 = b0c0
a1=b0c1+b1c0 etc.
Since we know that pIb0 or plco [since p is prime], but this time not
both since p2kao. Suppose WLOG that p Now not all the b,
are divisible by p or else all the a, would be, contrary to p k Let k be the
smallest suffix for which p k bk. Then 0< k r <n. Now ak = bock +
and pjak since k <n. Also p divides all the terms b,ck_, except the last [since
p ,? bk and p cc,]. This contradiction establishes the theorem.

Example 1.9.17
(i) f=k—3x+1r2—x3+Lv4+hx5 is irreducible in O[x]. For: We writef=
Noting that 313, 3145,. ..,3k2 and
32k 3 we see that Eisenstein's theorem tells us that the polynomial in brackets
is irreducible in Z[x]. Consequently so isf, in Q[x].

The following type of polynomial plays an important role in Gauss' theory

of constructible regular polygons (Section 4.6) and in the number theory
described in Sections 3.5 and 3.9.
(ii) Let p be a prime in it Then the pth cyclotomic (circle dividing)
polynomial C = +x +x2+. +x"1 is irreducible in Q[x]. For: If C =fg in
Q[x] then C+1 =f+ig+i in O[x] where, in each case, the polynomial ha is
defined as that obtained from h by replacing x by a +x. In particular C÷,
l+(1+x)+(1+x)2+. . '.This is most easily simplified by noting
1—f
that C can be written formally as and consequently
1 —x

1—(1+x) —x

Here, each binomial coefficient displayed (except (") = i) is divisible by

p (exercise 1.4.17) and so Eisenstein's test says that is irreducible in O[x].
Ihus there are no such h, k e O[x] (apart from trivial ones) whose product is
48 Numbers and polynomials

C Hence there are no h - 1' k_1 e Q[x] (except trivial ones) whose product
is C. Thus C is irreducible, as required. [All right?]
Remarks
(i) One might try to employ this sort of manoeuvre (exercise 10(c)) on any
given polynomial one suspected of being irreducible. One trouble is that one
wouldn't know for which integers a, if any, the replacement of x by x + a
would change the given polynomial into Eisenstein form.
(ii) Eisenstein's criterion shows that there exist in Q[x] infinitely many prime
(i.e. irreducible) polynomials of each degree 1.
(iii) As we did with the integers (1.3.9) we can use induction (but this time
on the degree) to show that every non-zero non-unit polynomial in Q[x] can
be expressed as a product of a finite number of irreducible ones. As implied
in note 1.9.4(i) an analogue of 1.5.1 can be established in Q[x]. This analogue
is:

Theorem 1.9.18 (Unique Factorisation Theorem for Polynomials) Let f be

a non-zero element of Q[x]. Then either f is a unit or f can be expressed as
a product of a unit and finitely many monic* irreducible polynomials. Further,
if f = ug1g2. . . = vh1h2. . . where u, v are units and g1, g2,.. . , h1,
Ii, are monic irreducible polynomials, then u = v, r = s and the and
the can be paired off in such a manner that paired polynomials are equal.

This result will finally be established in 3.7.13.

Exercises
1 Factorise x4—4 into irreducibles in Z[x], Q[x], R[x] and C[x].
2 Write down all the divisors of 2x2+6x +4 in Z[x]. [There are 16 of them.]
3 Is 2+2x +6x2 reducible or irreducible (i) in Z[x], (ii) in Q[x], (iii) in
(iv) in C[x]?
4 Find all the associates of 4x2+2x + I in Z[x]. Find five associates of
4x2+2x+1 inQ[x].
5 Does each prime in / remain prime when considered as an element of (i)
Z[x], (ii) O[x]?

6 Prove, by induction, that every polynomial of degree I in Q[x] can be

expressed as a product of irreducible polynomials.

7 Is 286x2— 187x +442 primitive or not?

8 Represent as a product of a unit in O[x] and a primitive polynomial (ii
2
2 X X
Z[x]) (a) 3x +6x +9; (b)

* A polynomial in O[x] is rnonic ill the coefficient of its greatest power of x is +1.
 The division algorithm 49

9 State (be careful!) and prove any relationship you can find involving degf,
degg, deg(f+g), g) wheref,geQ[x].
to Determine whether or not the following polynomials are irreducible in
Q[xl.
(a) x3+2x2+4x+2; (b) x3+2x2+2x+4;
(c) x4+6x3+12x2+12x±7.[Hintlookatf 1.]
11 Use Eisenstein's method and the method of example 1.9.17(u) to prove
that x4 + 1 is irreducible over 0.
12 Show that +x°' is irreducible in 0[x], p being a
prime in 71.
13 Prove the back-to-front Eisenstein test viz:
If plan, plan_i,... ,pta0 and in 71 then is
irreducible in Q[x]. [Hint: you might care to look at +. +a0y"
where y Hence prove that 7x4+12x3+12x2+6x +1 is irreducible in
0[x] (cf. exercise 10(c)).
14 Give a counterexample to show that the following converse to Eisenstein's
test is false. (Students often seem to believe this 'converse' at examination
time!)
Let . Suppose there is no prime p in 71
satisfying conditions (i) and (ii) of 1.9.16. Then F is not irreducible in Q[x].
15 Give an example to show that the conclusion of Eisenstein's test cannot
be replaced by: '. . then F is irreducible in Z[x]'. Identify where a proof of
.

this, along the lines of 1.9.16, breaks down.

16 Let F 1) be a monic polynomial in Z[x]. Show that the monic irreduc-
ible factors of F given by 1.9.18 all lie in Z[x].
17 Given f, g E 0[x] not both zero, show that there exists in S =
(sf+ ig: s, te Q[x}} a unique monic polynomial m dividing every member of S.
[Hint: look for the unique (?) monic polynomial of smallest degree in S—if
such an object exists at all.]

1.10 The division algorithm

As implied in 1.9.4(i), we shall leave a proof of the uniqueness of factorisation
in Q[x] until Section 3.7 where we shall get it as a special case of the theory
developed there. However we do obtain here one of the ingredients needed
for that proof, mainly as an aid to getting a second irreducibility test for
polynomials in 0[x] in Section 1.11.
The analogue in 0[x] of 1.4.5 is:

Iheorem 1.10.1 (The Division Algorithm) Letf, g E Q[x] be such that g #0.
50 Numbers and polynomials

Then there exist m, r E Q[x] such that

f=mg+r where either
(i) r=0
or
(ii)

Proof
(1) If f = 0 we write 0 = Og +0 which is clearly of the required form.
(2) Suppose f=ao+aix+. and where

(a) If d <0: we write f = 0g +f which is of the required form since degf<

degg.
We deal with the remaining cases using induction (see 1.2.9) and starting
with d =0.
(b) If d =0: Then m = n and we write f = g + r. Here r has its coefficient
of xm [recall rn = n] and, trivially, of all higher powers of x, equal to 0. Thus
even if r 0 at least deg r < m = n = deg g.
(c) Assuming the result valid for d = 0, 1, 2, . . . , k — 1 we take d = k: Set
= f — 1" "g. [fi is constructed to ensure deg fi <deg f (though it might
happen by accident that fi =0) so that we can apply the induction hypothesis
to the pairf1, g.] 1ff1 =0 we as required. 1ff1 and
if we set t=degf1—degg then Hence using case 2(a) [if
<0] or case 2(b) [if it = 0] or the induction hypothesis of case 2(c) [if
0< it k — 1] we see that, in each case, there exist and r1 in Q[x] such that
f1=m1g+r1 where either (i) r1=0 or (ii) and degr1<degg. It then
follows from the definition of fi that that is f

(mi +f xm_n)g +r1 exactly as required.

We give just one easy

Example 1.10.2 Given and g =tt2+x +é

we have, by long division,

22 ifs 24 3x3 —x
313
11 2 7 15
I

2
11 4 3 2
—TX TX flX
 Roots and the remainder theorem 51

14 3 1 2 1

14 3 7 2 7

52 5 1

5 2 15 15

125 19
x +
125 19
Thusm3X 3 2 7
11 15

Note that degr= lcdegg =2.

Exercises
Withf,gasgiven,findm,r€Q[x]asin 1.10.1.
f=4x 43332
1
2 1

(a)
(b)
6
f=3x —2x 5 13473929
—4X
3
—3x2 —x+4.
2 Give an example off, g e Z[x] (g 0) such that there do not exist m, r e 7L[x]
such thatf=mg+r with either r=0 or deg r<deg g. [Thus there is in general
no division algorithm in Z[x] except when g is monic.]
/
Show that there do exist M, R Z[x] and z E such that zf = Mg + R where
either R =0 or R and degR <degg. [Hint: work in Q[x] and get rid of
fractions.]
3 Imitate the process of the Euclidean algorithm as defined for / (p. 29) to
find the unique monic gcd in Q[x] of the following pairs of polynomials:
(a) x5+2x4+x3+x2+2x+1,3x4+9x3+10x2+Sx+1.
(b) x6—6x4+12x2—8,x3—x+2.
[The proper way to define the gcd concept should be obvious. It's a word for
word copy of 1.4.1. The existence of the gcd is given by exercise 1.9. 17.]
4 The Euclidean algorithm for Q[x] can be used, as in the case of 7, to
arrive at a proof of unique factorisation. Can unique factorisation in Q[x] also
be proved by copying the second proof of 1.5.1 on p. 34? [Recall that the
elements in C[x] can be ordered; see exercise 1.7.4.] If not, identify the first
place the proof breaks down. [Look, in due course, at exercise 3.11.1.]

1.11 Roots and the remainder theorem

Towards the end of Section 1.8 we noted that the only interpretation one
could put upon the equality x2 — 3x + 2 =0 between the two polynomials
mentioned is that 1 =0, —3 = 0 and 2 = 0 in 0. Thus the above equality is
meaningful but blatantly false. It is because such equalities are therefore not
worth investigating (they are all false!) that a totally different interpretation
is usually attached to the statement — 3x + 2 = 0. This interpretation is
expected to elicit from the reader the response: 'That's easy. x = 2 and x = 1
are the only solutions to this equation.' Any feeling of insecurity in the reader's
mind concerning this dual interpretation of x2 — 3x +2 = 0 is not so much due
52 Numbers and polynomials

to the perversity of the author but is rather due to the dual role that the
public at large expects x to play. In popular language these two roles are
those of 'indeterminate' and 'variable' respectively. This is perhaps best
illustrated by considering the assertion that 'one way to show that x —2 is a
factor of x2—3x +2 is to put x = 2 in x2—3x +2 and show that the result is
0'. We can 't put x = 2 or anything else for that matter: x is the infinite
sequence (0, 1, 0,. . .). [Nor can we apparently 'substitute 2 for x' since the
result, namely 2 2®(—3, 0,0,. .) C 2+(2, 0,0,.. .), is meaningless. On the
.

other hand we can get round these difficulties, so that we can talk (loosely)
as above without fear of any difficulties arising, by doing something which
looks very much like substitution. In what follows we shall let J stand for
any one of 7, Q, or C.

Definition 1.11.1
(i) Let us, for brevity, denote the polynomial a0+a1x in J[x]
byf For each c €J we define the value of f at c to be the element a0+a1c +
•
•+ We denote it briefly by f(c) and refer to the process of obtaining
f(c) from f in this way by the ill-chosen but universally familiar expression
'substitution of c for x in f' since on the face of it we have just replaced every
occurrence of x in f by c. It is then easy to check that, for f, g e J[x] and for
c e J, we have (f g)(c) =f(c)g(c) and (1+ g)(c) =f(c) + g(c).
(ii) If it J is such that f(t) =0€ J we call it a root or zero off in J.
(iii) The expression 'Solve the equation a0+a1x + . +a,,x" = 0 inf' means:
.

'Find all elements c €f such that f(c) = 0 in J.'

Notes 1.11.2
(i) What we are actually doing in 1.11.1(i) is to associate with the sequence
(a0, a1 0,. . .)and the element c€J the elementa0+a1 c+ in J.
(ii) In the expression a0 + a ix + the a1 are abbreviated notations
for infinite sequences; in a0+a1c + the a1 are actual elements
of J.
(iii) The assertions that one can solve x2 — 3x + 2 = 0 in 7 and that one cannot
solve x2 + 1 0 in 0 should now be completely unambiguous (and both are
true).

We can now prove a result which is often stated and proved in a rather
casual manner. In order not to obliterate its intuitive meaning, and yet satisfy
ourselves that we are not falling into a subtle trap, we denote the constant
polynomial (a, 0, 0,...) in J[x] once again by a. It might be quite instructive
for the reader to select an algebra text at random and see how many, if any,
holes he can pick in the proof of the following theorem given therein.

Theorem 1.11.3 (The Remainder Theorem) Let f J[x] and let a J. Then
there exists meJ[xj such that Further x—alf in J[x]
ifff(a)=0.
 Roots and the remainder theorem 53

proof By the division algorithm* we can surely find polynomials m, re.J[x]

such that f m (x — a) + r. Further, either r = 0 [the zero polynomial] or
and degr<deg(x—a)=1. In this case degr=O and r is (again) a
constant polynomial. Thus set r = b, where b EEJ. If we now, according to
1.11.1, substitute a for x in each side of the above equality we obtain, using
1.1 1.1(i), f(a)m(a) (x—a)(a)+b(a), from which one easily sees [if one
thinks carefully what the symbols mean] that f(a) = h. Hence r = h =j?j) as
required.
We leave the second part of the theorem to exercise 1.

Remark If in 1.11.3 we adopt the notation finally settled on in Section 1.8,

that is we drop the bars, we obtain the remainder theorem in its usual guise.
Notice that then the symbol a is being asked to play two roles simultaneously,
namely as an element of J and as a polynomial in J[x] — unless (see Fig. 1.1)
one is prepared to regard I and .1 (cJ[x]) not just as isomorphic but as
being one and the same (that is, identical) so that J is being regarded as a
subset of J[x],
Now we've shown that the sloppy approach can be put on a firm footing
we shall, because of its greater familiarity, lapse into it!
There now follows immediately

Theorem 1.11.4 Let J be any one of 0, R, C and let fef[x] such that
degf=n>0.Thenfhasatmostn rootsinJ.
Proof We proceed by induction on n. 1ff has degree 1 then f has exactly
one root in I. Now suppose f is a polynomial of degree k + 1. If f has no root
in I then since 0 k + I the result holds. Otherwise suppose b is a root of f
in J. Then, by 1.11.3, x—h is a factor off in J[x]. Write f=(x—h)'g, where
is chosen as large as possible and g(eJ[x])'has degree n — t. By the
induction assumption g has at most n — t roots in J. Hence f has at most
(n—O+l

Indeed, iffef[x] has degree n and distinct roots h1

f
Note 1.11.5 If, in 1.11.4, we call t the multiplicity of h in we can go further.
b5 with multiplicities
then For example,
x5+3x4—4x3—x2—3x+4=(x—l)2(x+4)(x2+x+l)
has roots 1, —4 with multiplicities 2, 1 respectively.
A second practical irreducibility test referred to in Sections 1.9 and 1.10
follows from

Theorem 1.11.6 (The Rational Root Test) If r/s is a rational root of the
Polynomial z0+z1x+ €Z[x]where
. then
*Why can we use it? Cf. exercise 1.10.2.
54 Numbers and polynomials

Proof If r/s isarootthen + zjr/s)" = 0. Multiplyingthrough

by s" yields z0s" =0. Clearly s then divides znr'2 and
hence SIZn [since (r, s) = 1] by exercise 1.5.5. In a similar manner one proves
rlzo.

Example 1.11.7 If x4+2x3 +2x2 +x —2 has a rational root r/S with (r, s) = 1
then rI—2 and sil. Hence r/S has four possible values, namely 1, —1, 2 and
—2. Substituting each of these in turn in the given polynomial never gives the
value 0. Hence x4+2x3+2x2+x —2 has no rational root.

Remark With regard to irreducibility, 1.11.6 is chiefly of use in showing

certain cubic polynomials in Q[x] irreducible, because if a cubic polynomial
is reducible at least one of the factors must have degree 1. Hence such a cubic
polynomial has a root in 0. The theorem is not of immediate use in checking
polynomials of degree 4 or more for irreducibility. Indeed the above quartic
which has no rootinG nonetheless factorises as (x2+x +2)(x2+x —1) in 0[x].

Whilst talking about reducibility of polynomials we mention a very impor-

tant result which we shall, using just a tiny bit of analysis, prove in full later
(Theorem 4.8.1).

Theorem 1.11.8 (The Fundamental Theorem of Algebra)* Let f E C[x] be

such that degf a Thenf has a root in C. It follows thatf factorises completely
into linear factors in C[x] so that if degf = n then f has, including repeats, n
roots in C.

Applying this to R[x] we get

Theorem 1.11.9 If fE then f can be expressed as a product of poly-

nomials of degrees at most 2 in R[x].

Proof By exercise 8 below we see that if a +ib is any complex root of f,

then a — ib is also a root of /1 Thus non-real roots of f (if there are any) occur
in complex conjugate pairs. The observation that
(x —(a +ib))(x —(a —ib)) = (x —a)2 —(ib)2 = x2 —2ax +a2+b2€ R[x]
then suffices.

And as a corollary,

Corollary 1.11.10 An irreducible polynomial in R[x] is either of degree 1

or of degree 2.
*
Gauss was the first to prove this. In all he gave six different proofs.
 Roots and the remainder theorem 55

Remark Given a polynomial f in 1.11.10 restricts possible irreducible

factors to having degree 1 or 2. This knowledge is unfortunately of little help
in actually factorising f and in general we have to be content with approxi-
mation methods. Investigations along these lines naturally belong to numerical
analysis.

Exercises
1 Complete the proof of 1.1 1.3.

2 Let feQ[x] be the polynomial .+a,,x". Define the

derivative f' of f to be the polynomial a1 +2a2x . Show that
if f has a double root a (that is f has a factor of the form (x — a )2) with a €0
then a is a root of f'.
3 Use exercise 2 to showthat neitherx4—4x3+4x2+ 17 norx4+4x2—4x —3
has a repeated root in 0.
4 Using the fact (see exercise 1.9.17) that the gcd d of any two polynomials
f, g 0[x] can be expressed in the form d = sf + tg for suitable s, t
show that the polynomial x4 + 4x2 — 4x — 3 of exercise 3 can't even have a

repeated root in C.
5 Let f and g be polynomials of degree n in 0[x]. If q1,q, . are
. .

distinct rational numbers such that f(q,) =g(q) for all i (1 i n + 1) show
thatf=g in0[x].
6 Let u1,. , u,,÷1 be distinct rationals and v1
. . be rationals (not
necessarily distinct). Show that there is exactly one polynomial f€ 0[x] such
that (i) degfn and for each i
[Try definingf to be
(x —a1) . .. (x —Ui 1)(x . . (x
—u1). . . . .

This is called Lagrange's interpolation formula.]

7 Find the roots, if any, in 0 of
(a) 4x4—2x3—2x2—x+1; (b) x4—x3—2x2—2x+4;
(c) 2x4+7x3+3x2+12; (d) 8x3—6x—1.

S Let f be a polynomial in Show that if a + ib (a, b R) is a root of f

then so is a — ib. [Hint: Write out f(a + ib) in full. Since f(a + ib) = 0 we have
f(a +ib) = = 0, the bar denoting complex conjugation. Show f(a +ib) is
identical tof(a —ib).]
9 The equation x4—3x3+2x2+x+5 has 2—i as one of its roots. Find all
its complex roots.
56 Numbers and polynomials

10 Using any method you find convenient determine whether or not

x4—x2+1 andx4+x+1 are irreducible in Q[x].
11 Kronecker's method for factorisingf L[x] is as follows. 1ff = g h where
g, h E L[x], then for each z EL for which 0 we have f(z)= g(z)h(z) so
that in /. If degf= n then degg may be assumed to be at most*
[i]. [Why?] Now, finding + 1 integers z1, z2,... for which f(z1) we
determine all possible values of the g(z1). We then use exercise 6 to find all
possible polynomials g. Next one simply tries all these in turn to see if any
of them divides f in L[x]. This process is then repeated as necessary.
For example, if f = x4+x + 1 then f has no linear factor in L[x]. [Why not?]
Suppose f=gh where degg=2. Then
g is such that g(—1) is I or —1, g(0) is 1 or —1 and g(l)
is 1 or —1 or 3 or —3. The finding and testing of all possible g is left to you!

*
[x] here denotes the greatest integer not exceeding x.
2
Binary relations and binary
operations
21 introduction
If we let X stand for any one of the sets 7, Q, R, C, Z[x], etc. mentioned in
Chapter 1, then the operations of addition and multiplication defined on X
may be described as binary operations on X in that, to each pair of elements
of X, both + produce a unique entity which is again an element of X.
As remarked in the prologue, present-day algebra might be defined as the
study of (n-ary) operations on sets (n = 2, but also 0, 1, 3, 4,... etc.).* This
would, however, be a little unfortunate since only relatively few n-ary oper-
ations are either interesting or important. In later chapters we shall investigate
in depth some of those binary operations which have, by their repeated
occurrence and usage, shown their importance. (For the sake of illustration
we shall also introduce some which, to say the least, are of little significance!)
The general concept of binary operation on a set is defined in terms of that
of 'function'. No doubt the reader can give what he feels is an adequate
definition of 'function'. (Indeed we invite each reader, here and now, to make
such a definition and compare his proposal with ours, given later.) However,
in keeping with the somewhat critical approach adopted in Chapter 1, and
because we also wish to indicate how our studies might be put upon a
set-theoretical base (the soundness of which is the concern of set-theorists)
we propose definitions rather more formal than many a reader might expect.
Any reader worried by the prospect of such formality will be pleased to see
how, as with polynomials in Section 1.8, the more informal notation to which
he is more accustomed is soon restored.
The formal approach requires that we start with a study of binary relations.

22 Congruence mod n. Binary relations

We begin with some simple number theory which will help us feel our way
and also lead us to some new and fascinating algebraic structures.
Let n be a positive integer, any one you wish, but fixed once chosen. Given
two integers a and b, it is the case that either (i) nb —b or (ii) n A'a —b.
[Equivalently: a and h (i) do have or (ii) do not have the same remainder (in the
range 0 to n — I) on division by n.] In the first case we shall write ba (mod n),
in the latter a*b (mod n). These we shall read as a is (is not) congruent to b
'2-ary' and are synonymous.
58 Binary relations and binary operations

modulo [more briefly, mod]n. Thus 7 —3 (mod 5) whilst 111 40 (mod 5).
Working modulo 5, then, we see that certain pairs of integers are related (by
being congruent mod 5) whilst other pairs are not so related.
There are other more familiar ways in which certain pairs of integers appear
to be related whilst other pairs are not. For example, since 3<7, the ordered
pair* (3, 7) is related by < whereas the ordered pair (7, 3) is not. The symbol
'=' yields another relation on 7, two integers being related this time if and
only if they are equal. As an example of doubtful mathematical value consider,
on the set P of all human beings alive at, say, 10.05 a.m. (GMT) on 7 April
1978, the relationship 'is greatgrandfather of'. Still thinking aloud it appears
that the relationships (mod 5), <, =, and 'is greatgrandfather of' can be
interpreted as subsets of 7 x 7, 7 x 7, 7 X 7 and P x P respectively. For
example, <can be identified with the subset of 7 x 7 comprising all ordered
pairs (a, b) for which a <b. Similarly = can be identified with the subset
{(a, a): a 7} of 7 x 7 and a (mod 5) with the subset of all those (a, b) for
which a a b (mod 5). That these identifications are circular (equality in 7 being
identified with the set of all pairs (a, b) for which a and b are equal!) is
unimportant at the moment since we are still informally trying to feel our
way. Now, being formal, but motivated by the above, we make

Definition 2.2.1 A binary relation on a set A is a subset of A x A.

One easily solved and one unsolved problem in number theory are
included in

Examples 2.2.2
(i) ThebinaryrelationR1 ={(x, y): XE t, ye y andf = yX}isafinite
set comprising just the two elements (2, 4)and (4, 2). [Challenge: Prove it!]
(ii) The relation R2 ={(x, y): X€! ,y€ 7 ,y =X +2 and X, y both primes}.
Thus R2 contains the pairst (3, 5), (5, 7), (11, 13), ..., (1706595x2"235-i-l),
(1706595
[It is an unsolved problem of number theory as to whether or not R2 is a
finite set. Each pair {X, y} is called a pair of twin primes.]
(iii) The relation R3 = {(X, y): X P, y P and X is greatgrandfather of y}.
It seems unlikely that many readers will find themselves as first member (that
is, in the x-place) of any element of R3. (The author would certainly be pleased
to hear from any such greatgrandfather! May 1991: I'm still waiting!)

Having shown (in 2.2.1) how we can give an unambiguous definition of

binary relation in set-theoretic terms we make the notation more readable.

*
See 0.3.1.
tAs of September 1990.
 Congruence mod n Binary relations 59

Thus

Notation 2.2.3 Let R be a binary relation on a set A. If (ai,a2)€R we

write a1Ra2 and say that a1 is related to a2 under (or by) R. Otherwise
we write a1Ra2 and speak accordingly.

On any given set few (if any!) binary relations will have mathematical
significance. Indeed, those binary relations occurring most fiequently in
practice possess special properties including some or all of:*

Properties 2.2.4
(r) VxeA xRx
(s) Vx,yEA
(t) Vx,y,zeA
(a) Vx,yEA xRyandyRx4.x=y
Definition 2.2.5
If R satisfies (r) we say that R is a reflexive binary relation.
If R satisfies (s) we say that R is a symmetric binary relation.
If R satisfies (t) we say that R is a transitive binary relation.
If R satisfies (a) we say that R is an antisymmetric binary relation.
If R satisfies (r), (s) and (t) we say that R is an equivalence relation.
If R satisfies (r), (t) and (a) we say that R is an order relation.

Note 2.2.6 To remember these latter two definitions the author uses the
mnemonics rEst and rOta. You will, of course, remember your own mnemonics
all the better.

Examples 2.2.7
(i) Since for all x, y, z El we have: (r) x (mod 5); (s) If x (mod 5)

we see that (mod 5) is an equivalence relation on 1. (mod 5) does not

satisfy (a).
(ii) The relation c on 1 satisfies (r), (t) and (a) and thus is an order relation.
[This basic example motivates the use of the words 'order relation' in 2.2.5.]
(iii) The relationship 'is brother of' on the set of all male human beings
satisfies (s) but, in normal parlance, not (r). What about (t) and (a)?
(iv) The relation {(a, a), (a, b), (/,, a), (b, b), (b, c), (c, c)} on the set {a, b, c}
satisfies only (r).
(v) On the set of all triangles in the plane the relationships of similarity and
of congruence are equivalence relations.
(vi) On 1 x (l\{O}) the relation R defined by (a, b)R (c, d) ill ad = bc is an
equivalence relation of a kind which we shall meet again in the proof of 3.10.3.
V means for all. See Section 0.4.
60 Binary relations and binary operations

Problems 1
(i) Equality (on 7, say) satisfies (r), (s), (t) and (a). Is it the only such binary
relation on 7?
(ii) Does the relationship 'is greatgrandfather of' on P satisfy (a) or not? [If
not there must surely exist x, y E P such that xRy and yRx but x y?]

Exercises
1 Show that on a finite set with n elements it is possible to define distinct
binary relations.
2 Let a,b,c,d,m,ne7L with n.>O. Show that if a=—c (modn) and if b—=
d(modn) then (fl (ii) (iii)
ma (mod n) show that a ac (mod n) if n)= 1.
(m,
Give a specific example showing this conclusion can be false if (m, n) 1.
Decide what it means to say that x +3x +x +5x
5 4 3 2
3
x4 + 2x + 1 (mod x2 +x + 1) and then decide if you believe this assertion.
4 For each of the following find, if possible, at least two solutions.
(i) (ii) 4xa7(mod8); (iii)
(mod 31).
5 Which of the properties in 2.2.5 are satisfied by the following relations R
on 7?
(i) aRb itla<b; (ii) aRb (iii) aRb ifiab=O; (iv) aRb if!

6 Find binary relations R1, R2 on 7 such that R1 satisfies (s) and (t) but not
(r), whilst R2 satisfies (r) and (t) but not (s).
7 Which of (r), (s), (t), (a) does R ={(1, satisfy?
8 An equivalence relation E on A = {1, 2, 3, 4} contains the pairs (1, 1),
(1, 2), (2, 3). Find E given that not the whole of A xA.
9 Find as small a non-empty subset R of 7 x 7 as possible so that R is not
(r), not (s) and not (t).
10 What is wrong with the following 'proof' that each symmetric and
transitive binary relation on a set A is an equivalence relation?
aRb 4 bRa (by (s)), aRb and bRa 4 aRa (by (t)), therefore aRa so that
R is (r). QED.
[Hint: look at your solution to R1 in exercise 6 above.]
11 Let X be a non-empty set and the set of all its subsets. Show that
the relation c of inclusion between the elements of 9P(X) is an order relation
on eP(X). Let X = {a, b, c, d}. How many elements (i.e. ordered pairs) does
the relation c on £?P(X) have? Find u, v e £?P(X) such that neither a c v nor
V U.
 Equivalence relations and partitions 61

12 Let R1, R2 be binary relations on a set S. Define their product R1 o

by: For a, b €5, a(R1 oR2)b if there exists c €5 such that aR1c and cR2z5.
If R1, R2 are equivalence relations is R1 oR2? If R2 are order relations
R10

2.3 Equivalence relations and partitions

Equivalence relations are often more readily comprehended and also
constructed in terms of the concept of partition.

Definition 2.3.1 Let A be any non-empty set and let be any collection of
non-empty subsets of A such that (i) the union of the sets in C is A and (ii)
each distinct pair of sets in C has empty intersection. Then C is called a
partition of A.

Examples 2.3.2
(i) The subsets of even integers (include 0) and of odd integers define a
partition of 1.
(ii) The sets of all males, of all females and of all joggers do not form a
partition of the human race.
(iii) The concepts of similarity and congruence define partitions on the set
T of all triangles in the plane.

Partitions give rise in a natural way to equivalence relations because of

Theorem 2.3.3 Let A be a non-empty set and let C be a partition of A.

Define a relation R on A by setting, for a, b A, aRb if a lies in the same
element of C as b. Then R is an equivalence relation on A.

Proof
(r) Each a in A lies in some subset of A belonging to 41 Then aRa follows
immediately [since a lies in the same element as does a!].
(s) is also immediate. [If a belongs to the same element as b then b
belongs to the same element as a.] Thus aRb 4 bRa.
(t) If a belongs to the same member of C as b and if b belongs to the same
member of C as c then... [can you finish it?]. That is, aRb and bRc 4' aRc.

On the other hand, each equivalence relation (e.r. for short) gives rise
naturally to a partition.

Theorem 2.3.4 Let R be an e.r. on the non-empty set A. To each a €A

define a to be the subset {x: x €A and xRa} of A. [Thus a is the subset of
A comprising all elements x of A which are related to a under R.] Then the
set 4H={d: a €A} is a partition of A.
62 Binary relations and binary operations

Proof Let a eA. We know that aRa [why?]. Hence [immediately from the
definition] a eá. [Thus satisfies condition (i) of 2.3.1.] Next suppose a and
b are such that a n b 0. Then there exists c E d n b. It follows that cRa
and cRb and consequently that bRc [why?]. Now, for each d e b we have
dRb. But then dRb and bRc and cRa 4'dRa [Why? See exercise 8] whence
d a. Since d was any element of b we have shown that b ç a. An identical
argument* shows that a c b and the equality a = S follows.

We have thus shown that if two elements have non-trivial intersection

then they are actually equal. In other words, distinct members of have
pairwise empty intersection. In particular if y a then a = 9, even though
a and y might not be equal (cf. 2.3.7).
There is a small point you might care to ponder.

Problem 2 We've seen that every partition on a set A gives rise to an e.r.
on A and vice versa, each in a natural way. Suppose I give you a partition,
you construct the corresponding e.r. and then I construct the partition corres-
ponding to your e.r. Do I obtain the partition I began with?

Terminology 2.3.5 Given an e.r. R on a set A the members of the corres-

ponding partition of A are called the equivalence classes of R.

Examples 2.3.6
(i) The relation R defined on the plane by setting (x1, y 1)R(x2, Y2) if! x +y =
+is an e.r. The equivalence classes are the circles with centre the origin.
(ii) The relation R defined on Q[x] by setting fRg if! f(O) = g(O) is an e.r.
on Q[x]. Each class comprises all those polynomials with a particular constant
term.

We now return to our most important example, that of congruence mod n

on 7L.

Example 2.3.7 For each n (mod n) is an e.r. In particular, when

it = 5 there are five equivalence classes, namely

O={...—1O,—5,O,S,1O,...}
1={... —9,—4,1,6,11,...}
—8,—3,2,7,12,...}
3={... —7,--2,3,8,13,...}
4={... —6,—1,4,9,14,...}
*
As an old schoolteacher of mine (DTC) used to say: 'Let symmetry work for
 Z,, 63

Note that
(i) two integers a, b lie in the same (equivalence) class if a (mod 5), that
ill —b or, again, if a =b±Sk for some k EL; and that
(ii) although the notation 0, 1, 2, 3, 4 is most natural, these equivalence
classes can also be denoted, respectively, by, for instance, ¶öì,
and

Exercises
Define on Z a binary relation (other than (mod 5)) which also has exactly
5 equivalence classes. [This exercise is supposed to help substantiate the claim
in the opening sentence of Section 2.3!]
2 Define, by means of a partition, an e.r. on 1 which has, for each positive
integer n, exactly one equivalence class with n elements. Describe this
equivalence relation in the form 'xRy if. . .'.

3 Describe the equivalence classes of the e.r. F in exercise 2.2.8.

4 Define in the form '(Xi, yi)R(x2, Y2) if ...' the equivalence relations on
R whose equivalence classes are:
(i) all circles with centre (1, 1).
(ii) all squares with vertices (r, r), (—r, r), (r, —r), (—r, —r); r EDt
5 How many distinct equivalence relations can be defined on the set
{a, b, c, d, e}?
6 On C\{O} define the e.r. R1 by z1R1z2 iff* Describe the
equivalence classes geometrically. Do ihe same for R2 defined by z1R2z2 if
z11z212 = z21z112.

7 Let R be the e.r. defined on by setting (xi,yi)R(x2,y2) if X1—X2

and y 1 — Y2 are both integers. Show that in each equivalence class there is
exactly one point lying in the region <1, <1.
8 Suppose that R is a transitive relation on A. If a, b, c, d e A are such that
aRb and bRc and cRd, show that aRd.

2.4 Zn
Let n be any positive integer. Form the set of equivalence classes determined
by the e.r. (mod ii). We call this set Zn. We now show how to manufacture
very interesting 'number' systems by introducing a kind of addition and
multiplication into These number systems are not mere curiosities. Indeed
certain of them have found application in coding theory and in statistics as
well as being important in geometry and in algebra itself. ([7], [61], [26] and
4.2.10.)
Using the notation of 2.3.4 let St and I be two elements of Zr,. Thus i and
1' are equivalence classes of integers mod n. We define their 'sum' and
*
For z = x + iy, z denotes the modulus of z, that is zI =
I
+ y2.
64 Binary relations and binary operations

'product' i 0 1 by
Definitions 2.4.1

set =s+t
and

S 'Li t = S • t

Note that, whatever else, f1'I and C? are certainly elements of!,1.
Taking the specific example of n = 7 we see that 456 = 10 = 3 whilst
6 06 36 1. Information such as this is conveniently stored in the form
of addition and multiplication tables as follows.

0
00123456
1234560 00000000
0 4 6 0 4 6
1 2 3 5 1 2 3 5

2
0123456
2345601 2 0246135 1

4 4560123 4 0415263
3

5
3

5
4

6
5

0
6

1
0

2
1

3
2

4
3

5
0

0
3

5
6

3
2

1
5

6
1

4
4

2
6 6 0 1 2 3 4 5 6 0 6 5 4 3 2 1

Here the values of a Sb and a 0 b are exhibited at the intersection of the

aft row and the bth column in the appropriate table.

Notes 2.4.2
(i) Our notation i is bad (but customary). It's bad because it does not
indicate which 7,, one is looking at. However the context usually makes a
more explicit notation (such as iJ unnecessary.
(ii) Corresponding tables for any other (fairly small) positive moduli n are
easily constructed. What has perhaps escaped your attention is that our
definitions of S and S might not make sense! The problem is this: In 77, say,
we have seen that 4$6=10=3 and that 606=36= 1. Now 4 and 6 also
go under the names and for instance. Thus, according to our
definitions,
=
whilst
606=—9250—925=855625
and so, unless and are just other names for and 1 we are in
the intolerable situation that a sum or product of two classes depends not
only on which the classes are hut also on what we choose to call them!
 65

fortunately everything works out all right (2.4.3), a situation we describe by

saying that $ and 0 are well defined. Thus

Theorem 2.4.3 The $ and 0 of 2.4.1 are well defined.

proof Suppose á=ê and b=d in 1,,. Then a—c±un, b=dtvn for
suitable u, v E 1. It follows that
a +b —c +un +d+vn =c +d+(u =c
whilst
ab =(c+un)(d+vn)=cd-f-(cv+ud+unv)n =cd
as required.

Problem 3 We could have avoided the well-definedness problem as follows.

Denoting the elements of in by O, 1,. we define sum and product
. .

unambiguously by =11 and 1 09 =ij where, using 1.4.5, x +y =

m1n+r1 and xy=rn2n+r2 with m1,m2,r1,r2Ei and where Ocr1<n and
o c < n. However, one rarely gets something for nothing in mathematics
and another small difficulty now replaces the well-definedness one. Can you
see what it is? [Hint: Try proving (a 0 b) 0 5 = a 0 (b 0 5).]

We now see which of the (analogues of the) axioms Al, A2,. . , M3, M4, .

and property Z, listed in Section 1.2, hold for and 0 on in. (We leave
axioms D, P and properties C, M of Section 1.2 to exercise 11.)
For Al we take any two a, b in in and ask: Is â$b =b®á? Indeed it
is. for a +b =b +a
in i, we have 21TB = bTi in in as required. The proof of Ml is similar. Let
us offer a streamlined proof that M2 holds. We leave to the reader the
explanation as to why each of the asserted equalities holds.
For all a, b, S in we have
(a
o a 0(1;
The proof of A2 is similar.

Note 2.4.4 The rather pedantic observation that (ab)c = a(bc) is included
here only to emphasise that the associativity of 0 depends heavily on that
of in i. Otherwise such pedantry relating to i was left behind as
long ago as 1.2.5.

Since for all dE7Ln we have OEf3d=ác33O=d and lOa=aOl=a

We see that the analogues of A3 and M3 hold in in. So does A4. For, given
0e4,
66 Binary relations and binary operations

We move on to M4 and Z. A quick glance at the 0 table for 17 shows

that both hold in 17. Consider however the 0 table for 16. It is, omitting
circumfiexes just for simplicity,

012345 0012345
0 012345
123450 0 000000
012345
2 345012
1

3
234501 2
1

3
024024
030303
4 450123
5 501234 4
5
042042
054321
It is clear that O and I are again the (unique) elements required by axioms
A3 and M3. Further: (i) and O whilst 0 = = O, and there (ii)

is no element 2 €16 for which 202 = 1. Thus both M4 and Z fail in 16.

What about M4 and Z for other 4s? It is pretty clear that the reason Z
fails in 16 is that 6 is a composite integer.* It is equally clear that Z fails in
4 whenever n is composite. Further, if n =n1n2 with n1>1 and n2>1 it is
immediate that there can be no 2 €4 such that e11 02 1.
We have, therefore,

Theorem 2.4.5 Let n be a composite positive integer. Then M4 and Z both

fail in 74.

On the other hand we can prove

Theorem 2.4.6 Let p be a positive prime. Then 4 satisfies both M4 and Z.

Proof
(a) First let a 4
where a O. Then p t a in 71. [Why not?] Hence (p, a) = 1
[why?]. Consequently, by 1.4.9 there exist r, s €71 such that rp + sa = 1. There
thenfollowsl=rp+sa=sa[why?]=sOa.ThusaOs=sOa=lsothat
71,, satisfies M4.

(b) Now suppose b, e i,, are such that 6o =0. Then either b =0 [in

which case there is nothing left to prove] or b 0. In this case we can, by

part (a), find f€i, such that tOb=1. Then 0=100=i0(b0ê)
(f0Ô)Oê=lOê=ê. Thus the assumption bOê=0 leads either to
S = O or to ê = O whence property Z is seen to hold for 4.

Problem 4 You may, even at first reading, have an uneasy feeling that the
proof of (b) was not quite what you were expecting and that a more direct
proof should be available. Indeed there is such a proof. Can you find it? (The
reason for giving the present proof will appear in exercise 3.3.5.)

*
A composite integer? Any integer which is different from 0, 1 and —1 and is not a prime.
 74 67

Remarks We have just seen that in some ways the number systems 7,,,
where p is a prime, are more like Q, R and C than is 7 itself in that all of
Al, A2,.. . , M3, M4, D, Z hold in each of them whereas M4 does not hold
in 7. Indeed 74, 0, R, C are all examples of fields (see 3.2.2(10)) whereas 7
is not.

Exercises
1 Compute the ® and 0 tables for 12, 73, 74. In the tables for 72 replace
every occurrence of O by the letter F and every occurrence of I by the letter
0. Compare the result with exercise 1.2.15. Are you surprised? Should you
be?
2

3 Find the number of distinct solutions to the equation x2eTt= O in

in717 and in 78. In the light of 1.11.4 do you find the answer in the case of
78 surprising? Why or why not?
4 (a) Carry out the proof of 1.11.4 with 74, replacing every occurrence of
J. Does the argument go through? What conclusion can you draw about roots
of elements in
(b) Comment on the assertion that the element x4e3x3® 2x S7 of 171[x]
has 9 roots in 171.
S Find the number of solutions in 77 to (i) X2$X ®4 = O; (ii) x2$2x $4 =
0; (iii) x2$3x$4=O.
6 Show that in Z4[x] there is a polynomial P such that
OP = 1.
7 Find all x such that (mod 7) and all x such that (mod 7).
[Hint: For the second part, use the first part and the last but one sentence in
exercise 2.2.2.]
8 Find, if possible, an element ci such that O, a, a2, a3, d4, a5, a6 exhaust all
of 77.

9 Find an n and elements á, 6,ê in 74 such that none of ci Oh, hOe and
C 0 ci is equal to 0 and yet ci 0 b 0 C =0.
10 Find, if possible, a multiplicative inverse for in each of (i) (ii) 113;
(iii) 121; (iv) 734; (v)
11 True or false?
,— -.-——---... —,
k1) 17 has elements —3, —2, —1, 0, 1, 2, 3 and the elements 1, 2, 3 show
that 77 satisfies P of Section 1.2.
(ii) Each 74 satisfies property M and axiom D.
('ii) Each 74 satisfies property C. [Hint: try n composite.]
68 Binary relations and binary operations

12 The binary relation R on 0 is defined by setting a/bRc/d itT a/b — c/d e 7.

Show that R isan e.r. on 0. For equivalence classes 217b and define
a/h c c/d = ac/hd. Show that is not well defined.

Carl Friedrich Gauss (30 April 1777-23 February 1855)

Gauss, called by his contemporaries the Prince of Mathematicians,
was one of the greatest scientists ever. Not only did he do
stupendous work in many areas of pure mathematics but he also
devoted much time to probability, theory of errors, geodesy,
mechanics, electromagnetism, optics and even actuarial science! He
was born in Braunschweig (Brunswick) Germany, to a poor family. His
mother was intelligent but not fully literate, his father at various
times a gardener, general labourer and merchant's assistant. It is
said that, aged three, Gauss corrected an error in a wages list whilst,
aged eight (some say ten), he wrote down in moments the answer to
the following problem set in class: add together all the integers from
I to 100. (One must presume he saw the answer had to be one half
of bOx 101.) Various sources claim that before age 20 he had,
amongst other things, rediscovered and proved the law of quadratic
reciprocity, discovered the double periodicity of elliptic functions,
proved that every positive integer is a sum of 3 triangular numbers
(n+1)
(these are integers of the form n where n E formulated the
2
principle of least squares and conjectured the prime number
theorem (itself not proved until 1896). But what made him choose to
devote himself to mathematics was his finding in 1795 of the
criterion for a regular n-gon to be constructible using only compass
and straightedge. Then for his doctoral thesis (1799) he gave the first
ever proof of the Fundamental Theorem of Algebra (Section 4.8).
In 1801 he became famous in the eyes of the general public by
locating the exact position of the newly discovered (and
subsequently lost!) planet Ceres, this from very little information and
where the best astronomers had failed. In 1807 he was appointed
professor of astronomy and director of the Göttingen Observatory, a
post he held until his death. 1801 also saw the publication of his
classic book Disquisitiones Arithmeticae.
 Some deeper number-theoretic results concerning con gruences 69

As one might expect, Gauss didn't collaborate much with others.

One exception was a collaboration with Wilhelm Weber which
produced in 1833 the first operating electric telegraph. In his
unpublished notes Gauss anticipated major work of several
mathematicians. In particular he considered non-Euclidean geometry
before Lobatchevsky and Bolyai, quaternions before Hamilton, elliptic
functions before Abel and Jacobi as well as much of Cauchy's
complex variable theory. His revelations of priority were painful and,
in the case of Bolyai, nearly disastrous.
It is reported in [134] that Gauss's friends found him cold,
uncommunicative, ambitious for security and fame, and very
conservative. It is said that he tried to dissuade his sons from
studying mathematics so that the name of Gauss would remain
synonymous with excellence.
There is a story that at the 50th anniversary celebrations of the
award of Gauss' doctorate, Gauss was about to light his pipe with a
page from the original Qisquisitiones. Dirichlet, appalled by this,
grabbed the page from Gauss and treasured it for the rest of his life.

2.5 Some deeper number-theoretic results

concerning congruences
In this section we present (essentially) two number-theoretic results of which
much use is made in many areas of mathematics. The second (2.5.4) may
even surprise and delight you if you've not seen it before. We use 2.5.4 later
(see 3.8.1) and 2.5.3 to solve exercises 2 and 3 in such a way as to show the
power generated by a mere change in notation from n a — b to a b (mod n).
It was Gauss who first disclosed this power in his Disquisitiones Arithmeticae.
The notation was suggested to him because of the similarity of the properties
of the relations of congruence and equality. The reader might care to recast
all congruences in terms of divisibility and see if the results we obtain are
then so obvious and their proofs so transparent. In particular he may care to
flex his muscles by proving, without the use of congruences, that
(i) (ii) 6411225+1;
(iii) the equation x2+y2— 15z2 =7 has no solutions in integers x, y, z.

We begin by stating explicitly the famous Little Theorem of Fermat an-

nounced (but not proved) in a letter dated 1640. Euler* proved it in 1736
by the method of exercise 1.4.17 and later generalised it. Apparently Leibnizt
was also in possession of a proof around 1683.

Theorem 2.5.1 (Fermat's Little Theorem) Let p be a (positive) prime and

n a positive integer. Then — n; in Gauss' notation n" n (mod p).
*
Leonhard Euler (15 April 1707 — 18 September 1783); probably the greatest mathematician
of the 18th Century. Even total blindness during the last 17 years of his life did not stop his
tremendous output of mathematics.
*

Gottfried Wilhelm Leibniz (1 July 1646— 14 November 1716). Great mathematician and
Philosopher. Regarded as being a cofounder with Newton. his contemporary, of thc calculus.
70 Binary relations and binary operations

Note that if (n,p)=1 we may deduce, using exercise 2.22, that

n" 1 (mod p) whenever (n, p) = 1, it follows from exercise

2.2.2(u) that n" —=n (modp) whenever (n,p)= 1. Since, trivially, n"—
n (mod p) whenever pjn, we see that 2.5.1 is equivalent to

Theorem 2.5.1' Let p be a positive prime and let n be a positive integer

such that (n, p) 1 then 1 (modp).
To get Euler's generalisation we first need

Definition 2.5.2 Let m be a positive integer. We denote by 0(m) the number

of integers t such that (i) 1 t m and (ii) (t, m) = 1.

Thus qS(l)=l, Ø(2)=1, cf(3)=2, qS(4)=2, b(2520)=576 [clearly?] and

qS(p)=p—l for each prime p.
q5 is called Euler's $-function or totient.
Euler's Theorem can then be stated as:

Theorem 2.5.3 Let a and m be positive integers such that (a, m) = 1. Then
I (mod m).

Proof Let b1 (1 i f (m)) be the integers from 1 to m inclusive which are

coprime to m. Since a is coprime to m so are all the ab1 (1 c i [why?].
If now for each i we choose c1 so that c1—ab1 (modm) and
1 c, m, we see that each of the c1 is coprime to in [why?]. Further, if for
we have CI=Ck, then (modm). But then
bk (mod in) [why?] and hence b1 = bk [why?]. It follows that if bk then
c1 Ck. Since there are qS (in) distinct b1s there are qS (in) distinct and so the
are just the b1s in some order.

Finally from all the —c1 (mod in) (1 i we get on multiplying

these congruences together,
ab1ab2. . . (mod in)
Since the b1s are the c1s in some order and since the are coprime to m,
we may cancel each with the unique c1 equal to it to yield
(modm)
thatis (modm)
Yet another proof accredited to Euler is found in exercise 9.
In 1770 the English mathematician Edward Waring* published his
Meditationes Algebraicae. In it he announced, without proof, several new
*Edwardwaring)b. 1736)?)-d. 15 August 1798).
 Some deeper number-theoretic results concerning congruences 71

results. One, which asserts that to each s E there corresponds k (s) 7 such
that each positive integer is expressible as a sum of at most k (s) positive integral
5th powers, was first proved by Hilbert (in 1909!). There is a nice article on this in
the Amerkan Mathematical Monthly, Vol. 78, 1971, pp. 10—36. We mentioned
in Section 0.4 that Lagrange proved k (2) =4 in 1770. In 1771 he proved another
of Waring's assertions suggested to Waring by one of his students, John (later, Sir
John) Wilson.*

Theorem 2.5.4 (Wilson's Theorem) Let p be a positive prime. Then

To make the proof easier to follow we first prove it in the case p = ii. In
we have

and, of course,

10=—i
Hence

151=1 .
. . .
. P
It follows that (mod 11) as required.

The method of proof in the general case just copies this.

Proof of 2.5.4 The cases p = 2, 3 are easily dealt with directly. So we may
assume p 5. We consider the p —3 elements of the setS = {2, 3,. . , .

and we prove
(i) if deS then there exists 6€s such that andt db = i, and
(ii) if d6 = = I then 6 = That is, the element b of (i) is unique.
We may then deduce that the elements of S resolve themselves into
pairs, whose product in each case is 1. And since p — 1 = —i in 4 we shall
be finished.
So let us prove (i) and (ii).
(i) Given ii €5 we know from 2.4.6(a) that there exists b 4 such that
db = 1. Clearly 6 I and 6 [why not?] and so 6 ES. Further b a. For
SUppose it were. We should then have âá = 1, that is a2 I (mod p) or, again,
pIa2—i. Thus pla+1 or pla—i in 7. Recalling that we
see that the alleged divisions are impossible. This contradiction ensures b ci.

*
John Wilson (6 Augusi 174 1—18 October 1793).
+tFor brevity we omit the multiplication sign 0.
72 Binary relations and binary operations

(iij From ag=ae we get Since (a,p)=1 exercise 2.2.2

tells us that b c (mod p). In other words b ê in 1,,.
Once again it seems that Leibniz knew this result 100 years before.
For an enjoyable journey through elementary number theory and its history
the reader may consult [43]—or, perhaps, even [42]!

Exercises
1 (i) Find x such that 360=29rn+x where [Hint: 328_9
(mod 29).]
(ii) Find x such that cl8 and xm2'°° (mod 19). [Hint: 2100=
264 . 232. 2k.]

2 Show that, mod that

228 = 1+232.Thus 641[,225÷ 1. (This example, found by Euler,*
shows Fermat's belief that F(n) = 2 + 1 is prime for each n E t u {0} is false.
Fermat knew F(n) to be prime for n = 0, 1,2, 3,4. No other prime F(n) has
ever been found.)
3 Show that 1 (mod 11) and that 1 (mod 31). Deduce that
1 (mod 11), that I (mod 31) and hence that 1 (mod 341). (This
provides a counterexample to the supposed claim of ancient Chinese
mathematicians that n is prime if and only if n 12" —2. On the other hand no
smaller n provides a counterexample.)
4 Show that if n is an odd integer then n2 I (mod 8). Deduce that the
sum of 3 (integer) squares cannot be congruent to 7 (mod 8). Show that if
x2+y2—15z2=7 then Deduce that the equation
x2+y2—15z2=7 has no integer solution.
5 Find the multiplicative inverses 4, 5 and 6 in 113. Hence verify
Wilson's Theorem for p = 13.
6 Show that if n 2 is an integer and if n 1(n — 1)! + I then n is prime. (This
converse of Wilson's Theorem is useless as a means of finding large prime
numbers.)
7 We know from 2.5.1' that, for all a such that pta, (modp).
Show that if d is the smallest positive integer such that a" 1 (mod p) then
dlp—1. [Hint: write p—1=md+r where and note that a'—
(modp).]
8 Let p be a positive prime. Show that 44p5 = pr '(p — 1). [Hint: count!]
It can be proved that if m, n e 1 + and if (m, n) = I then Ø(mn) = 4(m)4(n).
*
How on earth did Euler find this divisor? In fact he knew a theorem that told him that each
prime divisor of 225+ must be of the form 64k + I (see [42, p. 851).
1
 Functions 73

Use this to prove that if t = .n then = t(1

\
—
PiJ\
I
il PS!
Hence show that fr(2520) = 576.
P2

9 Prove Euler's Theorem directly from Fermat's as follows.

Let m and let (a, in) = 1. Let p0 be any of the factors
.

1 +h2p2. Prove by induction on that l)m 1 (mod pa). Deduce that

1 (mod for each i (1 Then deduce that 1 (mod m).

2.6 Functions
We have so far met several examples of what the reader would probably
regard as a 'function', but we have as yet given no definition, formal or
otherwise, of the concept. The reader may feel he can get along quite nicely
without any definition! If so, we should first remind him that we shall be
interested here in functions other than those between two sets of numbers.
Further, he might find it interesting (and salutory) to read (see, for example,
[80], [131]) how such great mathematicians as Euler, d'Alembert* and Daniel
Bernoullit came, around 1750, to arguing about their respective solutions
to the 'vibrating string' problem essentially because their ideas as to
what constituted a function did not coincide. In 1755 Euler wrote: 'If some
quantities depend upon others in such a way as to undergo change when the
latter are changed then the former are called functions of the latter.' For
Euler the quantities were numbers: today we have need to discuss functions
between more general sets.
The study of functions was originally an offshoot of the study of properties
of curves, geometrically defined, so it is interesting to see how a modern
definition of function, expressed in terms of the set concept, is equivalent to
what you would, in the case of a real valued function of onereal variable,
naturally think of as the graph of the function.
One definition of function often found in present-day texts is: 'A function
is a rule which associates with each element of some set A a single element
of a second set B.'
You might care to discuss with your friends:
(i) Is the word 'rule' any more self-explanatory than the word 'function'?
(ii) Can different rules lead to the same function?
We thus eschew the word 'rule' and opt for a moret precise definition of
the function concept.
*
le Rond d'Alembert (17 November 1717 - 29 October 1783).
Jean
Daniel Bernoulli (8 February 1700— 17 March 1782). Son of John Bernoulli; nephew of James
Bernoulli. The Bernoullis were probably the most illustrious family mathematics has produced.
t A somewhat stricter definiton of 'function' can be found on p. 5 of L21, vol. II.
74 Binary relations and binary operations

Definition 2.6.1 A function (also called map, mapping or transformation)

from a non-empty set A to a non-empty set B is a non-empty subset f of
A x B such that for each a e A there exists exactly one b B for which (a, e f.

Remarks
(i) The words map, mapping and transformation reflect the geometric origin
of the function concept.
(ii) The words map and mapping seem to have gained greater favour in
algebra than the word function; from Chapter 3 on we shall almost invariably
use one of these two words.

Examples 2.6.2
(i) The set {(x, x2+ 1): XE R} is a function from R to It
(ii) The set {(x, X2+ 1): x E is also a function from R to
(iii) The set {(w, first letter of w): w E W}, where W is the set of all words
in the English language, is a function from W to the set L of all letters of
the alphabet.
(iv) The set {(X, x3): x E R} is a function from R to R.
(v) The set {(a, 4), (b, 2)} is a function from {a, b} to {1, 2,3,4, 5}.
(vi) If J is any one of 1, Q, R, C, 4, i[x] etc., addition and multiplication
are functions from J x J to J.
(vii) {(a, 4), (b, 2)} is not a function from {a, b, c} to {1, 2,3,4, 5}. [Why not?]
(viii) {(a, 4), (b, 2), (a, 1)} is not a function from {a, b} to {1, 2, 3, 4, 5}. [Why
not?]

Several technicalities arising from 2.6.1 will be needed throughout the text.
We do not try to motivate them but simply gather them together for ease of
reference.

Notations 2.6.3
(i) If f g A x B is a function from A to B we emphasise this by writing

f:A->B orA LB.

(ii) Given (a, b) Ef the uniquely determined b is most frequently denoted
by either f(a) or af.

Notes 2.6.4
(i) Which of the notations introduced in 2.6.3(u) is used is merely a matter
of personal preference.* On the whole, algebraists seem to have a slight
preference for the latter whilst analysts tend to prefer the former, perhaps
because of tradition. The latter has an advantage when discussing compositions
of functions (see 2.6.7) but as both are in common use we shall encourage
the reader to use both by using both ourselves! We invite the reader, in his
*
In 1.2.7 we have a function symbol (namely I I) which straddles its in 2.7.3 we
introduce function symbols which bisect their arguments!
 Functions 75

verbal communications, to refer to the symbols f(a) and af as the value off
at a or, simply, f of a and the image of a under f respectively.
(ii) It then becomes rather palatable to describe f as the function given by
a b or again as the function given by af = b, for all a e A.
(iii) 2.6.3(u) emphasises the difference between the symbol f denoting the
function and the symbol f(x) denoting the value of the function f at the
elementx of A.

Definitions 2.6.5 Let f A -* B be given. A is called the domain off If S c A

we write Sf for the set {sf: s E S} and call this set the image of S under f. The
subset Af of B is often called the range of f (Note that Af B is permitted.)
If Af B we frequently say that f is onto B. If we don't know (or if we know
but don't particularly care) that f is onto B, we often just say that f maps A
into B. (Thus 'into' is just another word for 'to'. Please note that 'into' does
not mean 'not onto'.)
The mapf :A -*B is one to one (briefly 1—1) if from (a1, b)ef and (a2, b)ef
it follows that a1 = a2. That is, f is 1—1 if distinct elements of A map to
distinct elements of B. In the contrary case where there exist a1, a2 e A such
that a1 and yet aif= a2f we sometimes sayf is many—one.
Two functions f: A -. B and g: C -* D are equal if the sets f and g of
ordered pairs coincide. That is, f = g if A = C and af = ag for each a E A
(= C). Note that f= g is possible even if B D.
Let f:A ->B and let S be any non-empty subset of A. The subset fs
{(s, sf): s e S} is a subset of f and is a function from S to B. (Informally fs is
just f except that the action of f on elements of A lying outside S is ignored.)
It is called the restriction of f to S and is commonly denoted by f IS. We shall,
incorrectly, write fs as f. No confusion will result.
If b0 e B is such that af = b0 for all elements a in A we say that f is a
constant function. If A c B and if f : A -÷ B is such that af = a for all a in A
then f is called the inclusion function of A in B. 1ff : A A is such that af = a
for all a A then f is called the identity function on A.
Again letf: A -. B. If T B we write for the set {s: sf E T}. [Remember
this definition by 'equating' Sf T with s in your mind.]
T is 1—1 and onto B then, for each
b B, {b}f' comprises a singleton {a}, say, where a is the unique element
of A such that af=b. The subset {(af, a): a €A} of B xA is a 1—1 function
from B onto A called the inverse of f. We denote it (there is only one such
inverse; see exercise 12) by f1.
If f: A -. B is 1—1 and onto B, f is sometimes called a 1-.t correspondence.
A i—i correspondence f: A A is called a permutation on A.

Note 2.6.6 The terms surjection and injection are sometimes used to describe
maps which are respectively onto and 1—1. If f : A -* B is both 1—1 and onto,
f is then called a bijection.
76 Binary relations and binary operations

All examples relating to 2.6.5 are relegated to the exercises.

Now suppose f : A -÷ B and g B -÷ C are functions. We define a new function
from A to C which we denote by f o g to indicate its dependence on f and g
by: for all aeA, a(fog)=(af)g, the image of af under g. This may be
represented pictorially by Fig. 2.1.
I

Fig. 2.1

We have

Definition 2.6.7 The function f ° g is called the composite of f and g. The

operation o combining f and g in this way is called composition.

Example 2.6.8 Let -* R and g : R -* Fl be the functions given by

x x cosx for all x. Then

x(fog)=(xf)g =(x3+1)g =cos(x3+1)

whilst
x(g of)=(xg)f=(cosx)f=(cosx)3+1
Thus +1. It follows that fog of since, in
particular, of).

Note 2.6.9 We now see one advantage of placing the function symbol to
the right of the element it operates on, namely that the combined effect of f
followed by g is naturally denoted by f o g. Putting the function symbol on
the left leads to writing g(f(a)) as (g o f)(a) so that g of denotes, once again,
the action of f followed by g. This 'backwards' notation seems somewhat
unnatural when dealing with permutations (see 5.3.6).

Exercises
1 Which of the following subsets of R x are functions?
(i) {(x,y):x+y=5}, (ii) {(x,y):x2+y=5};
(iii) {(x,y):x+y 2=5}; (iv) {(x,y):x +y
. 2 2
5}.
 Functions 77

2 (a) Let A {1, 2, 3}, B {a, b}. Which of the following subsets of A x B
are functions?
(i) {(3,a),(1,b),(2,b)}; (ii) {(1,a),(2,a),(2,b),(3,a)};
(iii) {(1, a), (2, a), (3, a)}.
(b) How many functions can be defined from A to B? From B to A?

3 Let h=
{(x, x ): x €1 }c 1 xl . Determine which of the functions f, g, h are 1—1
and which are onto. What are the ranges of f, g and h?

4
and h:R-÷R given by xh =x if x is a rational number, xh = 1—x if x is
irrational. Which of the functions f, g, h are 1—1, and which are onto?
[Hint for g : xg = (x — + 3 so its graph (over R) is not unlike that of
•1

5 Given f : A B some authors call B the codomain of f. Explain why,

according to our definition, the use of the word 'the' is inappropriate.
6 Letf:7L3-*Z3begivenbyxf=x4+x2+I,xg=2x2—2.Showthatf=g.
(Functions defined by polynomials are called polynomial functions.)
Suppose f, g E Q[x] with f g. Prove that the corresponding polynomial
functions are distinct. [Hint: Use exercise 1.11.5.]
7 Let be given by Find 3f, rf, {O,7}f, 3f1,
and {O, 7}f'.
8 Let f : A -* B, let 5, T be subsets of A, and U, V be subsets of B. Prove
(let's change notation!): (i) f(S u T) =f(5) uf(T); (ii) f(S n T) cf(S) nf(T);
(iii) r1(U u V) =f1(U) uf1(V); (iv) f'(U n V)
Show f1(f(S)) 25. = U? Give a specific example to show that
c may not be replaced by = in (ii).
9 Letf:A-*B,g:B-*Cbefunctions.
(i) Prove that f o g as defined above is indeed a function. Show that f ° g =
{(a, c): there exists b e B such that (a, b) €f and (b, c) g}. (ii) Show that if
f and g are 1—1 (respectively onto) then f o g is 1—i (respectively onto).
10 Let and g={(x,x2)}cRxR. Show that gof=
{(x,2x2+1)}cRxR.
11 Letf:A-.*B be 1—landonto.Showthatf1 ={(af,a): a €A}isafunction
from B to A and thatf of1 = 1A andf' of = where 1A, 1B are the identity
functions on A and B respectively.
12 Letf:A-*B and g:B->A be functions. Show that iffog = 14 thenf is
1—i and g is onto. Deduce that if, also, g of= then f is 1—1 and onto B
and g =f1 is the unique inverse off.
13 Let A be a finite set. Show thatf:A -*A is 1—1 ifff is onto.
78 Binary relations and binary operations

14 For x E R let [x] be the greatest integer less than or equal to x. Let
f ={(z, 2z)} and g ={(z, [z/2])}c 7 x 7L. Showf ° g = 1, but that g of$ 1,. Can
you find a similar example in which 7 is replaced by a finite set?
15 It is clear that two finite sets A and B have the same number of elements
if! there exists a bijection f A -* B. Taking this over to infinite sets we say
sets A and B are equinumerous if there is a bijection g A -* B. Show that
the sets 7 and 7 = {x x E 7 } are equinumerous. (This was observed by
+ (2) 2 +

Galileo, 1564—1642.)
Any set equinumerous with is called countable. It can be shown that U is
and R is not countable (see [5, pp. 359—61]). Show that 7 is countable.
Show that and the interval Ocx <1 are equinumerous.
Hint:*

2.7 Binary operations

At last we arrive at

Definition 2.7.1 Let A be a non-empty set. A mapping is

called a binary operation on A. If S c A and if (S x S)p c S we say that S is
closed under (or, with respect to) p.

Examples 2.7.2
(i) Addition, multiplication and subtraction on each of 7, 0, R, C, 7[x],
0[x], 4, etc. are binary operations. Division is not a binary operation on
any of these sets. [Why not?] Regarding as subsets of C, each is
closed under the addition and multiplication on C but none is closed under
subtraction.

Notation 2.7.3
(i) According to 2.7.1 and 2.6.3(u) we should find ourselves writing
(3, 5)+ = 8 instead of the more usual 3+5 = 8. Since, due to familiarity, we
are more at home with the second notation than the first we shall adopt it
for all binary operations. That is, given p as in 2.7.1 we shall, from now on,
write apb rather than (a, b )p.
(ii) If, as in 2.7.1, S is closed under p it follows that pISxS is a binary
operation on S. For psychological reasons and to stop pedantry getting out
of hand, we replace p5 x S by the technically incorrect but intuitively more
*
A diagram is worth ten thousand words (AW, Physics master).
 Binary operations 79

palatablesymbol p and speak accordingly of p as being a binary operation

on S. Thus for s and t in S we write spt whether s and t are regarded as
orinA.

Examples 2.7.2
(ii) On the set V of all vectors in 3-dimensional space define p by: for all
Vt, V2E V, V1PV2 = V1A v2, the vector product of v1 and v2.
(iii)

On the set of all functions from R to IR, define f =fl®f2. (Here

f' is that function from to defined by x (fi ef2) = xfi + xf2 for all x E Fl
(v) On F, the set of all functions from the non-empty set X into itself,
define p byfipf2 =fi °f2 (as in 2.6.7).
(vi) Onldefinep byxpy=x+y—7.•
The following chapters of this book deal with a variety of sets upon which
are defined either one or two binary operations. In the most important cases
these binary operations will satisfy several of the analogues of the axioms
postulated for 1 in Section 1.2. To ease the reader into the following chapters
but, more important, to establish 2.7.6 and 2.7.7 we introduce, for general
binary operations,

Definition 2.7.4 Let p be a binary operation on a set A. If for all a, b eA

apb = bpa we say that p is commutative. If for all a, b, c A we have (apb)pc =
ap (bpc)we say that p is associative. If there exists in A an element e, say,
such that ape = epa = a for all a A we say e is an identity (or neutral)
element for p.

Examples 2.7.5 In examples 2.7.2, parts (ii) to (vi), p is commutative in

parts (iv) and (vi), is associative in (iv), (v) and (vi) and possesses identity
elements in (iv), (v) and (vi).
For example in (vi):
(a) xpy =x+y—7=y+x—7=ypx;
(b) (xpy)pz =(x+y—7)pz =(x+y—7)+z—7
whereas xp(ypz) xp(y +z —7) = x +(y + z
Finally
(c) xp7=x+7—7=x=7px.
(Thus is 7 an identity element for p.)

Much more important is the verification that o (in 2.7.2(v)) is associative.

Theorem 2.7.6 Composition of functions, when defined, is associative.

80 Binary relations and binary operations

Proof Letf:X—÷Y, g: Y—+Z, h:Z—+T so thatfog, go h etc. all exist. Thenfog

is a function from X to Z and so (fog) o h is a function from X to T. Similarly
fo (g o h) is a function from X to T. Directly from the definition one finds, for
all xeX,
x{(fog)oh}=(x(fog))h =((xf)g)h.
In a like manner x{fo(goh)}=(xf)(goh)=((xf)g)h. Thus (fog)oh and
fo(goh) act identically on all of X. Hence (fog)oh=fo(goh) by the
definition (see 2.6.5) of equality of functions.
That o of 2.7.2(v) is associative follows on setting X = Y = Z = T.

Next a remark about associative binary operations. These comments have

already been made in exercise 1.2.1 in connection with the binary operation
+ on 7.
Let p be a binary operation on a set A. For each ordered triple (a1, a2, a3)
of elements of A we can evaluate their product in that order either as (a ipa2)pa3
or aip(a2pa3). If p is associative then, by definition, these two evaluations
are equal and we can unambiguously denote the result by a1pa2pa3. In the
case of four elements there are five ways of evaluating their product, all of
which yield the same result (copy the argument of exercise 1.2.1). We now
prove generally

Theorem 2.7.7 (The Generalised Associative Law) If p is an associative

binary operation on the set A then every way of evaluating the product of
the ordered n-tuple of elements a1,a2,. , of A taken in that order leads
. .

to the same result.

Proof * We proceed by induction on n starting with the trivial case n = 3.
Now suppose the desired result proved for all n such that 3 n k —1
and let a1, a2,.. , a,, be elements of A. The final step in any evaluation of
.

a k-fold product is the evaluation of a product of the form

(a1pa2p. . .pa1)p(a1÷1p.. .pak) for some i such that (see
exercise 5). Because of the induction hypothesis, the products inside the
brackets are unambiguous. Thus to prove the theorem we only need to show
that for each i, j such that 1 i cj k — 1 we have

(atp. .pa1)p(a÷ip.
. . . .pa1)p(a,+ip. .pa,,)
. (2.7.8)

Ifi=fthereisnoproblem.Ificfwesetx=a1p...pa1,y=a1+1p...pa1and
z = a1÷1p.. . Now x, y, z are unambiguous by induction hypothesis. Thus
2.7.8 is valid provided that xp(ypz) = (xpy)pz. But this follows from the
associativity of p, and the proof is complete.
*
At the back of this proof is another logical subtlety. See the remarks concerning the definition
of the Fibonacci sequence in Section 1.2 tp. 16).
 Binary operations 81

Thuswe have shown thatforeach n theproductofn elementsa1, a2,...,

can be defined unambiguously and can be denoted by a ipa2pa3. . without .

brackets
Finally we establish the so-called laws of exponents for positive integral
powers. Writing a2, a3, etc. instead of apa, apapa, etc., 2.7.8 leads to

Theorem 2.7.9 Let p be an associative binary operation on a set A and let

a EA. Setting a" to be the (unambiguously defined) n-fold product
apap...pa

and

Notes 2.7.10
(i) Of course, when A = /
and p is multiplication, the above merely repeats
results we've known and used for a long time. The same is true when p is
taken as addition except that, of course, a" is then written na.
(ii)2.7.9 will in general fail if the hypothesis of associativity is withheld. An
example, already known to you from examples 2.7.2, is given in the following
exercises.

Exercises
I Show that the subset of odd integers of 7 is closed under multiplication
but not under addition. Can you find a subset of 7 which is closed under
addition but not under multiplication?
2 Confirm the assertions made in 2.7.5.
3 With p as in 2.7.2(iii), show that (xpx)px =xp(xpx) iffx = 1 orx =2.
4 With p as in 2.7.2(v), show that p is commutative if X has exactly one
element.
5 Check the first assertion made in 2.7.7 about multiple products by
removing all but the last two pairs of brackets from
((a ip(a2pa3))p(a4pas))p(a6p(a7pas)).
6 Show by giving actual concrete examples that subtraction is neither
commutative nor associative on 7.
7 For each of the following binary operations on say whether or not it
is associative, commutative, and whether or not it has an identity element.
(i) a *b=2a+b2;(ii) a *b=thelargerofa andb; (iii) a *b=thesmaller
ofaandb;(jv)a * b= 1;(v)a * ba;(vi)a * b=Ia—bI+1;(vii)a * ba2b.
82 Binary relations and binary operations

8 The following table defines part of the binary operation r on the set
{a, b, c, d}. Given that r is commutative complete the table. [Hint: arc = a so
cra =7]
rabc d
abbad
b 7cc c
cd?77db
? ? a
9 Show that there are, on a set containing n elements, precisely binary
operations. How many of them are commutative operations?
10 Let S be the set {a, N, x} and let the operation * on S be given by the
table below. Determine whether or not * is commutative, associative and
whether or not it has a neutral element.
* a Nx
a a Nx
Nx a N
x Nx a
11 £ is a binary operation on the set W. Given that £ is commutative is it
true that for all a1, a2, a3, a4 in W? A
proof or a counterexample is required.
12 On /xZ define & by (a,b)&(c,d)=(ac,bc+d). Check & for
associativity, commutativity and identity element.
3
introduction to rings

3.1 Introduction
In the Prologue we indicated some of the advantages of employing the
abstract-axiomatic method in attempting to gain anything more than a
superficial understanding of the many concrete examples (of numbers, of
polynomials, etc.) which arise in an algebraic setting. A famous application
of this method is found in the long and important paper The Algebraic Theory
of Fields written by E Steinitz in 1910. In this paper Steinitz, beginning with
an abstract definition of the concept of field,* attempted to bring some order
to the multitude of concrete fields previously studied and set himself the task
of finding all possible types of fields and the relationships between them.
Aside from 0, C and all the 74, the known fields included the algebraic
number fields of Dedekind and the rational function fields of Kronecker
(exercises 3.2.14 and 3.2.15), as well as the algebraic function fields of
Dedekind and Weber and the fields of p-adic numbers of Hensel which we
shall not consider here.
We observed in Section 2.4 that the set 1 of all integers does not qualify
as a field and there are many other concrete examples, each equipped with
two binary operations of types akin to + and on 7 which also fail for one
reason or another to be fields. Amongst these are the sets Z[x], Q[x], etc., of
polynomials, all the Lm (m composite), Dirichlet's algebraic integers (exercise
3.2.14), the quaternions of Hamilton and the matrices of Cayley (Section
3.2), the algebras of Lie (exercise 3.2.3), Hensel's p-adic integers and the
more general algebras of Grassmann and Peirce mentioned in the Prologue.
Motivated by Hensel's and by Steinitz' work, A A Fraenkel in 1914
inaugurated a general investigation of the abstract structure underlying several
of these examples. Such a system he called a ring,* following Hilbert's use of
the term Zahfring (number ring) for sets of the form Z[E] = {a + bE: a, b E L}
where E2 +AE+B = 0 with A, BE 7. (Since = —AE—B we see that
'closes up' (like a circle!) under multiplication.)
We shall begin our axiomatic study with a look at rings. After the remarks
at the beginning of Section 2.6, it should not surprise you that little in
mathematics is of totally permanent nature; even definitions are susceptible
to change! Thus the notion of ring as currently defined is more general than
that considered by Fraenkel. Since, however, technically speaking, the very
*
Definition 3.2.2.
84 Introduction to rings

important Lie algebras do not qualify as rings (more accurately they are called
non-associative rings) it may only be a matter of time before the present
definition of ring is widened to incorporate them too.
To summarise briefly some features of this chapter: we begin with a formal
definition of the concept of a ring and of several of its derivatives, including
that of field. A long list of concrete examples indicates, in the spirit popular
in the period 1900—1910, the independence of the axioms listed; although
the real reason for giving this list is to show the reader that our theory does
cover a multitude of essentially different concrete examples. In Section 3.3 we
see first how a single abstract theorem can both summarise and verify a host
of facts already intuitively known (3.3.1); and at a different level of sophistica-
tion supply an incisiveness hard to achieve by considering concrete examples
alone (3.3.4). Later we shall see how the process of abstraction, by eliminating
inessentials, suggests new concrete results which might otherwise have been
overlooked, hidden by a mass of detail. We then apply these results to establish
three assertions of the great number theorist P de Fermat. Finally, in Section
3.12, we shall see how the abstract method encourages us to ask the question:
'In what sense, if any, are the integers unique?' and how it helps us to supply
a precise answer.

3.2 The abstract definition of a ring

As implied above, the set 1 of integers is one of the tmodels' for the abstract
theory of rings. It is therefore no surprise that we shall make use of many of
the analogues of the axioms listed in Section 1.2. It is apposite at this point
essentially to repeat that list if for no other reason than to emphasise, by
making a couple of notational changes, that we shall no longer necessarily be
dealing with numbers or polynomials. Thus, in the remainder of this chapter,
we shall be interested in sets of elements, often of an unspecified nature,
upon which are defined two binary operations denoted (in deference to our
chief example 7 but also for want of anything better!) by + and and called
(for the same reasons) addition and multiplication. For any particular set 5,
say, the operations + and on S may satisfy none, some or all of

3.2.1 The Axioms For any three elements a, b, c e 5, distinct or not,

Al a+b=b+a Ml a
A2(a+b)+c=a+(b+c)
A3 BzeSsuchthat M3 BeES such that

z+a=a+z=a
e To each a a z
Ba*EESsuchthata+a*=a*+a=z
 The abstract definition of a ring 85

D a (b+c)=a and (a +b).c=a 'c+b

z Ifa bzthena=z orb=z (orboth)
Remarks The axioms often go under the same names as in Section 1 2. For
instance, M2 is the associative law of multiplication. For brevity, however,
we use the words zero and unityt (rather than additive •and multiplicative
identity) to describe the elements z and e respectively. Any element s of S
(including s z) for which there exists a non-zero element t of S such that
= z or t s = z is called a zero divisor. Some familiar systems satisfy all
the above axioms, others only a few. Later, in Section 3.12, we shall consider
analogues of axioms P and I.
For the present we make the following

Definitions 3.2.2 Let R be any non-empty set equipped with binary oper-
ations +

an d s atisfy the triple (H, +, )

A4 M2 D Ml M3 M4 Z iscalleda
1 SI SI SI SI SI

SI SI SI SI SI SI — — — commutativering
3 SI SI SI SI SI SI' — SI — — ringwithunity
4 SI SI SI SI SI SI — — — SI

5 SI SI SI SI SI SI SI SI — — ringwith unity
6 SI SI SI SI SI SI SI — — SI commutative ring with no zero divisors
7 SI SI SI SI SI SI I
— SI — SI ring with unityand nozerodivisors
8 SI SI SI SI SI SI SI SI

SI SI SI SI

SI SI SI SI SI SI SI SI SI Field*

Many concrete examples are given in 3.2.7 and in the exercises below.

Notes 3.2.3
(i) A ring, then, is a triple comprising a set R and two binary operations +
and . satisfying at least the six axioms indicated. Frequently one 'forgets' the
+ and . and talks of the ring R. This is bad since R is only the underlying
set. Further, R could well be the underlying set for two different rings (exercise
2). However this laziness is customary and rarely leads to confusion.
(ii) We have temporarily replaced the symbols 0, 1, —a, a -1, of Section 1.2
by z (for zero), e (for einheit—a German word), a* and a' initially to help
us avoid unconsciously attributing unproven properties, based on our intuitive
feelings about the numbers' 0, 1, —a, and a -1, to corresponding elements of
a general ring (exercise 3.3.9(a)). On the other hand these changes should
not be overdone (exercise 3.3.9(b)). After 3.3.2 we shall return to using 0,
Do not confuse the words unity and unit (see 1.3.5(i) and 3.6.1(iii)).
+ Meaning: no zero divisors apart from z itself.
§See 3.2.3 (iv).
86 Introduction to rings

I, —a, a1 for all rings. Theorems 3.3.1 and 3.3.2 should encourage us to use
our intuition, but with care.
(iii) Any ring R satisfying M4 (respectively A4) must of necessity satisfy
M3 (A3). It is less obvious that if R satisfies M4 then it must of necessity
satisfy Z (exercise 3.3.5). These two remarks explain why the above table
contains only 10 rows and not
(iv) In any ringR satisfying A3 and M3 we usually insist that e z. Otherwise
R is a ring with only one element (exercise 3.3.4). In particular, integral
domains, division rings and fields are always assumed to have at least two
elements.
(v) Despite the examples listed in Section 3.1 we do not insist that a ring
must satisfy M3 (see 3.2.7(v)).
(vi)Why are the axioms Al, A2, A3, A4, M2, D taken as basic? Why not
some other set? The answer is that a large enough number of important
examples satisfy each of these axioms. The somewhat 'smaller' number of
examples satisfying all ten axioms are of sufficient importance and richness
for a similar independent study to be worthwhile. At the other extreme the
theory of systems satisfying Al and M3 (and possibly no more) would be
very wide ranging but would include few more, if any, important examples
than does the more restricted but structurally richer class of all rings.
As a matter of fact, at this very moment a theory of those systems satisfying
A2, A3, A4, M2 and one of the axioms in D (systems which go by the name
near-rings) is in the process of establishing itself as worthy of independent
study. See [30] and [33], for example.
(vii) The reader will perhaps have noticed that in A3, M3, A4, M4 the word
'unique' has been dropped (cf. Section 1.2). In fact each uniqueness is given
to us, gratis, with the compliments of the other axioms (Theorem 3.3.1).

Relationship 3.2.4 The relationship between the 10 types of rings listed in

3.2.2 can be summarised, without further explanation, by Fig. 3.1.

Fig. 3.1

Using 10 concrete examples we now show that none of these arrows is

reversible. Perhaps the most interesting is the demonstration that there exist
division rings which arc not fields; that is 9 10. The example, presented in
a manner a little different from the original, is that of Hamilton's Quaternions.
— The abstract definition of a ring 87

William Rowan Hamilton (4 August 1805-25 September 1865)

Hamilton was born in Dublin, the fourth child of nine. When he was
aged only one year his parents put him in care of his uncle, James.
By the age of five he was reading Latin, Greek and Hebrew! In April
1827, whilst still an undergraduate at Trinity College, Dublin, he
presented a paper which together with later supplements won him
the 1835 Royal Medal of the Royal Society. In June 1827 he was
appointed Astronomer Royal and Andrews professor of astronomy at
Trinity. A few days after his 30th birthday he was knighted.
In 1833 he had presented a paper in which he eliminated the
symbol 'i' from the complex numbers by defining them as 'algebraic
couples' (see Section 4.4). For many years he tried unsuccessfully to
find an extension to triples. Eventually the idea struck him—use
4-tuples (quaternions) and relinquish the commutative law of
multiplication. He wrote several texts promoting the use of
quaternions in physics. However, Gibbs' vector analysis was
eventually preferred, perhaps not least because of the fact that much
of Hamilton's work was difficult to read.
In undergraduate mathematics Hamilton's name appears in the
Hamiltonian and Hamilton's equations in mechanics, and in the
Cayley—Hamilton theorem in linear algebra.

First we make

Definition 3.2.5 A matrix with complex number entries is a square

(711 712),
array enclosed by brackets, of 4 complex numbers. Given two
721 722

such matrices,A =
an) andB =(flui P12)wedefinetheirsumA®B
a21 a22 /321 /322
88 Introduction to rings

and their productA GB by

A$B = (aii+/311 a12+1312'1 andt
\a21+132I a22+f3221

(allfSll+a121321 a11012+a12$22
AGB \a21j3i1 +a221321 a21/312+a22322
The set of all these matrices we denote, fairly naturally, by M2(C).
Theorem 3.2.6 (M2(C), e, 0) is a ring.
Proof The most arduous part of the proof is the checking of axiom M2. To
(711
do this, let A and B be as above and set C = Then
721 722/
1'1

\a21131l+a22/.321 a211312+a2213221 '721 722/ \Z1 T1

where X1
Y1 =(ail/311+a12/321)712+(a11j312+a121322)722,

Zi = (a211311+a22f321)yll+(a211312+a221322)721,
T1 = (a211311 +a221321)712+(a211312+a221322)722,•
whereas

\a2i a221 \1321711+1322721 1321712+13227221 \Z2 T2

where X2 = +012721)+a 12(1321711+1322721) =

Yi,
Y2=all(13l17l2+13l2722)+al2(1321712+1322'y22)=

a21(131171i +1312721) +a22(1321y11 +1322721) =

T2=a21(1311y12+1312y22)+a22($2iy12+1322y22)= T1.
Hence (A OB)OC=A 0(B CC), as required.
It is easily seen that axioms Al and A2 are satisfied. Further, the matrix
02 satisfies A3 whilst, given A above, the matrix A* =
=
a11
is clearly such that A+A*__A*+A=02. The checking of
( —a2i a22
axiom D constitutes exercise 3.2.12.
We now consider the subset Q (for quaternions) of M2(C) comprising all
s+it u+iv
matrices ( u+iv . $ —.it
) where s, t, u, v ER. It is not difficult to check [you
should do it!] that Q is closed under and 0 as defined above so that ®
tThe reason for the definition of®? Given x2=cy1+dy2 and y1=Az1+Bz2.
Y2 = Cz1 +Dz2 find x1 and x2 in terms of z1 and z2 and then youil see. For an authoritative
account of the introduction of matrices see [1051.
 The abstract definition of a ring 89

and 0 can be taken as binary operations on o (see 2.7.3(u)). Further (Q, ® 0)

satisfies axioms Al, A2, M2 and D because these axioms already hold for
any three elements out of M2(C). Clearly 02= and '2 belong
1)
to 0 and satisfy A3 and M3 for 0. Further, A4 holds. We show that M4
(and hence Z) holds and Ml fails in 0.
a—ib —c--id

For M4: given A =

(a+ib c+id setA'=
Lc+id a—ib c—id a:ib
S S

where s=a2+b2 +c2+d2. Then A OA'=A'OA =12 (except when s=O;

that is, when A =02).

whence (Q, 0) is certainly not a field.

It is now an easy matter to give ten concrete examples which between them
show that all the arrows of 3.2.4 are irreversible.
Examples 3.2.7
(i) 0 is a field. It satisfies all the axioms.
(ii) 0 is a division ring and not a field, i.e.
(iii) 1
(iv) Q[x] (polynomials with quaternion coefficients)
(v) 27 (the ring of even integers) 6jt'8
(vi) 74
All elements of Q[x] with zero constant term
(vii)
(viii)* M2(l)
(ix) The set {O, 4, 6} mod 8
(x)* M2(27L) lj*'2
Remark You should convince yourself that the above examples, which were
the first to occur to the author, do what is claimed of them.
Problem 1 Can your replace each of (i) to (x) by an alternative concrete
example?

Exercises
1 With the usual definitions of addition and multiplication, do the following
Sets form rings, integral domains, fields? Check the axioms in the order given
in 3.2.2 and stop as soon as you come to the first (if any) which fails.
(a) The set
(b) The complex fourth roots, 1, i, —1, —i, of 1. [What is 1+1?]
(c) The set of all a+ib where a, beQ.
*
See the end of the Notation Section.
90 Introduction to rings

(d) The set of all a +ib where a, b E/.

(e) The set of all a/b in 0 where (a, b)= I and b is odd.
2 On the set I define two new multiplications a and S by: for all a, b e 1,
a ob = 0, a Sb = 1. Show that (1, +, a) is a ring with zero divisors. Is (1, +, Eli)
a ring?
3 Show that vector multiplication A on the set V of all vectors in 3-
dimensional space is not associative. [Hint: Consider i A (1 A j) and (i A 1) A j.]
Deduce that (V, +, A) is not a ring, but that it satisfies axioms Al, A2, A3,
A4 and D. [V is essentially a type of non-associative ring called a IJe* algebra.]
4 Let S ={s}, a set with one element. Define s +s=s and s =s. Prove
that (S. +, ) is a ring in which s satisfies both A3 and M3 (see exercise 3.3.4).
5 Let S be any non-empty set. On (see 0.3.4), the power set of S, define
® and 0 by A$B=(AuB)\(AnB) and AOB=AnB. Show that the
empty set 0 satisfies axiom A3 and that for each A e9'(S), A ® A = 0 and
A 0 A = A. Go on to show that <p(S), 0> is a ring. [That e is associative
is easily 'verified' by using a Venn diagram.] Which elements of 99(S) are zero
divisors? [<a(S), 0> is an example of a Boolean ring. Such rings are of
interest in particular to logicians.]
lf we define by AEBB = Au B, is 0) a ring?
6 Let F be the set of all functions f:R-* It Defining 0 and 0 by: for all
XER, (f0g)(x)=f(x)+g(x), (fOg)(x)=f(x)g(x) show that (F,®,0) is
acommutative ring satisfying M3 but not Z. Which elements of F have
multiplicative inverses? If 0 is replaced by ° (i.e. 'composition') is (F, +, a) a
ring?
7 Which elements of (la, 0, 0) are zero divisors? Which have multiplicative
inverses?
8 On the setS ={0, 1} define 0 and Shy
9J91
010 001
101 111
Show that (5, 0, 0) is a ring. Is it a field?

9 Given rings and define p and a on RxS by

(r1, s1)p(r2, s2) = (r1, s1)a(r2, s2) = (r1 r2, s1 S s2). Show that
(R x 5, p, a) is a ring. This ring is called the direct sum of R and S and is
denoted briefly by R OS.
10 Find a ring R and elements a, b, c all distinct from z such that a b = a c
and yet b c.
*
Marius Sophus Lie (17 December 1842- 18 February 1899).
 Ring prope flies deducible from the axioms 91

ii The equivalence classes O, 2,.. . , SO, 32 of even integers mod 34 form

a ring with respect to the usual definitions of ® and Q. Show that 18 is a
unity element. Is it the only one?
ii Show that axiom D is satisfied in (M2(C), $, 0).
13 Is either of the rings (M2(Q[x]), $, 0), (M2(12), $, 0) commutative?
14 A complex number a is called an algebraic number (respectively, an
algebraic integer) if a is a root of some monic* polynomial in O[x]
(respectively, Z[x]). As particular examples show that, if del, if a, be 0 and
if a, v e Z, then a + b 'Ld is an algebraic numbe and a + V -'JdJs an algebraic
integer. [Hint: a +b'Jdis a root of {x —(a +b'id)}{x —(a —b'Jd)}eO[x].]
Let be square-free, that is: ifxel and x2ldthen x2= 1. Denote by
the set of all a+b"/ with a, beQ and by the set of all
algebraic integers in Let = {u + v'Jd; u, v It can be
Z}.
shown (see 1 [49, Vol. 2, p. 54]) that if d*j (mod 4) then Q['/d] =
whereas if d 1(mod 4) u, v e Z, a v(mod 2)t}. Show
that is a field whereas is a ring but not a field. [Hint for
— 1
-=
a+b'Jd a2—db2

15 Show that the set of all 'fractions' 1 wheref, g O[x] forms a field under
g
the obvious definitions of +

(. f
\I.e.g
hfk+gh f
k gk 'g
[This example will be considered more formally in Section 3.10.]

3.3 Ring properties deducible from the axioms

Having seen that there is a limitless supply of concrete rings and even a large
supply of different types of rings, it certainly appears worthwhile to try to
develop a theory relating to abstract rings, the results of which will, by
specialisation, be applicable to each and every concrete ring then encountered.
Although we are allowed to draw on intuition in formulating statements of
theorems and their proofs, we ensure that our results are logically sound by
employing the axiomatic method in which hypotheses are clearly stated at
the outset and a proof comprises a sequence of statements logically derived
therefrom
As a first example: It is clear that each concrete ring so far encountered
contains exactly one zero element. Axiom A3 tells us that there is always at
*S 1.9.18.
t In other words u, v are both odd or both even.
92 Introduction to rings

least one. But is there always exactly one? In every ring? Surely! How could
it be otherwise? But where is the proof? Here it is; and a little more besides.

Theorem 3.3.1 Let (R, +, be a ring. Then R contains exactly one element
satisfying axiom A3. Further, to each a E R, there corresponds exactly one
element a* as given by A4.

Proof Suppose z and z1 are elements of R each satisfying A3. [Thus z + a = a

and b + z1 = b for each a and b in R. In particular, taking a to be z1 and b
to be z we see that] then
z +z1 = z1 (since z satisfies A3)
and
z +z1 = z (since z1 satisfies A3)
It follows immediately that z1 = z, as required.
For the second assertion, first note that R has at least one element of the
required kind, by hypothesis. Next, suppose, given a E R, that a * and at are
elements of R each satisfying A4. [We start from the equality (a * + a) + at =
a*+(a +at) which is true by A2.] Now
(a*+a)+at =z+at (sincea*satisfiesA4)
= at (since z satisfies A3)
Similarly
a*+(a +at)=a*+z (since at satisfies A4)
= a* (since z satisfies A3)
Thus a* = at, as required.

Similar results relating to axioms M3 and M4 can be proved (exercise I

below). These assertions should not be regarded as obvious. Exercise 3.2.11
should emphasise this.
Without apology and further motivation we establish further consequences
of the axioms. [Please read carefully and slowly, making sure you understand
and believe each step in]

Theorem 3.3.2 Let (R, +, .) be a ring. Then, for any a, b ER:

(i) (ii) (a*)*=a
(iii) a*b=a .b*(a .b)* (iv)
a unity element e then
(v) e*.b=b.e*=b* (vi) e*.e*=e.
Proof [Recall that wc're only allowed to use axioms Al, A2, A3, A4, M2
and D.]
 Ring properties deducible from the axioms 93

(i) z+z=z [Why? Which axiom?]

a '(z +z)=a z [Propertyof = sign.]
[Why?]
Now there exists in R an element (a . z )* such that a z + (a . z
)* = z. [Why?]
Also

{a .z}+(a .z)* =a •z+(a .z)* [Why?]

a.z+{a.z+(a.z)*}__a.z+(a.z)* [Why?]
z z [Why?]

a z is similar.
[How are you going? Read the proof of (i) again if there was any step you
didn't follow.]
(ii) By definition, (a*)* denotes an additive inverse, unique by 3.3.1, of a*
mR.
But a*+a =a +a*=z. Hence a also satisfies axiom A3 for a*. Thus
a=(a*)*, by uniqueness.
(iii) By definition, (a b)* is the unique additive inverse for a b in R. We
show a* 'b is too! Now
a .b+a*.b(a+a*).b [Why?Whichaxiom?]
= z [Why?]
= z [Why?]
One can prove similarly (or use axiom Al) that a* b +a b = z. Thus
a* b is a (and hence the) unique additive inverse of a b. That is, a* . b =
(a .b)*.
We leave proof that (a b)* = a b* to the reader.
.

(iv) a*.b*=(a .b*)*((a .b)*)*a [Why? Why? Why?]

(v) Replacing a in (iii) by e we get e* b =(e . b)*. But e b =b. [Why?]
Hence e* b = (e . b)* = b*, as required.
(vi) Replacing b in (v) by e* we get e* . e* = (e*)* = e. [Why?]

Did the statements and proofs just given leave you asking 'What does all
this mean? How did he think up the statements in the first place and how did
he then find the proofs?' The answers are easily seen if, despite the fact that
Our elements may not be numbers—indeed their nature may not be specified—
we replace the z, e and a * notation by the symbols 0, 1, —a. In
this notation statements (i) through (iii) above become (i) 0 a = a 00;
(ii) —(—a)=a; (iii) b =a (—b)—(a . b); whilst (vi) becomes
(—1) . (—1) = 1. This illustrates how I allowed my knowledge of Z and the
94 Introduction to rings

proofs in 1.2.1 to suggest the statements and proofs of 3.3.2 and shows the
advantages (when treated with care) of a suggestive notation. One disadvan-
tage (for the beginner) is highlighted in exercise 9(a).
We summarise our new notation and introduce more in:

Notation 3.3.3 From now on the symbols z, e, a*, a' of 3.2.1 will be replaced
by 0, 1, —a, a1. Further, a +(—b) will be shortened to a —b. [It is then easy
to show that c (a—b)=c ja—c for all a, b, ceR (exercise 1(b)).] We
shall also lapse into writing ab in place of a b.

We are now (already!) in a position to demonstrate the power of the abstract

method. In 3.2.7(iii) we gave 7 as an example of an integral domain which
is not a field and in the subsequent problem you were asked to find a further
such example. What did you offer? 7[x], 02 = 0: b is an
odd integer}? I can only guess. But one thing I know: each (valid) offering
just had to be a ring with infinitely many elements. More explicitly,

Theorem 3.3.4 Each finite integral domain J is necessarily a field.

Proof [Since an integral domain satisfies all the field axioms except possibly
M4 we only need show that J necessarily satisfies M4.]
Let the elements of J be labelled a0(=0), a1(=1), a2,. . . , Select any
element a1 other than a0. Consider the list a0a1, a1a1,.. . , of elements,
all in J. Suppose, for suffices f, k (where 0<1 k n) we have a1a1 = akal.
[Since J is a domain] it follows that a1 = ak [exercise 1(b)]. This means that
the list of a1a1s comprises n + 1 distinct elements of J; that is, all of J. Since
1EJ there is an 1 such that a1a1=1. But then aa1=a1a1=1 and a1 has a
multiplicative inverse in J. [Thus axiom M4 holds, as required.]

Note 3.3.5 A much deeper result is: Every finite division ring is a field. This
result, proved by Wedderburnt in 1905, supplies a proof of the geometrical
assertion that every finite projective plane which is Desarguian is necessarily
Pappian. (Prologue, p. xxiii.)

Exercises
1 (a) In the manner of 3.3.1 and 3.3.2 prove that in any ring (i) the equality
a + c = b ± c implies the equality a = b; (ii) a + x = b has the unique solution
x = a * + b; (iii) the elements e and a' of axioms M3 and M4 are unique—when
they exist; (iv) if a' exists then (a')' exists and (a')' = a.

± Joseph Henry Maclagan Wedderburn (26 February 1882—9 October 1948).

 Ring properties deducible from the axioms 95

(b) Prove carefully, using the less suggestive notation, that c (a —b) =
(That is, prove that c.(a+h*)=c.a+(c.h)*.) Deduce that ift
c an integral domain J then a=b in J.

2 Let (R, ) be a ring. For a R and n we define (cf. 2.7.9) na and

a'1 to be respectively the sum and the product of n copies of a. We define
Oa to be the elementt °R (here 0€ l, R) and, for k (—k)a to be the
element k(—a).
Prove that, for all a, b €R and m, n El we have (i) (—m)a = —(ma);
(ii) (m+n)ama+na; (iii) (mn)a=m(na); (iv) m(a+b)—ma+mb;
(v) (vi) (ma)(nb)=(mn)(a'b).
Assuming R to have a unity element e and a to have a multiplicative
1)k
inverse define a°e and, for State and prove the
(a
multiplicative analogues of (i), (ii) and (iii). Show by example that the multi-
plicative analogue of (iv), namely (a b)tm =am 'btm, might not hold in R.
.

Does (v) suggest any further identity? Still assuming that R has a unity element
e show each ma (where mel, aeR) can be written as r a for suitable reR.
[Hint: r=me is suitable.]
3 Let(R, +, a ring. On R xl define $ and® by:
(r1, zi)®(r2, z2) = (r1 +r2, z1 +z2)

(r1,z1)Q(r2,z2)=(r1 . r2+z1r2+z2r1,z1z2).
Show that <R xl, $, 0> is a ring with unity element (OR. I).
4 Prove that if R satisfies (A3 and) M3 and if e = z then R is a ring with
one element. [Hint: For a R, a = a e = a z =?] (See exercise 3.2.4.)
S Prove that if R satisfies M4 then R satisfies Z. [Hint: generalise the proof
of 2.4.6(b).)
6 Fraenkel,t in his definition of a ring, insisted that R should satisfy M3
but not necessarily Al. Expand (e +e) . (a 4-b) in two ways and using A3,
A4, D show that a +b = b +a. That is, Al is automatically satisfied in
Fraenkel's rings.
7 Let (I?, be a ring in which x2=x x =x for all x eR. Prove that R
is a commutative ring in which 2x = x + x = [Hint: expand (a + a )2 and
(a +b)2.] Note: exercise 3.2.5 gives an example of such a ring.
8 Prove, or give a counterexample to, the assertion: for all x, y in any ring
R, (x+y)2=x2+2xy+y2.
0.)
9 (a) Use the elements a and b = (? of Q to show that
=
(a . C9b'. Hence pinpoint the error in the following
denotes the zero element or the ring J(R).
t Adolf Abraham Fraenkel (17 February 1891 15 October 1965).
96 Introduction to rings

argument* Writing in place of a' we get

(b) To banish all possibility of making intuitive but unwarranted assumptions
about + and in an arbitrary ring let us replace + and by p and 'r.
Is a'r(bp(c'r(apb))) = ((arc)ra)p((ap(arc))'rb) in (R, p, r)?
Isa
Which was easier to decide? Why? So is it easier to use p, r or +,
10 Let R be a finite ring with no (non-zero) zero divisors and at least two
elements. Show that R is a division ring.

3.4 Subrings, subfields and ideals

Of the rings mentioned in Section 3.2, six fall into three pairs, the first in
each pair being contained in a natural way inside the second. Thus the ring
2! is naturally thought of as being contained in the ring 1, M2(21) as being
part of the ring M2(Z) and, finally, Q as being in M2(C). Further, with a little
bit of care (see Remark (ii) on p. 40) we can think of 0 as being contained
in Q[x] and of 0{x] as being inside 0[x,y] (exercise 1.8.2), etc. With the
same care (Section 4.4) we can think of the succession 0 c c C of fields.
Thus on the grounds of convenience alone it seems appropriate to introduce
a terminology to describe the sitting of one ring nicely inside another. Note
that in each case the elements of the 'smaller' ring form a subset of those of
the 'bigger' ring and the results of adding and multiplying each pair of elements
within the smaller ring coincide with the corresponding results when the same
two elements are regarded as being in the bigger ring.

Definition 3.4.1 Let (R, be a ring and suppose that S is a non-empty

subset of R. We say that S is a subring of R if and only if (i) S is closed
under + and . (so thatt + and are binary operations on 5) and (ii) (5, +,')
is a ring.

In addition to the motivating examples above we note that, in the ring R x 1

of exercise 3.3.3, the subset R x {0}={(r,0): reR} is a subring. Note that R
itself is not a subring of R x since R is not actually contained in R x 1. This
state of affairs causes the algebraist no embarrassment at all (see Remarks in
Section 3.10 and exercise 3.10.9).
How can one recognise when a non-empty subset S of a ring R is a subring
of R? One can check 3.4.1(i) and (ii) of course; but easier is

Theorem 3.4.2 Let S be a non-empty subset of a ring R. Then S is a subring

of R if and only if both (a) and (f3) below hold.
*Due to a former student whose name I have conveniently forgotten!
More accurately, the restrictions of + and to S. See 2.7.3 (ii).
 Subrings, subf/elds and/deals 97

(a) lfa,beSthena+beSanda'beS
(13) JfaeSthen —aES (Here —a is the inverse ofain R.)
To prove the 'only if' part of this we shall need

Lemma 3.4.3 Let S be a subring of R. Letting 0s and °R denote the zero

elements of the rings (5, +, and (R, +. ) we have Further, if a e S
then (—a)s = (—a)R.

Remark This lemma is so obvious it doesn't need proof. Right? And just
as obvious is the remark that if S and R have unity elements and then
= Ig. Since this latter assertion is false (see 3.4.4 (iii) and exercise 7) we'd
better take 3.4.3 seriously and offer proof.

Proof of 3.4.3 Let s S. Inside S we have s + = s. Inside R we have

S+OR=S. Hence, in R, S+Os=S+OR. Thus 05=°R [by exercise 3.3.l(a)(i)].
Next, in 5, a + (—a )s = 0s whilst in R, a + (—a )R = 0R' Consequently, in R,
a +(—a)5 =Os = = a +(—a)R, whence (—a)5 =(—a)R. [Why?]

We can now give the

only if
Proof of 3.4.2 Since S is a subring of R, (5, +,) is a ring. Hence for
a, beS we have a+b€S, a and (—a)5€S. But (—a)s—(—a)R——a.
This is enough.

'& [We assume conditions (ci) and (/1).] First, (cx) implies that S is closed
under + and Since there exists s€S. By (j3), (—s)g€S. By (a),
= s + (—s )R €5. Clearly is the element required, by A3, to be in S. Also
foreach a €5, (—a)R ES by (f3). Clearly (—a)R is the element required by A4
tobe in S. Finally, each of Al, A2, M2 and D holds for all elements of R;
hence, in particular, all elements of S. [Thus we have shown that + and are
binary operations on S and that (5, +, is a ring.]

Examples 3.4.4
(i) Q is a subring of (M2(C), 0). For, given x, y Q, it is easily seen that
x®y, x Dy and ex are also in Q.
(ii) On 2/ define + as usual, but ED by: for all a, b 2/, a ED b 0. Then

(2/, +, 0) is a ring but it is not a subring of (Z, since a ED b a b for

all a,b€2/.
(iii) S = {O, a, g, is a subring of Z10. Note that = I whereas = &
(iv) In the ring (F, 0) (exercise 3.2.6) of all functions from to R the
subset of all differentiable (respectively continuous) functions forms a subring,
essentially because the sum and product of differentiable (continuous) func-
tions is differentiable (continuous).
98 Introduction to rings

We now use 3.4.2 to establish the important

Theorem 3.4.5 Let S1, S2 be subrings of the ring R. Then the set-theoretic
intersection nS2 is also a subring of R.
Proof Since 051 = = 052 we see that 51 n 52 0. Now suppose x, y E
S1nS2. Then xe51 and y and —x all lie inS1 [by
3.4.2]. Similarly x +y, x y and —x all lie inS2 and hence inS1 nS2. By 3.4.2,
51 nS2 is a subring of R.
This result clearly extends to intersections of arbitrary (possibly infinite)
sets of subrings of a ring I? (exercise 5).
There are important analogues of 3.4.1, 3.4.2 and 3.4.5 for fields. We have
Definition 3.4.1(F) A subfield of a field F is any non-empty subset T of F
such that (i) T is closed with respect to the binary operations + and defined
on F and such that (ii) (T, +, ) is a field.
Theorem 3.4.2(F) Let T be a non-empty subset of a field F. Then T is a
subfield of F if T has at least two elements and both (a) and (/3) below hold.
(a) Ifa,bETthena+bETanda
(/3) 1eT
Theorem 3.4.5(F) Let {Ta: a e A} be a set of subfields of the field F.
Then the set-theoretic intersection fl Ta is also a subfield of R.
a
In particular the intersection P of the set of all subfields of F is a subfield
(clearly the unique, smallest one in F) called the prime subfield of F. Clearly
°F and lp lie in each subfield of F and hence in P. [Aren't you just a bit
suspicious about iF? See Example 3.4.4(iii) above and exercise 3.4.9.] The
exact nature of P is somewhat restricted (see 3.10.9).
In the theory of rings a special kind of subring arises in two disparate ways.
Asking you to wait until Section 3.9 and 4.2.7 for motivation we make
Definition 3.4.6 Let I be a non-empty subset of a ring R. Then I is called
an ideal of R if and only if (i), (ii) and (iii) below hold.
(i) Ifa,belthena+beI
(ii) Ifaelthen—aeI
(iii) IfaelandreRthenraandarel
Clearly each ideal is a subring (in (iii) r can be any element of I). Note that
a r and r a may well be unequal; we demand that both belong to L
Further, since each product ar and ra gets swallowed up by I we can think of I
as a sort of ring-theoretic 'black hole'!
Examples 3.4.7
(i) Let s e 7. Then the set {sz: z e Z} is an ideal of 7. In fact every ideal—
indeed every subring—of 7 has this form, for suitable s. (See 3.7.16.)
 Subrings, sub fields and ideals 99

(ii) Let feQ[x]. Then the set {fm: m €O[x]} is an ideal of Q[x]. In fact
every ideal, though not every subring, of Q[x] has this form. (See 3.7.16
and exercises 1(b) and 13.)
(iii) In any ring R the subsets {OR} and R are ideals. If F is a field then {O}
and F are its only ideals (exercise 14).
(iv) Let R be any ring and let a1, a2,.. . , am R. Then the set of all ele-
ments of the form z1a1 + L s,a1 + a1t1 + ukaevik) where
m, n1 Z, s1, 4, Uj,k, R is an ideal. In fact it's the smallest ideal
of R which contains a1,... , am. Hence it is called the ideal generated
by a1,... , am.
R is commutative and has a unity the above set reduces to the set
If

{airi . . r1 R}. [Why can z1a1 be written in this

form?] This ideal we denote briefly by [ai,. . , am]. If m = 1 the ideal
.

[a1] is called the principal ideal generated by a1. In particular [IR]= R.

(v) The subset E of Z[x] comprising all polynomials with even constant
term is an ideal of l[x]. In fact E [x, 2] [are you sure?] and is not
principal. (See exercise 22(b).)
Remark A commutative ring in which every ideal can be generated by a
finite number of elements is called a Noetherian ring (after Emmy Noether
who undertook deep investigations of such rings). A theorem central to the
subject of algebraic geometry is the so-called Hubert Basis Theorem:* Let R
be a Noetherian ring. Then the polynomial ring R[x1,x2,. . ,xj in the n .

commuting letters x1, x2,... , with coefficients in R is Noetherian. In particular

in the ring R[x1, x2,... , xj where R =Z, 0, R, or C, each ideal is expressible
in the form , fj, only finitely many polynomials
. . being required.

Amauie Emmy Noether (23 March 1882- 14 April 1935)

Emmy Noether, the first child of Max and Ida Noether, was born in
Erlangen, Bavaria, where her father, a famous algebraic geometer,
* The theorem Gordan called 'Theology, not mathematics'. For a proof see [9, Vol. 2, p. 401].
100 Introduction to rings

was a professor of mathematics. Although a discussion of the

so-called Noetherian rings would be out of place in this book, we
include this biography because she did so much in the 1920s and
1930s to point algebra in the direction it faces today. As the Russian
mathematician Alexandroff said of her: 'Emmy Noether taught us
to think in a simpler and more general way; in terms of
homomorphisms, of ideals—not in terms of complicated algebraic
calculations. She therefore opened a path to the discovery of
algebraic regularities where previously they had been obscured by
com plicated specific conditions.'
Despite the prejudices in the early years of the century against
women in universities, she obtained her Ph D in 1907. However,
despite great efforts by Hubert, she had, by 1922, still attained
nothing more than an honorary professorship at Göttingen. And
more difficulties were to come. Along with other Jewish colleagues
she was dismissed from Gättingen in 1933. The USA gladly offered
her a home and she spent the last two years of her life at Bryn Mawr
and Princeton.
Emmy Noether was one of the greatest of algebraists and most
probably the greatest of all women mathematicians.

Exercises
Subrings
1 Establish, using 3.4.2, whether or not the following subsets of the given
rings are subrings:
(a) All positive integers in (1, +, .).
(b) All polynomials with integer constant term in (Q[x], +,).

(c)

and °): zE in (1
1k): a, b, CE R}, e,
(d) All integers not divisible by 3 in (2, +,.)
(e) All polynomials of degree in (Q[x], +, .).
(f) All numbers a +bi where a, b €2 in the field (C,
(g) The set {75a +30b: a,b €l}in (2,
(h) All zero divisors in (214, 0); in $, 0).
2 Find the smallest subring S of U which (i) contains 1 [Hint: Clearly
1 7 S. But 7 is a subring.]; (ii) contains and
3 Let S be a non-empty set. if A show that {ø, A} is a subring of
®, 0) (see exercise 3.2,5).
4 Show that a non-empty subset S of a ring R is a subring if and only if for
alla, b €S, a—b €S and a b €5 (cf. 3.4.2).
5 Copying the proof of 3.4.5, extend it to arbitrary sets of subrings.
 Subrings, sub fields and ideals 101

6 Show that the union {2m: in eZ}u{3n: n €Z} of two ideals of 7 is not
even a subring of 7.
7 Copy the proof of 3.4.3 to identify where a proof that = 1R would break
down.

Fields
8 Prove,using3.4.2(F),thatthesubset{a + Oi: a eR}isasubfieldof(C, +,•).
9 Let T be a subfield of a field F. Show that iT = iF and that, for a T
(a$OT), (a')T=(cC1)F (cf. 3.4.3 and exercise 7 above.)
10 Prove 3.4.2(F) and 3.4.5(F).
11 How many different subfields can you find in (i) (Q, (ii) (li, +,
[For (ii) recall all the a where a, b EQ.]
Ideals
12 Showthat,inl,[15,21]=[3]. Is[15, 35, —77]=Z?
13 Which subsets in exercise 1 above are ideals? [(b) tells you that, unlike
in Z, not all subrings of Q[x] are ideals.]
14 Show that if I is an ideal in the field F then either I = {O} or I =F. [Hint:
If I {O} then there exists x E I such that x 0. Then x1 exists in F and
I = xx 'EL] Let R be a commutative ring with unity. Show that R is a field
if and only if R has 32 elements and {0} and R are the only ideals. Exhibit
infinitely many subrings of 0 which contain 7.
15 Let I be an ideal in the ring R. Show that 1[x]=
{ao+a,x+. . .+a/:aEJrEZ+u{0}}isanidealofR[x]
16 Prove all the assertions in example 3.4.7(iv).
17 Foridealsl1,I2inaringRdefinef1+I2tobetheset{a+b:aeI,,b€12}
and '1'2 to be the set {a1b1+' . nc7C, b1c12}. Show that
+ '2, 1112 are ideals of R. If R is commutative with
and I, n 12
unity, if 11=[a,,... ,am] and if 12=[b1,... show 1112=
[a1b,,a1b2, a2b,,. ,
. . . . .

18 In Z[x] set I2=[x,2] and 13=[x,3].Showthatl={ab:aEI2,bEI3} is

not an ideal in Z[x]. [Hint: x2EI, (Thisexplains the definition
of '112 given in exercise 17.)
19 Let I, ç 12 ç 13 be .a possibly infinite sequence of ideals in a ring R.
Set U ={a: a ER and a U is an ideal of R. [Hint:
given a, b E U there exists t such that a and b are both in Ii.]

20 Show that o o = in M2(R). After finding

three similar equalities deduce that if I is an ideal in M2(R) then either

102 Introduction to rings

or I=M2(lt). [A ring with no ideals other than {OR}

)}
and R is called a simple ring.]
21 Let S be a set {(a1, b1)} of points in the plane R x It Show that the set
{p(x, y)} of all polynomials in R[x, y] for which p(a1, b,) =0 for all (a1, b1)eS
is an ideal of R[x, y]. (Such examples are at the heart of that part of algebraic
geometry which studies the geometry of curves by algebraic means.)
22 (a) Show that in Q[x, y] the subset S = {xf+yg: f, g Q[x, y]} is an ideal.
Show that S is not a principal ideal. [Hint: If S = [hi then x = hu and y = hv
for some u, v Q[x, y].]
(b) Show that in Z[x] the ideal [x, 2] is not principal.

Pierre de Fermat (20 August 1601-12 January 1665)

Fermat's father was a prosperous leather merchant, and his mother
came from a family of high social standing. As a young man he
studied law and took a law degree in 1631, becoming a magistrate.
He was fluent in Spanish, ltalian, Latin and Greek. Serious interest in
mathematics seems to date only from his late twenties. Fermat can,
with Pascal, be regarded as a founder of probability theory. He also
invented analytic geometry independently of Descartes, although he
did not develop it so extensively, and contributed towards the
invention of the calculus. But his main claim to fame is as the 'Father
of modern number theory'. Most of his results were communicated
by letter or written in his copy of Diophantus' Arithmetica, and
almost invariably without proof. Iormally he published almost
nothing. It appears he must have had an amazing intuition because
all of his definite assertions, except one, the famous conjecture (or
Last Theorem), have been proved true. Even his belief that every
integer of the form 22Th-1 is a prime (this is false) was accompanied
by his admitting that he couldn't actually prove it. (How much more
intriguing does this make his unqualified assertion about the FC!)
 Fermat's conjecture (FC) 103

Besides work mentioned above he is credited with finding the

f' (x) criterion for maximum and minimum values of functions, and
in optics he enunciated the principle named after him.

3.5 Fermat's conjecture (FC)

In 1847 G Lamé* announced to the Paris Academy that he had finally proved
the FC, otherwise known as Fermat's Last Theorem, dating from the 1630s.
This is the assertion that to each integer n > 2 the equation x" + y = z has
no solution in integers x, y, z for which xyz 0. Fermat, given to writing
statements, but not their proofs, in his copy of Bachet's 1621 translation of
Diophantus' book Arithmetica, had asserted that 'I have found a truly remark-
able proof but the margin is too small to contain it.' Fermat actually left a
proof which covers the case n = 4 and an inadequate proof by Euler (1770)
for the case n=3 can be salvaged [46]. 1825 saw proofs by Legendre and
Dirichlet for the case n = S and in 1839 Lamé gave a proof for the case n = 7.
Even today no proof of, nor counterexample to, FC is known and it is
generally assumed that Fermat's proof contained a subtle error. Lamé's idea
(and possibly Fermat's) in the case n = p, an odd prime (see exercise 3), is to
factorise z" +yn as (x +y)(x (x +('y), 1 being a complex
pth root of 1. Since this product is a pth power, and since a greatest
common divisor d of any pair x x can be shown to be a common
divisor of all the x +4a<y, it allegedly followed (as it does in t; see exercise
1) that each x is itself a product of d and a pth power. From this a
contradiction is to be obtained by Fermat's so-called method of descent (see
3.8.2 and exercise 3.8.6). The chief objection to Lamé's proof is in his implicit
assumption that a theorem analogous to the Fundamental Theorem of Arith-
metic (1.5.1) remains valid for the more inclusive sets of complex integers
under consideration here. We amplify this objection in Section 3.6.
In fact, the unique factorisation theorem is not valid for numbers of the
above kind, at least for p = 23.t This had apparently been known to Kummer
since 1844 (see [46]). But this failure is a most fortunate one as far as progress
in mathematics is concerned, for much high powered mathematics has resulted
from attempts to circumvent it.
Kummer's efforts (c. 1847—57) extended the range of prime exponents p
for which FC is known to be true from the set of all primes p <23 to the set
of all so-called regular primes (see [49, Vol. 2, p. 97], the first three irregular
primes being 37, 59 and 67. Sadly, it is still not known if there are infinitely
many regular primes though it has been known for almost 80 years that there
are infinitely many irregular ones! Conditions sufficient to prove the FC in
special cases have been discovered over the years. For example in 1909
*
Gabriel Lamé (22 July 1795 — 1 May 1870) was considered, by Gauss, the foremost French
mathematician of his generation.
t In fact, the unique factorisation theorem holds for all primes <23 and fails for all other primes.
See Crelle's Journal, vol. 286—7, p. 248.
104 Introduction to rings

Wieferich showed that if x" 9 and if p txyz then the

first prime satisfying this being 1093. Aided by results of this kind (and
computers) it is now known that the FC is true for each exponent n which
has a prime divisor less than 125000 (!) (see [53]).
In addition, recent work for which Gerd Failings won a 1986 Fields Medal
(p. 266) shows that has, for each n, at most finitely many integer
solutions.

Exercises
1 Show that if z1, z2,. . ,. z, n e1 are such that (z, z,) = I when i
and if ZIZ2. .. = z' then each z1 is an nth power.
2 Look up a proof (you'll find one in, for example, [42] on pages 149150) of
the assertion that every solution of the equation x2 +y2 =z2 is of the form
x = s(a2 — h2), y = 2sah, z = s(a2 + h2), or the same with x and y interchanged,
where s, a, be!.
3 Show that to prove FC one only (!) needs to prove it for n =4 and for n
an odd prime. [Hint: If (x0, y0, z0) is a solution when n = u c then (xc,
is a solution for n = v.]

3.6 Divisibility in rings*

In order to be more explicit about the remarks made in Section 3.5 concerning
non-unique factorisation we must define the concept of division in a ring. We
restrict ourselves to commutative rings and following Sections 1.3 and 1.9 we
make, for any commutative ring R whatever,
Definition 3.6.lt Let a, b e R.
(i) We say a divides b and write a lb if and only if there exists c e R such
that ac = b.
(ii) If c, deR, if dia and dib and if, from cia and cib, it follows that cid
then d is a greatest common divisor (gcd) of a and b. We denote any
one of these gcds by (a, b).
Assuming, in addition, that R has a multiplicative identity:
a e R is a unit if and only if there exists v e R such that uv = vu = 1.
(iii)
a, b are associates if and only if there exists a unit a such that a = ub.
(iv)
n e R is irreducible if and only if ir 0, ir is not a unit and from ir = afi
(v)
(with a, j3 e R) it follows that either a or is a unit.
(vi) ir e R is prime if and only if n 0, ir is not a unit and from irial3 (with
a, f3 eR) it follows that either lTIa or irj/3 (or both).
(vii) a, b e R are relatively prime if and only if I is a gcd of a and b.
*
Divisibility theory in fields a rather dull topic. Why?
+ 3.6.1 is easily adapted to rings which are not necessarily commutative hut we shall have no
need of such.
 Divisibility in rings 105

We have already given (Sections 1.3, 1.9) examples of these concepts in i

and in O[x]. Here we do the same for another type of ring of central importance
in number theory, namely the ring +, .), where d is square-free in
j, as introduced in exercise 3.2.14. It is easy to check that +,.) is
a ring, indeed an integral domain.
To help determine the units in we formulate (in see exercise
3.2.14).

Definition 3.6.2 Let ct = a + The norm of cx, denoted by

N(cr), is defined by N(cr) = a2—db2j.

There follows easily

Lemma 3.6.3
(i) If cx then N(cx) is a non-negative integer.
(ii) N(cx) = 0 if and_only if cx = 0.
(iii) For cx, $e QhJd], N(cxfl)=N(cx)N(fl).

Proof Exercise 11.

Now suppose u = s + t'Jd is a unit in Then for some v

we have It follows that N(u)N(v)=N(uv)=N(1)=1. By
3.6.3 (i),N(u)=1.
Ifd=—lthenN(u)=s2+t2=1.Thisimpliesthats=1,t=Oors=—1,
=0 or s 0, t = 1 or s = 0, t = —1. Hence u = I or —1 or i or —i.

Ifd<—lthenN(u)=s2—dt2=1.Thisimpliess=1,t=Oors=—1,t=O.
Hence u = 1 or —1.
If d>0 the situation is quite different, for it is known (see [42]) that the
equation s2 — =I has (for d not a perfect square) infinitely many solutions.
For example, u is a unit in since 1.

Further, the powers u" are distinct and yet each is a unit (exercise 3).
Noting that for d <0 the elements listed are indeed units in the appropriate
Z['JdJ we have, summarising the above,

Theorem 3.6.4 In the rings where d is square-free, the units are:

0)1, —1, i, —i if a' = —1; (ii) 1, — I if a' <—1; (iii) 1, —1 and infinitely many
Others if d> 1.

in Section 3.5 we asserted that the unique factorisation theorem fails in

Some rings of complex numbers. The example mentioned there, with p = 23,
15 too involved to reproduce here, so we describe the same phenomenon using
an easier example,
In Z[i—3] the equality (1 + — =2 . 2 holds. Suppose that
(1 -4-..JJj)= (a in Taking norms and using
3.6.3(iii) gives
+b'J—3)N(c +dI—3)=(a2+3b2)(c2+3d2)
106 Introduction to rings

Since a2+3b2 = 2 has no solution in! we see that a2+3b2 = 1 or a2+3b2 =4

In the former case a = 1 or —1 and b = 0, whence a is a unit. In the
latter case c2 + 3d2 = 1 and c + is likewise a unit. Thus, in expressing
1+ as a product, one of the factors is necessarily a unit. Since 1 + 'JTi
is clearly non-zero and is not a unit [why not?] it follows that I +
irreducible in Similarly 1 — and 2 are proved irreducible in
We deduce that in the number 4 has two distinct decompo-
sitions into products of irreducibles. In other words, the analogue of the
Fundamental Theorem of Arithmetic fails in (For other deductions
see exercise 12. In view of Lamé's approach exErcise 12(c) is especially
interesting.)

Exercises (Assume here that all rings mentioned, except matrix rings,
are commutative.)
1 (a) Prove that if aib and bic in the ring R then aic in R.
(b)
2 Let c e R c REx], R being an integral domain. Prove that c is a unit
(respectively prime, irreducible) in REx] if and only if c is a unit (respectively
prime, irreducible) in R.
3 Let u =2—13. Show that the set {uM: n el} is a set of infinitely many
distinct units in
4 Find all the units in M2(IU, in M2(l) and in l[x,y]. Is a
unit in 14k]? Show that, for any field F, is a unit in
the power series ring (exercise 1.7.5) if and only

5 In show that 3 ,l' 1 + 21—5. Deduce that 3 is irreducible but not

prime in l[1—5]. [Hint: 31(1 —21—5).] Show that if a2+5b2
is a prime in 1 then a +bl—5 is irreducible in ![1—5], but not conversely.
[Hint: 3 is irreducible.]
— a+b'i—3
6 Let denote the ring of all complex numbers where
2
a, bE! and a, b are both even or both odd as introduced in exercise 3.2.14.
±1±1-3
Show that the six numbers 1, —1, and are precisely the units of
Deduce that, in 2 and 1 + are associates and that
hence each divides the other.
7 If a lb and b a in a ring R must a and b be associates? Prove, or give a
counterexample. What if R has a unity element?
8 Identify the irreducible and the prime elements in (i) £12; (ii) 0;
(iii) (iv) the ring of exercise 3.2.6.
 Euclidean rings 107

9 Show that if ir a prime then irla or or

Generalise this from 3-fold to n-fold products of elements of R.
io Is 2 or any of its associates a square in Show that associates
of primes (irreducibles) in are again primes (irreducibles).
Prove Lemma 3.6.3.
12 (a) Prove that in an integral domain each prime element is necessarily
irreducible. [Hint: Copy the proof of 1.3.7.] —
(b) Give an example of an irreducible in which is not a prime.
(c) Show that, in a = 1 + %J—3 and /3 = 1 — are relatively
prime. Show that af3 =22 and yet neither a nor /3 is a square in
(cf. exercise 3.5.1).
(d) Show that 2 and 1 + are each (common) divisors of 4 and 2+ 2J1
in ZN! —3], but that 4 and have no gcd in
13 Show that in the integer 8 can be expressed both as a product
of two and as a product of three irreducibles.
14 Show that 1 is a gcd of x and y in Q[x, y]. Do there exist r(x, y) and
s(x, y)eQ[x,y]such that r(x,y)x+s(x,y)y =1?

15 Let denote the subring of Put A=

B Show that there exist infinitely many Ce R2(R) such

=
that AC = B but no D E such that DA = B.

3.7 Euclidean rings, unique factorisation domains

and principal ideal domains
The fact that the analogue of the Fundamental Theorem of Arithmetic fails
in 1N1—3] should make us all the keener to isolate the reasons why it holds
in 1. Uniqueness of factorisation in 1 was proved as follows:
(A) Existence of factorisation: either a e 1 cannot be factorised or it can
into smaller factors, and this process terminates after finitely many
steps.
(B) Uniqueness depended upon: (a) every irreducible is a prime; (j3) if p
is a prime and if plab then or Further, (a) was a consequence
of the Division Algorithm.
This analysis indicates that we might be able to establish uniqueness of
factorisation in any ring satisfying

Definition 3.7.1° Let R be a domain. Then R is called a Euclidean Ring

(briefly ER) if and only if to each element a of R there corresponds a
108 Introduction to rings

non-negative integer hail such that

(1°) For all a, b eR, hail hlbil=IIabll

and
(110) Foralla,b €R withb 0 there exist m,reR such thata =mb +r
and

Of course 7 is an ER if we define hizhl to be Izi. Unfortunately 3.7.1° is not

so obviously suitable for use with Q[x] since, for the most natural definition
of namely lifli = degree 1011 has no meaning. Further Ilfhl in
general.
Remedies are available (see exercise 2) but it is somewhat more natural to
modify 3.7.1° without injuring 7. Taking these points and 1.10.1 into account
we arrive at

Definition 3.7.1 Let R be a domain. Then R is called a Euclidean Ring

(ER) if and only if there exists a map 8, called a norm or valuation, from
R\{0} into! u{0} such that
(I) For all non-zero a, b €R,
(11) Let a, beR with b Then there exist in, reR such that a =mb+r
where either (i) r = 0 or (ii) r 0 and 8(r) <8(b).

There then follow easily

Examples 3.7.2
(i) 7 is an ER if we set 8(z) = Izi for each non-zero z E/.
(ii) Q[x] is an ER if we set 8(f) = degf for each non-zero fE Q{x].
(iii) Each field F is an ER if we define 8(a) = 1 for each non-zero a E F.

We present less trivial examples in 3.7.4 below.

Notes 3.7.3
(i) A ring may well be Euclidean with respect to more than one norm
(compare 3.7.2(i), (ii) with exercises 1(d), (e) below).
(ii) Another way of defining Euclidean rings, this time with Q[x] as its model,
is given in exercise 2.

Further examples of ERs are given by

Theorem The rings and l[p]={u +vp: u, v eZ}

—1
where are all ERs.
= 2
1Ev—li is called the ring of Gaussian Integers. (See Prologue, p. xx.)
 Euclidean rings 109

proof For each a in or in we set 8(a)=N(a) as in 3.6.2.

condition (I) of 3.7.1 holds by 3.6.3 (iii) and the fact that if #0 then
1.
For Z[p] we define 8(u+vp) be (u+vp)(u+vp2)=u2—uv+v2
to
232 (recall

p3=l and 1+p+p2=0). Further, +—f->o unless u=

v=0 whilst 8(a/3)=8(a)8(13) since 8(a), as in the other two cases, is in fact
nothing more than the square of the modulus* of the complex number
So once again condition (I) holds.
Let us show that condition (II) holds in Z[p]. Thus we are given a = u + VP
a u+vp (u+Vp)(s+tp 2
and We form —= —1+mp for
/3 s+tp 2
s —st+t
2

suitable!, m eQ. Now choose L, Mel such that* and

and write where K=(l—L)+(m—M)p. Now
(1—L)2—(1—L)(m —M)+(m 1 because both

1—Li and lm —Mi are

If we now write a — (L +Mp)fl +Kfl we see that both L +Mp and K1 = Kf3
are in 7L[p}. (Why is K1?) Further 6(K1) =1K 26(P) <6(fl) and condition (II) of
3.7.1 has been shown to hold.
Similar working applies to ZN/—i] and and is left to exercises
1(a) and 1(b).

We now set about obtaining those results which will indeed show that every
ER has the uniqueness of factorisation property (3.7.13). Keeping one eye
on the proof of the corresponding result (1.5.1) for 1 we begin with

Theorem 3.7.5 (cf. 1.4.4) Let R be an ER with respect to the norm 8. Any
two elements a, b not both 0 have a gcd c, say, expressible in the form
c = so + it for suitable s, e I?. If, further d is a also a gcd of a and b then c
t

and c/ are associates.

Proof Let S denote the set of all elements of R expressible in the form
ma +nb where m, n eR. Since la +Ob and Oa + lb belong to S we see that
S does not comprise 0 alone. From all the non-zero elements of S choose
one, c = m0a +n0b, say, with 8(c) as small as possible. Now suppose w =
ua + ob eS. By 3.7.1(11) there exist k, r eR such that w = kc +r where either
r=0 or 8(r)<8(c). Now r=w—kc=(u—kmo)a+(V—kn0)beS. Thus the
possibility r 0 and 6(r) c 6(c) is untenable [why?] and we are forced to deduce
that r=0. But then w=kc. Consequently c divides every element in 5; in
particular cia clb. Thus c is a common divisor of a and b.
and
Next, if Ci R is such that c
e and c 11b then c iimoa + n0b = c. This shows
that c is a gcd of a and b.
Finally if c and d are gcds of a and b then c = xd and d = yc for suitable
x, y e R. It follows that c = xd = xyc, whence 1 = xy. Thus x is a unit, as
required.
See footnote p. 63.
110 Introduction to rings

Notes 3.7.6
(i) Applied to Q[x], R{x], etc., 3.7.5 lies at the heart of the proof of the
well known results on partial fraction decomposition (see [32, pp. 165—70]).
(ii) For actually finding gcds in specific cases the analogue of the Euclidean
Algorithm in Section 1.4 and exercise 1.10.3 is more useful than 3.7.5.

Example 3.7.7 Find, in Z[.J—i], all the gcds of 10 + lii and 8+ i.

To find (10+ lii, 8+i) we look at
10+111 10+lli 8—i 91+78i 126 13
8+i = 8+i 65

91+ 781
(Here we write as (1 +i)+ since 1 and 1 are the nearest integers
65
to and respectively.) Multiplying up by 8+ i we get

10+lli=(1+i)(8+i)+(3+2i) (i)

Next, to find (8+i, 3+2i) we look at

8+i 8+i 3—2i 26—13i
=2—iexactly.
3+2i3+2i3—2i 13

Hence
8+i=(2—i)(3+2i) (ii)

Thus, as in the case of 7, the last non-zero remainder, in this case 3 + 2i, in
the sequence (i), (ii),... of division algorithms is a gcd of 10 + lii and 8 + i.
The others are, of course, —2+ 3i, —3— 2i and 2— 3i [because. . why?]. .

From 3.7.5 we can deduce

Theorem 3.7.8 (cL 1.4.10) Let R be an ER. If ir ER is irreducible then ir

is prime.

Proof Suppose b, c ER and irlbc. If irlb we are finished so suppose ir tb.

Since ir is irreducible its only divisors are of the form u and uir where u is
a unit. Since ir A' b the only divisors common to ir and b are the units, one
of which is Thus (ir,b)—IR and by 3.7.5 there exists, t€R such that
sir + it = But then sirc + tbc = c whence, clearly, inc as required.

This is the theorem which will tell us that decompositions into products of
irreducibles are unique. All we need now is to prove that such decompositions
exist! For this we shall need

Lemma 3.7.9 Let a, b be non-zero elements in R, b being a non-unit. Then

8(a)<6(ab).
 Euclidean rings 111

proof By 3.7.1(I) we know 8(ab). Suppose 8(a) = 8(ab). By 3.7.1(11)

there exist tn, rEER such that a =m(ab)+r where either r=O or else
and 8(r)<8(ab). If then 8(r)<8(a). But r=a(1—,nb) and so 3.7.1(I)
implies 8(r) e8(a). These inequalities show that r 0 is untenable. Hence
r = 0, whence mb = 1 and b is a unit.
We can now have (cf. 1.3.9)

Theorem 3.7.10 Let R be an ER and let be an element of R.

Then a is expressible as a product of irreducibles.

Proof Assume the desired result false! Then from amongst the non-empty
set S of non-zero, non-unit elements a of R which are not expressible as
products of irreducibles select one m, say, for which 8(m) is as small as
possible. Now m is certainly not irreducible [why not?] and so m = uv for
suitable non-units u, v ER. By 3.7.9 we find 8(u)<8(uv)=8(m). Similarly
8(v)'8(m). Hence and v are expressible as products
of irreducibles. But then so is m [being equal to the product of these two
products!], a contradiction.

We are now ready to prove the analogue for ERs of Theorems 1.5.1 and
1.9.18. Notice the change in format compared with the statements of 1.5.1
and 1.9.18. The reason for the change is given in the Remarks following the
proof.

Theorem 3.7.11 (The Unique Factorisation Theorem for ERs) Let R be an

ER andleta beanon-zeronon-unitin R.
where the ir1 and i4 are irreducibles then m = n and there is a 1—i correspon-
dence between the ir such that corresponding elements are
associates.

Proof If the theorem is false there exists a e R with decompositions as given

but in which the and ir do not pair off. From the set of all such 'nasty'
elements a, choose one for which m is as small as possible. Since . ..
and since lTm is prime (by 3.7.8) we see (exercise 3.6.9) that lTmIlr for some
1. Since ir,,, and ir are irreducible they are associates [proof?], rr = ulrm, say.
But then, in R, a = where

b is not a nasty element.* Hence the ir1 (1 i m — 1) and

the remaining pair off as associates. In particular m — 1 n — 1, so that
m = n. Since irm and ir were also paired off we see that a is not nasty after
all. This contradiction proves the theorem.
*
What if in = 1? Certainly b won't be nasty' since it will be a unit and not a product of
irreducibles. Does this affect the argument?
112 Introduction to rings

Remark The format of 1.5.1 and 1.9.18 could have been preserved in the
statement of 3.7.11 if we had, in advance, selected within each set of associate
elements in R a representative. In Z and in Q[x] obvious representatives
stand out—namely positive integers and monic polynomials. In a general ER
no such natural choice presents itself and the slight change of format seems
preferable to having to choose a system of representatives at random.

We can now summarise 3.7.10 and 3.7.11 by introducing

Definition 3.7.12 Let D be an integral domain. D is a Unique Factorisation

Domain (UFD) if and only if (i) every non-zero non-unit element a of D can
be written a = ITIIT2. . ir,,,, the being irreducibles, and (ii) if a =
. . .. =
IT'i . . . (with irreducibles ire, ir) then m = n and the and ir pair off as
associates.

Then 3.7.10 and 3.7.11 combine into

Theorem 3.7.13 Every ER is a UFD. (In particular, 7, ZN/—i], ZN/—2],

Z[p] and* F[x], F being any field, are all UFDs.)

It must be admitted straight away that 3.7.13 is not the best general
theorem In example 3.4.7(iv) we introduced the notion
that can be proved.
of principal in which every ideal is a principal ideal
ideal. An integral domain
is called (naturally enough!) a Principal Ideal Domain (PID). We give a very
quick proof (see if you can follow it through) of

Theorem 3.7.14 Every PID is a UFD.

Proof
(i) Let a,beR, R being a PID and Then [a,b]=[d] for some d=
a a and b. Following 3.7.8 we see that
every irreducible in R is a prime.
(ii) Let a e R be a non-zero non-unit. If a is not a product of irreducibles
then, certainly, a is not itself irreducible and so a = a 1b1 where not both the
non-units a1, b1 can be (products of) irreducibles. If a1 (say) is not a product
of irreducibles then a1 = a2b2 where not both the non-units a2, b2 are (products
of) irreducibles. Continuing in this way produces a sequence [a]c [ai] c [a2]c
of ideals in R. The set-theoretic union of this sequence is again an ideal
(exercise 3.4.19) and therefore principal, [z], say. But z E [z] and hence z E [a1]
for some f. Thus [z] [a1] c [a1+1] [z] and [a1] = [a1+i] follows. In particular
a141 = ap (for some r ER). But a1 = a1+1b1+1, hence a141 = a14 1b1±1r. It follows
that b1+1 is a unit—contradiction. Hence every non-zero non-unit a E R is a
product of irreducibles.
Repeating the proof cf 3.7.11 we obtain the desired result.
* F[x] is pretty clearly an integral domain and we define 8(f) = degf just as for Q[x].
 Euclidean rings 113

Is there a relationship between ERs and PIDs? Indeed there is.

Theorem 3.7.15 Every ER is a PID.

Proof Let R be an ER and I an ideal of R. If I = {0} then I = [0] and is

principal. If I {O} we choose in I an element a for which 8(a) is as
small as possible. We claim I —[a]. If now 5 El then by 3.7.1 we can find
m,r€R such that b—ma+r where r=0 or and 8(r)<8(a). In this
latter case we note r b — ma El [why?] and yet 8(r) < 8(a), contrary to the
choice of a. Thus r =0 is the only possibility. It follows that b = ma and hence
b E [a], as required. [Is there nothing else to prove?]

Confirming assertions made in examples 3.4.7(i), (ii) we have

Corollary 3.7.16 In each of the rings 1, F[x] where F is any field,

and 7L[p] each ideal is a principal ideal.

Proof 3.7.2 and 3.7.4 indicate that all these rings are Euclidean.

Remark If you re-read the proof of 3.7.15, keeping in mind exactly what
demands are made on the ring R, you may be in for a surprise. For nowhere
do we seem to have used the fact that condition 3.7.1(I) holds in R. That is,
every integral domain for which 3.7.1(11) holds is necessarily a PID and hence
(by 3.7.14) a UFD; the specially incorporated condition 3.7.1(1) is redundant!
Observations of this kind have been at the basis of research papers written
as recently as 1971 [119].

Problem 2 Let R be a commutative ring with no (non-zero) divisors of zero

and suppose R satisfies condition 3.7.1(11). Confirm that R necessarily has a
multiplicative identity, hence that R is an integral domain and consequently
an ER.

In 3.7.15 and 3.7.14 we proved that every ER is a PID and every PID is
a UFD. These assertions demand that we consider the converse questions as
to whether or not every UFD is a PID and every PID is an ER. In fact the
answer to each question is 'no'. Indeed we have already seen (exercise 3.4.22)
that the ring 7[x] is not a PID whilst exercise 14 below invites you to prove
that Z[x] is a UFD.

Problem 3 Here is a very short proof that Z[x] is a UFD. Is it a valid proof?
Proof: Z[x] is a subring of the UFD Q[x]. Hence Z[x] is a UFD immediately.

To see that there are PIDs which are not ERs (no matter how skilfully one
tries to choose the mapping b) one can consult an expanded account [130]
of a result of Motzkin [113] which says that the set of all
114 Introduction to rings

algebraic integers in the field (see exercise 3.2.14) is a PID which

is not an ER.
Concerning the number fields €P(\/d) where d is any (non-zero) square-free
integer we have the following information. The algebraic integers lying in
Q&Jd) form a subring denoted by If d 2 or 3 (mod 4), then
= {a + b 'Id: a, b e Z}, a set we have previously denoted by If
a
(mod4), then :a,bE7L and both are odd or both
2
are even}. It is known that is an ER itT d = —11, —7, —3, —2, —1, 2,
3, 5, 6, 7, 11, 13, 17, 19, 21, 29, 33, 37, 41, 57, 73. Further is a
UFD for many positive values of d (it is conjectured to be a UFD for infinitely
many positive d), whereas ford <0, it is a UFD itT d = —1, —2, —3, —7, —11,
—19, —43, —67, —163 (see [123, dated 1969]). Finally, if d.cO and if
is Euclidean then it is Euclidean with respect to the norm function as defined
in 3.6.2. For d >0 it seems to be unknown if Q{'Id] can be Euclidean without
being Euclidean with respect to the norm (see [118]).
The inquisitive reader will already be asking for values of d for which
is a UFD but not a PID. This is readily answered: there is none (see
Theorem 3.9.4).
After seeing the relative ease with which 3.7.15 and 3.7.14 were obtained
the reader may ask why we didn't go straight to those theorems and omit the
long journey to 3.7.13. The answer is partly one of learning to walk before
one can run. The professional mathematician is usually quite happy to let an
author pluck general definitions and theorems seemingly from thin air, so
long as familiar examples do appear as special cases. But this text is not
written for one with such maturity.

Exercises
I Show that the following rings together with the given norms are ERs:
(a) +, where 8(a +ib)=a2+b2;
(b)
(c) +, .) where 8(a = 1a2—2b21;
(d) (1, +, .) where* 8(z) is the number of digits in a binary representation
of zI (thus 8(—6) = 3 since, in binary notation, 110 represents 6);
2degf;
(e) (Q[x], +, .) where 8(f) =
(f) where aix') is the smallest ifor which a1 0;

(g) (Q['I—7], where Q['I—7]= {a + b(l+ a, b z} and where

'I—7))
8(a +b(1+ =a2+ab+2b2;
*
Taken from [119, p. 289].
 Euclidean rings 115

2 The following definition, clearly modelled upon Q[x], has been given ([32,
p. 147]): Let R be an integral domain. Then R is called a Euclidean ring if and
only if there exists a map {0} such that, for and h#0 in
1?, (i) i(ab)=ij(a)+q(b); (ii) ii(a+b)Cmax{n(a),i,(b)} if also (iii)
there exist m,reR such that a=mb+r and either r=0 or (n#0 and
Is it true that a ring Euclidean in this sense is also Euclidean in the
sense of 3.7.1? What about the converse question? That is, are the definitions
equivalent?

3 Apply the technique of 3.7.4 and 3.7.7 to find in, r satisfying 3.7.1(11) when
(i) a=29+20i,b=2+l9iinZ['J—1];
(ii) a= 10+ 181—2 in i[.J—2].
Find the gcds of each of these pairs of elements.
4 Show that —4+i(--1+i)[5+3i]+j i)=—1[5+3i]+(1-4-4i). Deduce
that, in the division algorithm for in and r are not unique. (They
are in 1: exercise 1.4.8.) Can you explain this?
5 Prove that if f,geO[x], where degffl and deggel, and if (f,g)=1
then there exist r, s e Q[x] such that rf+ sg = I and deg r 'C deg g and deg s C
deg f.

6 Show that cannot be made into ERs no matter how

skilfully you may try to choose 5.

7 Verify that, in (5 + — 12) = (11 — + each of

these elements being irreducible since their norms are primes in /. Surely
this contradicts the assertion of 1(c) above that Z[12] is an ER and hence
a UFD?

8 Corollary 3.7.16 says that 7 and Q[x] are PIDs. Find two distinct values
of z and a monic polynomial f such that (i) [z]={77z1+91z2+143z3:
zi,Z2,Z3e7}, (ii) [f]{(x3+2x2+3x+2)g+(x4+x3+3x2+4x+l)h:
g, Ii eQ[x]}.

9 Let R be an ER. Show that (i) in 7 for all a in R; (ii)

3(a) = 3(1) if a is a unit in R; (iii) 3(a) = 3(b) if a and b are associates in R.

10 An element in in a commutative ring R is a least common multiple (1cm)

ofelementsa,beR mR in!? imply
mjk in R. Show that in a UFD each pair of elements, not both zero, has a
gcd and an 1cm.
11 It can be shown ([6, p. 96]) that in a domain each pair of elements (not
both zero) has a gcd ill each pairjpot both zero) has an 1cm. Show that in
the elements 2 and 1 +1—3 have gcd 1 and yet have no 1cm at all.
[Hint: Use norms.] this contradict the theorem just referred to?
116 Introduction to rings

12 Let D be a domain in which each non-zero non-unit element has at least

one factorisation into irreducibles. Prove that D is a UFD if each irreducible
mD isaprime.
13 Let Q,[x] {f: f€ Q[x] and f has integer constant term}. Given f, g
Qz[x] let d be a common divisor of maximum possible degree in Qz[x]. Show
that fo =f/d and go = g/d (both in Q,[x]) have gcd 1 in Q[x]. Deduce that
z =mfo+ngo for suitable z€7L and in and n in Qz[x]. Write fo=fi+a,
go = +j3 where a, j3 are the constant terms in to and g0. Show that a =
to_(ñ)z lies in the ideal [fo, go] in Qz[x] and similarly for 13. Set c = (a, 3)
in 1. Deduce [to, go] = Ed and hence that [f, g] = [cd], a principal ideal in
QzIx]. Deduce from x = '2= 22=. that x cannot be expressed as a
product of irreducibles in Q,[x] so that Q,[x] is not a UFD. This surely
contradicts 3.7.14 (every PID is a UFD)? [This example is due to P M Cohn.]
14 Prove that l[x] is a UFD as follows: Use exercise 3.6.2 and the concept
of 'content' to write every non-zero non-unit element f as a product
u1u2 . . .
. .15
of irreducibles in l[x] such that deg u1=O and I.
Now use the solution to Problem 3 of Section 1.9 on theft and then exercise 12
above.
15 Show that a subring of a UFD need not be a UFD. [Hint: c l[p].]
Compare this with Problem 3 in Section 3.7.
16 Try to copy the second proof of 1.5.1 (p. 34) to prove that and
are UFDs. (Careful!)

3.8 Three number-theoretic applications

We interrupt our discussion of uniqueness of factorisation to give three
applications of the theory. To be sure these results can be obtained without
leaving the ring of integers (see, for example, [42], [44]). Since, however,
number theory has given rise to a great deal of present day algebra it is nice
to repay the compliment and re-obtain these number-theoretic results by
algebraic means. First we prove

Theorem 3.8.1* Let p be a (positive) odd prime in 7L. Then p = a2 + b2 for

suitable a, bE! if and only if p is of the form 4k +1. Further if p =a2+b2=
c2+d2 then either (i) a =±c and b =±d or (ii) a =±d and b =±c. That is,
the representation of p as a sum of squares is essentially unique.
only if
Proot
(exercise 1).
*
Stated but not proved by Fermat in a letter to Mersenne dated 25 December 1640. Euler gave
a proof in 1754.
 Three number-theoretic applications 117

Suppose p is a (positive) 4k +1 prime. Then, by exercise 2, pjl +x2 in

z where x = Thus pjl +x2 = (1 +ix)(1 —ix) in Since, clearly,
r1 + ix and p t 1— ix in p is not prime and hence not irreducible
in Thus we can write p = (1+ bit )(u + iv) as a product of non-units
in Z['J—fl. Since S(1+im)S(u+iv)=8(p)=p2 we see that 12+m2=
3(l+im)=p in 7, asrequired.
'Further': Fromp = (a +ib)(a —ib)we find 8(a +ib)= 8(a —ib)=p, a prime
in 7. Hence (cf. exercise 3.6.5) a +ib, a —ib are primes (i.e. irreducibles)in
Similarly for c +id, c —id. By uniqueness of factorisation in
(Theorem 3.7.13) the equality (a +ib)(a —ib) =(c +id)(c —id) yields either
(a) a+ib=u(c+id) or (j3) a+ib=v(c—id) where u,ve{1,i,—1,—i} are
units. Conditions (i) and (ii) above follow immediately.

Next we give, essentially, proof* of the FC for n = 3.

Theorem 3.8.2 The equation x3 + + = 0 has no solution in non-zero

.

elements of 7[p] 7[ In particular it has no solution in non-zero

2 j.
integers (see exercise 5).

Proof Suppose that a, /3, yE 7[p]dosatisfyx3+y3+z3 = 0. We maysuppose

a, j3, y have no common prime (i.e. irreducible) factor. [Otherwise cancel it
and start afresh.] Then (a, /3) = (/3, -y) = (y, a) = 1 in 7[p]. [Why?] Set a =
/3+y, b=y+a, ca+/3. Then (a+b+c)3—24abc=(b+c—a)3-t-
(c+a—b)3+(a+b—c)3=0. Setting ir=l—p, we get 8(ir)=3 (see proof
of 3.7.4). Hence it irreducible (i.e. prime) in 71[p]. [Why?] Also —p21r2=
is
—p2(1—p)2=—p2(1—2p+p2)=3. Hence irI3 in 7[p], and so irl24abc=
(a +b +c)3. Consequently ir3I(a +b +c)3 = 24abc, whence irlabc. [For, if
ir3124 then ir3124—33= —3 = .ir2p2, whence a contradiction since p is a
unit in Z[p].] if then Since trI3
we have irl/33+y3=—a3. Hence irla, whence ir%$ and ir%y.
Now it can be shown (exercise 8) that there are c, d e 7 such thatt /3pC 1
or and or in Z[p]. Since and
(d)3 = y3 we may assume that the /3 and y given at the outset satisfy /3 1
or and or Now we cannot have and For then
(mod 3) in 7[p]. Now ii'ja, hence irja3. Since it13 we would
haveirla3+/33+73—2=O—2=—2,andhenceirl—2+31,acontradiction.
The congruences /3 —1 y lead to a similar contradiction. Thus WLOG we
assume that f3e1 and -ya—l, that is /3=1+3A, y=—l+3p. where
*
Published posthumously.
txy (mod z) in the ring R means: there exists w€R such that wz =x —y. That is, z!x —y
in It
118 Introduction to rings

A,p€Z{p]. It follows that fl3+y3=(i+3A)3+(—i+3ji)3 is a multiple of

=
A C = 0 whilst ABC =
Recalling that we see that irIABC and hence ir3IABC.
Next /3 = —pA +p2B and -y —p2A —pB so that (A, B)= tin Z[p].
wise we would have in Z[p].] Thus from A+B+C=0 we find
(B,C)(C,A)1 in Z[p]. =
From (A, B) (B, C) = (C, A) = 1 and ir3IABC we deduce that (i)
except for the possible presence of units, A, B, C are all cubes in Z[p] (cf.
exercise 3.5.1) and (ii) IT divides exactly one of A, B, C— let us suppose C.
Set A = u1b3, B = u2x3, C = u3tfi3 where u1, u2, u3 are units in Z[p]. From
A+B+C=0 we get, on multiplying through by u71, q53+u4x3+u51130
where u4, u5 are units. Since ir3lC we have (mod it3). Since
and irtx we may write (mod ir3) and (mod ir3). (For
each element v, say, of Z[p] is of the form z1 + z2p and hence of the form
Z3+Z4IT where Zl,Z2,Z3,Z4EZ. From ir,l'z3+z4ir we deduce that -rr-4'z3.
Since -irl3 we see that z3 is not a multiple of 3, in 7. Hence or
—i (mod 3) in 7. It follows that z3, and hence s, is congruent to 1 or
—1(modir) in Z[p]. From u=±i+air (uEZ[p]) we find p3=±i+3uir±
(mod it3), as required.) We deduce that, mod ir3,
But, by exercise 3.6.6, u4=±1, ±p or ±p2. The conS
gruence (mod it3) shows that u4— I or U4= —1 are the only
possibilities. Since u4=u11u2 we find u2=±u1. But u1u2u3 is a unit and a
cube. Hence U1U2U3=±i whence
u1u2 U1
a"3
Thus the equalities A + B + C 0 and ABC = ( —) yield the equations
\IT/
3
3 3 3
q5 ±(±x) +(±tfi) =(—
'IT
Finally note that if the total number of irs (or associates of it) occurring
in the decomposition of af3-y into a product of primes is n then, since they
all occur in a,.their total number in is n —1. Repeating the above process
a further n—i times we arrive at a solution to x3+y3+z3=0 in which no
irs are present. But this contradicts the remarks of the first paragraph of this
proof.

Remark The method of 'reducing' a supposed solution to one which is in

some sense 'lesser' was invented by Fermat and is called the method of descent
(see exercise 6).
 Three number-theoretic applications 119

The third result, again due to Fermat and mentioned in a letter of 1657 is
Theorem 3.8.3 The only positive integer solution to the equation x2 + 2= y3
isx =5, y =3.
Proof We consider the uationinside the ER where we may
factorise x2+2 as (x +'J—2)(x Assume iris a prime
ible) in which divides both x and x Then
irl(x = = Since is irreducible in
we see that IT=%JTi (or —V—2). [Why no other possibilities?]
This means that = ii-Ix whence and hence 2jx2, all in
Since x e 71, the assertion 21x2 in implies the assertion 21x2
in 1. [Why?] Thus x E 271 whence x2 E 41. This means y is even so that y3 E 87/.
Since x2 + 2 871 [why not?]Ljhe assumption on ir is false. That is,
(x + — = 1 in l[v'—2LNow using uniqueness of factorisation
in it follows that x and x are cubes* in
Setting for suitable a,be7L we find x—'J—2=
It follows [why does it?] that x =
a3—6ab2=a(a2—6b2) and that —1 =3a2b—2b3=b(3a2--2b2). From this
latter equality b = ± 1 and hence 3a2 —2 = T 1. Then, necessarily, b = —1 and
a = ±1. We deduce that x = ±1 (—5) so that x, being positive, is equal to 5,
as required.

Exercises
1 Show that if n = u2+v2 is an odd integer then n 1 (mod 4)..
2
2 Show that tor each positive prime p of the form 4k + 1, 1/p—i\
(Use 2.5.4.)
3 Show that each positive prime in 71 of the form 4k +3 remains prime when
regarded as an element of Is the same true of any of the positive
primes of the form 4k + 1?
4 For a=a+ib and in find N(c43), N(a) and N(13).
Deduce that if x and y are integers each expressible as a sum of two squares
then so is xy. Hence write 11 009 as a sum of two squares. [Hint: Factorise
II 009 as a product of primes.]
5 Show that (xo, is a solution to x +y +z =0 if (xo, Yo,
.
. 3 3 3
—zo) is a
solution to x3+y3 =
6 Using the fact that every solution of the equation x2+y2 = z2 is of the
form given in exercise 3.5.2 prove that the equation x4 + y4 = z2 has no solution
*
Does ii? What about the possibility that x and x are of the form ux3 where
XE and u is a unit in
120 Introduction to rings

in integers (other than those for which xyz 0) using Fermat's method of
descent as follows.
Suppose u4+v4= w2 where u, v, w €7, uvw and w is as small as pos-
sible. Then
(i) (u, v) = (v, w) = (w, u) = 1 [otherwise a solution with smaller ii' can be
found].
(ii) We may assume u, v, w and that not both u, v are odd [no square
is of the form 4k +2].
(iii) Assuming u is odd, use exercise 3.5.2 on (u2)2+(v2)2 = to deduce
u2=a2—b2, v2=2ab, w =a2+b2.
(iv) (a, b) = 1 and one of a, b is even, the other odd [w is odd].
(v) a must be odd [a even and b odd implies —1 (mod 4), whereas
u2 1 (mod 4), a contradiction].
(vi) Useexercise3.5.2onu2+b2=a2togetu=12—m2,b=21m,a=12+m2
where (1, m) = 1.
(vii) v2=2ab=41m(12+m2)and(l,m)=(m,12+m2)=(12+m2,l)=1.
(viii) Henceeachofl,m,12+m2 isasquare, l=r2, m 12+m2=t2, say.
(ix) r4 + = t2. But rst 0 and t C w so we have a new 'smaller' solution of
the given equation, a contradiction.
Deduce the FC for n =4. [Hint: rewrite x4+y4=z4 as x4+y4= (z2)2.]
7 Following the line of argument given in exercise 6 try to prove directly
that 1' + y4 = z4 is impossible when x, y, z €7 and xyz 0. That is, start with
a solution u4 + = w4 and try to obtain a smaller. [This is another
example where a desired result is more easily obtained as a corollary of one
stronger than that sought. An earlier example ocurs in exercise 1.2.18 which
can be proved by establishing I + + = 2— first.]

8 Show that there exist integers c, d as in the proof of 3.8.2 as follows.

we see 3tu in 7. Hence u or—i (mod3). Suppose j3 al+vir(mod3)in

7[p] (the other case being similar). Since 31ir2,
(mod3) in 7[p]. Taking f=v does the trick.
(Any problem if v is negative?)
9 Show that the only solution of x2 + 4= y3 in which x and y are positive
and x is odd is x=ii, y=S as follows: Write x2+4=(x+2i)(x—2i). If
njx +2i and irk —2i then N(ir)1x2+4 and N(ir)IN(4i) = 16 (see 3.6.2). Thus
N(ir)=1 and ir is a unit. Thus (x+2i,x—2i)=i. It follows that x+2i (and
x —2i) is a perfect cube*, (a -Fbi)3, say. Then x = a3—3ab2, 2 = 3a2b —b3.
10 In verifying Fermat's Conjecture for n = 3 Gauss considered the set
G(p)={ao-i-a1p+a2p2: a0,a1,a2e7L} where p Define the norm of
a=a0+aip-1-a2p2 by Show that
*
Might it not be, say, i(a +bi)3? Cf. the proof of 3.8.2.
 Three number-theoretic applications 121

G(p) = Z[pJ = Q[V'—3] (see 3.7.4 and exercise 3.2.14 for notation). Show
that if a=ao+a1p+a2p2=b0+b1p+b2p2=f3 then IlaII==IIPII and that if

then

M =ö(u+vp)=

11 Euler's proof of the FC in the case n = 3 relied on his assertion that if

a,be7L are such that (a,b)=1 and if a2+3b2 is a cube in 7 then a+'I—3b
and a are both cubes in 1N7—3]. Whilst this is in fact p.
54]) it cannot be proved using uniqueness of factorisation in since
is not a UFD! Give an example to show that if El, if (a, b) = 1
and if a2+3b2 is a square in then neither a+sI—3b nor a -4—3b
need be a square in (see exercise 3.6.12).

Julius Wilhelm Richard Dedekind (6 October 1831-12 February

1916)
Richard Dedekind, born in Braunschweig (Brunswick), Germany, was
the youngest of four children of a law professor. He studied with
Gauss at Góttingen and took his PhD in 1852. After teaching in
Gättingen and Zurich he moved in 1863 to the Technische
Hochschule in Braunschweig where he remained until he retired in
1894. As a young man his first main interests were in chemistry and
physics but he turned to mathematics because he appreciated its
more precise arguments. Some of his early mathematical writings
were in analysis and probability. In the late 1850s he became
interested in algebraic number theory and was also disturbed by the
lack of foundation to the real number system. The former interest
found an outlet in his theory of ideals originally published in his
supplements to Dirichlet's lectures on number theory. It was
Dedekind who introduced the word field (KOrper) and who gave the
122 Introduction to rings

first university course on Galois theory. Amongst his great work is

his algebraic treatment, with H Weber, of algebraic functions. His
interest in the real numbers arose in 1858 when, on teaching
calculus, he saw there was little by way of solid basis to the various
number systems necessary for a rigorous theory of limits. His
solution to the problem, using 'cuts', was published in 1872. His
methods made him a natural supporter of Cantor's set theory.
An accomplished musician—he even composed a small opera—he
never married and lived with his sister until his death.

3.9 Unique factorisation reestablished. Prime and

maximal ideals
In Section 3.5 we indicated why Lamé's 'proof' of the Fermat Conjecture
was incomplete. In fact, the complex numbers appearing in Lamé's factorisa-
tion of x" + had already been investigated by Kummer* in his attempts to
extend the reciprocity laws of Gauss beyond the quadratic, cubic and
biquadratic cases. In more detail: Kummer had considered 'complex integers'
of the form . p being an odd prime, a
pth root of 1 and the a1 Z. For a the set Z[C] of all such integers
forms an integral domain but, for p = 23, not a UFD.
Now one can restore uniqueness of factorisation to the ring ZNJ—3] by
enlarging it to the UFD Q[V'—3]. Kummer tried the same kind of thing with
the by introducing extra divisors in the form of his so-called ideal factors.
In fact Kummer did not define the ideal factors explicitly but rather gave
conditions under which a complex integer shall 'contain an ideal prime factor
of the real prime p'.
We shall now outline very briefly the change in approach indicated by
Dedekind around 1870.
Exercise 3.6.5 shows that the domain D = = (a +
is not a UFD; in particular where and
3, 7, 1+20, 1 —20 are all irreducibles. Letting a1 be the ideal factor common
to 3 and 1+20 and letting a2, 132 be defined analogously we see that we
restore uniqueness of factorisation if we write a = 3 a1a2, j3 = 7 = 131132,
p.1+20=a1j32, v=1—20f31a2. If cox+yO is any element of D it
appears to follow that if w is divisible by ai (a2, /32 respectively) then
3k'w (31gw, 71gw, 7jvw respectively). Now vw=(1—20)(x+yO)=
(x+lOy)+(y—2x)0 whilst gw=(x—lOy)+(y+2x)0 and so the above
divisions yield and hence (3jx—y,
respectively). Thus from ailw x +yO we find 31x +y, that is x = 3z —y for
suitable z €1. It follows that w = 3z +(0 — 1)y.
Now Dedekind turns attention from the somewhat enigmatic ideal factor
a1 to the concrete subset a1 = {3z + (0 — l)y: z, y of elements of D. We
*
Ernst Eduard Kummer (29 January 1810 — 14 May 1893).
 Unique factorisation reestablished. Prime and maximal ideals 123

denote a1 more briefly by (3, 0—1). Similarly a2, /3i, 132 are replaced by
a2 "(3,0 + 1), b1 (7,0+3), b2 '(7, 0—3). One then naturally defines
a1a2=(3,0—1)(3,0+1)=(9,30+3,30—3,—6)
={9z1 +(30 +3)z2+(30 —3)z3+(—6)z4: z1, z2, z3, Z4E l}
and this is easily seen to coincide with {3y1 + 30y2: Y2 = 3(1, 0) = 3D.
Thus 3D =a1a2. Similarly Dedekind finds 7D =b1b2, (1+20)D =a1b2,
(1 — 20)D = b1a2, so that uniqueness of factorisation is restored via
a1a2 b1b2—3D (1—20)D
(3.9.1)
Dedekind noted that for each pair of numbers x, y e a1 and for each d e D
the numbers x +y, x —y and xd again lie in a1. -That is a1, and similarly a2,
b1, b2, aD, $D, ,aD, vD, satisfies the properties called for by 3.4.6. Because
of their close relationship to Kummer's ideal numbers Dedekind called his
subsets ideals.
To describe Dedekind's main result we first observe, given the ideal [n] in
7, where n 0, 1, —1, that n is a prime in 7 if and only if whenever a, be 7
are such that abe En] then either a E [n] orb e [n] (exercise 1). This motivates

Definition 3.9.2 Let R be a commutative ring. An ideal R) is prime if

and only if from a, b e R and ab e P we may always deduce a C P or b e P.

Examples of prime ideals are given in the exercises. First, Dedekind's

Theorem 3.9.3 Let 0 be an algebraic integer*, F the fieldt of all algebraic

numbers of the form where n and r,eQ, and R the
ringt of all algebraic integers in F. Then each ideal {0}, R) of R is
expressible uniquely, except for the ordering of the factors, as a producU
. of powers of prime ideals in R.

If, in particular, R is a PID and if a e R then we may write [a] = . . .

where each is principal ideal, [ir1] say. It follows that [a] = [rrt]k . . . [7r1]k =
and hence (exercise 3) that, apart from units, a is uniquely
expressible as a product of irreducible [?] elements in R. Since for rings of
algebraic numbers the converse can also be proved we have

Theorem 3.9.4 ([49, vol. 2, p. 66]) Let R be as in 3.9.3. Then R is a UFD

if and only if R is a PID.
*
See exercise 3.2.14 for the definition.
t That the sets F, R are a field and a ring respectively is not obvious. See exercise 3.2.14 and
Theorem 4.3.4.
* See exercise 3.4.17 for the definition of product.
124 Introduction to rings

To illustrate 3.9.3 we return to the above example in D = where

we have seen for example that (3, 0 — 1) is the ideal [3, 0 — 1] of D. Further,
3D is the principal ideal [3] of D. Thus in terms of ideals (3.9.1) becomes
[7]=[1+20][1—2o]=[3, O—1][3, O+1][7, 0+3][7, 0—3]
Is this the required decomposition into prime ideals? One way to prove these
last four ideals prime is to use the following definition and theorem.

Definition 3.9.5 Let R be a ring. An ideal M R is a maximal ideal of R

if and only if R has no ideal I such that Al I R.

Again, examples are given in the exercises.

Theorem 3.9.6 Let R be a commutative ring with unity. Each maximal ideal
of R is a prime ideal.

Proof Suppose Al is maximal in R but not prime, so there exist a, b e R

such that but ab eM.
Let S (respectively T) be the smallest ideal of R containing M and a
(respectively M and b). Since we have S = R. Similarly T = R. Hence
S T=R R =R [why?]. However S={m1+ar1: m1EM,r1ER} whilst T=
by definition each element of ST is a sum of
{tn2 + br2: ,n2 E M, r2 E R } and
elements of the form (m1+ar1)(m2+br2)=mim2+in1br2+ar1tn2+abr1r2,
each of which is in M. Thus ST c M R, a contradiction.
We leave the checking that [3, 0 — 1] is a maximal ideal of D to exercise
7. We shall say a little more about prime and maximal ideals in exercises 4.3.

Exercises
1 Show that if nEZ and if n 1,—i then n is a prime integer if? [n] is a
prime ideal in 1. Show that [0] is a prime ideal, but not a maximal ideal in 1.
2 Show that in Z[x,y,z] the ideals [x]c[x,y]c[x,y,z] are all prime, but
none is maximal. [Hint: For the first part work directly from the definition.
For the second, consider the proper ideal [x, y, z, 2].]
3 Let R be an integral domain and let a, b eR. Show that [a] = [b] if and
only if a, b are associates. Deduce that if [a] is a prime ideal then either a = 0
or a is irreducible.
4 Let R be the direct sum Z®Z (see exercise 3.2.9). Let a be the element
(1, 0). Show that [a] is a prime ideal in Rand yet a=(l, 0) is the product
of two non-unit elements. Doesn't this contradict the last assertion in exercise
3?
5 Show that in Q[x] a non-zero ideal is prime if? it is maximal. [Cf. 3.9.6.
Corollary 3.7.16 might help.]
 Unique factorisation reestablished. Prime and maximal ideals 125

6 In check:
(i) 1 are all irreducibles;
3, 7,
(ii) that [3, 1] is not a principal ideal. [Hint: suppose it is, [a], say.
Then 3 = ay and 1 =at5 for suitable y, S E
(iii) setting A = [3, —1], B = [3, ij, c = [7, +3],
D=[7,v'—5—3] it follows that [3]=AB, [7]=CD,
[1

7 Show that A = [3, 1] is maximal (and hence prime) in as

follows.
Every element of A is of the form 3(a
(i)
(c+d'I—5)=(3a—Sd—c)+(3b—d+c)'I--5=X+Y'J—5,say.
(ii) —
(iii) Deduce 1 A so that A Z[%J—5].
(iv) Show that if x 72[sJ—5]\A then X, Y can be chosen so that
Y=y andX—x=1 or2.
(v) Deduce that any ideal T of which contains A and x + y
must contain 1 or else 2 (and hence 3—2= 1). Since 1 T we find T =
8 Show that:
(i) in a UFD, if x is irreducible then [x] is prime;
(ii) in a PID every prime ideal other than [0] is maximal;
(iii) in a PID, if x is irreducible then [x] is maximal;
(iv) in a PID x is a prime element if and only if [x] is a non-zero prime ideal.

9 Show that in the ring of even integers the ideal {4z:ze7L} is maximal but
not prime. (Thus 3.9.6 may fail if the condition that R has a unity is omitted.)
10 Show that if 12 are ideals in a ring R and if '2 and '2 then
L n12 is not prime in R.
11 Let A, B, P be ideals with P a prime. Show that: (AB c P A c P or
B c P). [This condition is actually equivalent to 3.9.2 in the commutative
case. Furthermore it is taken as the definition of prime ideal in non-
commutative rings where 3.9.2 iS too restrictive.]
12 Let (7, +, s) be the integers with the usual +, but with multiplication
a b e 7. Show that the subset 27 is a maximal ideal which is
not prime. Doesn't this contradict Theorem 3.9.6?

13 Show that 7 has infinitely many distinct maximal ideals.

14 Theorem .3.9.6 says that in each 7,,, each maximal ideal is prime. Do
there exist integers m for which 7,,, has primes which are not maximal?
15 Prove that each non-zero ideal in has the form [x"] for some
nG 7' u {0}. Deduce that has just one maximal ideal.
126 Introduction to rings

310 Isomorphism. Fields of fractions. Prime

subfields
Having finally confirmed in 3.7.13 that for each field F the polynomial ring
F[x] is a UFD, there is no need for apology if nothing more than intellectual
curiosity leads one to ask if each polynomial ring F[xi, x2,. . , on n letters
*

is also a UFD. One feels instinctively that the answer must be 'yes' (how
could it be otherwise?) but since, for n F[xi,x2,. . . ,xj
is not a PID (cf.
exercise 3.4.22), there is no chance of applying 3.7.14 directly. On the other
hand Z[x] is equally not a PID (exercise 3.4.22) and yet it can be shown to
be a UFD (exercise 3.7.14) essentially by journeying into the UFD Q[x]. We
answer the question concerning F[x1, X2,.. . , in Section 3.11. In this
section we set up the apparatus for the job.
Now we can think of F[xi, x2 x,,] as the polynomial ring where
R =F[x1,x2,. . . For n R is, like 1, an integral domain which is
not a field. So we are led to ask: Can we enlarge R to some field R°, say,
just as I can be enlarged to 0? We show that it can. To help us do this
properly we introduce the following idea which is of fundamental importance
throughout algebra.
Suppose (R, +, ) is a ring with elements a, b,.. . and that (R', $, 0) is a
ring with elements a', b',. . in 1—1 correspondence (see 2.6.5) with those of
.

R so that a corresponds to a', b corresponds to b', etc. It may be that for

all a, b eR the elements a +b and a and a'Ob'
also pair off under this correspondence. If so, then each calculation performed
in R can be changed immediately to one in R' simply by 'dashing everything'!
That is, (R, +, .) and (R', 0) are structurally identical (in other words
'essentially the same'—see Remark (ii) in Section 1.8) even though their
elements and their binary operations have different names.
Formally this idea of sameness becomes

Definition Let (R, and (S, ®, 0) be rings. We say that R and S

(or, more accurately, (I?, +,•) and (5, ®, 0)) are isomorphic iff there exists a
1—I map i/i:R—,S from R onto S such that for all r1, r2 in R: (i) (r1 +r2)çU==
and (ii) . r)il, = r1iji0r2/i.

If R and S are isomorphic we write R S. The map fr is called an

isomorphism.

Fig. 3.2
 lsomorphism. Fields of fractions. Prime subfields 127

It may help you to look at Fig. 3.2.

Thus tb isa 1—i and onto map which addition and multiplication.
Examples 3.10.2
b)
U) Let he the ring [is it one?] of all matrices of the form (
Y)
where Then the map given by is

an isomorphism between the field of complex numbers and N2(IU. In particular

N2(R) is a field (exercise 7(v)).
(ii) The map 0 from the ring Zo to the direct sum 12$13 (see exercise 3.2.9)
given hy* = (th2, th3 shows that the two rings mentioned are isomorphic.
For any algebraic system an investigation of the set (later, the group) of all
the isomorphisnis of the system with itself can be illuminating. Such an
isomorphism, called an automorphism, exhibits a kind of internal symmetry
of the system. As examples we mention
(iii) For any ring R the identity map i: R -* R given by ri = r for all r E R is
clearly an automorphism of R. The fields 0 and R possess no automorphisms
apart from the identity (exercise 8) but the field C certainly has at least one
other automorphism, namely that given by (a + ib = a — ib. [Can you think
of any others?]
(iv) The map of Q[xi, 12, 13, 14] to itself in which each occurrence of x2,
13, 14 is replaced by, say, 14, 1i, 13 respectively is clearly an automorphism
of O[xt, 1:, 13,14].
In this section the isomorphism concept is used in two ways: (i) to enlarge
rings to bigger (but nicer) rings (Theorem 3.10.3); (ii) to elucidate the structure
of special subrings of given rings (Theorem 3.10.9). These uses depend upon
the following remarks.
(v) If T is a subring of a ring S and if O:R -* T is an isomorphism we say
that 6 is an embedding of R in S. Of course R itself need not be contained
in 5, hut something structurally identical to R, namely T, is. The algebraist
tends to think of R as replacing the structurally identical subring T, often
going so far as to rename each element rO of T by calling it r, As an example,
given any ring R, we can define the ring R[x] of polynomials just as we did
for0inl.6.land,usingthemapA:R-*R[x]givenbyrA=(r,0,0,...),think
of R itself as a subring of R[x]. In particular, we see how each of F, F[x1],
12],... may he regarded as a subring of its successors and in particular
of F[xi, 12
(vi) Let S be the ring Rx7L of exercise 3.3.3. The map ji:R-
{(k, 0): k e R}cS given by rp. = (r, 0) is an isomorphism. Thus can every ring
be embedded in a ring with a unity element. Exercise 9 strengthens this
statement slightly.
(vii) Without going into details the map 6: 7 -* 27 given by zO= 2z is clearly
not an isomorphism. Why not? What does 7 have that 27 hasn't? (See exercise
7(u).)
*
Here denotes the element th of 'k• Cf. 2.4211)1
128 Introduction to rings

Remarks It is clear that, given any concrete ring, an infinite number of

distinct but structurally identical rings can be constructed merely by altering
the names of the elements, but keeping their relationships intact. For this
reason the algebraist, a person more interested in relationships between
elements (i.e. in structure) than in the elements themselves, tends to collect
all algebraic systems into heapst of isomorphic ones, regarding all systems in
the same heap as mutually indistinguishable. (One may consequently definet
an algebraist to be a person who can't tell the difference between isomorphic
systems!) The algebraist thus sees his task as the describing, in some manner,
of a member out of each heap and the finding of a procedure which will
determine, given two systems, whether or not they belong to the same heap.
In particular we shall show in Section 3.12 that any two rings satisfying the
axioms Al through to I of Section 1.2 lie in the same heap whilst in Section
4.5 we find a very simple criterion for the isomorphism of two finite fields.
Some basic properties of isomorphisms are given in exercise 7.
We can now prove the main result of this section. Concrete examples are
given in 3.10.6.

Theorem 3.10.3 Let D be an integral domain. Then there exists a field F0

containing a subring D* isomorphic to D and such that every element of F0
is of the form where u, v ED* (and, of course, v 0).

Note 3.10.4 In case D the field F1, will be nothing more than C. Indeed,
we model the proof of 3.10.3 on our understanding of how C is obtained
from /. Intuitively, elements of C are of the form where r, se7L and

Of course = is possible even though r1 r2 and S1 $ But then r1s2 = s1r2.

Si 52

Thus a rational number is really a representative of all those for which

5 Sj
rs1 = sr1. [If you do get a bit lost in the following abstraction, work through
the entire proof assuming that D = 1 so that the element {a, b} introduced
below may be thought of as

Proof of 3.10.3 Consider the subset P ={(a, b): a, b ED, b of D xD.

Write (a, (c, d) if ad= cb in D, Then is an equivalence relation on P.
[This needs proof!] We the equivalence class containing (a, b) by {a, b}
and the set of all such classes in P by F0. On F0 we define ® and C by
d+b c,b . d} and {a,b}C{c,d}={a c,b . d}. [To what
A better name for heap' is equivalence class! (See exercise 7(vii).)
t The definition of a topologist as one who can't tell the difference between a tea cup and a mint
with a hole is much more fascinating!
§ Temporarily: Confusion with the set containing just a and b is unlikely.
 Isomorphism. Fields of fractions. Prime sub fields 129

do + and . refer?] An immediate problem is: Are and C well defined?

(Cf. Section 2.4.) That is, if {a, b} = {A, {c, d} = {C, D}, do
B} and if

we have and
{A . C, B D}? In fact everything turns out all right. [Prove it!]
We next check that (F0, $, C) is a field.
Al: Let {a,b}, {c,d}eF0. Then {a,b}${c,d}={ad+bc,bd} whilst {c,d}®
{a, b}={cb +da, db}. [Why?] Since ad + bc = cb +da and bd = db in D [why?]

we see that $ is commutative as required.

M2: For {a,b}, {c,d}, {U,V}EFD we have ({a,b}C{c,d})C{u,v}=
{ac, bd} C {u, v} = {(ac)u, (bd)v} = {a (cu), b(dv)} [why?] = {a, b} C {cu, dv} =
{a, b}C({c, d}C{u, v}) as required.
A3: For {a,b}eF0 we have {a,b}${0, l}={al+bO,bl}={a,b}. Similarly
{0, l}${a, b} = {a, b} whence {0, l} is a zero element for F0.
M4: Given {a,b} l})eF0 we see that {a,b}C{b,a}={ab,ba}, the
element required by M3. Note that {b, a} is in F0 since a 00 [why not?].
Ml, A2, M3, A4 and the distributive laws are left to you.
Now D itself is clearly not in F0: the elements of F0 are equivalence classes
{a, b} and none of these belongs to D. The algebraist can however see D in
FD. First, the subset D*={{d, l}: deD} of FD is easily seen to be a subring
of FD. Second, the map 0: D_.D* given by dO = (ci, I } is an isomorphism.
For,given d1,d2eD we have (d1+d2)O={d1+d2,l}={d1,l}®{d2,l} [are
you sure?]=diO$d20. Similarly (d1 . d2)O ={d1 l}{d1, l}C{d2, l}
d10 d20.
. is clearly onto, is also 1—1. For, if d10 = d20 then
Further 0, which
{d1, 1}={d2, l} which means [by definition] (d1, l)—(d2, 1) and hence d11 =
d21 in D.IThus D aD* as required.
Finally, given {a,b}EF0 we may write {a,b}={a,l}C{l,b}=
{a, l}C{b, l}' with both {a, l} and {b, l}ED* as required.

The observant reader will note that 3.10.3 does not strictly fulfill our promise
to 'enlarge D to a bigger (but nicer) ring'. The algebraist is happy since he
can't see any difference between D and D*. Exercise 9 below should satisfy
the uneasy reader.
Notes 3.10.5
(i) The field F0 is called the field of fractions or quotient field of the integral
domain D. If we think of D rather than D* as lying in F0, we can then denote
the element {a,b}={a, 1}C{b, 1}1 of F0 by the symbol aCU' or, more
briefly, by ab1, or even by
(ii) If D1 and D2 are isomorphic domains then their corresponding fields of
fractions F0, and F02 arc isomorphic. (How could it be otherwise?) Further,
if F is any field and if D is a subring of it then the subset F = {ab 1: a, b ED,
b 0} of F is a subfield of F such that P Summarising: a given domain
D can always be embedded in a field F, say. If D cannot be embedded in any
propert subfield of F then F F0. F0 is unique (up to isomorphism).
A subfield (subring) is called proper 1ff it is not the whole field (ring).
130 Introduction to rings

(iii) Even when applied to 1 itself our construction is not devoid of interest.
For in Section 3.12 we show that 1 is the unique (up to isomorphism) well
ordered integral domain. Thus (ii) will allow us to define 0 as the (unique)
field of fractions of the unique well ordered integral domain.
(iv) Dropping the commutativity of D it seems clear 3.10.3 will prove for
us that every 'non-commutative domain' can be embedded in a 'non-
commutative' field, that is, a division ring. In fact this assertion is false (exercise
15). Moreover there exist non-commutative domains which cannot be embed-
ded in division rings no matter what method may be tried. The Russian
mathematician A I Malcev* found such an example in 1937.
(v) Thinking of D as a subring of FD leads one easily to think of D[x] as a
subring of FD{x].

Examples 3.10.6 According to 3.10.5(i) the fields of fractions of Q[x], Q[x, y]

and Z[-12] comprise respectively the sets

and

In this last example we may write

a+b'J2 (a+b\12)(c—d'12) C
—= 2 2 =r+sV2
C 2d

for suitable r, s EQ. We thus see that, in the notation of exercise 3.2.14, the
field of fractions of is the field Notice that is the
smallest (sub)field (of which contains 0 and Similarly one sees readily
that the field of fractions of = {a + b +c a, b, c E Z} is the
smallest (sub)field (of R) containing 0 and What is less obvious (but
true, see 4.3.4) is that this field of fractions coincides with the set {u +vJ2+
w(J2)2: u, w E Q}. The big surprise here is that multiplicative inverses of
numbers of this form are again of this form. (Try writing {2+J2-4(J2)2}'
in the form u ± vJ2 + where u, v, w EQ.)

Notation 3.10.7 Given fields F c E and elements a, j3, y,.. . E E it is

customary to denote the smallest subfield of F which contains F, a, j3, 'y,
by F(a, p, .). In this notation the fields of fractions of Q[x], 0[x, y],
are 0(x), Q(x, Qe12), Q(J2), whilst the smallest subfield
of C containing, for example, 0, and i is denoted by i).

A nice little application of a field of fractions to solve a problem posed

solely in terms of domains is
*
Anatoly Ivanovich Malcev (27 November 1909 —7 July 1967).
 lsomorphism. Fields of fractions. Prime subfields 131

Example 3.10.8 Let D be an integral domain, a and b elements of D and

+ in
m,n€7L ,where (m,n)=1. If a =b and a
rn it it
b then a=b in D. For:
Since (m, n) = 1 there exist s, t €1 such that sm + tn = 1. Form F0. Assuming
thats cOandt>Owefind(a")t =(b")t irIDcFD and(am)s =(bm)5 inF0. But
then a = a'"a ins = bfhbms =b in F0. Hence a =b in D.

Although we shall not need it until 4.5.7, now seems an appropriate time
to present the following embedding theorem first proved in E great
paper of 1910. It shows that the vast totality of all fields is built around a
rather restricted collection of 'basic' fields, namely the prime (sub)fields.

Theorem 3.10.9 Let F be any field and p(F) the intersection of all subfields
of F. Then p(F), the prime subfield of F, is isomorphic to 0 or to one of the
finite fields 74.

Sketch of proof Let D ={m iF: m Z}. Since 1F €p(F) we see that D cp(F).
There are two possibilities: (i) for m, n 1, m n in 1F n 1F; (ii) for some
pair in, neZ, with m#n, we have in lF=n 'F'
In case (ii) suppose WLOG that m > n. Since (in — n )1F = °F there exists a
smallest positive integer r, say, such that riF = O,. It follows that r is a prime
in 1. For, if not, then r = St where 1 <s t <r and we would have (s lF)(tiF) =
rlF=OF. Since F is a field it would follow that either or =°F,
both of which would contradict the choice of r. Consider then the subring
D ={m iF: in e Z}={klp: k cr} of F. It is not difficult to check—mind the
well-definedness—that the map 0: 74—÷D given by ñO=n 1F is an isomorphism
from Zr onto D. Since r is a prime Z,, and hence D, are fields. Since D cp(F)
and since p (F) cannot have any proper subfields [can you prove this?] we see
that p(F)=D
In case (i) it is not difficult to check—no well-definedness problem arises—
that the map 0:1 -÷ D given by zO = z 1 F is an isomorphism between 1 and
D. Then, by 3.10.5(u), the subfield F={(,nlF)(niFYt: m,n eZ,n is a
subfield of F isomorphic to field of fractions, namely 0. Further, since
p(F) is afield, we have Pcp(F). Thus, as above, as required.

Exercises
1 Are the following maps isomorphisms between the given rings?
(i) given by (a =a
(ii) 0:21-'3lgivenby(2n)0=3n;
(iii) 0:C-Cgivenby(a+ib)0=b—ia;
(iv) 0:K5-+K5 given by (ao+a1A +a2A2+a3A3)0 =(ao—a2)+(a3—a2)A +
(a1—a2)A2—a2A3=a0+a1A2+a2A4+a3A whereA isacomplexfifthrootof
1 and the a1 1. [K is for Kummer—see Section 3.9];
*
Ernst Steinitz (13 June 1871 —29 September 1928). Crelle's Journal, Vol. 137, pp. 167—309.
132 Introduction to rings

(v) O:Zoo-+16$Z10 given byt 01600 =(016, where <60;

(vi) Let U, cr be the real and one of the complex cube roots of 2 respectively.
Define 0: Z[u]-* 4u] by (a + bu +cu2)0 = a +bcr +cu2.
2 On Z define ® and C by aEBb=a+b+1, aOb—ab±a+b. Show
(Z, $, C) is a ring with unity element. Is (1, ®, C) (1,
3 Let R1=M2(Z) and R2=M2(R1). Is denotes n Xn
matrices.]
4 Explain intuitively why Z[J3]. Back your intuition with a proof.
[Note: this example not only says that a + b 4 a + b is not an isomorph-
ism. It says that no isornorphism can be found at all—no matter how clever
a choice of mapping you might try to make.]
5 Let H {(a, b, c, d): a, b, c, d e R}. Define W componentwise and C by
(a,b,c,d)El(A,B, C,D)=(aA—bB—cC—dD, aB+bA+cD—dC, aC—
bD + cA + aD + bC — cB + dA). Show that (H, E, C) is isomorphic to the
ring of quaternions (Section 3.2). [H is, of course, for Hamilton.]

6 Show there exists A =

(X )1)
such that A '2 and yet A'7 = '2.
[Hint: think of 17th roots of 1 in C and use 3.10.2.]
7 Let q5: (R, +, . ) -* (5, ®, C) be an isomorphism. Show that
(i)
(ii) If R has a unity 1R then is a unity for 5;
(iii) (—a)cfi = 9(aq5);
(iv)
(v) If R is commutative so is 5; if R is a field so is 5;
(vi) qY1:S-*R exists and is an isomorphism;
(vii) If i/i:(5,$,GD)-*(T,m,C) is an isomorphism then so is
Deduce that 'isomorphic to' is an equivalence relation on the class of all rings.
[Hint: for (i) look at ORqS = (OR +OR)qS 0R4 $ORqS in S—or see 4.2.6.]
8 After noting that Q g c C and that C each possess
non-identity automorphisms show that the only automorphisms of 0 and of
the identity maps. [Hint: By 7(u) For It if reR4 then r4 =
'Ir4 Hence, if a<r<b with a,b in 0 then
in Rq5 =R.]
9 Let R, S be rings and 0:R-+S an embedding of R into S. Set R0=
{rO:reR}cs. Let be a set of symbols disjoint from R and in 1—i
correspondence with the elements of S\R0. Let R * 5* u R. If r E R, ifs * 5"
and if corresponds with s, define r = (rO$s)*. Taking the hint from
this show that R * can be made into a ring isomorphic to S. Then R * actually
contains the subset R as a subring. [In particular for each domain we can

See footnote p. 127.

 U[x) where U is a UFO 133

find a field actually containing it, whilst for each ring R we can find a ring S
with unity such that R c S.]
10 Describe other than in form the fields of fractions of the domains:

11 Show that if 9:D -D1 is an isomorphism then FD [Hint: try the

map Ô defined by {d1, d2}O = {d10, d20}.]
12 Show that between 7 and 0 there are infinitely many pairwise non-
isomorphic rings each with the same field of fractions.
13 Is [Such an isomorphism would surely have to be a non-
obvious one. Note that their fields of fractions coincide.]
14 Identify where in 3.10.3 commutativity and also the non-existence of
non-trivial divisors of zero is used. Show that the existence of a unity element
in D is not really required. What is the field of fractions determined by 21?
15 Obtain the result of 3.10.8 working entirely inside the integral domain D.
16 Show that the conclusion of 3.10.8 is false in M2(Z).
17 Let F be a field and S a subfield. Show that if S 0 or if S 74, then F
contains no other subfield isomorphic to S.
18 By definition the characteristic of a ring I? is the smallest positive integer
n (if there is one) such that na = °R for all a R. If there is no such integer
we say R has characteristic 0. Show that if R has a 1 then n is also the
smallest positive integer for which n 1 Deduce that in an integral domain
the characteristic is 0 or a prime p.
Show that, for all a, b in a domain D of characteristic p, (a ± b = a ±b
k t. (Including the case p = 2, k = 1?) Deduce that the map a4 a"
is a 1—1 homomorphism of D into D. Need 0 be onto?

3.11 U(x] where U is a UFD

We now indicate how to prove that, for any field F, the ring F[xi, X2,. . . , X,,]
isa UFD. Regarding F[x1, X2 as the polynomial ring where
= F[xi, x2,. . ,
. the statement about F[xi, X2 will follow from
the more general
Theorem 3.11.1 Let U be a UFD. Then U[x] is a LJFD.
One way to prove 3.11.1 is to copy, as far as possible, the corresponding
proof for 7[x]. Since the complete proof of that result is somewhat scattered,
we shall identify the lemmas, theorems, etc. needed in the proof of 3.11.1
indicating the modifications required. (One advantage of this to the reader is
134 Introduction to rings

that he will have the motivation to reread the appropriate proofs very care-
fully—to ensure the author isn't cheating!) An alternative proof is described
in exercise 1.
We begin with proofs of parts of exercises 3.7.12 and 3.7.10.

Theorem 3.11.2 In a UFD an element it is irreducible if it is prime.

if only if
Proof 4: is true in any integral domain (exercise 3.6.12(a)). 4) : let it be
irreducible in the UFD U and suppose irlab where a, b 6 U. Then iw = ab
for some c in U. Writing a = . b . .. ir, c =
. ir', products .

of irreducibles, we find it ir' = in .. . it..By definition of

. .

UFD, it is an associate of a or a ir. Thus iria or irib, as required.

[This proof needs a slight modification if a or b is a unit. Why can't both
a and b be units?]

Theorem 3.11.3 In a UFD D every two elements a, b, not both zero, have
a gcd unique except for units. So then does every set of n elements, not all zero.

Proof If one or both of a, h is a unit then 1 is the required gcd. Otherwise

where LetI={ii,...,im}andf=
{J',. . . , , r} and N =
be subsets, as large as possible, of M = {l,. . .

{1,. . . ,s} respectively

such that, for k (1 c k ct-n), ltik is an associate Uk7T.
of Then g = is a common divisor of a and h. Indeed it is a gcd
. . .

of a and b. [This seems pretty obvious. Let us see what is required by way
of proof.] If gig where gla and Sib then for suitable A, B and G in D we have
5 = Gg, a = Ag = AGg, b = Bg = BGg. Comparing these factorisations of a
and b with those above it quickly follows, by uniqueness of factorisation, that
either G is a unit or G = WIiTQ = . . . where A, ji 1, wl
. . .

and W2 are units in D, the 6 M\I and the e N\J. Applying uniqueness of
factorisation to G in this case we see that is an associate of some
But this contradicts the definition of I and J. Thus G is a unit in D and g is
a gcd, as required.
The uniqueness of g to associateW and the extension of the result to
sets of ii elements are immediate.

Question In connection with the above proofs, why did we not just say 'Copy
the proofs of 1.4.4 and 1.4.10'?

Notation 3.11.4 (cf. 1.9.7) For elements Z0, z1 EU we denote their

gcd, unique apart from occurrence of units (by 3.11.3), by (z0, z1,. . ,zj. .

Definition 3.11.5 (cf. 1.9.3) A polynomial .

isprimitiveif(z0,z1 zn)l.
 U[x] where U is a UFO 135

Theorem 3.11.6 (cf. 1.9.10) If F=y0+yix+' '+YmXm and G=

z0+ziX +' +z,,x" E U[x] are both primitive then so is FG.

Proof In 1.9.10 alter '(other than 1 and —1) in 1' to '(other than units) in
U' and note that every irreducible element in U is prime—by 3.11.2.
Definition 3.11.7 (ci. 1.9.11) Let F = z0+z1x + e U[x]. Each gcd
is called the content of F.

Theorem 3.11.8 (cf. 1.9.14) If FEU[x] and if F=gh where

then we can write F = GH where G, H E U[x], deg G = deg g and deg H =
degh.

Proof In 1.9.14 alter 1 and to U [what has happened to and deduce

ac = bdu where u is a unit in U.
Corollary 3.11.9 (ci. 1.9.15) If FE U[x] and if one cannot write F as a
product of polynomials of smaller degree in U[x], then one cannot write F
as a product of polynomials of smaller degree in the bigger ring Fu[x].

Theorem 3.11.10 (ci. the solution to Problem 3 in Section 1.9) Let FE U[x]
be irreducible (in U[x]). Then F is prime (in U[x]).

Proof Replace! andt by with (a,b)=1'

by 'a,bEU with (a,b)=1'.
To prove 3.11.1 all we need, after 3.11.10 and exercise 3.7.12, is to show
that every non-zero non-unit element of U[x] is expressible as a product of
irreducibles of U[x]. But this is easy. For if FE U[x] write F = cG where c
is the content of F and G is primitive* in U[x]. If c is not a unit in U then
write it as a product of elements which are irreducibles in U and hence in
U[x] (exercise 3.6.2). By an argument on degree we can clearly factor G
into a product of primitive irreducibles in U[x]. This does it.

Exercises
I Prove 3.11.1 by copying the second proof of 1.5.1 as follows. Assuming
U[x] is not a UFD let F P1P2. . Pr = qlq2 . . . q5 be a polynomial of smallest
.

degree with distinct factorisations into products of irreducibles in U[x].

Assuming in =degp1 n =degq1a and in set
.
G = p1x" q, and suppose a and b are the leading coefficient of Pt and
.

q1 respectively. Show that .. . q,. Now either

aF = bG or deg (aF — bG) C deg F; hence in either case p ijaqi. Deduce [care-
ful!] that p1jq'. Thus Pt and are associates, contradicting the obvious initial
assumption that they are not.
* G necessarily primitive if the content of F is factored out? Can you prove this?
Is
136 Introduction to rings

2 (i) Prove the converse of 3.11.1, namely: If U is a ring and if U[x] is a

UFD then U is a UFD.
(ii) Prove that if U[x] is a PID then U is a field.
(iii) What can you say about U if U[x] is Euclidean?
[Hint: for (ii) think first of Z[x].]
3 Let U be a UFD and Ietf=ao+aix+' Show that iff
is primitive and ifp is a prime in U such thatptao,plai,. . , p . andp2tao
then f is irreducible in U[x]. (That is, prove Eisenstein's test in U[x].)
Deduce that y4+3x2y2+4x7y +2x is irreducible in Q[x, y]. Is y6+3xy4+
2),2
3x + x3 irreducible in O[x, y]?

4 Use exercise 3 to show that x3 — 6x2 + 4ix +1 + 3i is irreducible in

1][x].

5 State the analogue for U[x]of 1.11.6. Hence write x3+yx2+(y —2y2)x —
as a product of a linear factor and a quadratic in x with coefficients in
Z[y]. Treating it as a polynomial in x with coefficients in Z[y] find the content
C of P—y2x3+y3x +y3x2—y4—2y4x—x3—yx2—yx+2y2x+y2. Hence
factorise P completely into irreducibles in Z[x, y].

6 Show that if F1, F2 U[x] and if F1, F2 have a common divisor of degree
greater than zero in Fu[x] then they have a common divisor of degree greater
than zero in U[x].
7 Let f, g be polynomials in R[x, y] having no common factor except units.
Show that there exist polynomials s, t y] and u R[x] such that u =
sf + tg. [Hint: Let K be the field of fractions of R[x] and think of f, g as being
in K[y]. By exercise 6,f, g have no common divisors other than units in K[y].
Thus there exist v, w K[y] such that 1 = vf + wg.]
Deduce that if a,b€R are such that f(a,b)=g(a,b)=O then u(a)=O.
Show that there are at most finitely many pairs (a, b) R x R such that
f(a, b) = g(a, b) = 0. (This result is used in algebraic geometry to determine
the so-called irreducible varieties in the plane R X It)
8 Is a UFD? What about and These latter two
rings are isomorphic, aren't they?

3.12 Ordered domains. The uniqueness of 1

In Section 1.2 we supposed that the set Z of integers, whatever they be,
satisfies the axioms Al through to I (excluding M4). Replacing axiom I by
axiom M4 it is easy to give many distinct examples, including 0 and which
satisfy axioms Al through to P (including M4). Now 0 and ff1 are clearly not
isomorphic [can you pinpoint one reason why not?] and so the question raised
in Section 3.1 regarding the essential uniqueness of 1 does seem to be
somewhat less trivial than it at first appears.
 Ordered domains. The uniqueness of 1 137

We close this chapter by proving that 7 is indeed unique up to isomorphism,

thus fulfilling our promise of 3.1O.5(iii).
A preliminary definition is helpful (cf. axiom P of Section 1.2).

Definition 3.12.1 Let (R, +, be an integral domain. R is said to be ordered

itT R contains a non-empty subset R4 such that
(i)
(ii) each element of R belongs to exactly one of the sets R {O}, R - where
R - = {—x: x E R
R + is called a set of positive elements of R.

Examples 3.12.2
(i) According to Section 1.2, 7 is an ordered domain. Intuitively, so are the
fields Q and It
(ii) The ring 7NJ2] of all a + where a, b 7 sustains two distinct sets
of positive elements as was noted in 1.2.3. So does the ring 7[x] (exercise 2
below).

Remark Each ordered domain must satisfy the analogue of property Z of

Section 1.2 (exercise 6).

Now let R be any ordered (integral) domain. The generalisation to R of

axiom I in Section 1.2 is easily written down. Do it by replacing N and 1 by
and 1R respectively. Furthermore if, just as we did for 7 in 1.2.4, we
write a <b whenever b — a E R the generalisation to R of condition W in
Section 1.2 reads
W: Every non-empty subset of R4 contains a least member. That is, if T c R +
and T0 then T contains an element t such that t <z for every other z e T.
Following the terminology of Section 1.2 we say that R + is well ordered.
It is not difficult to follow through the proof of 1.2.9 to show that these
generalisations of I and W are logically equivalent. Thus it is reasonable to
make

Definition 3.12.3 Let R be an ordered domain. If R satisfies axiom I

(equivalently axiom W) we call R a well ordered (integral) domain.

We can now establish

Theorem 3.12.4 Let (C, +,.) and (D, +,) be two well ordered domains*.
Then (C, +, +,•)..
Proof Since C is a domain, C has a unity element 1c such that 1c
Since C is ordered, C contains a subset of positive elements. Hence either
K
For simplicity we use ± and . for each ring.
138 Introduction to rings

or In this latter case 1c = (—ic)(--lc)E so that in any

event 1c Now set Uc ={n n t}. We have just seen that
furthermore is closed under addition. Thus Uc c Likewise, given
x=n Uc, we see that x + 1c = n + = (n + lyic for some n €7.
Since C satisfies axiom I we deduce that Uc = Ct Since, by axiom P,
C__C*u{0}uC,sinceocOlcandsince_(nlc)__(_n)lc(exercise3.3.2)
we see that C = {m lc: in Z}.
Next suppose ini > rn2 inland set t = rniic — in2lC = (in1 — m2)lc. Because
and That C.
Now all remarks made so far apply equally to D. In particular D =
{in in 7}.Further,sincem1 = in2lC andm1l0 = tn2lD,
the correspondence 4) : C -÷D given by (in = in is indeed a map (i.e.
it is well defined). Further it is i—i and onto D.
Finally, if c1,c2€C then C2=in21c for suitable m1,m2€Z and
we have
(cl+c2)q5=(mlic+m21c)qS=((rnl+m2)ic)q5 =(ml+m2)iD
=mlin+m2iD=(mlic)q5+(m2lC)q5=clqS+c2q5.
Similarly
(c1c2)qS m21C)qS = (inlm2lC)cb = mlm2lD
=mjlD
It follows that 4) is an isomorphism, as required.
We thus see that whatever the integers are, there is, up to isomorphism,
only one algebraic system satisfying precisely the axioms Al, A2, A3, A4,
Mi, M2, M3, D, P and I. A similar sort of result relating to the real numbers
is given in 4.4.2 where it is far from QbviOus that each of two natural
constructions of R based upon the rational numbers leads to the same result.

Problem 4 It is not difficult to check that the only automorphism of 7 is the

identity mapping. Suppose D is an ordered domain with just two units I,
and — 'D and that the only automorphism of D is the identity. Is D necessarily
isomorphic to the ring of integers?

Exercises
1 Is the (integral) domain of Gaussian integers (p. 108) an ordered domain?
2 Prove the assertion made in 3. 12.2(u) about Z[x]. [Hint: Letf a1x' +
a1x' where a1 0, a1 0 and where 0 i j be a typical element of Z[x].
Define subsets P1, P2 of Z[x] by P1 = {f: a1 >0 in 1} and P2 = {f: >0 in 7}.
Show that P1 and P2 are subsets of positive elements of l[x] in the sense of
3.12.1. Show that P1 P2.] Can you find a third such subset distinct from Pi
and P2?
 Ordered domains. The uniqueness of 7 139

3 Can /7 be ordered? (See exercise 2.4.11.)

4 We know can be ordered. Can it be well ordered?
5 Show that 1D is the smallest positive element in a well ordered domain
D. Show that this assertion is generally false if D is only assumed to be ordered.
6 Let R be an ordered domain. Show that R has characteristic 0.
7 Can a well ordered domain sustain two distinct sets of positive elements?

8 Let D be an ordered domain and FD its field of fractions. For e FD define

°<Fh iff0.c0ab. Show that the set Fj={XEFD:0<Fx} is a set of positive

elements as defined by 3.12.1. (A field which is ordered as a ring is called an
ordered field. Setting D=7L we see that 0 is an ordered field.). Write if
Show d<Fh if Is it true that if bc<0ad?

9 7 can be ordered in only one way (1.2.3). Is the same true of 0? (One
way to order 0 is given in exercise 8.)

10 An ordered domain D is called Archimedean if for each r E D there

exists n E 7 such that r <n Show that 7 and 0 are Archimedean but that,
with P2 as defined as in exercise 2 above, 7{x] is non-Archimedean. Deduce
that its field of fractions 7(x) is a non-Archimedean field (see exercise
8). Is there any way to order /[x] (bcsides using P1) to make it Archimedean?
II Let D be an ordered domain in which there is no element t such that
OD<t<ID. IsD/?
4
Factor rings and fields

4.1 Introduction
In Chapters 1, 2 and 3 we have given examples of fields. In this chapter we
introduce the concept of ring homomorphism to help us make a deeper study
of field structure. Our main objectives are:
(i) to describe the structure of all finite fields and then to complete the
algebraist's dream as far as these fields are concerned by giving a simple
criterion for telling any two of them apart (see 4.5.8);
(ii) to prove the impossibility of solving affirmatively the old Greek problems
of angle trisection and cube duplication (see Section 4.6);
(iii) to give an almost totally algebraic proof of the Fundamental Theorem
of Algebra (see 1.11.8 and 4.8.1).
On the way we shall take time off to show (Section 4.4) how one can
construct the field R without making any assumptions other than that we
already have to hand the field 0 of rational numbers and how, in the same
spirit but very much more easily, one can produce C using only known
properties of It Having already indicated (Section 3.10) how to construct 0
from / we shall have built C on the same foundations as those of /. Further
discussion can be found in [48] and [112].

4.2 Return to roots. Ring homomorphisms.

Kronecker's theorem
Kronecker, whose attitude to mathematical 'existence' has been noted in the
Prologue, had no difficulty in accepting the 'reality' of each individual integer.
The creation of the integers he ascribed, in a much quoted statement,* to
God. He could also accept the existence of each rational number since it can
be created (by man*) in a finite way using only integers. But irrational numbers
such as and even when constructed by Dedekind's method (see Section
4.4), could not be said to have an existence since their definition required
prior acceptance of infinite subsets of rational numbers. In fact Kronecker is
reported to have asked Lindemannt after the latter had proved (1882) that
ir is not an algebraic number, 'Of what use is your beautiful investigation
*
Die ganzen Zahien hat der liebe Gott gernacht, alles andere ist Menschenwerk.
C L F Lindemanri {12 April 1852—6 March 1939).
 Return to roots. Ring homomorphisms. Kronecker's theorem 141

regarding ir? Why study such problems when irrational numbers do not exist?'
Nevertheless Kronecker showed how to create irrationals like in a way
acceptable to him by employing the same method Cauchy had earlier used
to introduce (see Section 4.4). The same ideas can be taken further
and so the following theorem is usually credited to Kronecker.

Theorem 4.2.1 (Kronecker's Theorem) Let F be any field and let f =

a0+a1x +' +aj." eF{x] be such that f has no root in F. Then there exists
a field E containing F as a subfield such that f has a root in E.

Terminology 4.2.2 E is a good letter to use for this new field since one may
regard F as an extension (field) of F.

To obtain 4.2.1 in a modern way we introduce the concept of ring

homomorphism.

Definition 4.2.3 Let (R, and (5, $, 0) be rings and 0 :R -+S a map. 0
is called a homomorphism or homomorphic mapping from the ring (R,
(in)to the ring (5, e, 0) if! the following conditions hold for all a, be R:
(i) (a+b)0=aoebo;
(ii) (a

Remarks
(i) The requirements of a homomorphism are thus similar to those of an
isomorphism (3.10.1) except that there is no insistence on 0 being 1—1 nor onto.
(ii) Even if 0 maps R onto S we do not say (cf. 3.10.1) that 'R and S are
homomorphic' nor that 'S is homomorphic to R'. The symmetrical nature of
the binary relation 'is isomorphic to' is missing in the case of homomorphisms.

Examples 4.2.4
(i) The map 0 1-÷ 1,, given by mO = th defines a homomorphism from
(1, +, onto (Zn, S, 0). The only homomorphism from (Zn, $, 0) to (1,
is the trivial homomorphism, that is the map i/i 1,, -. Z given by =0.
(ii) The map 0 : 118 -*13 given by rh = ph3 (where rh1 denotes, temporarily,
the element th of defines a homomorphism from $, 0) onto
(13, $, 0).
(iii) The map 0 1 -* I given by zO = 2z is not a ring homomorphism from
(1, +, .) into itself. [Why not?]
(iv) The maps 0,tfr:Q[x]-*Q defined by .

+an are homomorphisms. (In the notation of

1.11.1, is the map given byfili =f(1).)
(v) The map 0:Q[x]-*R given by
a0 + + a homomorphism of (Q[x], onto
+,.)
142 Factor rings and fields

(vi) The map 0 :Z-*Q given by zO isa homomorphism of (7, +, into

= (f)
lc)g -
(1 + i)!(s3J2)k
(vii) The map 0 Q[x, y]-. C given by 'y = is a
homomorphism.
The homomorphisms in (vi), (vii) are not onto and only that in (vi) is 1—1.

Notes 4.2.5
(i) 0 R -÷5 is above all a map from the set R to the set S. We shall however
talk briefly of 'the homomorphism 0 from the ring R to the ring 5' or write
at length 0 (R, +, -÷ (5, $, 0), whichever seems more hdpful.
(ii) The subset RU = {rO: r e R} of S is called the (homomorphic) image of
R under 0. RU is easily shown to be a subring of S (exercise 4).
(iii) The reason for insisting on some 'preservation of structure' as in (i) and
(ii) of 4.2.3 (rather than looking at, say, arbitrary mappings from R to 5) is
dealt with in the following Remarks and in exercise 7.
(iv) One can describe a homomorphism 0 :R pictorially as in Fig. 4.1.

Fig. 4.1

Compare this with the pictorial description of isomorphism given in Section

3.10. The shaded region in R represents the kernel of 0 (see 4.2.8).

Remarks
(i) Homomorphic images have well been likened to photographs ([18, p. 511,
[37, p. 69J). Just as in a photograph of some object certain information
present in the original—that round the back of the object—is lost, so in a
homomorphic (as distinct from an isomorphic) image RU of a ring R some
information present in R (essentially that contained in ker 6—see 4.2.8) is
lost. On the other hand; the information remaining in RU might, if R0 is a
'simpler' ring than R, be more readily extracted than from the original R. Of
course, to be useful, RU will have to bear some structural similarity to R—hence
the demands made for the preservation of structure in 4.2.3(i), (ii). The point
 Return to roots. Ring homomorphisms. Kronecker's theorem 143

being made is that, for investigating a ring R, the investigation of a different

but not totally unrelated ring will generally prove more beneficial than staring
blankly at R. At the level of this text observations of this sort show up rather
better in the investigation of finite groups but 4.2.10 below provides a relatively
non-trivial ring-theoretic instance of the above remarks.
(ii) Some texts give the impression, by introducing them together, that the
concepts of homomorphism and isomorphism play roles which are mild
variants of one another. In fact they serve somewhat dissimilar purposes. The
isomorphism concept brings some order to the plethora of algebraic systems
(see Remarks in Section 3.10) by identifying 'algebraically identical' ones.
(More fundamentally it defines the rather loose term in inverted commas!)
Clearly the homomorphism concept is unsuitable for this purpose. As noted
above, homomorphisms act in an investigative capacity. Although it may
appear that isomorphisms cannot play such a role (surely any abstract system
is just as difficult to penetrate as an isomorphic one?) an isomorphism between
two systems, concrete or abstract, can often be revealing. See for example
5.9.5, 5.9.6 and 6.4.4.
Because of the similarity of 3.10.1 and 4.2.3 it is only to be expected that
homomorphisms will have at least some of the properties possessed by isomor-
phic mappings. There is, for example (cf. exercise 3.10.7),

Theorem 4.2.6 Let 6: (R, +, .)-+ (5, $, 0) be a ring homomorphism. Then

(i) °R6=O5; (ii) for each aeR, (—a)6=e(aO); (iii) (a—b)6=aOebo.
Proof
(i) Hence ORO=(OR+OR)O=ORO®ORO, in S. Thus, immedi-
ately, ORO = [Why?]
(ii) = ORG = (a +(—a))6 = aO®(—a)6, in S. Thus, immediately, (—a)6 =
9(aO). [Why?]
(iii) Use part (ii).

At this point ideals (see Section 3.4) enter again. We see that they are
intimately related to homomorphisms by proving

Theorem 4.2.7 Let 6 :(R, +, $, 0) be a ring homomorphism. Then

the subset K = {k e R: kG = is an ideal of R.

k1,k2eK. Then (k1—k2)6=k169k26=0500505. Thus

Proof Let
k1—k2eK. Next let keK and rER. Then (k .r)9k90r9050r005
Thus k r K. In a similar manner one proves r k K, so that K is an ideal,
as asserted. [Question: Is K non-empty?]

Definition 4.2.8 The above ideal K is called the kernel of 6. We often denote
it by ker 6.
144 Factor rings and fields

Examples 4.2.9 The kernels of the homomorphisms given in 4.2.4 are

(i) For 0: the ideal [n]={kn: ke7L). the whole ring /,,.
(ii) The ideal comprising {O, 9, 12, l} in
(iv) For 0: the ideal [x] = {xf:fe Q[x]} in Q[x]. For i/i: the ideal comprising
all f in Q[x] whose coefficients have sum zero, that is
{(x— l)f:feQ[x]}=[x— I].
(v) The ideal [x2—2] in Q[x].
(vi) The subset {0}.
(vii) The least ideal of Q[x, y] containing (in other words, the ideal generated
by) the polynomials x2—2x +2 and y3—2. [Can you see why?]

Applications 4.2.10
(i) A nice application of homomorphisms and kernels yields another proof
of 1.9.10.
Suppose that and . are
primitive in i[x] but that every coefficient in their product is divisible, in 1,
by p. The homomorphism 0 :1 -+74 defined by zO = I clearly 'extends' to
the homomorphism defined by (a0+a1x +.. =
74+7f1x Clearly ker 6* comprises all polynomials in Z[x] whose
every coefficient is divisible by p. Apply 0* to the product fg. Since p divides
all the coefficients of fg we see that =0 in But (fg)9* zf0*gO*.
Since is an integral domain (same proof as for Q[x]; see p. 112) we
deduce that 10* = O or gO* = O. Thus f (or g) belongs to ker 0*. That is, in
1, each of the as (or bs) is divisible by p, a contradiction.
(ii) f = + x3 +1 is irreducible in i[x]. For if = gh in Z[x] then (notation
as in (i)) fO*=gO*h0* in 12[x]. Clearly x5+x3+I has no root in £2 so
h0* must be of degrees 2, 3 respectively. But this is impossible by inspection.
(Try it!) Hence is irreducible in 12[x], whence so is f in 71[x]. [Why is it
better to map i[x] onto 12[x] rather than onto 13[x]?]

Problem 1 Can you prove Eisenstein's test using the same technique?

Remark For a quick proof of exercise 1.9.12 using the abovç methods see
[38, p. 181].

Ideals are very useful in helping us construct new rings from old, as our
proof of 4.2.1 will show. The following theorem is invaluable. The proof will
take a while.

Theorem 4.2.11 Let (R, +,.) be a ring and I an ideal of R. Then there
exists a ring (S, $, C) and a homomorphism 0 : R S such that ker 0 = I,
exactly.

Thus we need (i) to construct 5; (ii) to define 6; (iii) to check ker 0 = L We

first construct S.
 Return to roots. Ring homomorphisms. Kronecker's theorem 145

Notation 4.2.12 Let (R, be a ring, I an ideal and r0 an element of R.

We (rather naturally) denote by the symbol r0 + I the subset {ro + i: i I} of R.

Note 4.2.13 It is perfectly possible for r0 r1 in R and yet r0+I = r1 +L

Indeed r0+I =r1+I if and only if r1—r0€I (exercise 13). As an example,
suppose I is the ideal {5k: k Z} in the ring 1 of integers. Then 3+1 =
{3+5k: ke71}={—22+5k: k€Z} [can you see why?]=—22+I whilst, of
course, —22. Note that —22—3e1.
You may recognise 0+1, 1 +1,. . . , 4+1 as being new names for the ele-
ments 0, 1,.. . , 4 of 74. In fact the following construction of S from R is
merely a generalisation of our construction of 1,, from 1 (Section 2.4) using
the notation r + I instead of, say, or just P. In particular, replacing R and
I by 1 and {Sk: k €Z} in 4.2.14, 4.2.16 and 4.2.19 you'll retrieve familiar
results about

Next we need

Lemma Let r1, r2 €R. Then

4.2.14 either r1 + I = r2 + I or else
(r1+I)n(r2+1) = 0, the empty set.

Proof Suppose that (r1 + I) n (r2 + I) 0. Then there exists an element t in

both r1 +1 and r2+L By definition this means that it = r1 +i1 and that it =
for suitable i1 and i2 in L But then r1 = r2 + i2 — i1 and hence, for each i €1,
r1+i—r2+i2—i1+i. Since i2,i1, i€I we see that r1+i€r2+I. It follows that
I r2 + L An identical argument shows that r2I + r1 + I whence r1 + I =
r1 +
r2+I, as required.

Notes 4.2.15
(i) r1 +1 is called the coset of I in R determined by r1. We shall meet a
similar concept in the theory of groups in Section 5.7.
(ii) 4.2.14 says that any two cosets are either identically equal or else totally
disjoint.
(iii) Since each r R lies in the coset r +1 which it determines, we see that
the various cosets of I in R entirely 'cover' R and, by (ii), do it in a pairwise
non-overlapping manner. Does this remind you of something studied in
Chapter 2? If not read Section 2.3 now. For an alternative way of presenting
the coset concept see exercise 15.

Denoting by R/I the set of all (distinct) cosets of I in R we show below

how to turn R/I into a ring in a natural way.

Problem 2 Why would the last sentence be redundant if its last four words
were omitted?
146 Factor rings and fields

Theorem 4.2.16 Let (R, and Ru be as above. Then the set Rh can,
by appropriately defining operations of 'addition' and 'multiplication', be
made into a ring (later to be taken as the ring S in 4.2.11).

Proof [Proceeding somewhat naively] we attempt to define binary operations

® and 0 on R/1 as follows. Let a + I, b + I e R/L Set
(a +I)®(b+J)—(a +b)+J
and
(a+I)0(b+I)=a
b +1 both lie in R/I [but once again there is the
problem—see Section 2.4—of the well-definedness of and 0. The problem
here is this: If X =a +1 =c +1 and if Y=b +1 =d+1 then according to the
definition proposed above XE) Y=a 'b+I on the one hand and c
on the other. Similar remarks apply to X Y. Consequently, we must check
that, from the equalities a + I c + I and b +1= d+ 1, there follow the equali-
ties
a +1 =c +1 we may deduce (exercise 13) that a =c +1 for
suitable I 11 Similarly from b +1 = d + I we deduce b = d + k for suitable
keL Then a since
why?
it follows, as in the proof of 4.2.14, that a b +1—c d+1, as required.
The corresponding problem of showing that (a +b)+1 = (c +d)+I is left
to the reader in exercise 14.
To check that (Rh, ®, 0) is a ring we only have to check the six axioms
Al, A2, A3, A4, M2, D as required by 3.2.2. We check only the axioms Al
and M2 leaving the others to exercise 14.

AxiomAl Fora+I,b+IeR/Jwehave
(a+1)$(b+I)—(a+b)+1=(b+a)+J=(b+1)$(a+1),as required.

AxiomM2 Fora+1,b+1,c+IeR/1 we have

{(a+I)Q(b+J)}Q(c+f)={ab+J}O(c+J)=(ab)c+J=a(bc)+1=
(a +1) O{bc +I}= (a +1) 0{(b +1)0 (c +I)}, as required.

Definition 4.2.17 The ring R/1 is called the factor ring* of R by L

Examples 4.2.18
(i) If R 0[x] and if! =[x2—2] then every element of RI! can be written
in the form a+bx+I (where a, bcQ). Further, is clearly (?;
exercise 20) isomorphic to the field of all real numbers of the form a +
b a
* Some authors prefer the quotient ring. We avoid this since it would otherwise be too
easy to confuse this concept with that of quotient field. (3.10.5(i).) In any case quotient fields
comprise quotients (i.e. fractions); factor rings, despite the (standard) notation Rh, do not.
 Return to roots. Ring homomorphisms. Kronecker's theorem 147

(ii) If R = l[x, y] and if I = [x2, y2± 1]—the ideal of l[x, y] generated (see
3.4.7(iv)) by x2 and y2+ 1—then every element of has the form a +bx +
cy+dxy+I. Here KR/I, H, El> is isomorphic to the ring of all quadruples
(a, b, C, d) where a, b, c, del, where addition H is defined componentwise and
multiplication by (a, b, c, d) H fi, y, ö)=(acx—cy, ba+afl—dy—c5,
ccx +ay, dcx +c/3 +b'y+aS).

Prohlem 3 Am I right here in (ii)? Can you see how I can be fairly sure I'm
right even before I check anything. (At first glance it might not be apparent
that the quadruples even form a ring under the given H and H!)

As stated in 4.2.16 we shall take the ring S required in 4.2.11 to be the

ring RI! constructed above. Thus we are left to prove

Lemma 4.2.19 Let KR, +, ), land (R/I, 0> be as in 4.2.16. The mapping
o:R -* RI! given by rO = r +1 is a homomorphism from R onto RI! and
ker 0 =! exactly.

Proof Since every element of RI! has the form a +! where a e R and since
aO=a+l by definition, it is clear that 0 maps R onto R/I.

One shows, similarly, that (a +b)0 = aO Thus 0 is a homomorphism.

Finally, let a e ker 0. Then, on the one hand, aO = °R + I whilst, on the other,
aO = a +!, by definition of 0. Consequently, °R +! = a +! and so a e! follows.
Thus ker 0 c L Conversely if i e! then iO = i +! = °R +!, the zero element of
RI!. Thus ! c ker 0. It follows that ker 0 =7, as required.

With the completion of 4.2.19, we also complete the proof of 4.2.11.

Remarks The passage from the ring (R, to the ring (RI!, 0) (via
the homomorphism 0) achieved by ! to zero' can be colourfully
described as the process of killing off I. The elements of!, and they only, are
sent to the zero element of RIl, but do note that this does not necessarily
imply that if a and b are distinct elements of R which are not in! then aO, bO
are distinct in RI! (see Fig. 4.1 on p. 142). For it may be that a, b lie in the
samecosetof!inR inwhichcaseaO=a+!=b+!=bO.
We now give the

Proof of Theorem 4.2.1 [Although f has no root in F it may still factorise

in F[x] into (irreducible) factors (each of degree at least 2).] Suppose that
f = gg2. . . g. where the g are irreducible in F[x]. Let ! be the (principal)
ideal of F[x] generated by g and set S = [Thus the elements of S are
cosets p ± 1 where p e Fix 1.]
148 Factor rings and fields

Now S is a homomorphic image of a commutative ring with 1 (that is, with

unity) and is thus itself a commutative ring with 1 (exercise 6). We show that
S is a field.
Suppose g = b0 + b 1x + bmxm and let p + I be a non-zero element of
S. Then p I and so p is not a (polynomial) multiple of g. Since the only
divisors of g in F[x] are its associates and the units of F[x] we see that, in
F[x], (p, g) = 1. It follows from 3.7.5 that there exist, in F[x], polynomials r,
s such that
rp+sg=1 inF[x]
Thisimplies rp+sg+I=1+I inS. [Why?]
Itfollowsthat (rp+I)$(sg+I)= 1+1 in S, [Why?]
whence (r+I)C(p+I)=1+I inS. [Why?]
Thus r +1 is a left (and hence two-sided [why?]) multiplicative inverse for
p +1 in S which is thus proved to be a field.
Clearly F is not itself contained in S but an isomorphic copy is (see exercise
'1'

19). Hence, by exercise 3.10.9, we can construct a field S such that F

itself is actually a subset of Let* y denote the element of S which corresponds
under ifr to the element x + I of S. The element b0 + b iy + b,,ym in S then
corresponds to the element
Thus
that is, y is a root, in of the given polynomial g. It follows immediately
that y is a root, inS, off.

Remark Part of the above result can be represented pictorially by Fig. 4.2.
F[xl

Fig. 4.2

The remarkable thing about the proof is not that we can make a structure
S in which b0+b1x + is zero—this is easily arranged by killing off
b0 + + brnx is that, when one does this killing, S is a field which
furthermore contains an undamaged replica (i.e. isomorphic copy) of F as a
subfield.
*
This use of rather than S is not common, but the author feels it renders this particular proof
more easily understood than that usually given.
 Return to roots. Ring homomorphisms. Kronecker's theorem 149

Problem 4 0 has no roots of either x2+1 or x2+x +1. Let I be the ideal
of 0[x] generated by x2 + 1 and x2 + x + I. Then surely 0[x]/I is a field
containing the required roots. Am I right?
Although we shall have no need of it until Section 4.5 it seems now
appropriate to obtain the
Corollary 4.2.20 Let E F[x], F any field. Then there exists a field F such
that F 2F and in E[x] f factorises completely into linear factors.
Proof (Outline) If f = . . expresses f as a product of polynomials
irreducible over F, find F1 as in 4.2.1 so that has a root a1, say, in F1.
Then, in Ei[x], f has at least one linear factor, x —a1. Factorise f into
irreducible polynomials in Fi[x] and repeat the process, with F1 in place of
F. After at most deg f steps we arrive at a field F with the desired properties.
Is the field F of 4.2.20 uniquely determined? By no means: take f = x —1
and F =0 whence F can be 0 or or C.... But if we insist that F be as
small as possible then we do indeed have
Theorem 4.2.21 Letf E F[x], F any field, and let F1, F2 be extension fields
of F such that
(i) f factorises into linear factors* (x—cx1)(x— . (x— in E1[x] and
. .

(ii) F1 (respectively F2) is the smallest subfield of F1 (respectively F2) which

contains F and all the roots (respectively
Then F1

Proof We ask the reader to be patient—until 7.2.5.

This (essentially unique) extension field F1 is called the splitting field
of f over F.

Exercises
Homomorphisms
I (a) Let 0 and i/i be maps 0[x]-* 0[x] defined by
"
(a0 + a1x + )0 = a0 + a3 (x + 2) . + a,, (x +

2n
'+a,,x )ili=ao+aix 2
Determine whether or not 0, ifr are homomorphisms.
(b) Are the maps zO——z on Z and (x+iy)ifi=x from C to R
homomorphisms?
*we also say thatf splits in E1[xl (or over E1).
150 Factorrings and fields

2 Describe all possible ring homomorphisms (i) from onto 74; (ii) from
712 into 78; (iii) from "2 into (iv) from 7 into 7. Can you, from (i), (ii),
(iii), decide precisely when there exists a homomorphism from 7,,, onto 7,,?
(a b)
3 Is the map 0 from (M2(7), e, 0) to (7, +, by =ad—bc =
d
det a homomorphism?

4 Let 0 :(R, +, ')-÷(S, ®, 0) be a homomorphism.

(i) Prove that RO is a subring of S. Give a concrete example where RO is
not an ideal of S.
(ii) Prove that, for r1, r2 ER, r10 = r20 ill r1 — r2 E ker 0.
5 Let
0 : (R, +, a

be a a homomorphism
from R onto (5, $, 0). Prove that (5, ®, 0) is a commutative ring with a 1.
Is this proof still valid if 0 is not assumed to be onto 5?
7 Let (R, +, be a ring and let (5, ®, 0) be a ring which is not commutative
and does not possess a 1. Given that there exists a mapping from the set
R onto the set S what can you say about (R, +, Can you say any more if
you are told is a homomorphism?
8 Let L2(l) be the set of matrices of the form ( °) where a, b, c e 7.
Define 0 :L2(l)-*l by ( °) =a. Show that 0 is a homomorphism onto 7
and that L2(1) is not a commutative ring.
9 DefineO:Q[x]-*Cby
(a0 + a1x + = a0+ . + + 7.)n
Find ker 0. If Q[x] is replaced by l[x]—so that all the a1e 1—how is your
answer changed?
10 Show that a homomorphism 0: <R, +, )—*<S, $, 0) is I — I if and only
if ker 0={OR}.
11 Let F be a field. Use exercise 3.4.14 to describe all possible homomorphic
images of F.

Cosets
12 Write the coset 9 + 2x + 2x2 +7x3 +4x5 +1 where I = [x2 + 5] in the form
a +bx +1. Write down three more elements of this coset.
13 For the ideal! and the elements r,, r2 of the ringR show that r, +1 = r2 +1
iff r,=r2+j for suitable icL Deduce that if a+bx+cx2+dx3+I=
 The isomorphism theorems 151

A +Bx +Cx2+Dx3+Jwherej [x4+2x +7]inQ[x]thena =A,b =B,c C

andd=D mO.
14 Complete the proof of 4.2.16.
15 Let I be an ideal in the ring R. For r1,r2eR define iffr1—r2eI.
Show directly that — is an equivalence relation on R and that the cosets of
I in R are the corresponding equivalence classes.
16 Let S be the subring of O[x] [is it one?] comprising all polynomials
whose coefficients of odd powers of x are all zero. For each p Q[x] denote
the set {p+s:s€S} by p+S. Show that if then (pi+S)n
(p2+S) =0. Attempt to make a ring out of these subsets of O[x] by defining
(pi+S)®(p2+S)=(pi+p2)+S and (pi+S)O(p2+S)=pip2+S. Is your
attempt successful? Demonstrate that it is—or explain why it is not. [Hint:
x+S=(l +x)+S. Hence their squares are equal? Cf. exercise 2.4.12.]
Factor Rings
17 To which more familiar ring is RI! isomorphic when (i) R = O[x], I = [x];
(ii) R =O[x], I=[x +2]; (iii) R =Z[x], I=[x,4]; (iv) R =1[x], I=[x2+1];
(v) R = Z[x], I = [2x + 1]. Which of these factor rings are fields? [Hint: In
(ii) think of O[x] with x +2 'put equal to 0'.]

18 Let g E F[x]. Show that is a field if and only if g is irreducible in

F[x]. Is a field? With how many elements?

[x +x+1]
2

19 Show that if g €F[x] and if g is irreducible in F[x] then in the map

C the subring F of F[x] is mapped isomorphically into
[g] [g]

20 Establish the claim made concerning isomorphism in 4.2.18(i).

21 Show that x3+x + 1 is irreducible in O[x]. Find r, S€ O[x] such that
r(x3+x + 1)+s(x2-f-2x +2)= 1. Hence write in the form a +bx +cx2+I the

x2.
Q[x]
multiplicative inverse of 2+ 2x + +1 in the field —y---'' being the principal

ideal [x3+x + 1]. [See the proof of 4.2.1.]

Extension Fields

22 Find inside C an extension of the field 0 in which x4 + 2x2 —4 has a root

but over which x4 + 2x2 —.4 does not split into a product of four linear factors.

4.3 The isomorphism theorems

Because of the extreme importance of the homomorphism concept, not only
for rings and groups (see Section 5.10) but for all algebraic systems, we offer
152 Factor rings and fields

the reader an opportunity to strengthen his understanding of the concept in

the case of rings by proving three quite involved results concerning
homomorphisms. Their collective name, 'The Isomorphism Theorems', indi-
cates their standing. In Chapter 6 we shall make non-trivial use of the group-
theoretic versions of these theorems. It must be admitted that, strictly speak-
ing, we do not require 4.3.1, 4.3.2 and 4.3.3 in this text, the applications we
give being obtainable just as naturally by other routes. On the other hand
the application 4.3.4 of 4.3.1 is useful later.
We begin with

Theorem 4.3.1 (The First Isomorphism Theorem) Let C : (R, +, .) (5, +,.)
be a homomorphism from R onto S with kernel L Then the ring (5, +, .) and
the factor ring (Rh, €, C) are isomorphic.

Proof [There is an obvious mapping which, if the proposed assertion is true,

one might expect to prove the theorem: namely the map if' : Rh -. S given
by (r +I)i/i = rO. The first thing we must check is . what?*]
. .

Suppose r1 + I = r2 + I. Then r1 — r2 €1. It follows [since I ker 0, that

(r1 —rjO=O5 and hence] that r10=r20. Thus the proposed mapping i/i is
well-defined.
Now i/i is a homomorphism. For {(ri+I)$(r2+I)}Ø ={(ri+r2)+I}cfr =
(r1 + r2)O = r10 + r20 = (r1 + í)ç& + (r2 + I)ifi,
whilst {(ri + I) C (r2 + fl}if, =
{rir2+I}ifi =(rir2)O =r10 - r20 —(ri+I)qi .
Clearly i/i maps R/I S. Finally, if (ri +I)ifi = (r2+I)Ø then r10 = r20
onto
whence (r1 — r2)O = that r1 — r2 EL That is, r1 + I = r2 + L Thus Rh and
so
S are isomorphic [via cli] as required.

Remark This result tells us that all homomorphic images of a ring R are of
a special kind—they are all (up to isomorphism) obtainable as factor rings R/L

Next, let A be an ideal and B a subring of the ring R. Defining A + B to

be the subset {a+b:aEA,bEB} of R one checks easily that A+B is a
subring of R, that A is an ideal of A +B and that A nB is an ideal of the
ring B. We then have

Theorem 4.3.2 (The Second Isomorphism Theorem) Let A, B, R be as just

described. Then the rings (B +A)/A and B/(A n B) are isomorphic.

Proof Define O:B—+(B+A)/A by bO=b-i-A. Since b+a+A=b+A, it is clear

that U maps B onto (B + A)/A. We leave you to check that U is a homomor-
phism. Suppose xekerU. Then x+A is the zero element O+A of (B+A)/A.
Thus xeA. But xeB, whence xeAnB. It follows that ker OcAnB.
*
Well-definedness.
 The isomorphism theorems 153

Conversely, ifxeA n B then xO=zx+A=0+A since xe A. Thus ker 0= A n B

and from 4,3.1 we deduce

Theorem 4.3.3 (The Third Isomorphism Theorem) Let I, K be ideals in the

ringR such that IcK. Then I is an ideal in K, K/I={k+I:kEK} is an
ideal in R/I and

Proof We let the reader check the statement about L Define the map
0 :R/I-*R/K by (r+I)0 =r+K. 0 is a well-defined map since from r1+I =
r2+I we get r1—r2e1 c K whence r1+K=r2+K. Clearly 0 is onto. Again
we leave you to check that 0 is a homomorphism. From 4.3.1 it follows
that R/K Now every element of Rh is of the form r + I (r e R) and
ker 0
(r+I)ekero if! r+K=0+K=OR/K. But this is the case if! rEK. Thus
ker 0 ={k +1: k eK}=K/L In particular K/I is an ideal of R/I (by 4.2.7)
and R/K by 4.3.1.

These three theorems can be represented pictorially as in Fig. 4.3.

'B n A

{O}
Fig. 4.3

Applications of the isomorphism theorems We have already applied 4.3.1

twice in proving 4.3.2 and 4.3.3. Other applications appear in Section 4.4
and the exercises below. Here we give just one, substantially extending the
examples given in 3.10.6 and 4.2.18(i).
Let F and F be fields with F c F and let a e F. a is called algebraic over
F (cf. exercise 3.2.14) if! a is a root of some (monic) polynomial in F[x]. We
let F(a) denote (cf. 3.10.7) the least subfield of F containing F and a. We
then have

Theorem 4.3.4 Let E, F, a, and F(a) be as above and let Ma e F[x] be a

monic polynomial of least degree having a as a root. Then Ma is unique and
154 Factor rings and fields

isirreducible* in F[x]. If, further, degMa = n then every element of F(a) is

expressible uniquely in the form fo+fia +' for suitable €F.
F[x]
Moreover F(a)

Proof If a were a root of two distinct monic polynomials Ma, Na each of

degree n then a would be a root of Ma Na and hence of some (monic)
polynomial of degree n — 1, contradicting the minimality of n. Next, Ma is
irreducible. For, if Ma = RaSa where deg Ra, deg Sa <deg Ma then we should
have OMa(a) Ra(a) Sa(a). Then at least one of Ra(a), Sa(a) would be
o since E is a field. Hence a would be a root of at least one of Ra, Sa, a
contradiction [why?].
Now consider the map U :F[x]-* F given by gO = g(a) (cf. 4.2.4(iv)). Clearly
o is a homomorphism. What is its kernel? Let k E ker 0. Use 1.10.1 to write
k = mMa + r where either r =0 or r 0 and deg r <deg Ma. It follows that
0 = kG = m (a )Ma (a) + r(a) = r(a) [why?]. But this is untenable unless r is the
zero polynomial of F[x] [why?]. Thus r =0 and k = rnMa, a polynomial
multiple of Ma. This shows that (F[x])O, which clearly contains both F and
a and is itself contained in the field F(a) is, by 4.3.1, isomorphic to
But this factor ring is a field (exercise 4.2.18) since Ma is irreducible. Thus
(F[x])0 =F(a), the smallest field containing F and a. Using 1.10.1 again we
see that every element of (F[x ])0 is of the form s (a) for suitable s of degree
n — 1 in F[x] as claimed. Finally the uniqueness of this form is proved by
an argument similar to that proving Ma = Na.

Examples 4.3.5 — —

(i) Taking F = F = Q and a = we see that Q(J2), the least subfield

of l1 containing 0 and J2, comprises precisely all elements of the form
a where a,b,c EQ. —

(ii) (See 3.10.7 for notation.) Taking E = i), F = and a =

one sees that Ma = + 1. Thus every element of F is expressible uniquely
in the form fo +fii where fo, f' F. But similarly every element of F is
expressible uniquely in the form q0 where q0, EQ. Hence every
element of F is expressible uniquely (uniquely?) in the form r0 + r1i + riI2 +
where r0, r1, r2, r3 E 0.

As an application of 4.3.2 we reobtain the content of the Remark on p. 148.

Corollary 4.3.6 Let F be a field, p a polynomial of degree 1 in F[x] and

let a be the natural homomorphism from F[x] onto = (F[x])a. Then
F aFa (F[x])rn
*
Ma is called the minimum polynomial of a over F.
 The isomorphism theorems 155

Proof Set I=[p]. Then Fa={aa:aeFcF[x]}={a+I:aEFcF[x]}=

(F +1) F[x] (F +1) F
By 4.3.2 But FnI=[0] [since the non-zero

elements of I have degree >0]. Hence

Genuine and significant applications of 4.3.3 at the level of this text seem
less easy to find (although see exercise 6.2.12 and 6.5.8). One sees immediately
from the statement of 4.3.3 that an ideal M of a ring R is a maximal ideal
(see 3.9.5) if! RIM is a simple ring (exercise 3.4.20). But it really is more
natural to deduce this assertion from exercise 4.

Exercises
1 Let C : R -. S be a homomorphism not necessarily onto S. Show that

ker C

2 Let A1, A2 be ideals of the ring R. Show that the map*

given by rO = (r +A1, r +A2) is a homomorphism. Deduce that has

R
a subring isomorphic to Setting R =1, A1 [in] and A2 = [n] where
(A1 nA2)
(m,n)= 1, show that 7LmlI=([] (ci 4.2.9(i)).
[nrH$[nrlm$mn

3 Let O:R-.S be an onto homomorphismt with kero=I and let

R/K be the natural map where K c L Show that there exists a homomorphism
rj :R/K-+S such that'O=i/corj.
This result, which may be described by the following picture, generalises
4.3.1.

4 The Correspondence Theorem Let 0 : R -. S be an epimorphism with

ker 0 =L Let U ={A: A isasubringofR andl cA}and V ={B: B isasubring
of S}. Show that the map®: U -* V defined byA® =A0 is a 1—1 map of U onto
*
$ denotes the direct sum (exercise 3.2.9).
t Also called an epimorphism.
156 Factor rings and fields

V. [Hint: 1—mess is not too hard. To prove onto show that, given B c 5, BC1
(see 2.6.5) is a subring of R containingl and that (BC1)€J B.] Show that, under
this correspondence, ideals of R which contain I correspond to ideals of S.
Deduce that M is a maximal ideal of R if RIM is a simple ring. If R is
a commutative ring with unity show that M is a maximal ideal if RIM is a
field. [Exercise 3.4.14 might help.]
5 Let R be a commutative ring with unity. Show that an ideal P R) is a
prime ideal if RIP is an integral domain.
6 Use exercises 4 and 5 to reprove Theorem 3.9.6.
7 (i) Let R be a commutative ring with unity. Show that if P is a prime
ideal such that RIP is finite then P is a maximal ideal.
(ii) Show that in F[x, y] the ideal [x] is prime but not maximal. Is the ideal
[x, y] aaximal in F[x, y]? [Cf. exercise 3.9.2.]

8 .Useexercise
[x +x+1]

9 Let F c F be fields and let a, /3 e F be such that Ma = M8. Show that the
fields F(a) and F($) are isomorphic.

4.4 Constructions of R from 0 and of C from 11%

We have seen in Section 3.10 how one can construct the field 0 from the
ring 7 using only the properties assumed, in Section 1.2, to hold in 7L. Indeed
O can be formally defined as the unique field of fractions of the unique well
ordered integral domain (see 3.10.5(iii)). Furthermore exercise 3.12.8 indi-
cates how 0 can be made into an ordered field, again using only the axiom
P of Section 1.2. The question arises as to how we might similarly replace
our intuitive and possibly vague ideas about real and complex numbers by
formal constructions. Whilst employing some ideas from Sections 4.2 and 4.3,
this section is meant to be easy reading—for information, if you like—and
we shall only outline the formal procedures involved.
Informally we recognise that, whilst R is algebraically deficient (x2 + 1 = 0
has no solution in R), 0 is even worse: it is analytically deficient too. For
instance, in 0 one can find infinite sequences (there is, for example, one whose
firstfew terms are 14110, 1411100, 141411000, 14142110000,
141421/100000,...) which appear to 'home in'—technically converge—to
non-rational quantities (in this instance A geometrical interpretation
of this is given by considering the (intuitive) real numbers stretched out along
an infinite straight line. (fhink of the x-axis in the plane.) Marking off the
points on this straight line which correspond to rational numbers leaves
 Constructions of R from Q and of C from R 157

infinitely many unmarked points on the line and it is possible, by alighting

only ona sequence of rational points, to home in on the unmarked (irrational)
point Thus 3 can be thought of informally as 'R-with-gaps'—so that 0
is analytically incomplete—with ñ, e, ir, being amongst the
infinitely many points at which such a gap is located. —

The above intuitive picture leads one to propose defining to be the

above sequence. The elements of 0 present no difficulty; we identify with

the sequence In fact this proposal won't quite suffice since,

for instance, 1,
14141 \
and (7, 3,
6aaa \
also seem to be
.j
candidates for '12 and To identify precisely which infinite sequences of
rationals are of interest to us and to get round the insufficiency we proceed
as follows.
The infinite sequence (a1, a2,.. .) of rational numbers is said to be Cauchy
convergent if and only if, to each positive rational number h >0, we can find
a positive integer H (which will in general depend upon h) such that for all
integers i,j>H we have <h. Two convergent infinite sequences
(a1,a2,.. .), (b1,b2,...) are then said to be equivalent if and only if,
given any rational k >0, we can find a positive integer K >0 such that
<k for all s> K. [If these ideas are new to you, you may find you
I

need to read them very carefully twenty times or more before they begin to
stick.]
The set § of real numbers then comprises, by definition, the totality of all
equivalence classes (see exercise 3) of Cauchy convergent sequences. Addition,
$, and multiplication, 0, are defined as you would expect [and how is that?]
and S can be shown to be a field containing (an isomorphic copy of) 0 as a
subfield. In S the analytic deficiencies apparent in 0 are absent.
This method, using Cauchy sequences, is due to Cantor. Using the ter-
minology of Section 4.2 we may proceed as follows. Under componentwise
® and 0 the set C of all Cauchy convergent sequences becomes a ring in
which the subset of all those sequences which are equivalent to the zero
sequence (0, 0, 0,. . .) form an ideal N (N for 'nought' or 'null'?). The above
set S is nothing more than the factor ring C/N.
A second method of defining R, this one due to Dedekind, is based on the
informal observation that every real number r appears to split 0 into two
disjoint subsets: (i) all rationals less than or equal to r; (ii) all those greater
than r (!!) This dissection is called a Dedekind cut, the subsets (i) and (ii)
being called, respectively, lower and upper sections. Clearly r corresponds to
the upper section of all those rationals greater than r but, of course, one
cannot define r to be this upper section since such a definition (of r in terms
of r) would be circular! Thus we have to find an honest way of defining the
term upper section in 0 solely in terms of 0. Assuming the familiar ordering
158 Factorrings and fields

which exercise 3.12.8 allows us to place on 0 we make

Definition 4.4.1 An upper section in 0 is a subset U of 0 such that

(i)
(ii) ifx, y E 0, if xe U and if x y then
E U then there exists y E U such that y <X.
(To see what motivates this definition, try thinking pictorially of the upper
sections corresponding to and to, say, 5k.)
By defining,* suitably, and C on the set U of all upper sections one can
show that (U, $, C) is a field. (This is an arduous task, the proof that C is
associative requiring 9 subcases! See [50].) U, like 5, can be shown to contain
a subfield isomorphic to 0 and, again like 5, be analytically complete. Before
reading the next sentence the reader may care to ponder the question: Which
of S and U more accurately reflects his intuitive conception of R?
The answer is that neither should: for S and U are isomorphic!
Although we haven't properly defined the terms (see exercise 1) we quote,
noting the close analogy with 3.12.4,
Theorem 4.4.2 Any two complete ordered fields (S and U being two
examples of such) are isomorphic [50].
Any such field is called the field of real numbers and is denoted by R.
One result we shall require in Section 4.8 is

Theorem 4.4.3 LetfE R[X] be of odd degree. Then f(c)= 0 for some c ER.
Proof See exercise 7 below and then [50].

Problem 5 Why is it at this point unfair to say: 'Theorem 4.4.3 follows

immediately from Theorem 1.11.9'?
This is not as easy as in the Cauchy sequence case.
 Constructions of from 0 and of C from R 159

Augustin-Louis Cauchy (21 August 1789-22 May 1857)

Cauchy was the eldest child of a barrister. In 1807 he entered the
Ecole des Ponts et Chausees intending to become a civil engineer. In
1812 he proved Fermat's assertion that every positive integer is a
sum of n n-gonal integers. By 1816 he was teaching at the Ecole
polytechnique and a member of the Académie des Sciences. After
the July revolution of 1830 Cauchy, an ardent Royalist, refused to
take an oath of allegiance to Louis Philippe. He left for Turin where it
appears he taught Latin and Italian, later returning to Paris in 1838.
The records on his subsequent career are astonishingly confused for
a person of his eminence.
Cauchy was a prolific author, producing about 800 papers ranging
over many areas including optics, elasticity, ordinary and partial
differential equations, mechanics, determinants, theory of
permutations and probability. He wrote the classic textbooks in
which analysis was made more rigorous. The first developments in
complex-variable theory are all due to him.
In his undergraduate career the reader can hardly fail to meet
Cauchy's root test, ratio test, convergence criterion, inequality,
integral theorem, integral formula as well as the Cauchy—Riemann
equations.

We now come to the formal construction of the field C of complex numbers.

As in the construction of R from C, we offer two approaches (each based on
the assumption that P is already 'known'). Above we recognised that P (even
though it contains substantially more elements* than C) is still not sufficiently
rich to contain all roots of all polynomials with real coefficients, x2 + 1 provid-
ing the simplest example. Informal introduction of the missing roots was made
by A Girard as early as c. 1620, but it was not until Gauss in 1831 (unpublished)
and Hamilton in 1833 put them on an apparently firmer (i.e. arithmetic)
footing that the complex numbers met with a fair degree of approval.
Gauss himself said (c. 1825) that 'the true of is hard' whilst
Cauchy (1847) said: 'We repudiate the sign completely because one
does not know what the sign signifies, nor what meaning one should attribute
to it'. Simply introducing the symbol i and asserting that i2 = —1 and that i
behaves as do real numbers raises problems (see exercise 16). Euler, in his
book Complete Introduction to Algebra (1768) had said that 'Since all conceiv-
able numbers are greater than, less than or equal to zero, square roots of
negative numbers cannot be included amongst the possible numbers'. Now if
i is not a number of previous acquaintance, on what grounds can one assert

that 2 i = 2? Gauss and Hamilton got round the problem by noting that
each complex number a + ib determines and is determined by the (ordered)
pair (a, of real numbers. Noting that we would ideally like a + ib = c + id
ifandonlyifa = c andb = d,(a +ib)+(c +id)= (a +c)+i(b +d)and (a +ib)x
(c + id) = (ac — bd) + i(bc + ad) they defined a complex number to be a pair,
*
In technical terms, 0 is countable, is uncountable (exercise 2.6.15).
160 Factor rings and fields

(a, b) of real numbers and defined equality, addition $ and multiplication

0 of complex numbers by:
(i) (a,b)=(c,d)ifandonlyifa=candb=d;
(ii) (a,b)$(c,d)—(a+c,b+d);
(iii) (a, b) 0 (c, d) = (ac —bd, bc + ad).
It is not difficult to check (exercise 8) that the set C of all complex numbers
(a, b) forms a field with respect to ® and 0. One observes that the mapping
a (a, 0) shows that C contains (an isomorphic copy of) R as a subfield.
Denoting by i the pair (0, 1) one finds i2 = (0, 1) 0 (0, 1) = (—1, 0). Finally
(a, Li) = (a, 0)${(0, 1) 0 (Li, 0)} as is easily checked. Thus, identifying each
r E with (r, 0) in C, the ordered pair (a, Li) can be more recognisably written
as a $ {i 0 b } or a + ib if we resort to the more usual notations for addition
and multiplication.

Note 4.4.4 The reason for returning to the a + ib notation is one of con-
venience only. It seems easier to work with than does the ordered pair notation,
simply because most of us have been used to working with it for so long
(recall exercise 1.6.4 and see exercise 10 below).

An alternative way of constructing C is to follow Cauchy and extend Gauss'

notion of congruence from 7 to Given f, g, h e Cauchy wrote
(mod h) if in He then replaced each x inf and g by a new
symbol i and wrote f(i) = g(i) rather than f g (mod h). In particular when
h= +1 we find, for example, that

which Cauchy wrote as

2i4+3i3+3i2—i+4= —4i+3
Of course from it follows that i2+1=0. In modern
terminology this amounts to nothing more than forming the factor ring
=
. (Kronecker's method mentioned in Section 4.2 amounts to
+1]
[x
forming factor rings of Q[x] in a similar way.)
Since x2+ 1 is irreducible over P the proof of 4.2.1 tells us that 13 is a field
with typical element a +bx +1 where I is the ideal [x2+ 1] of P[x]. We leave
it to the reader to show (exercise 9) that C and 13 are isomorphic fields each
containing a root of the polynomial x2 + 1.
C was introduced in order to produce a solution to the equation x2 + 1 =0.
But there are many other equations [how many?] with real coefficients which
do not possess any real roots. Even worse, we are now morally obliged to
consider polynomials with complex coefficients. Surely to find roots for such
polynomials we shall need to construct some sort of super-complex numbers?
That indeed we do not is thc content of the Fundamental Theorem of Algebra
(Theorem 4.8.1).
 Constructions of R from 0 and of C from R 161

Exercises
1 Based on our intuitive picture of 0 as the real line R with gaps we define,
formally, for any ordered field (F, <) a gap to be a pair of subsets A, B
ofF such that (i) A 0, B 0; (ii) A uB —F; (iii) reA, s EB4'r<s; (iv)

maximum element and B no minimum element). Prove that if B is the set

of all positive rationals t such that t2> 2 then the pair A, B (where A = 0\B)
is a gap in 0. (An ordered field without gaps is called complete.)
2 (i) Show that between any two rational numbers there exists one and
hence infinitely many others.
(ii) Show (using the less cumbersome notation but remembering what is

meant and hence avoiding proofs by intuition) that if 0 and then

y S

there exists n such that (Cf. exercise 3.12.10.)

S 3'

3 (i) Show that the relation of equivalence on the collection of all Cauchy
convergent sequences is indeed an equivalence relation.
(ii) How should one define ® and C on the set § in order that the result is
a field modelling our intuitive grasp of
(iii) Let s = (x1, x2,. . .) be the sequence defined by x1 = 1 and, for each n 1,
= Show that s is a Cauchy sequence. Which real number is
represented by the equivalence class containing s?
4 As stated in the text, the set C of all Cauchy convergent sequences in 0
can be turned into a ring by defining $ and C componentwise. Show that the
subset N of C comprising all sequences equivalent to the zero sequence forms
an ideal. Show that the factor ring C/N is a field, by showing each non-zero
element of C/N has a multiplicative inverse in C/N.
5 Let V,Wbetwouppersectionsin0.Define V+Wtobe{v+w:veV,
w e W}. Show that V + W is an upper section in 0. In order to make a
field out of this set of upper sections we shall need to identify a zero element
and an additive inverse for each upper section V. Can you do that? [Be careful
over e V. Keep upper sections corresponding to and to —12 in mind.]
6 Working informally show that (i) 0(12) can sustain two sets of positive
eleriients (cf. 1.2.3); and that (ii) 0 and R can each sustain only one set of
positive elements. [Hint: All of 0 is determined by the fact that 1 E Ot In
r ill r is non-zero ahd a square.]
7 . .+a0€R[x]andn is odd then

f(s) <0 ifs <minimum of {—1, . .

162 Factor rings and fields

Hence deduce that 4.4.3 follows at once if only we can prove: if g is a function
continuous on the closed interval [a, b] and if g(a) <0 and g(b) >0 then there
exists c such that a <c <a5 and such that g(c) = 0. [How might you prove this
result? Bolzano* (1817) found he didn't know enough about the real numbers
to prove this to his satisfaction.]
8 Check that with respect to ® and C as defined above the set C of ordered
pairs (a, b) of reals do form a field.
9 Show that the field of exercise 8 is isomorphic to the field ( ®, a).
[x +1]
10 Solve for x,y; (i) (3,5)C(x,y)=(1,1) formally; (ii) (3+5i)(x+iy)=
I + i informally. Which way is easier? Why?
11 On the set S of all ordered pairs (a, b) of reals define ® and C by:
(a, b)®(c, d) = (a +c, ,5 +d), (a, b) C (c, d) = (ac, bd). Is(S, ®, C) a field?
12 With S as in exercise 11 define $ as in exercise 11 and C by
(a, b) C (c, d) = (ac + bd, bc — ad). Is (5, $, C) a field?
13 Apply the definitions of® and C of the text to ordered pairs (a, b) where
a, b lie in the fields (i) C; (ii) (iii) Is the resulting system always a field?
14 Does the map O:R-*C defined by aO=(0,a) embed in C? (See
3. 10.2(v).)
[x]
15 Let f = ax 2 + bx + c 6 R[x] be irreducible. Then is necessarily a field.

Must

16 Show that C cannot be ordered. [Hint: Assume it can. Then i2 will be

in Ct]
17 Let N denote u {0}. On the set N x N define the binary relation — by
(rni,ni)—(rn2,n2) if rn1+n2=rn2+n1 in N. Show that is an equivalence
relation on N. Denote the equivalence class containing (rn, vi) by (m, vi) and
the set of all classes by S. On S define $ by (a,b)$(u,v)=(a+u,b+v).
Show that under $ the subset = {(m, vi): vi = 0} is structurally identical to
N under +. (Technically (Si, $) and (N, +) are isomorphic semigroups.) Denot-
ing (in, 0) by rh show that for all in, vi 6 N the equation x $th = vi has a solution
in S (whereas, of course, x + m vi may not have a solution in N). Can you
see what we are doing here? See Section 0.1.

4.5 Finite fields

Having indicated in Section 4.4 how one can use 4.2.1 to generate C from R
we now use its corollary (4.2.20) to prove, as part of the first objective posed
in Section 4.1,
*
Bernard Boizano (5 October 1781 — 18 December 1848).
 Finite fields 163

Theorem 4.5.1 Let p be a positive prime and vi a positive integer. Then

there exists a field with exactly p" elements.

Proof In 4.2.20 take F = 7,,, f=x" —x and E to be any field containing 7,,
and in which —x factorises into p" linear factors, f' —x =
x(x —t1). .. (x say. Now all the are distinct. For, as in exercise
1.11.2, we can argue that if = t1 where i then is a root of the formal
derivative f' —1 of f. But then f' = —1 since 7,,, £ and hence £[x]
have characteristic p (see exercise 3.10.18). Thus t1 cannot possibly be a root
of f.
Let T be the above set of roots (including 0). For t, e T we see easily
= tr —tr = t,—t1 and that (t1t
-1 = =
(exercise 3.10.18) that (t1
Hence and t,t71 T and thus T is a subfield of F (by 3.4.2(F))
containing exactly p" elements.

Theorem 4.5.1 raises more questions than it settles. For example

Question I Do there exist fields with other than a prime power number of
elements?

Noting that if f is a polynomial of degree vi which is irreducible in

then is (see the proof of 4.2.1) a field with exactly p" elements (in
which f has a root), we naturally ask

Question 2 Is every field with p" elements obtainable in this way?

Or at least

Question 3 Is there for each prime p and vi a polynomial f of degrec

vi which is irreducible in 7,,[x]?

Finally

Question 4 What relationship, if any, is there between two fields with the
same number of elements?

To help answer these questions we introduce some concepts which are used
time and again throughout the whole of mathematics and its applications to
problems in the real world; for example in differential equations, statistics,
linear programming and even in modelling an economy (see [67]). The proper
mathematical setting for these concepts is the subject of Linear Algebra, of
which they, along with matrices, constitute the life blood. For reasons of space
we take the development only as far as required in this text. Since, however,
the machinery to be developed will also find application in Section 4.6, we
164 Factor rings and fields

save effort by working, even here, in terms of general fields rather than with
just finite ones. We begin with

Definition 4.5.2 Let F be a field and F a subfield of F.

(i) The set u1, u2 u. of elements of F is said to span F over F iff every
element e of F can be expressed in the form e =fiU1 that is, iff e
is a linear combination of the U1 using suitable fi,. . , f. in F. .

(ii) The set v1,. .. , of elements of F is said to be linearly independent

over F if the only way of writing °E in the form f1v1 . +f5v5, where the
.

f1eF, is by taking to be OF.

(iii) The set w1,.. ., of elements of F is said to constitute a basis for F
over F iff it spans F over F and is linearly independent over F.

Notes 4.5.3
(i) It is clear that if F is a finite field and if F is any subfield then a spanning
set (as in (i)) for F over F certainly exists; the set of all elements of F
constitutes such a set. —

(ii) {1, i} forms a spanning set—indeed a basis—for C over R; {1, forms

a basis for over 0 (see 3.10.6); {1,i,'12,ish} forms a basis for
0(i,'12) over 0 (see 4.3.5); {1+[f],x+[f],. ..,x"1+{f]} forms a basis
for the field over 0, if f is irreducible of degree n in Q[x]. (See exercise
4.2.18 and the proof of 4.2.1.)
(iii) Let w1,.. ., w, be a basis as in 4.5.2(iii). From 4.5.2(u) it follows
immediately that, for Ii,. . . , ft, . . , e F we have

if and only if f' = gi, = g2,.. . = (exercise 2). (Roughly stated: in terms
of a given basis different looking elements are different.)

Problem 6 Does R have a finite basis over 0?

Our first result relates the respective sizes of spanning sets and independent
sets.

Theorem 4.5.4 If {Ui,.. . , UT} spans F over F and if {vi,. . . , v5} is linearly
independent over F then s r.

Proof We may suppose that no proper subset U of U = {Ui,. , UT} spans . .

E. (If so, replace U by a possible such U at the outset.) Now suppose

rccs. Since {Ui,. . , UT} spans F over F there exist fll,f12, . . . ,f'T in F such
.
 Finite fields 165

that v1 =fiiui 4" +firur. Since v1 °E [why not?]f1, 0,- for at least one i.
WLOG we can assume i = 1. Then we have firur).

It follows [prove it now] that {v1, u2,.. . , uj spans F over F and hence that
there aref21,f22 f2r in F such that v2=f21v1+f22u2+.. +f2,Ur. Here, at
least one of the 121 with j 2 is non-zero. (Otherwise we should have v2 Sf21 V1 =
0 which is impossible whether or not 121 = °F [why?].) WLOG, we suppose
Then we may write u2=f {—f21v1+v2—f23u3—. so that

Vi, V2, U3 Ur} spans F over F. One can continue in this manner replacing,
at each step, a u1 by a v1 [why never a vk(k <i) by a v,?] until all the u1 have
been replaced by v1, v2 Vr, thus showing that {v1,... v,} spans F over
,

F. There then exist . , in F such that v5 =f51v3 +. But this

equality can be rewritten . . =0,
which is an impossibility since the v1,. .. , v. were supposed to be independent
over F. This contradiction shows that r <s is untenable and we deduce that
s r as required.

Corollary 4.5.5 Let F E F be fields such that F has a (finite) spanning set
over F. Then F has a (finite) basis over F and any two such bases contain
the same number of elements.

Proof Let S ={u1, u2,.. , be a spanning set for F over F. We define a

.

subset B of S as follows. For each k (1 k r) we take Uk e B if and only if

Uk cannot be expressed in the form fk 1u1 - IUk_1. (For k = 1 this is
interpreted as: u1 B if and only if u1 0.) [Intuitively we are relieving S of
redundant elements.] It is easy to check (exercise 2(u)) that the subset B still
spans F over F and that it is linearly independent over F. That is, B is a
basis for F over F.
If B1 is another basis and if B, B1 have t, elements respectively then,
thinking of B as an independent set and B1 as a spanning set, t Similarly
t [why?]. Thus t = t1 as asserted.

Notation and Definition 4.5.6 This unique number of elements in any (finite)
basis forE over F is denoted by [E:F]. We call [E:F] the dimension of E over F
and say that E is a finite-dimensional extension of F.
It is now an easy matter to answer Question 1 above.

Theorem 4.5.7 Let F be a finite field. Then F has pt elements where p is

some prime and t some positive integer.

Proof Since F is finite the prime subfield of F must be 74 for some prime
p (3.10.9). By 4.5.3(i) and 4.5.5, F has a basis w1,. . ., say, over 74. Then
166 Factor rings and fields

4.5.3(iii) tells us that the pt elements fiwi +f,w1 where the are in 74
are pairwise distinct.
To answer (affirmatively) Questions 2 and 3 raised above we first observe
(exercise 4) that in any field F with p" elements every non-zero element
satisfies the equation x — 1 = 0. Using exercise 5 we deduce that the set

of all non-zero elements of E is exactly the set {a k: 0 k <p" — 1} for some

suitable a E E. Let Ma be the (unique irreducible) minimum polynomial (see
4.3.4) of a over 4 (c E). [How do you know Ma exists?] Suppose deg Ma = m.
Then, by 4.3.4, the smallest subfield 4(a) of E containing 4
and a, namely
F itself, contains ptm elements. It follows that (i) p" = p1t1 whence n = m and
A At A ("\ C' 7I \ pL J

This still leaves Question 4. The answer is quite amazing: Every two fields
with precisely p" elements are isomorphic! Given 4.2.21 this is not difficult:
for each field with p" elements is easily seen (exercise 4) to be a splitting
field for the polynomial x'" —x over 4 and the uniqueness then follows from
4.2.2 1.
To give a specific example we note that f= x3 + x + 1 is irreducible over
(Being a cubic it suffices to check that f has no root in (cf. Remark
after 1.11.7).)
12[x].
Thus is a field whose elements are of the form a0+ aix +a2x2 +[f]
where a0, ai, a2 e Z2. Denoting this element briefly by aoaia2 we find that
12[x].
the multiplication table for is [can you complete it?] as follows.

000 100 010 001 110 101 011 111

000 000 000 000 000

100 000 100 010 001 101
010 000 010 001
001 000 001 110
110 011 111
101 010 001 110
011 100 110
111 011 001

The entry in 'row' 101 and 'column' 011 is 110 since

(0+x 4-x2+[f]) +x2+x3+x4+[f]= I +x +[f].
Finite fields were introduced by Galois in 1830 during number-theoretic
investigations. Wishing to solve congruences f(x) 0 (mod p) where p is a

*
We denote the elements of hy I) and I.
 Finite fields 167

Picture by courtesy of the American Mathematical Society

Eliakim Hastings Moore (26 January 1862 30 December 1932)

Moore was born at Marietta, Ohio, into a family with strong university
connections. His interest in mathematics was strengthened by his spending a
summer, whilst still at school, as assistant to the astronomer Ormond Stone.
After Moore had completed his PhD in 1885, the Yale professor, H A Newton,
financed Moore's visit to Göttingen and Berlin where he made contact with
Weierstrass, Kronecker and Klein. After holding positions at Northwestern and
Yale, Moore was appointed acting head of the mathematics department at the
University of Chicago when it opened in 1 892. By persuading the university to
appoint Bolza and Maschke to the department, Moore began a pursuit of
excellence which was one of his characteristics. Of the twenty nine students
whose doctoral theses were supervised by Moore, three, L E Dickson, 0 Veblen
and G D Birkhoff, all mathematicians of the very top rank (the first two coming
from Iowa) followed Moore as President of the American Mathematical Society.
Moore's teaching methods were also revolutionary. Instead of lectures of fixed
length he ran classes as 'laboratories' whose durations were independent of the
clock and meal times. Sessions ended when he felt the subject of the day was
exhausted.
Moore's research, characterised by rigour and generality, took in algebra,
number theory, integral equations and analysis. His algebraic work included the
results mentioned on finite fields.
Moore's crusade to promote precision in mathematical argument won him at
least seven honorary doctorates. At one ceremony he met an old friend, the
physicist Carl Barus, who had instructed Moore in his student days in the art of
organ pumping. 'Who knows', asked Barus later, 'what my instruction in organ
pumping may have done for Professor Moore?'

prime and f is irreducible of degree n (over 74) he introduced, in analogy

with the introduction of i to 'solve' the equation x2 + 1 =0, a hypothetical
symbol j to act as a root for f and considered the p" formally distinct
expressions where z1 are integers such that 0EZ1 <JY.
With the obvious definitions of addition and multiplication these expressions
can be turned into a field—a Galois field, GF(p"). (Thus GF(p) 4.)
168 Factor rings and fields

The objection to this kind of introduction of j is the same as that raised

over the introduction of i (Section 1.6). The above construction of GF(p") is
replaced in present day mathematics by forming the factor ring

The answer to Question 4 was first obtained in 1893 by the American

E H Moore. Together with the answers to Questions 2 and 3, part of the
algebraist's dream of classifying all algebraic systems comes true. We have

Theorem 4.5.8
(i) Every finite field is of the form where f is a polynomial irreducible
in In particular every finite field has p" elements for some prime p and
it El.
(ii) To each such p and n there exists a field with p" elements.
(iii) Two finite fields are isomorphic ill they have the same number of
elements.

For applications of finite fields see, for example, [7], [61].

Exercises
1 Show that the set Li, i, is linearly independent over 3. Do the
same for Does one retain independence on extending this
set to include the various products ho, 130?
2 (i) Prove 4.5.3(iii).
(ii)Prove that if {ui,..., Urn } spans F over F and if Urn is a linear combination
of UI,.. , Urn_i then {u . .,
. spans F over F.

3 Prove that [3('Y2): 0] = 3. (See 4.3.5(i).)

4 Prove that in a finite field with p" elements each non-zero element satisfies
the equation — I = 0. [Hint: Follow the method of proof of 2.5.3.]
Deduce that each such field is a splitting field for xr —x over 74.
5 Prove that the multiplicative group of non-zero elements in a finite field is
cyclic as follows. Let F be a finite field with q =pfl elements and let e F where
$0. We know (exercise 4) that 2q - = I. Let k be the smallest positive integer
such that cxk= 1. Show that I. (Exercise 2.5.7 might help.) We call k the
order of Show that if fi have coprime orders k, I then has order ki
(exactly). Now suppose F has no element of order q — 1. Let y be an element of
largest order e, say, in F and let ö be any element of F, of order m, say.
Suppose rn-Fe. Then, for some prime r, rn=r"rnt and e=re1 where b>c, r+rn1
 Finite fields 169

and Deduce that 5miyPc has order r"e1(>e), a contradiction. Thus mle. It
follows that 5C = 1. Thus the equation f = I has q — solutions in the field F, a
1

contradiction (see exercise 2.4.4). Thus F has an element a, say, of order q — 1.

Hence

6 Show that x3—x + 1 and x3—x —1 are irreducible over 13.Is

[x —x+1]

[x —x—1]

7 Show that P=x4+x+1 and Q=x4+x3+x2+x+1 are both irreducible

in 7L2[x]. Show that but that x + [P] and x + [Q] have multiplica-
tive orders 15 and 5 respectively. Deduce that the 'obvious' mapping between
12[x] 12[X]
and namely f+[P]-*f+[Q], is not an isomorphism.

8 H= [x13[x]
+111

9 (i) Show that if mjp" —1 then C —1 = 0 has exactly m distinct roots in

GF(p").
(ii) If tin show that has exactly one subfield with pt elements. [Hint:
tin implies p' — —1.]

10 Find all monic irreducible polynomials of degree 2 over GF(3). Show

that their product divides x8 — 1.
11 Is GF(4) (isomorphic to) a subfield of GF(8)?
12 Let f be irreducible of degree t in 4[x]. Show that fixt" —x in 4[x].
[Hint: let g=(f,x"—x) in There exists ueGF(p') such
thatx—uifandx—uix"—x in
4[x ].]
13 Try to convince yourself of the following. Let be the ring ot
polynomials in the letter t and let F =4(t) denote its field of fractions (see
Section 3.10). Consider the polynomial f = x" — tin F[x]. Let F be a splitting
field for f over F.
First claim: f is irreducible in F[x]. Second claim: Since f' = 0, f
has repeated roots in F. Thus: whilst it is true that irreducible polynomials
cannot have repeated roots in their splitting fields F when the base field F
is finite or of characteristic zero they can have repeated roots if F is infinite
of prime characteristic.
170 Factorrings and fields

4.6 Constructions with compass and

straightedge
Turning now to the second of the objectives listed at the start of the chapter,
we show how the algebra we have developed can help in discussing problems
of a purely geometrical nature, namely in deciding whether or not certain
constructions can be carried out (theoretically) using only a compass and
straightedge, a straightedge being an ungraduated ruler.
Some initial remarks are in order. Most readers will recall from schooldays
how to bisect a given angle using only a ruler and a compass (no use being
made of the markings on the ruler, the often-used term 'ruler and compass
construction' thus being erroneous). However, few will have given much
thought to the apparently similar problem of trisecting a given angle. In the
5th Century BC the Greeks were studying this problem, although the precise
reason appears unknown ([84, p. 92]). Other famous construction problems
were also under consideration at this time. Two of them were: using only
straightedge and compass construct (i) a square with area equal to that of a
given circle; (ii) a line segment equal in length to each edge of a cube of
volume 2 units given the length of an edge of a cube of unit volume. It does
not seem clear how or why problem (i) arose ([80, p. 70]), but there are stories
relating to the origin of problem (ii) ([84, p. 90]). Now, it is the case that
these problems can be solved if one is allowed to introduce fixed curves other
than circles and straight lines (Archimedes' spiral r = all, for instance) or a
marked straightedge ([84, p. 93]). The restriction to use of compass and
straightedge only might well reflect a supposed primitiveness of the straight
line and circle. Whatever the reason, repeated attempts to solve the above
problems led to the introduction of an immense amount of good (and bad!)
mathematics (just as did attempts to establish Fermat's Conjecture). One
other point. The problems posed are wholly theoretical ones requiring exact
answers. The physical drawing is to be seen merely as an aid to understanding.
Further, whilst we shall prove that all three problems above are insoluble, it
is easy enough, even theoretically, to solve them to within any required degree
of approximation (example 11).
We first recall how to bisect a given angle AOB. Figure 4.4 should suffice
(but see Problem 8).

Fig. 4.4
 Constructions with compass and straightedge 171

In any construction problem of the above type we are given points, lines,
circles, etc. and are required to construct (or show the impossibility of con-
structing) further such objects. To prove that a construction can be performed
one only needs to describe a method for doing it. To prove it cannot requires
us first to list the complete 'rules of the game'. We adopt the following rules.
Let a finite set P1 of points be given in the plane. Joining all pairs of points
in P1 by straight lines of arbitrary length and drawing each circle which has
a point of P1 at its centre and a point of P1 on its circumference, we augment
P1 with the (finitely many) points of intersection of all these straight lines and
circles. Calling this new (finite) set P2 we repeat the procedure to obtain sets
P1 c P2 c P3 c. of points in the plane. The totality P of points so obtained
.

we call the set of points constructible from P1.

In the above problems we are given an angle, a circle and a unit line segment
and we are required to construct, respectively, another angle, a square and
another line segment. Each of these configurations determines and is deter-
mined by a finite set of points and so it is seen that problems concerning
geometrical constructions are equivalent to problems concerning constructible
points.
Problem 7 If basing our work on sets of points rather than on complete line
segments, complete circles, etc. seems pedantic, contemplate construction (ii)
above where a 'unit line segment' is given. The question arises: Given a line
segment are we also given all points on it? If not, which exactly are we given?
Problem 8 According to the rules we have adopted, the only circles we can
draw are those whose centres lie at and which pass through points already
given or constructed. We do not allow for drawing circles of arbitrary radius
nor for transferring radii using fixed-opening compass, from one centre to
another. Both these 'moves' were used above to bisect a given angle, so
shouldn't we allow them here? Does it make any difference?

Since in every construction problem we are given at least two points we

can take these to be the points 0, 1 on the straight line (the 'x-axis') joining
them. We then easily construct (exercise 1; see Fig. 4.5) a 'y-axis' through 0

Fig. 4.5 Fig. 4.6

172 Factor rings and fields

perpendicular to the x-axis. This permits us to give each constructible point

coordinates (a, b) determined by the unit of length from 0 to 1.
It is easily seen (exercise I) that all points (u, v) where u, v e 0 are construct-
ibl! and that problems (i) and (ij) above become: (i) construct the point
('fir, 0); (ii) construct the point (42, 0). Further, from Fig. 4.6 the problems
of trisecting the specific angle of 60° is clearly equivalent to: construct
(cos 20°, 0).
Before we can complete the solutions to the three classical problems, we
require one more technical result, namely
Theorem 4.6.1 Let F c E c K be fields such that* [K : E] = r and [F : F] = s
are both finite. Then [K :F] is finite. Indeed [K :F]= rs =[K :E][F :F].
Proof Let a1,. . , and v1,. . ,v. be bases for K over F and F over F
. .

respectively. Then each element k of K is expressible in the form k =

e 1u1 + for suitable e1 E F. But each e1 (1 i r) can be written as
e, =f1ivi . It follows easily, on substituting for the e, in k, that
. the
v,u1 certainly span K over F.

To prove that the v1u1 are also linearly independent over F one essentially
reverses the above procedure. The details are left to exercise 2.
In the notation of 3.10.7 (see also 4.3.5(u)) we have
Examples 4.6.2
(i) [0(i, 0] = [0(i, 'h): 0] = 2 2 4.
(ii) [0(42,43): 0] = [0(42,43): 0(42)][0(42): 0] =2 3=6.
Returning to constructible points, suppose we have labelled two points in
P1 as Pi=(O'O) and P2=(l'O) and suppose is the
(n + 1 )th point of a sequence constructible from P1— including the given points
Pi,...,PrOf Pi.
The straight line joining and (1 I 4 n) has equation
(a1 —a,)(y — b1) = (b1 — b,)(x — a1) whilst the circle centre Pk and passing through
2 2
has equation (x — ak) + (y — bk) = (a, — ak) 2 + (b, — bk) 2 . These equations can
be rewritten ax +by +c = 0 and x2+y2+2fx +2gy +h = 0 where a, b, c, f, g,
h =tO(a3, b3,. , .ba). Now it is easy to check (exercise 3) that if Pp.41
.

is obtained as the intersection of two distinct lines then ap.÷1, bp. eQ whilst
if Pn÷i is obtained as the intersection of a line and a circle, or of two distinct
circles, then either + + e Op. or—at the worst—an ÷ +1
where deOp.. Thus for each we have or 2. There then
follows
Theorem 4.6.3 If p1 =(a1, b1) is a point constructible from P1 = {Pi pr}' as
above, then there exists a sequence Or+ ç. 0, of subfields of l1 such
. .

that a1,b,eO, and for each I (rzcie—l).

*
For notation see 4.5.6.
tSee 3.10.7 for notation: note that
 Constructions with compass and straightedge 173

We can now prove the impossibility of solving the three classical problems
of the Greeks.

(A) Duplication of the cube Here we are given a line segment of unit length.
Equivalently_we are given P1 {(0, 0), (1, 0)}. We are asked to construct the
point p = (J2, 0). If p is constructible then according to 4.6.3 we see that
'ö2 lies in some subfield of 1k where [0,:O]=2"' for some m. But by 4.6.1
2m=[Ot:0]=[Ot: 0] whilst = 3 (exercise 4.5.3).
This contradiction shows that p is not constructible from (0,0), (1,0) alone.

By similar means we investigate

(B) Angle trisection Of course some angles can be trisected, 90° and 27°
for example.* The point is that, if we exhibit just one angle which cannot be
trisected, this suffices to quash the assertion that every angle can be trisected
by straightedge and compass. We claim that 60° is not trisectible. Thus (see
Fig. 4.7) we are given three points (0, 0), (1, 0) and one other which we may
take as v-). [Why may we? Cf. Problem 9.]
r3
1 fi2' 2
/

(0,0) (1,0)

Fig. 4.7

We show that p = (cos 20°, 0) is not constructible. We argue as above that,

if it were, we would be able to find a field as in 4.6.3 such that cos 20° E
and :0] = 2" for some m. Setting 0 = 20° we find cos 30 = cos 60° = L But
cos 30 =4 cos3 0 —3 cos 0 so that cos 0 satisfies the equation 8?— 6x — 1 = 0.
Since this is irreducible over 0 (exercise 1.11.7(d)) we see that
[0(cos 20°): 0] = 3. Thus, from 4.6.3, we deduce that p is not constructible
rrom (0,0), (1,0) and (see exercise 4.)

Problem 9 Is cos 20° constructible if we are given the points (0, 0), (1, 0),
(cos 20°, 20°)? What if we are given this latter point—but not told its
cos
coordinates? Does this mean we can trisect 60°?
* can't prove it yet!
27°? 1 bet you
174 Factor rings and fields

(C) Squaring the circle To show that p = ("Jir, 0) is not constructible (given
P1 = '[(0, 0), (1, 0)}) we argue that if it were constructible so would be (n, 0).
[Why? See Fig. 4.8.] We now cheat by stating Lindemann's result (see p. 140)
that it satisfies no polynomial equation at all with rational coefficients. As a
consequence it cannot lie in any of the fields 0,. (For a proof see [38, p. 74}.)

Fig. 4.8

A fourth straightedge and compass construction which occupied the Greeks,

partial positive solutions to which appear in Euclid's Book 4, was the problem
of constructing regular n-sided polygons. It was known how to construct
polygons of 2", 2" . 3, 2" 5, 2" 15 sides and a construction for a regular
heptagon (n 7) was sought. In 1796 Gauss constructed a regular polygon
with 17 sides, the first new such polygon in 2000 years! (It is said that this
discovery made him decide to devote himself to mathematics rather than
linguistics.) In his famous Disquisitiones Arithmeticae of 1801 he shows that
polygons with p sides are constructible if the prime p has the form + 1.
(Fermat conjectured that all numbers of the form + I are primes. See
exercise 5.) He also warns the reader that for no other prime p can a
regular p-sided polygon be constructed—but offers no proof. He further asserts
that an n-sided polygon is constructible if n = 2°p1 . Pr where the are
. .

distinct Fermat primes. The necessity of this condition was not proved by
Gauss. It was first proved in 1837 by Wantzel* in a paper in which he also
showed the impossibility of the trisection and duplication problems by the
methods indicated above (viz, every constructible quantity satisfies an equation
of degree 2'for some t). (See exercise 8(b). For the sufficiency use exercise 8(a)
and 7.6.6).
For details about alternative construction problems, for example those
using a 'rusty' compass (!) or no compass at all see, for example, [84].
Exercises
1 (a) Given points (0, 0) and (1, 0) show how to construct (0, 1), Use ideas
from Fig. 4.8 to construct o) and then where 0.
*
There is an article on Wantzel in the Bulletin of the American Math. Soc., Vol. 24, 1917/iS,
p. 339.
 Symmetric polynomials 175

(b) Show that (1, in) is constructible from {(0, 0), (1, 0)} where

I i— - and m=

(c) Prove that the set {a:ae l1 and (a,b) is constructible from {(0, 0), (1, 0)}}
is a subfield of lit

2 Complete the proof of Theorem 4.6.1.

3 Complete the proof of Theorem 4.6.3. State and prove its converse.
4 In problem (B) shouldn't we be looking at [Q(cos20°): Q(13/2)]—rather
than [Q(cos2o°):Q]—in order to apply 4.6.3?
S (See exercise 2.5.2.) Show that, if 2k+ lisa prime integer, then k must be
a power of 2. [Hint: Assume where v is odd. Now use: x+ I f+ I in
Z[x] and put x=22"j
6 Construct a regular pentagon. [Hint: Use the formula
and the fact that is a root of y4+y3+y2+y+l to deduce that
2cos(2ir/5) is a root ofx2+x—l.]
7 Using the same technique as in exercise 6, show that the regular
heptagon is not constructible. [Hint: Write y6+y5+ +y+ I =
y3(x3+x2 —2x— l)=y3f(x) and show thatf(x) is irreducible in Q[x].]

8 (a) Show that, if a (regular) m-gon and a (regular) n-gon are constructible
and if (m, n) = 1, then a (regular) mn-gon is constructible.
(b) Let p be an odd prime. Show that: (i) no p2-gon is constructible; (ii) if a
p-gon is constructible then p = + 1 for some n. [Hint: Use the technique of
I
exercises 6, 7 to deduce that constructibility implies: (i)P(P2 must

both be powers of 2. See 1.9.17 and exercise 1.9.12.]

9 Construct, or show that it is impossible to construct,
(a) Regular polygons with 8 sides, 9 sides, 10 sides. (Do this directly—don't
use Wantzel's result.)
(b) Angles of 15°, 2°, 3°, 9°.

10 Construct, or prove it is impossible to construct,

(a) A square whose area is the sum of the area of 2 given squares.
(b) A cube whose volume is the sum of the volumes of 2 given cubes.

11 Using the equality to show that any angle

= 1

can be trisected to within any required degree of accuracy.

176 Factor rings and fields

4.7 Symmetric polynomials

Here we prove for use in Section 4.8 a basic result concerning certain special
polynomials in several letters which arise naturally in the problem of factorising
polynomials in one letter into linear factors.
LetF be a
field in which f factorises, f=(x—a1) ... (x—aj,
into a product of distinct linear factors. From the two expressions for f one
sees that

Now consider the ring F[xj, x2,. . . , x,j of polynomials in the (commuting)
letters x1, . , Setting
. .

Si =

XiXj
I t it

L X1X

Sn=X1X2...xn
we call the ith elementary symmetric polynomial in F[xi,. . , x,,I.
By a symmetric polynomial in F[xi,. . , x,,] we understand any element I
.

of F[xi,. . , which is left unaltered by every permutation* of the set

.

{xi,. , x,,}. In particular each of the (indeed every polynomial in the

. . is
a symmetric polynomial in F[xi, . . . , In Q[xi, x2, x3], xix +x2x +
x3x +x1x — 4x1 — 4x2 —4x3 is a symmetric polynomial whilst
X1X2+X1X3, are not.
The basic result referred to above is

Theorem 4.7.1.t (The Fundamental Theorem on Symmetric Polynomials)

Let P be a symmetric polynomial in F[xi,.. . , Then PeF[si,. .. ,
* See 2.6.5.
t This theorem also holds if F is just an integral domain.
 Symmetric polynomials 177

That is, P can be written as a polynomial, with coefficients from F, in the

elementary symmetric polynomials .. Sn.

Proof P is a sum of monomials a €F and the a1 are

non-negative integers. We can order the monomials of F[x1,. . , in a $

manner reminiscent of 'dictionary ordering' by writing ax 71 . x . .

ill a5 <b, where s is the least integer for which a1 b1. Thus, for
example, in Q[x1, x2, x3, x4] we would have
6 (1 3 ii 6 0 4 2
7x1x2x3x4 <—3x1x2x3x4

where zero exponents are used in the obvious way.

Suppose that, with respect to this ordering, ax 71 .. . x is the largest
monomial in P. Let cr be a permutation of {1, 2,. . , n}. Since P is symmetric, .

the monomial ax . x is a summand of P. It follows that a2

Now as7l_d52s72_a3. . = say, is symmetric in x1
. Writing gi
as a sum of monomials in the x1, the largest of these is evidently
ax71"2(xix2)"2'*3 . . is, it is precisely the largest monomial
. . . .

in P. Consider the (obviously symmetric [why?]) polynomial P1 = P —

P1 possibly involves more monomials than P but the largest monomial of P1
is smaller than that of P. Clearly this gives scope for a formal induction proof
(which we ask the reader to supply for himself). Informally: we repeat the
above process with P1 replacing P to obtain a new symmetric polynomial P2
whose largest monomial is smaller than that of P1. This process clearly stops
after at most N steps where N a1 [is this estimate correct?] when we
arrive at Then with each g
similar in form to g1, as required.
Note 4.7.2 Clearly this procedure is one which can be put to practical use.
Alternative methods are to be found in the exercises.
Example 4.7.3 In f= eF[x1,. . . , xj, the largest monomial is
Then (for what happens if n=2, 3?)
/fl \2/
x) xx1)=—
11 (

in terms of elementary symmetric polynomials, the following

symmetric polynomials: (Assume those in (a), (d) e Z[x1,.. . , xn] with n big
enough.)
(a) (b) (c) (d) L
1,11

2 Given that the equation x3 + 3x2 + 7x —2 has roots a, b, c find the poly-
nomial with roots
(i) a2, z52, C2; (ii) a +b, b +c, c +a; (iii) 1/a, 1/b, 1/c.
178 Factor rings and fields

3 Prove that the are algebraically independent, that is: For each n 1, if
f(y1,.. , y,,) is a non-zero polynomial in O[x], then f(s1,. . ,
. is also .

non-zero. [Hint: Assume false for some h(y1,.. hjy1,. .

.
.
+ . with k as small as possible. Assuming, for this k,
that m is also chosen as small as possible, prove: (i) h0(,y1,.. . 'n- and,
putting Xk =0, (ii) 5k— where the are (elementary) symmetric
in x1, . . ,
. Deduce that the polynomial in the mentioned in 4.7.1 is
unique.
4 Express f = x?x1 in terms of elementary symmetric polynomials as
1 i,j
follows:
(i) Show that f is expressible as a sum As + Bs + Cs3 where A, B, C are
constants.
(ii) Find A, B, C by choosing (a) Xi = 1, x2 x3 = 0; (/3) Xi = X2 1, x3 = 0,
etc.
5 For each keVu{0} put Use the equality (t—x1).. .(t—xj.
in to show that, for
that, Deduce for
For lCr<n, put
+(— l)rs,r. From the above conclude Ur(Xi,.. . , 0,. . . , 0)=0. Now, using
ideas from exercise 3, try to deduce

implies
.. ,x,,. . . , 0, .. . , 0) =0
Use these results to find r0,'r1,r2,r3,-r4,where x1,x2,x3 are the roots of
z3+3z2+7z—4.

4.8 The fundamental theorem of algebra

Here, using the minimum of analysis and quite a bit of our earlier develop-
ments, we prove the third result mentioned at the start of the chapter, the
Fundamental Theorem of Algebra. The first reasonable proof of this was
given by Gauss in 1799 (where others, including Euler and Lagrange, had
failed). In a second proof in 1815 Gauss assumed a result we assume here,
namely 4.4.3. It is interesting to note ([85, p. 598]) that proofs of this fact
then given would not pass today's more rigorous demands.
Theorem 4.8.1 Let* f(x)=ao+aix . eC[x]. There exists in C a
number a such that f(a) =0.
Remark The proof is quite long but is not so difficult if you take it steadily.
Proof Clearly we may assume that f(x) is irreducible over C since if f(x)
were reducible we should only need to prove the theorem for one of f(x)'s
irreducible factors.
*
For the readers convenience we write f(x) rather than [throughout 4.8.1.
 The fundamental theorem of algebra 179

Letus suppose initially that n is odd. Then if all the

a real, and hence complex, roo! by 4.4.3. If at least one of the is
not real we introduce the polynomial f(x) = + J1x + and form
the polynomial h(x) = f(x)f(x) of degree 2n. It is easily seen that f(x) is
in C[x] [since f(x) is] and that h (x) E [since h(x) =
f(x)f(x)f(x)f(x) = h(x)]. If h(x) factorises in R[x], h(x) —gi(x)g2(x),
say, where degg1<degh and degg2<degh, then since h(x)=f(x)f(x)=
gi(x)g2(x) in C[x] and since f(x), f(x) are irreducible and_hence prime in
C[x] we must conclude that gi(x)=cif(x) and g2(x)=c2f(x) (or g1(x)=
c17(x) and g2(x)=c2f(x)) for suitable units c1, c2eC[x]. Thus c1, c2eC. In
particular deg = deg = n. Since n is odd we can infer that gi(x), and
hence f(x), has a real root as required. Thus we may suppose that h (x) is
irreducible in R[x].
Now, using 4.2.20, we construct a splitting field E for h (x) over It In E,
h (x) factorises into a product of 2n distinct [why?] linear factors, h (x) =
—t1).. (x . Consider the polynomial k(x) = fl (x —t1 —t1)
2n(2n—1)
in E[x]. Clearly k(x) has degree , an odd integer. Further one sees
2
easily that any permutation of the t (1 i 2n) simply rearranges the factors
of k (x). Hence, on multiplying out the product, the coefficients of the various
powers of x are seen to be symmetric polynomials in the t1. Consequently
each of these coefficients is expressible as a polynomial (with integer
coefficients) in the elementary symmetric polynomials on the t,. But each of
these being a coefficient (or its negative) in h (x) is a real number. It follows
that each coefficient in k(x) is a real number, that is k(x) R[x] c E[x]. Using
4.4.3 again we deduce that k(x) has a real root: r = t1 +t1, say.
Now consider the polynomials h 1(x) = h(x h2(x) = h(—x Clearly
+f).
h1(x), h2(x)€IR[x] and are irreducible in R[x] (cf. example 1.9.17(u)). If
(h1(x),h2(x))=1 in R[x] then there exist* u(x), such that
u(x)h 1(x) + v(x)h2(x) = 1 in R[x], an identity which holds equally well in E[x].
Evaluating each side at the element in F we obtain, on noting that

and =h(t1)=0, the equality 0=1 in F. Thus

(hj(x), h2(x) in R[x]

both have the same leading coefficient we deduce = h2(x).

From this it follows that h1(x) = h(x = h(—x is a polynomial m(x2)

of degree n in x2. Since n is odd it follows from 4.4.3 that tn(f3) = 0 for some
/3€l%. Hence where lies in C. Thus
*
See 3.7.5.
180 Factor rings and fields

is a root of h (x) and hence of f(x) or 7(x). This completes the proof
of the theorem in the case where n is odd.
So assume n where w is odd and s and let t1, . , be the
. .

distinct roots of f(x) in the splitting field F for f(x) over C. We proceed by
W —1)
induction on 5. Forming k (x) as before we have deg k (x) =
2
and k (x) e C[x]. Consequently, by induction hypothesis, there exists c 6 C
such that t1 + = c for some i, f.
Formfi(x) =f(x andf2(x)=f(—x Thenfi(x),f2(x)are irreducible
over C and as above we can deduce [can we?] that (f1(x),f2(x)) 1 in C[x].
It follows that fi(x) = ef2(x) where e 6 C. Thus f(x ef(—x and

comparing leading coefficients we find = that is e = 1 (since n is

even). Thus again, as above, f(x =M(x2) is a polynomial in x2 in C[x].
Since deg M = 2Sw we can conclude that M(y) =0 for some -y 6 C. But then
and =0, as required.

Notes 4.8.2
(i) The complex numbers play a somewhat lesser role in algebra today than
in their heyday (namely the majority of the 19th Century) and in the alge-
braists' armoury the fundamental theorem has been replaced by the result
that to every field F one can find an algebraically closed extension field E.
To say that a field F is algebraically closed is to say that each polynomial
f e E[x] has a root in F. (This is equivalent to saying that each polynomial
in E[x] factorises into linear factors in E[x].) Thus 4.8.1 says that C is
algebraically closed.
(ii) C is not the smallest subfield of C which is algebraically closed. In fact
one can prove [49, Vol. 2, p. 40] that the set of all algebraic numbers (exercise
3.2.14) forms a field which is algebraically closed. Of course C is the smallest
algebraically closed field containing It
(iii) The above algebraic proof of 4.8.1 is quite long. There are short (once
the machinery has been set up!) proofs using complex analysis. For a nice
heuristic argument see [5, p. 107]. See also [101].

Exercises
None! If you've worked hard on this section you deserve a break.
5
Basic group theory

5.1 Introduction
This chapter introduces, and Chapter 6 examines more deeply, the third of the
three main types of algebraic systems that we study in this text, namely that of
group. We begin by outlining how Lagrange's investigations into the problem of
finding an algebraic formula yielding the roots of the general polynomial of
degree n led, naturally, to the introduction of what we now call permutations.
The culmination of these investigations, namely the remarkable result of the 20-
year-old Evariste Galois, which explains precisely when a polynomial equation
is soluble by radicals, is obtained in Theorem 7.10.5.
Following Section 5.2 we first introduce to the reader many concrete
examples of groups and then the basic tools and concepts of group theory.
Several of these concepts are the analogues of concepts already introduced
for rings and fields. This is not really so surprising. Homomorphisms, kernels,
cosets, subsystems, etc. are concepts of a general algebraic nature: we are
merely meeting their specialisations to rings, fields and now groups. We
close Chapter 5 with a rather sketchy description of the kinds of ideas
involved in the first application of group theory to a problem from outside
mathematics.
Within the next two chapters there are the occasional references to Chapters
3 and 4. The reader who has not studied Chapters 3 and 4 will lose nothing
if he simply ignores references to earlier definitions and takes on trust the
very few references to earlier theorems. On the other hand the reader is
assumed to be acquainted with the concept of binary operation (see 2.7.1).
In addition, reference to Section 3.1 and Remarks in Section 3.10 may help
the reader to understand the algebraist's general philosophy.

5.2 Beginnings
We give here a brief account of the ideas involved in Lagrange's explanation
of why, by 1770, the quest for a formula for the roots of the general quadratic,
cubic and quartic equations had proved successful whilst that for the quintic
had not.
In the introduction to his paper Réfiexions sur la Resolution Algébrique des
Equations (published in two parts in the years 1770 and 1771) Lagrange
182 Basic group theory

Joseph Louis Lagrange (25 January 1736-10 April 1813)

Lagrange was born in Turin, the eldest of eleven children, into a
family of French origin. On entering the University of Turin his initial
interests were in physics and geometry but he soon dropped the
former in favour of analysis. At about the time of his appointment at
the Royal Artillery School in Turin he began researches which made
him the founder of the Calculus of Variations.
In 1766 Euler left Berlin for St Petersburg. Frederick the Great
urged Lagrange to fill the vacancy with the words 'The greatest king in
Europe wishes to have at his court the greatest mathematician in
Europe'.
Between 1764 and 1788 Lagrange several times won the biennial
prizes offered by the French Académie des Sciences for solutions to
problems in celestial mechanics. From 1787 onwards he resided in
Paris and in 1788 he wrote his famous Méchanique analytique, a
beautiful work providing a unified treatment of mechanics and
which, he proudly announced, 'contains no pictures'.
He devoted much time to number theory. He was the first to prove
that
(I) for each square-free a e Z÷ the so-called Pell equation
x2 — ay2 = 1 has infinitely many solutions (1768);
(ii) every positive integer is the sum of 4 integer squares (1770);
(iii) Wilson's Theorem (1771) (see (2.5.4).
Apart from his work in algebraic equations (Section 5.2) he did
sterling work in numerical equations.

writes (*R206)
'I propose in this memoir to examine the different methods that have
been found for the algebraic solution of equations, to reduce these
to general principles, and to show a priori why these methods succeed
for the third and fourth degree and fail for higher degrees.'t
*
Rxyz indicates page xyz of Volume 3 of Lagrange's Oeuvres edited by T-A Serret.
± This doesn't mean that the quintic can't be solved, merely that it can't be done this way.
 Beginnings 183

Lagrange found that all the methods proposed by del Ferro, Tartaglia, Ferrari,
Descartes,* Tschirnhaus, Euler and Bezout found previously for the solution
of the cubic and quartic equations depended essentially on the same principle.
As he wrote at the beginning of the fourth section of his paper (R355)
'...all the methods are reducible to the same general principle,
knowing how to find functions of the roots of the proposed equation,
which are such: 1° that the equation or the equations by which they
are given, that is of which they are the roots (equations which are
commonly called the reduites), have themselves degree less than that
of the proposed equation or are at least decomposable into other
equations of a degree less than that.

Let us, by following Lagrange's discussion of two methods of solving the

general cubic, see how these reduites and the consequent appearance of
permutations of roots arose.

First note that the substitution y = x + M/3 reduces the equation x3 + Mx2 +
[six + P =0 to y3 + ny + p =0 for suitable n and p. The method of solution
proposed by Hudde (R207) tells us to put y = z +t whence y3 + ny +p =0
becomes z3 + + p + (3zt + n) (z + t) =0 and this will certainly be satisfied if
we arrange for z3+t3+p =0 and 3zt+n =0. From these two equations we
get =0, that is This equation, the reduite of

sixth degree (R213) is a quadratic in z3 with solution z3 = where

we have put q = + for brevity. Thus z can take any one of six possible
values,_namely a, wa, w2a, j3, o43 and w2f3 where a is one of the cube roots

(3 is one of the cube roots and cv is a complex

cube root of 1. Thus y = z + t = z — appears to be six-valued whilst being

a root of a cubic equation! (R208) However,_Lagrange notes that if we choose

a, j3 so that then

we have

(i)

or wa+w2(3 (ii)
3wa 3w/3
* René du Perron Descartes (31 March 1596— 11 February 1650).
 ___

184 Basic group theory

2
or y=y3=wa— 2
2 =w/3————=wa+w/3 (iii)
3wcr 3co/3

so that y is 3-valued, after all.

Lagrange goes on to observe that just as we have written the roots yl,
y3 of the given cubic in terms of the 6 roots of the reduite so the roots of the
reduite can be expressed in terms of the roots Yi, y3. Indeed from (i), (ii)
and (iii) above we soon find
1 2

w
2

1 2
w y1+coy3)

Thus one sees why (R215) the reduite is necessarily of degree 6. (There is
one root corresponding to each of the six permutations of the 3 'letters' Yi,
Y2, y3.) Further, one sees that the reason the reduite is a quadratic in z3 is
that the expression (y 1+ w 2Y2 + coy3)3 takes on only two distinct values when
the roots yi, Y2, y3 are permuted in all six possible ways.
A second method, in which we meet another 2-valued function, is that due
to Tschirnhaus (R222). The idea here is to determine constants a and b such
that, on solving the equation x2 = bx + a + y simultaneously with the given
equation x3 + mx2 + nx +p =0, we are reduced to an equation of the form
+ C =0. Then y can be found directly and x from the quadratic above.
Making the substitution required it is a few lines work to check that a and b
have to satisfy the equations
3a—mb—rn2+2n=0 (i)

and

3a2 — 2a (nib + in2 — 2n) + nb2 + (mm — 3p)b + — 2mp =0 (ii)

Substituting for a from (i) into (ii) we find a quadratic which must be satisfied
by b (whence the necessary value of a is immediate from (i)).
We suppose that the roots of y3 + C = 0 are labelled
Yi Y2 y3 = —co2 If the corresponding xs are x1,
X2, x3 we have

= bx1 +a

—wJC
x3=bx3+a—w 2 3C
2
 Beginnings 185

#+wx22+w2x32
whence it follows that b = so that a, too, is found. The six
X1+WX2+CU 2 X3
apparent values of I' (all of which are equally valid since b is surely independent
of the order in which we choose the roots of the given equation!) reduce, in
fact, to two.
Thus we see once again how the solution of the cubic depends upon being
able to find functions of the roots x1, x2, x3 (of the given equation) which
yield only two distinct values when the x1, X2, x3 are permuted in all 3! = 6 ways.
As regards the quartic equation x4 + nix3 + nx2 +px + r = 0, it is soluble (by
radicals) essentially because there exists a 3-valued function of four variables,
namely x1x2+x3x4. Under the 4! = 24 different permutations of x1, x2, x3, x4
the only values this function takes are =x1x2+x3x4, —x1x3+x2x4, g3 =
x1x4+x2x3. For then the cubic (y—gI)(y—g2)(y—g3) is a cubic with
coefficients which can be determined from the coefficients of the given equation
(exercise 1).
If we then set Zi =x1 we see that
z = (x, + ÷ x3 +x4)2 —4(x1x2 + x1x3 1x4 -'- x2x3 + x2x4 + x3x4) +4(x 1x2 + x3x4)

—4,'

Hence the two possible values of z, can be found. In a similar manner one
can find the two possible values of Z2=Xi—X2+X3---X4 and z3=
Recalling, finally, that —m =x1+x2+x3+x4 we can deduce

x1= —*m+*(z1+z2+z3)

x2= —Im+t(z1—z2—z3)

x3= +z2—z3)

x4= —Im+1(—z1 —z2+z3)

Each Xi has 2 = 8 different values but once again it can soon be deter-
mined which the correct four are. The point to be observed is that solving
the given quartic reduces to solving a cubic (because there is a 3-valued
function of x1, x2, x3, x4 and hence a cubic whose coefficients are obtainable
from the coefficients of the given equation) and three quadratics, whose
coefficients involve (i) the coefficients of the original equation (m and n above)
and (ii) a quantity which is expressible, using radicals, in terms of the
coefficients of the original equation since it is a root of the above cubic.
Thus Lagrange saw that to obtain, in like manner, the solution (by radicals*)
of the general quintic, he would have to begin by finding a function (equivalent
to above in the case of the quartic) of x1, x2, x3, x4, x5 which under the
* That is, by means of a formula involving +, -' —, ± and various / and the coefficients of the
given equation. See 7.7.2 for an 'official' definition.
186 Basic group theory

5! 120 permutations of x1,. . . , x5 yields at most four formally distinct values.

Having found such a function he'd then need to find the equivalent of the
zs above. Maybe a third (or more) step would be needed in order finally to
arrive, as above, at the required general formula.
Lagrange remarked (R403):
'Here, if I mistake not, are the true principles of the solution of
equations and the most suitable analysis to guide us to it; all is
reduced as one can see to a kind of calculus of combinations, by
which one finds a priori the results one might expect. It would be
appropriate to apply these to equations of the fifth degree and higher
degrees whose solution is at present unknown; but this application
requires too large a number of trials and combinations, whose success
is as yet much in doubt, for us to be able to spend all our time on
it now; we hope to be able to return to it at another time and we
content ourselves here with having laid the foundations of a theory
which seems to us new and general.'
In fact Lagrange never returned to the study of the general equation. He may
even have suspected the general quintic to be insoluble by radicals for he
succeeded in finding a 6-valued function of five variables but not a 4-valued
one. Indeed such a function cannot exist as was proved by Ruffini in 1799
not only for n—S but for all n
For further details relating to Lagrange's work the reader might profitably
consult [85], [103], [108], [114], and, of course, Lagrange's paper itself!

Exercises
1 Given that X2, 13,14 are the roots of x4+mx3-Fnx2+px+r=0 find
the coefficients of the equation whose roots are 1112 + 1314, 1 1X3 + 1214, x t14 +
1213.
2 Show that under the 24 permutations of x1, x2, X3, 14 each of the functions
(x1-i-x2)(x3+x4) and (xl—x2+x3_x4)2 is three-valued.
Find, in each case, the corresponding cubic equation.

5.3 Axioms and examples

Having seen in Section 5.2 a little of the mathematics which essentially gave
rise to the theory of (permutation) groups, we now begin an account in modern
style of the rudiments of that theory. Although in practice an abstract concept
tends to emerge only when there are sufficiently many interesting examples
to warrant it, we reverse this ordering here and begin with

Definition 5.3.1 A group is an (ordered) pair (G, o) where G is a non-empty

set and ° is a binary operation on G satisfying the following axioms:
 Axioms and examples 187

(A) The associative law: that is, for all a, b, c e G we have (a o 1,) ° c =
a ° (b ° c);
(N) Existence of neutral or identity element: that is, there exists in G an
element e, say, such that for all a G, e o a = a ° e = a;
(I) Existence of inverses: that is, to each a 6 G there exists in G an
element denoted by such that a =a1 oa =e.
Remarks
(i) (cf. 3.2.3(i)) It is common practice to talk of the group G rather than
the group (G, o) We shall follow this practice except when seeking extra clarity.
(ii) (cf. 3.2.3(u)) The letter e (an abbreviation for the German word einheit)
is used here rather than the symbol '1' for the identity element to remind
the reader that the most important applications of group theory are to systems
of elements which are not numbers.
(iii) According to our definition of binary operation on G (see 2.7.1) it is
automatically the case that, if x, y 6 G then x ° y 6 G. One often emphasises
this fact, however, by saying that G is closed under If one denotes this
property by C (for closure) a mnemonic for recalling the group axioms is
given by the name CAIN.* This is particularly appropriate in view of

Definition 5.3.2 If for all x, y in G we have x ° y = y ° x, the group G is

called commutative or ahelian (after N H Abel who showed the importance
of such groups in the theory of equations).

Notation and Terminology 5.3.3 (cf. 3.3.3) It is usual, in dealing with groups
in general, to replace o by , or even to omit the dot altogether and simply
juxtapose the elements being combined. The method of combination is called
multiplication with a b (or a b or ab) being called the product of a and b.
When the group one is dealing with is known to be abelian one sometimes
(but not always) replaces o by +, and refers to the binary operation as addition
and to a +/, as the sum of a and b. In this latter case we also write 0 rather
than e, —a rather than a' and, of course, (a+b)+c rather than or
(ab)c.
To show that there do exist objects to which the theory we are going to
develop does apply, we now give an extensive list of specific groups of different
kinds. The examples given can be used by the reader as test cases on which
to try out the theorems we prove as well as the various questions which might
naturally occur to him.

Examples 5.3.4
(a) (1, +), where + indicates the usual addition on 7, is an abelian group.
The identity element is, of course, 0 and the inverse of, say, —3 is +3.
(b) +) and where is ordinary multiplication, are abelian groups.
* This is taken from the book [65, p. 291.
188 Basic group theory

Remark (a), (b) are rather trivial examples and there would be no need of
a theory of groups if these were the only examples. Indeed we learn nothing
we didn't know already about these three groups from our theory, although
there is an interesting structural similarity, which we shall uncover in 5.9.5
between the two groups in (b) which is perhaps not apparent at first glance.

Examples 5.3.4
(c) (Generalising (a) and (b).) Let (R, +, be a ring and (F, +, be a field.
Then (R, +), (F, +) and (Fx,.) are all abelian groups. (For any field F the
set of non-zero elements of F is usually denoted by F><.)
It might be wise, before going further, to give examples which fail to be
groups.
,
(d) (7 —). In fact it's a bit unfair even to write down (t, —) since — is not
a binary operation on For instance,
(e) (7, —). This time — is a binary on the given set 7 but axiom
A fails. [What about N and I? Do they hold or not?]
(f) (7, +). Here N fails since the only conceivable candidate, namely 0, is
not in 7'-.
(g) Here I fails. For instance 2€?'- but there is no x €7 such that
2 x= 1(1 being the clear candidate to satisfy N).
(h) Let n €t and let M(n) denote the set of all equivalence classes, modulo
n, of integers which are coprime to n. Then (M(n), 0) is an abelian group
with qS (n) elements. As we saw in Section 2.4, 1 acts as the identity element
forOand,ifn
Remark We do not use the additive notation with this abelian group. The
multiplicative notation is much more suggestive and certainly less confusing.

In Section 5.1 we talked of permutations. We recall from 2.6.5 the

Definition 5.3.5 Let X be any non-empty set. A permutation on K is a

function f: K -* K such that f is 1—1 and onto.

Reminder 1ff, g are two 1—i functions from K onto K then their composition
f ° g is defined by x(f g) = (xf)g for all x K and is itself i—I and onto. (See
2.6.7 and exercise 2.6.9(u).)

Example 5.3.4
(j) If P(X) denotes the set of all permutations on K then (P(K),°) is a
group. For, we have just remarked that is a binary operation on P(X);
composition of functions is associative (see 2.7.6); the identity function,
I:X—.X given by xI=x, for all x€X, acts as the identity element of P(X)
with respect to and, finally, given f P(K) the function acts as an inverse
tof (see 2.6.5 and exercises 2.6.11, 2.6.12).
 Axioms and examples 189

Remarks
(i) When K is a finite set with n elements the set P(X) is often denoted by
The group (more accurately, the group (Sn, o)) is called the symmefric
group on n symbols since leaves fixed each of the n (formal) elementary
symmetric polynomials on the elements of X (see Section 4.7). Clearly Sn
contains exactly n! elements.
(ii) (h) gives the first (specific) example in this chapter of a finite group, that
is a group with only a finite number of elements. ((a), (b) give examples of
groups with infinitely many elements, otherwise called infinite groups.) (j)
gives, in the case when X has more than two elements, our first example of
a non-abelian group. For example, if X = {a, b, c} and if
af=c ag=b
bf=a bg=a
cf=b cg=c
then a(fog)=(af)g =cg =c whereas a(g of)=(ag)f=bf —a. Thusfog and
g of have different effects on a E K; hence f ° g g of, immediately.

For practice, and to assist with 5.3.6, you might care to confirm that
b(fog)=b whilst c(fog)=a.
Notation 5.3.6 If we express f and g in the form suggested by Cauchy,
a b c\ a b c\
namely f and g then their product f ° g is, accord-
= ( c a lv ( b a cl
(12 IT).
ing to the above calculations, expressible as f o g can be calculated
quite quickly according to the following self-explanatory scheme
•

b)
f '= (' whkh may be expressed as if desired.
(A still more economical way of working with permutations will be developed
in Section 5.5.)
Definition 5.3.7 The number of elements in the group (G, o) is denoted by
GI and is called its order.

Thus we can talk of 'a group of order 6' (for instance the group 53 just
dealt with) or 'a group of infinite order' (as for example in 5.3.4(a)).
190 Basic group theory

To show that there exist groups of every finite order n (ii E we note

Examples 5.3.4
(k)

complex nth roots of 1. With respect to multiplication forms a group of

order n. [This is essentially de theoremj
To get hold of some more examples of non-abelian groups which are
group-theoretically significant and to connect group theory, in albeit a rather
trivial way, with the symmetry of geometrical figures we look at
(1) The group of isornetries of the plane. This is the set F of all those
permutations! of the points in the real plane x P such that for all x, y e P x P
the distance between x and y is equal to that between xf and yf. Such
permutations include rotations, translations and reflections (about a line or
point). Given a geometrical figure in the plane the number of such motions
which leave the figure 'unchanged' is clearly a measure of the figure's sym-
metry, larger numbers corresponding to greater symmetry.

Fig. 5.1

Consider, for instance, the regular n-gon (drawn in Fig! 5.1 for ii = 8). We
label the vertices solely to keep track of them under various rigid motions.
Now any isometry of the plane which preserves the overall position of the
n-gon clearly must send the vertex 1 to one of the vertices 1, 2,. . , n. If we .

suppose that 1 is mapped to the vertex K then the isometry must, since it is
distance preserving, send 2 and n to K — 1 and K +1 (or K +1 and K — 1)
respectively. Whichever of these two possibilities occurs the positions of the
remaining vertices (and indeed all points in the planet) are then automatically
determined. Thus there are exactly 2n distinct isometries of the plane which
'preserve' the regular n-gon. This group of isometries is called the dihedral
group of order 2n. We denote it by
We look a little more closely at this example in the case of n = 8. Let a
1 2 3 4 5 6 7 8'.
denote the isometry determined by the permutation (2
3 4 6 7 8 i)
*
Abraham de Moivre (26 May 1667 —27 November 1754).
t Since an isometry is completely determined once the images of three non-coilinear points are
known ([68, p. 140]).
 Axioms and examples 191

that a is an anticlockwise rotation through in4 and let b denote the reflection
in the vertical line through the mid-point of the sides 1—2 and 5—6. Thus b
is completely determined by the permutation (2
12345678 Using the
1876 4
same letters for the permutations as for the isometries they correspond to,
we see, using the notation introduced in 2.7.9, that

a 2 =a°a=ç3
f12345678\ a
3712345678\ etc.,
4 5678 1 2)' 5678 1 2 3)'
whilst a8 is the least positive power of a equal to the identity permutation
(or isometry). It is not difficult to check that the 16 (=2n) symmetries of
the regular octagon can be expressed (omitting the a signs) as

I,a,a2 a7,b,ba,...,ba7
where I is the identity permutation and ba for example, denotes (the isometry
determined by) the permutation
2 3 4567 2 3 4 5 6 7 8\_(1 2 3 4 5 6 7 8
½ 1 8 7 6 54 3R4 5 6 7 8 2 3)k5 4 3 2 8 7 6
1 1

One final remark: It is not difficult to check—see the demonstration in Fig.

5.2, but also check it from the permutations given above—that b1ah = =
a7 a. it follows that ab ba so that the group of symmetries of the regular
octagon is not abelian.

Fig. 5.2

As a final example on permutations

fX y z t\ fX y z t\ y z t\ yz t
(m) LetV= xytz yxtz
xyzt yxzt
Then each element of V gives rise in an obvious manner to an automorphism
of the ring Z[x, y, z, tJ (see 3.1O.2(iv)). We leave the reader to check that
(1", a), wherea denotes composition, is a group.

Applying these four permutations to the polynomial xz + y yields the four

distinct 'values' xz +y, yz +x, xt+y, yt+x. Thus, under the group V, xy +z
is a 4-valued polynomial which is fixed only by the trivial subgroup of V.
Similarly xy + zt is a one-valued function whilst xy + z is two-valued. As we
have seen in Section 5.2 these kinds of considerations were of interest to
Lagrange.
192 Basic group theory

Problem 1 Does V give rise to a 3-valued function on x, y, z, t?

Next we present three more specific groups of order 4.

Examples 5.3.4
(n) (i) Let V1 = {1, —1, i, —i} and let ° denote (ordinary) multiplication of
complex numbers.
(ii) Let
(1 (? with o taken as

matrix multiplication.
(iii) Let V3
={(i n 14} with o taken as matrix multiplication.
(p)* The set of all invertible (i.e. non-singular) n X rz matrices with
coefficients in the set X, where X=Z, C, C or 4, forms a group with
respect to matrix multiplication. (By definition, the element A of MJX) is
invertible if! there exists B e such that AB = BA = h, the identity
matrix of Thus A is invertible if! detA = I or —1, and
A€Mn(lm) is invertible if! detA is a unit in 7,,,. For X=Q or or C,
A is invertible ill.. . what?)
(q)* For each X above the subset SL,, (X) of M,,(X) comprising those
matrices with determinant + 1 forms a group with respect to matrix multipli-
cation.
Groups of this kind are, for X = R or X = C, of interest to physicists and
chemists. They are special instances of so called Lie groups. These are
essentially groups L in which (i) the elements can be labelled by r-tuples of
continuously varying real number parameters so that (ii) some sort of 'near-
ness' condition (a 'near' c in L and b 'near' d in L is to imply ab is 'near' cd
in L) holds. As another example
(r) The set all mappings of R to of the form
Ta,,, (x) = ax + b constitutes a group with respect to composition of mappings.
Each real number pair (a, b) with a 0 provides a group element. The identity
element is T10. For products and inverses see exercise 10.

Exercises
I Are the following sets, together with the given multiplication, groups? If not,
list the first of the axioms C, A, N, I which fails.
(a) M2(75): matrix multiplication;
(b) all elements of M2(4) with determinant I: matrix multiplication;
(c) (Z,°)wherea°b=a+b—37;

*
If anything in this example is unfamiliar to you look it up now. For cf. 3.2.5. GL and
SL stand for general linear (group) and special linear (group) respectively.
 Axioms and examples 193

(d) (R,o)whereaob=a+b_ab;
(e) ({2, 4, 6, 8}, 0) where 0 is multiplication mod 10;
(f) the set of all rotations about the origin of 3-dimensional space which
permute the unit vectors ±i, ±k: composition of rotations;
(g) the set of all isometries of 3-dimensional space leaving a given cube
'unchanged': multiplication of isometries of 3-space (cf. 5.3.4 (1));
(h) the set of all permutations on {1, 2, 3, 4} which send 1 to 3 or send 2
to 3: composition;
(i) the set of all vectors in 3-dimensional space: vector product;
(j) {a +bv'2: a, b Q}: ordinary addition;
(k) the non-zero elements of exercise (j): ordinary multiplication;
(I) the six functions fi(x)=x, f4(x)=1, f5(x)=

1 — x, f6(x) = defined on the set of all real numbers other than 0

and 1: composition of functions.

2 For those parts of exercise 1 which are groups state which are abelian and
which have finite order.
3 Let(G,o)and(H,*)begroups. on GXH
(g1 a g2, h1 * h2). Show that KG x H,) is a group. (Cf. exercise 3.2.9.)
4 Let P be your favourite pop record and H your favourite hammer. Show
how to make the set {P, H} into a group by defining 'multiplication' suitably.
Deduce that there exist infinitely many distinct groups of order 2.
5 Let(G,°)be a group. Define * on G bya *b =b oa. Is(G,*)a group?
6 An affine geometry comprises a set X whose elements are called points
together with various subsets of X called lines such that
(i) Each pair of distinct points is contained in exactly one line.
(ii) Each pair of distinct lines has at most one point in common.
(iii) Given a line 1 and a point P not on it there exists exactly one line n
which contains P and has no point in common with L
b
,
, 'N

— — 'N
csrc'r_
Fig. 5.3

(iv) There are at least two lines. Figure 5.3 gives a pictorial representation
of an affine geometry with 4 points and 6 lines.
A collineation of an affine geometry is a permutation of the points
of X which maps lines to lines. Show that the set of all collineations
194 Basic group theory

forms a group under composition. What is the order of the collineation

group of the above 4-element affine geometry?
7 (An example for those who have read Section 3.10) An automorphism
of a field (F, is a I — mapping t/, from F onto itself such that, for all a,
1

be F, (a + b)t/i = at/i + bc/i and (a b)t/i = at/i bc/i. Show that, under composition,
.

the set of all automorphisms of F forms a group G, say. Let K be a subfield

of F. Show that the set of all automorphisms r of F which are such that kr = k
for all k e K is also a group. It is called the Galois group of F over K.
2
a +b 2
8 Let P be the set of all positive rationals of the form 2 2 where a, b,
c
c, d 7. Is (P, a group, being ordinary multiplication? (Exercise 3.8.4 might
help.)
9 Find the inverse of in (M(11),G) and in (M(12), 0).
10 Show that, in example 5.3.4(r), 0 = T0C.0d÷b. What is the inverse
in T of T0,b?
11 How many elements has the group of isometries of the plane leaving
unchanged (i) the regular 27-gon?; (ii) the circle of unit radius?
12 Let fez [xi,x2,x3]. List all the elements of 53 which
leave this polynomial fixed.
13 Let S be any set with more than one element and define for all x, y €5,
x y = x. Show that ° is an associative binary operation on S. Is (5, o) a group?
(A pair (5, o) where S is a set and o is an associative binary operation on S is
called a semigroup.)
14 Let S be a non-empty set and an associative binary operation on 5,
such that, for every a, b S the equation a o x = b has a unique solution for
x in S. Is (5, o) necessarily a group?

5.4 Deductions from the axioms

We now parallel the development of Section 3.3 by making some logical
deductions from the axioms listed in 5.3.1. As in the case of rings, no
assumptions are made concerning the nature of the elements involved. Con-
sequently any conclusions we draw will be applicable to any of the specific
groups mentioned in Section 5.3—and to any other groups we come across.
Our first observation is that the axioms of 5.3.1 conceal the fact that the
elements given to exist in (N) and (I) are unique. They also conceal some
redundancies, as exercise 9 shows.

Theorem 5.4.1 (cf. 3.3.1) In any group G there is exactly one identity
element. Further, to each a e G there corresponds exactly one inverse.
 Deductions from the axioms 195

Proof Axiom N of 5.3.1 assures us of at least one element e such that

ea = ae = a for all a in G. 1ff e G also satisfies axiom N we havefb = bf = b for all
b in G. Taking a and b to be the particular elementsf and e we get ef =fe =f and
fe = ef= e. Combining these we see that e = ef=f thus proving the uniqueness of
the identity element.
Axiom I assures us that to each a E G there corresponds at least one inverse
element, namely a - Let b e G be any element such that ab = ba = e. Then
(a1a)b=eb=b 1.Since(a'a)b=a'(ab)we
have b = a -' and uniqueness is proved.
An immediate consequence is that the inverse of the inverse is the original.
1)
Theorem 5.4.2 (cf. 3.3.2(u)) For any group G and any a e G, (a
1
= a.

Proof We know a a a1 a = e. From this it follows immediately that

a is an inverse for a But is the notation for the (unique) inverse
ofa'.Thus(a')'=a asrequired.
Another consequence shows that, unlike the case in the theories of rings
(exercise 3.2.10) and of semigroups (exercise 7) cancellation is always possible
in a group.

Theorem 5.4.3 If a, b, cCG are such that ab = ac, then b = c.

Proof From ab = ac we deduce a '(ab) = a1(ac) [why?]

whence [why?]
so that eb = ec [why?]
thatis b=c [why?]

A final consequence is a slight surprise (unless you've done exercise

3.3.9(a)).

Theorem 5.4.4 Let a, beG. Then = (and not, in general,

&

Proof 1a')=((ab)b)a'=(a(bb'))a1—(ae)a=aa'=e.
= (ab) follows.
1

Similarly (ab) = e and

We confirm this in one special case.

Example 5.4.5 Let A B then A, B e GL2(Q).

= =
4)

(ABY'=(?
196 Basic group theory

whereas

Remarks
(i) The fact that, in general, a is just a fact of life that the
beginner must learn to live with. In particular the reader should refrain from
using the notation preference to a ', as many beginners are then tempted

to write = —. — —a quite natural reaction since the only occasions on

a b
which most readers will have used this equality is in the case of rational, real
or complex numbers where of course the equality is valid.
(ii)The equality (ab)' =a 'b1 is universally true in a given group G when
and only when the group is abelian (see exercise 2). That is, this property
characterises abelian groups within the class of all groups.
The brackets were left in the proof of 5.4.4 in order to exhibit clearly
exactly which axioms were being employed at each stage. They could have
been left out because (see 2.7.7) the generalised associative law follows from
axiom A of 5.3.1.

For completeness and amusement we note, without proof,

Theorem 5.4.6 (cf. 2.7.7) Let a1, a2,. . . , be elements of G. There are
(2n —2)!,
ways of evaluating the product of a1, a2,.. . , in that order (see
n!(n —1).
exercise 1.2.1 and [20, Vol. 1, p. 19]). All these ways yield the same group
element which can thus be denoted unambiguously by a1a2. . .

Exercises
1 Show that if G is a group and if xe G is such that x2 = x then x = e.
2 Show that a group G is abelian if for alla, beG we have =a
3 Let G be a group such that X2 = e for all X e G. Prove that G is abelian.
4 Given a, b, c elements of the group G, show that there exists in G a
unique element x such that aXb = c.

5 Show that if the finite group G comprises the elements g1,

g is one of these, then the list g1g, g2g contains each of the elements
of G exactly once.
6 Let G be a group, a an element of G and m and n elements of
Defining a°=e, and =(a'Y show that a "=(aT1 and that the usual
 The symmetric and the alternating groups 197

lawsof exponents hold; that is, for all U, V e 7, = a" + V and (a7 = a"".
Show that if G is abelian and if a, b e G then (ab)' = a'!" for all tel.
7 Give a specific example of a semigroup S and elements a, b, c C S
such that ab = ac but b c [You've had examples already in this chapter!]
8 Try to prove the formula in 5.4.6.

9 Let G be a set and o an associative binary operation on G.

(i) Consider the axioms
NR: There exists in G an element e such that, for all aeA, a°e=a.
JR: To each a £1 there exists in G an element such that a ° a1 = e.
Show that (G, o) is a group.
(ii) Defining NL in an analogous manner, show, by exhibiting an example,
that if * is an associative binary operation on G and if * satisfies NL and
'R then (G, *) might fail to be a group. [You have already seen a suitable
example in this chapter.]

5.5 The symmetric and the alternating groups

In this section we look more closely at the groups 5,, introduced in 5.3.4(j).
There are three main reasons for this. The least important is that we thereby
offer the reader even more concrete groups on which to test our theorems
and his theories—indeed, in a sense, we offer him all groups! (See 5.9.6.) The
other two reasons are: (i) with each 5,, we shall associate a group A,, called
the alternating group on n letters which, for n 5, will prove to be a basic
building block in the theory of finite groups (see 6.5.9 and 6.6.1); (ii) the
problem concerning the existence of a radical formula for the roots of the
general quintic equation is resolved using S5 (Section 7.9).
We begin by replacing the 'double row' notation of 5.3.6 by something
easier.
It will probably help the reader if he reads through the following three
definitions and theorem with 5.5.5, below, in mind.

Definition 5.5.1 Let f be a permutation on the set X = {1, 2,.. ,n} and let
.

x1,x2,... ,x, be distinct elements of X. If for we have

x if Xrf = x1 and if yf = y for every other element of X, then f is called
a cyclic permutation or cycle and is denoted briefly by (x1, x2, . . . , xr). r is
called the length of the cycle.

Next, let f be any permutation on X and let a X be chosen arbitrarily.

Writing f" for the composition off with itself k times (k >0) in (Sn, a) consider
the subset
A = {a, a!, af2 ,.. . ,
n
af }
198 Basic group theory

of K As there are only n elements in X not all of these n + 1 elements of X

can be distinct. Reading the above list from the left, suppose af' is the first
element of X to be repeated. Interpreting af° as a we can then say that there
exists i e / such that 0 i <f and af' = af'. Applying the permutation f to
this equality* we find that (af')fl = (af')f, that is, a = af'. By the minimality
of I we see that i = 0. It follows (see exercise 7) that A comprises precisely
the set of distinct elements a, af,.. , af''. We make
.

Definition 5.5.2 The set A described above is called the orbit of f on X

determined by a.
Now suppose that the elements a, af,. . , af'' do not exhaust X. Then
.

there exists an element /3, say, in X but not in A. As we found A so we find

B, say, the orbit of f determined by j3. Clearly A n B = 0 (see exercise 8).
Continuing to obtain orbits in this fashion until all the elements of X are
exhausted, we can put all this information, and more, into a theorem if we
first make

Definition 5.5.3 To each orbit A (B,. . . , etc.) define the function

A (fB,. . , etc.) by
.

xfA=xf ifxEAl (and similarly forf8,. . . , etc.)

xfA=x
We then have the promised
Theorem 5.5.4 Let f be a permutation on the finite set X and let A, B,.. ., T
be the finitely many distinct, pairwise disjoint orbits of f on X. Then each of
A, fB,. .. , Pr is a cyclic permutation and f is equal to their product, taken in
any order. That is, f is expressible as a product of disjoint cycles.
Example 5.5.5 If
2 3 4 5 6 7 8 9 10 11 12 13 14 15
10 13 4 5 12 3 15 2 9 11 6 1 14 8
then, sincef maps ito 7, 7 to 3, 3 to 13 and 13 back to 1; 2 to 10, 10 to 9
and 9 back to 2; 4 to 4; 5 to 5; 6 to 12 and 12 back to 6; 8 to 15 and 15
back to 8; ii to ii and finally 14 to 14, we see that the various orbits are
{i, 7,3, i3}, {2, 10, 9}, {4}, {5}, {6, 12}, {8, 15}, {ii}, {14} and in the notation
of 5.5.1 we can write (omitting the o symbols)
f= (1,7,3, 13)(2, 10, 9)(4)(5)(6, 12)(8, i5)(ii)(i4)
$
According to the theorem we can express f in many other ways, one of which
is
f= (2, 10, 9)(8, 15)(4)(6, 12)(11)(1, 7,3, 13)(14)(5)
*Or using the i—mess of f on the equality
 The symmetric and the alternating groups 199

On the other hand, except for variants of this type, and those arising from
noting that, for example, 13)=(7,3, 13, 13, 1,7)=
(13, 1, 7, 3), the way of representing f is clearly unique.

Notation 5.5.6
(i) It is usual to omit cyclic permutations involving only one element. Thus,
one more way of writing f above is as follows:
f=(6, 12)(8, 15)(2, 10, 9)(1, 7, 3, 13)
(ii) Commas too are often omitted within the cycles.

To introduce the important concept of alternating group we first make

Definition 5.5.7 If the permutation f on the set X is a cycle which inter-

changes just two elements of X, leaving all the others fixcd, thcn f is called a
transposition.
We have immediately
Theorem 5.5.8 Every cyclic permutation (and hence every permutation) can
be expressed as a product of transpositions.
Proof One only has to check that
(x1x2) . (x1x3) .. . . = (x1, x2,.. . , xj.
Note that the way of writing f as a product of transpositions is by
no means unique. For instance, (12345)=(12)(13)(14)(15)=
(24)(45)(3 5)( 1 2)( 1 4)(45).

On the other hand we do at least have

Theorem 5.5.9 Let the permutation f on the set X = {1, 2,. . , n } be expres- .

sible in some way as an even (respectively odd) number of transpositions.

Then every way of expressing f as a product of transpositions requires an
even (respectively odd) number of transpositions.
Proof* Consider, in Z[x1, x2, . . . , xj, the polynomial
P= [T (x1 —x1) = (x1 —x2)(x1 —x3) . . . (x1

(x2—x3). . .

For a more direct proof see exercise 11.

200 Basic group theory

Now each transposition (4), say, sx gives rise to a mapping (indeed an

automorphismt) of Z[xi, x2,. . , x,,] onto itself, namely that determined by
.

interchanging x1 and x and leaving the remaining Xk fixed. Such a mapping

changes P to —P. Likewise each permutation f of S< gives rise to a mapping
of Z[xi, x2,. . , x,,] onto itself and similarly maps P either to P or to —P.
.

Clearly if f can be expressed as a product of an even number of transpositions

then f maps P to P. It is then immediate that f cannot be written in any way
as a product of an odd number of transpositions, for if it could f would
simultaneously map P to —P, an obvious contradiction. An identical argument
holds if f is given to be expressible in at least one way as a product of an odd
number of transpositions.

Because of this last theorem the following definition is unambiguous.

Definition 5.5.10 Let f be a permutation on the finite set X. f is an even

(respectively odd) permutation if f can be expressed as a product of an even
(respectively odd) number of transpositions.

Finally, since the product of two even permutations is now clearly even,
since the identity permutation is even and since the inverse of an even
permutation is even, we have

Theorem 5.5.11 Let X = {1, 2,. . , n}. The set of all even permutations on
.

X forms a group under composition of functions. This group, called the

alternating group on n symbols, is denoted by or A,, and has elements.

Proof The only part left to be proved is that A,,j = Now every permutation
on X is either even or odd. Let pi, P2,.. . Pr be the set of all even permutations
and . ., q5 be the set of all odd permutations [so that r +s = n !]. The
permutations (l2)pi, (l2)P2,. . . , (l2)p. are all odd, are pairwise unequal
[why?] and there are clearly r of them. Thus r s. Similarly the permutations
(12)qi, .. . , (12)q, are all even, pairwise unequal and there are s of them.
Thus s r. Consequently r = s
=

Exercises
I The set S of all functions from X = {1, 2, 3} to itself is, with respect to
composition, a semigroup (see exercise 5.3.13) with 27 elements. Show that
it is possible to find a EX andfeS such that af =af' with i Cf but af''
[Hint: Look for a, f such that a, af, af2 are distinct but af =af' if
*
5x is used as an alternative to P(X).
tSee 3.1O.2(iv).
 The symmetric and the alternating groups 201

2 Express as a product of disjoint cycles:

(a)
123456785); 1234567
(2 416837 (1 4 2 7 5 3
3 Express as a product of disjoint cycles:
(1 2 3 4 5 6 7\ (1 2 3 4 5\ (b)
(a)
64725 3

36 4 63
Using only the double row notation find XY, YX, Y 'XY. Can you see any
relationship at all between X and Y - 'X Y?
5 Repeat exercise 4 using the single row notation.
6 (i) Prove that if f is any permutation on X and if Xi, x2,.. ,.t, are distinct .

elements of X . ,xr)f=(xif,x2f,.
. ,xj). . .

(ii) Show conversely that if u and v are cyclic permutations of the same
length on X then f'uf= v for some fE
7 (i) Show that the orbit of the permutationf on X determined by a is
equally well determined by af, by af2, etc.
Ic
(ii) Prove that if k is the smallest positive integer such that at = a then for
each I El such that 1> k, af' {a, af,. , afk_i}. [Hint: use the division
. .

algorithm on I and k.]

8 Show that distinct orbits determined by any permutation are necessarily
disjoint.
9 Is a cycle of odd length an odd or an even permutation? Determine which
elements of S4 are even and which are odd.
10 Find two distinct even permutations onX = {1, 2, 3,. . , 9}, each of which .

has just one orbit with 4 elements and moves I to 6, 2 to 3 and 7 to 9.

11 Showthat(...Xay.. .)(ab) =(.. .Xbay...)
that(...Xaby.. .)(ab) =(.. .Xby...)
that(. ..xay. .zbt. .)(ab)
. . =(. .Xbt. . .)(...zay...)
. .

andthat(. ..Xay..j(. .zbt...)(ab) =(...Xbt...zay...)

.

Use these results to prove by induction on tin Let t1 tm be transpositions,

and let t1t2.. t,,, = .. . where c1 . are disjoint cycles. Then m

E (1(c1) — 1) (mod 2), where 1(c,) is the length of c. Deduce that a product of
202 Basic group theory

an even number of transpositions can never be equal to the product of an

odd number of transpositions.
12 Show that the permutation
1 2 3 4 5 6789101112131415
\is 1413121110987 6 5 4 3 2 1

is odd. It is this which lies behind the fact that the numbers 1, 2,. .., 15 in
the well-known '15-puzzle' cannot (without cheating!) be rearranged in the
reverse order.
In the 15-puzzle can the position

r 14 11

L
112

[6 5'lO 4

13 1
8 3 15

0
be changed (legally!) to

12 13

4 11 3 14
- 7
6 9 8 10

5 2 1

0
5.6 Subgroups. The order of an element
In studying rings and fields certain subsets, namely subrings and subfields,
arose in a natural way. A similar situation exists in group theory where,
amongst all the subsets of a group G, those which are groups in their own
right, the so-called subgroups of G, stand out as worthy of consideration. A
special type of subgroups, normal subgroups, plays a role analogous to that
played by ideals in the theory of rings at least as far as their connections with
homomorphisms are concerned. On the other hand it seems fair to say that
the concept of subgroup is more important in group theory than is the concept
of subring in ring theory.*
*
Of course in field theory subfields are more important than ideals (see exercise 3.4.14 and
Theorem 3.10.9).
 Subgroups. The order of an element 203

We begin with

Definition 5.6.1 (cf. 3.4.1) A non-empty subset S of the group (G, o) is called
a subgroup of G if! (a) the restriction of ° to S x S is a binary operation on
5, and (13) (5, 3) is a group.
If S is a subgroup of G we shall write S c G. If we know (and if we care)
that S G we can denote this fact by S C G.

Remarks (cf. 2.7.3 (ii) and 2.7.1)

(i) Although it is technically incorrect to do so, no danger will result if we
replace by Requirement (a) above can then be restated as: S is closed
under
(ii) Thus, informally, a subgroup is a subset which is itself a group with
respect to the binary operation induced by (or passed down from) the group.
(iii) We emphasise the requirements in italics above (just as we did for rings;
see 3.4.4(u)), by noting that whilst (R, +) and are both (abelian) groups,
and whilst cR, cannot be regarded as a subgroup of (R, +) since
the method of combination for the elements of is not that (+) handed
down from the containing group (R, +).

Examples 5.6.2
(i) Consider the group S4 of all permutations on the set X = {1, 2, 3, 4}. Thus
(1
54 comprises the 4! 24 elements d) where a, b, c, d are the
integers 1, 2, 3, 4 in some order. (The method of combination, composition
of permutations, is, from now on, understood.) According to example 5.3.4(m)
the subset V comprising the four permutations, I (the identity), (12), (34),
(12)(34), is itself a group with respect to composition of permutations. Thus
V<S4.
The subsets U={s:seS4andd=4} and W={s:seS4and{b,c}={2,3}}
are subgroups of orders 6 and. . . what?. . of S4. A4 is a subgroup of
. of
order 12.
(ii) In the group M = GL,, (C) the subset S of all those n X n matrices with
determinant +1 or —1 and the subset T of all matrices with determinant +1
are subgroups. Thus SCM and T CM. Further, regarding S as a group in
its own right, T <S. This raises the question: Is each subgroup T of each
(sub)group S of a group R necessarily a subgroup of R? We leave this to
exercise 7.
(iii) Let (G, o) be a group. The subsets G and {e} are two (extreme) subgroups
of G. {e} is called the trivial subgroup, any subgroup S {e} is called a
non-trivial subgroup of G and any subgroup S other than G is called a proper
subgroup of G.
k:
(iv) Let (G, o) be a group. Let x E G. Let C = {x k Z}. It is not difficult
to check that C is a subgroup of G. It is called the cyclic subgroup of G
generated by x.
204 Basic group theory

Note that we may have x r = x5 for some r, s E 7 with r > s (as, for example,
when G is a finite group). In this case x'5=x'x5=4_5—e. That is,
some positive power of x coincides with the identity element. This leads us
naturally to

Definition 5.6.3 Let G be a group and let a E G. If there exists a positive

integer m such that = e then the smallest such positive integer is called
the order of a. If no such integer in exists then a is said to be of infinite order.
Note that this use of the word order does not clash with that of 5.3.7. For,
clearly, the element x of G has finite order n if the cyclic subgroup {x": k e Z} of
G has finite order n.

Examples 5.6.4
(i) In (C, .), —1 has order 2, i has order 4, and de Moivre's Theorem tells

us that cos sin — has order n. 2 has infinite order.

(ii) In GL2(Z), (1 and have orders 4, 6 and infinity

respectively.
(iii) 5 has infinite order in (7, +); 5 has order 2 in $) and order 11 in

(iv) In a finite group every element has finite order.

(v) Theorderof(1 3 2 6 7)(4 5)inS7islO.
Let us return to the groups in examples 5.6.2. The reader who has carefully
checked all the group axioms for each of the parts of 5.6.2 will surely know
the group axioms like the back of his hand! And if he had read 3.4.2 and
3.4.2(F) he should prepare to kick himself! For, in those sections were given
theorems which showed how to minimise the work of establishing that a subset
of a given ring (or field) was a subring (subfield). The reader should expect
a similar result to hold in the case of any algebraic system, in particular in
the case of groups. Thus we prove

Theorem 5.6.5 (cf. 3.4.2 and 3.4.2(F)) Let (G, o) be a group and S a
non-empty subset of G. Then S G ill for all a, b e S we have both
(i) aobeS
and
(ii) a'eS.
Proof If S is a subgroup of G, so that (5, o) is a group, then clearly, from
a, b e S we deduce a o b e 5, since S is closed under [It is not quite so
straightforward to prove (ii).] Since S is a group it has an identity element f,
say, which is such that f of = f. It follows from exercise 5.4.1 that f coincides
 Subgroups. The order of an element 205

with e, the identity element of (G, o) Finally, a is the unique inverse of a

1

with respect to e in G. But, since S is a group, a has an inverse a, say, with

respect to f ( e) in S. It follows that a S [why?], as required.
Conversely, let S be a non-empty subset of G such that (i) and (ii) hold.
[We must check that (5, o) is a group.] From (i) S is closed under If a, b,
c€ S then a, b, CE G and so (a ° b) °c = a o (b cc) automatically. Given a
we know S [by (ii)] and then e = a ° S [by (i)]. Then, trivially,
e os=s ce =5 for all s€S [why?]. Thus e is in S and acts as an identity
element there. Finally, given a €5 there exists [by (ii)] an element (namely
a1) in S which is such that a ° a1 = a1 ° a =e. As we have just shown CAIN
for (5, o) we know that (5, o) is a group; that is S G.

Remarks
(i) We ask the reader to show in exercise 2 that if G is known to be a finite
group then even 5.6.5(u) can be dispensed with since it is then a consequence
of 5.6.5(i). That is, a non-empty subset S of a finite group G is a subgroup of
G if S is closed under the operation on G.
(ii) Some texts define a subgroup of a group G to be a non-empty subset
for which conditions 5.6.5(i), (ii) both hold. Since 5.6.5 is an if and only if
theorem such a definition is equivalent to 5.6.1.

We leave the reader to re-do 5.6.2 in the light of 5.6.5 and pass on to
several of its consequences. First we invite the reader to prove, with the
aid of 5.6.5, the group-theoretic analogue of exercise 3.4.5 and Theorem
3.4.5(F), namely

Theorem 5.6.6 Let {SA: A A} be any set of subgroups of a group G. Then

the set-theoretic intersection fl 5,, is also a subgroup of G.
A€A

(The reader who has not read Chapter 3 will lose little by taking this result on
trust. Equally it will not harm him to try to prove it!)

Example 5.6.7 Let and have the same meaning as in 5.3.4

(p) and (q). Then both are subgroups of and their intersection is the
subgroup of comprising all n x n matrices with integer entries and
determinant +1 (i.e.

Remark We leave it to thç reader to show (exercise 3) that the set-theoretic

union u S2 of two subgroups of a group G is itself a subgroup only under
special circumstances.

In each group G there exist subgroups of such general importance that

they bear special names. Definition 5.6.8(u) introduces one such.
206 Basic group theory

Definition 5.6.8
(i) Let a, b G be such that ab = ba. We then say that a and b commute.
(ii) Put C(G) = {x: x E G and xg = gx for all g E G}. C(G) is called the centre
of G. In words, ((G) is the subset of G comprising those elements which
commute with every element of G.

Theorem 5.6.9 C(G) is an abelian subgroup of G.

Proof Clearly e E C(G) so that C(G) is not empty. Let a, b C(G) and let
geG. [Then ag=ga and bg=gb.] It fol!ows that (ab)g=a(bg)=a(gb)=
l;
Also a that
is ga' = atg. Thus a' E C(G) so that C(G) is a subgroup of G. Finally, C(G)
is abelian. For, let a, b be any two elements of C(G). Then ag = ga for all
g E G; in particular when g = b.

Examples 5.6.10
(i) The group with 16 elements in example 5.3.4(1) has centre comprising
the two elements a4 and e. For, as observed there, every element of the group
can be expressed in one of the 16 forms a' or ba (i = 0, 1,. , 7). Now. .

btab = a 'for all integers t (see exercise 14). This means

that a'b =ba' unless a' =a', that is unless a2' —e. But a has order 8
exactly, 50 8 2t, that is 41 t. Thus the only (distinct) powers of a which could
possibly lie in the centre are a4 and a8 = e: and these do lie in the centre, as is
readily verified. We leave it to the reader to check that no element of the form
can possibly be in the centre.
(ii) If A is an abelian group then A =
At the other extreme:
(iii) C(S3) = {e}. The inequality (12)(123) (123)(12) shows that neither (12)
nor (123) is central in 53. Other elements can be dealt with similarly.

Any group whose centre is no bigger than the trivial subgroup is called a
group with trivial centre or even a group with no centre!

Another important deduction from 5.6.5 can be given after

Definition 5.6.11 Let G be a group and let U = {a, b, c. . .} be a non-empty

(possibly infinite) set of elements of G. We denote by (U) or (a, b, c,. . .) the
set of all elements g of G which can be expressed as a product, g =
where reV, where each and each e,e{—l,l}. We call
<U) the subgroup* of G generated by a, b, c,. . . (or by U) and {a, b, c,. .} .

a set of generators for (U). If (U) = 0 then {a, b, c,. .} is described as a set
.

of generators for G. If the set U is finite and G = (U) then G is called a

finitely generated group.
*See 5.6.12.
 Subgroups. The order of an element 207

Remark Every subgroup S of G (including G itself) certainly has a set of

generators: just take for U all the elements of S. In general G will have many
different sets of generators; one often tries to find as convenient a set as
possible (for example, some infinite groups do possess finite generating sets).

Our first task is to prove

Theorem 5.6.12 (U) is a subgroup of G.

Proof Since U is non-empty certainly (U) is not empty. Suppose u1 =

and Then, rather trivially, u1u2=
and This does it.

Examples 5.6.13
(i) (7, +) has, amongst infinitely many others, the generating sets {1}, {—1},
{2, 3}, {1, 2, 3, 4}. The first two are the only one-generator subsets; the third
has two elements, neither of which is redundant.
(ii) .) has, as one of its generating sets, the set of all (positive) primes.
has no generating sets with only finitely many elements (see
exercise 22).

A characterisation of (U) which is often taken as the definition is given by

Theorem 5.6.14 Let {SA: A e A} be the set of all subgroups of G which

contain the subset U. Then (U) = fl
Ac A

Proof By 5.6.6 fl
AcA
is a subgroup of G containing U and hence <U). The
remaining details are left to you (see exercise 24).

Remark We see that 'the cyclic subgroup generated by x' of 5.6.2(iv) has,
according to 5.6.11, a generating set comprising just {x}. Thus 5.6.11 gen-
eralises the notion of generator for a cyclic (sub)group.

Exercises
I Let S be a non-empty subset of a group G. Show that S is a subgroup of
G if for all a, b eS we have ab1 €5 (cf. exercise 3.4.4).
2 Let 5 be a non-empty subset of a finite group G. Show that S is a subgroup
of Q if for all a, b S we have ab €5. Show, using the simplest infinite group
you know, that this result is in general false for subsets of infinite groups.
3
specific example to show that G =A uB uC where A <G, B <G, C<G is
possible. [Hint: 5.3.4(m) has three suitable subgroups of order 2.]
208 Basic group theory

4 Show that in an abelian group G the set of all elements of finite order
forms a subgroup. (This conclusion may not be valid if 0 is not abelian; see
exercise 12.)
5 Show that the set of all elements of S15 which fix the symbols 3 and 7 and
permute 4, 9, 13 amongst themselves forms a subgroup T of S15. Find 1TI.
6 Let S be the subset of all elements of which map 5 to 7, 7 to 13 and
13 to S. IsS asubgroupof 0?
7 Given that (5, 3) (G, o) and given that (T, (5, 3) (where is the
restriction of 3 to T x T) show that (T, c (0, o) (This proves that a subgroup
of a subgroup of 0 is a subgroup of 0.)
8 (a) List all the subgroups of the group of order 16 in example 5.3.4(1).
[Lagrange's theorem in Section 5.7 would help a lot here!]
(b) List all the subgroups of order 4 in S4 and all those of order 5 in S5.
9 Show that S7 has an element of order 12, but 56 does not. [Hint: 12=3 . 4.
See 5.6.4(v).]
10 (i) Show that if for all a, b G we know a2b2 = (ab)2 then 0 is abelian.
[Hint: aabb = abab is given.]
(ii) Find elements a, b €53 such that a2b2 (ab)2.

11 Let G be an abelian group and let a, b €0 be such that a has order m

and b has order n. What is the order of ab? [Note: a'b' = (ab)' for all 1€!.]

12 LetA=(? Show that A has order 4, B

has order 3 and AB has infinite order.
13 Show that in a finite group, every element has finite order.
14 Show that for any elements x, y, a, b in a group G we have
(i) x, x andy have the same (finite or infinite) order. [Note: (y_lXy)r
y_lxry (prove it!).]
(ii) ab and ba have the same order.
If c 0 do abc and cba necessarily have the same order? [Hint: try taking
a, b, c out of S3.]
15 Show that if 0 is a group with an even number of elements then 0
necessarily contains an element of order 2. [Hint: Show that = e if a = a
Try pairing off elements of 0 with their inverses, beginning with e.]
16 Let 0/7 denote the set of all rational numbers x such that <1.
Define ® on 0/7 by
x$y=x+y
xEBy=x+y—1
Show that (0/7, ®) is an infinite group.
 Cosets of subgroups. Lagrange's theorem 209

Show that every element of this group is of finite order and that for any
E there exists an element of order exactly n.
17 Show that for eachn
18 Find the centre of 0L2(R). [Hint: If g is in the centre it commutes with

1) and (1

19 Find the order of

(i) the subgroup of S3 generated by the set {(12), (13)};
(ii) the subgroup of generated by (12) and (13)(45).
20 Find four distinct generating sets each with two elements for the group
of order 16 in 5.3.4(1).
21 For each n e find a set of n generators for (1, +) such that no subset
of n — of these generates the whole group. [Hint: for n = 2 see 5.6.1 3(u). Then
1

try to generalise the idea of that case.]

22 Show that (Ut.) is not finitely generated. [Hint: Suppose it is. Get a
contradiction.]
23 Prove that the group of exercise 5.3.1(1) can be generated by two ele-
ments.
24 Complete the proof of 5.6.14. [Hint: (U) is a subgroup which contains
U and is contained in every SA.]
25 Show is generated by (12) and (12... n).

5.7 Cosets of subgroups. Lagrange's theorem

We now come to a result, Lagrange's Theorem, described in [31] as 'the most
important theorem in finite group theory'. It relates the number of elements
in a subgroup of a finite group G to the number of elements in G. As most
of current interest in ring theory (as distinct from field theory) concerns rings
with infinitely many elements there is not much use in the theory of rings for
an analogue of this theorem. (Note, however, the dimension equality 4.6.1
in the theory of field extensions.)
In its original form Lagrange's Theorem related two functions
ofagiven
equation when the group of all permutations on the x which fix qS (say)
is also known to fix 0 (R373*). As part of the proof of this Lagrange
needed to observe that (in modern terminology) the order of a subgroup of
5,, divides n!
*
See first footnote on p. 182.
210 Basic group theory

We begin with

Definition 5.7.1 (cf. 4.2.15(i)) Let H c G and let g be a fixed element of

G. We denote by gH the subset {gh: h e H} of G and call this subset the left
coset of H in G determined by g. One defines analogously the concept of
right coset of H in G determined by g. Denoted by Hg, this is the subset
{hg: h eH}.

Remark (cf. 4.2.15(iii)) A more sophisticated way of introducing cosets is

given in exercise 5.

Examples 5.7.2
(i) The group S3 comprising the 3! = 6 permutations on the set X = {1, 2, 3}
contains the subset H = {I, (1 2)} as a subgroup. One easily checks that there
are three distinct left cosets of H in 53, namely:

H —IH = (12)H; (23)H = (123)H; (13)H = (132)H

and three distinct right cosets

H —HI —H(12); H(23) = H(132); H(13) =H(123)

(ii) In our construction of factor rings we had occasion to consider subsets

of the form a+I—{a+i: iel} where I is an ideal in a ring R and a is a
fixed element of R. Note that in this case a + I = I + a since the group (R, +)
is abelian.
(iii) Let P3 denote the set of all triples (x, y, z) of real numbers. Then of
forms a group under ® if $ is defined by (xi,yi,zj)®(x2,y2,z2)=
(xi-1-x2,yj+y2,z1+z2). If one considers the subset H=
one soon sees that H is a subgroup of G. In
geometrical terms H is a plane through the origin in 3-dimensional space.
The coset (3. Se, $H is, in geometrical terms, nothing more than the

plane K passing through the point (3. Se, 4v2) and parallel in of to H.

Remark Certain general facts about cosets are reflected in (i) above. There
we have:
(A) (23)H ={(23), (123)}#{(23), (132)}=H(23);
(B) H(23) = H(132) even though (23) (132);
(C) all the cosets—on right and left—contain the same number of elements
(in this case 2);
(D) H is always one of the left cosets and always one of the right cosets.
 Cosets of subgroups. Lagrange's theorem 211

That the behaviour discussed in (C) occurs in every group is shown by

Lemma 5.7.3 Let H G and let g E G. The mapping r from H to gH

defined by hr=gh isa 1—1 map of H ontogH.

Proof Clearly r is a mapping. Next if hir=h2r then gh1=gh2, whence

= h2 [why?]. Thus r is 1—1. Now let x E gH. Then x = gh* for some suitable
h*EH. Trivially h*r =gh* so that r is onto.

To obtain Lagrange's Theorem we need one more observation.

Lemma 5.7.4 (cf. 4.2.14) Let H G and let g2 G. Then either g1H =
g2H or g1H n g2H = 0. (That is, any two left cosets of H in G are either
identical or miss each other completely.)

Proof Suppose g1H n g2H is not empty and let c be an element common
to both cosets. In particular c giH and so c = g1h * for suitable h * H. Then
cH{ch: h €H}={gih*h: h €H}={gih: R€H} (see exercise 3)=giH. In a
like manner, since c g2H we deduce that cH = g2H. Thus g1H = cH = g2H,
as claimed.

Now it is trivial to note that each element of G lies in some left coset of
H in G; indeed g = ge gH. Thus we see that G is the set-theoretic union
of a number of left cosets of H. From 5.7.4 and 5.7.3 we see that distinct
cosets are mutually disjoint and contain the same number, namely HI, of
elements. Thus if G is a finite group and if G is the union of r distinct left
cosets of H in G we see (by counting!) that GI = rIHI. An identical argument
for right cosets shows that if G is the union of s distinct right cosets of H in
G then al = sIHI. It follows that r = s. This leads to
Definition 5.7.5 Let H c G with I°I finite. The number of right cosets of
H in G (which is equal to the number of left cosets of H in G) is called the
index of H in G. It is denoted by IG:HI.

We then have

Lagrange's Theorem Let H cG, G a finite group. Then IGI = G:HI .

Remark Lagrange's Theorem is often stated in the form: The order of a

subgroup divides the order of the group. Combining this with the observation
following 5.6.3 we deduce that: The order of an element in a finite group G
divides the order of G.

Problem 2 Here is a converse to Lagrange's Theorem: Let G be a finite

group and let n I IGI. Then G has at least one subgroup of order n. Do you
212 Basic group theory

think this assertion is true? If so, try and prove it; if not, try to find a
counterexample. [As the answer to the problem will be given later, don't spend
more than an hour at it. You might begin by looking at a few specific
examples—in theory you have infinitely many permutation groups at your
fingertips!]

Problem 3 Can you say anything significant about a group G which is known
to have order 107?

Exercises
1 List the left and right cosets of the subgroup S = {I, (123), (132)} in A4.
2 List the left and right cosets of the subgroup V = {I, (12)(34), (13)(24),
(14)(23)} in A4.
3
{gh: heH}=gH. Deduce that h*H=H=Hh*.
4 Show that, for c,dEG, cH=dH if C1ceH. Deduce that cH=H if
c E H.

5 Given H G define a binary relation on G by putting a b if a H.

Show that is an equivalence relation and that the corresponding equivalence
classes are just the left cosets of H in G.
Given a,b,cEG and a—b is it necessarily true that (i) a1—b';
(ii) ca —cb; (iii) ac bc?

6 Let A G and B G. Show that for x, y e G either xA n yB = 0 or else

xA n yB is a left coset of A n B. Deduce (even if G be infinite) that if H 0,
if K G and if IG :Hi and 10 :Kj are finite, then IG :H :HI . 1G:K1
and is finite. Can you describe any circumstances under which we have equality
here?
7

8 Suppose IG : HI = n and a E G. Show that there exists it

and a' €H. Need tin? [Hint: look at (12) and {I, (13)} in S3.]
t such that it n

9 Let ga,. . . , gr be a set of elements, one from each of the r left cosets
of H in G. Show that gi', . . . , g1 is a set of elements one from each
of the r right cosets of H in G.
10 Let G be a finite group. Show that for all g G one has = e. [Hint:
the order of an element....]
11 A finite group 0 contains elements of every finite order up to and
including 12. Find the least possible value of 10!.
 Cyclic groups 213

5.8 Cyclic groups

Since every group contains, along with each of its elements x, the whole of
the cyclic subgroup generated by x we see that (i) every group is built up
from (specifically, is the set-theoretic union of) its cyclic subgroups, and that
(ii) the simplest possible types of groups are those which comprise the distinct
powers of some one element. This section contains a few easy observations
about the family of groups described by (ii), the so-called cyclic groups.
Although it is in general very difficult to relate the structure of a group to its
cyclic structure,* we shall see in Section 6.4 how this is possible in the case
of abelian groups.
The reason for the terminology cyclic is that, at least in the finite case, the
prototype amongst cyclic groups of order n is the multiplicative group of the
n complex nth roots of 1. In the case n = 6 these roots can be pictured as in
Fig. 5.4. Figure 5.5 then presents itself naturally as a pictorial representation
of the abstract cyclic group of order 6 with generator x.
In the case of infinite cyclic groups the prototype is the group (7, +).

-i2 + 2
• +
2 2

—1

22
I

Fig. 5.4

=x8=.

..x_3=x3=x9=..

..x2=x4=x10

Fig. 5.5

* A group may have all its proper subgroups cyclic of prime order and yet be infinite. See
Mathematical Reviews, Vol. 80i, review 20013.
214 Basic group theory

Remark We show in 5.9.3 that the above prototypes are essentially the only
examples of cyclic groups of orders 6 and infinity respectively.

Other concrete examples of finite cyclic groups are

(i) P4—{I, (1234), (13)(24), (1432)} (usual multiplication).
(ii) All the groups (M(n), 0) of equivalence classes of integers coprime to
n (see 5.3.4(h)) for which n=2 or 4 or p2 or 2p2 where p is an odd prime
and a e (Actually, for all other positive integers n, the groups (M(n), 0)
are not cyclic. You can easily check a few individual instances, but the general
result just stated is a quite difficult and very important result in number theory.
In those cases when M(n) is cyclic any generating element for M(n) is called
a primitive root modulo n.)

We know (by 5.3.4(k)) that for each n E there exists at least one cyclic
group of order n. We illustrate the power of Lagrange's Theorem by showing
that, if n is a prime, then all groups of order n are cyclic. (In particular we
answer problem 3 above.)

Theorem 5.8.1 Any group of prime order is cyclic.

Proof Let G be a finite group of prime order and select x (x e) in G. Now

x generates a cyclic subgroup H, say. By Lagrange's Theorem HI I IGI = p.
Hence Hi = 1 or Hi =p. The former is impossible since H contains e and
x. Thus HI whence H = G follows. Consequently G is cyclic! [Isn't that
a neat proof?]

Remarks
(i) Since it is clear (exercise 1) that every cyclic group is necessarily abelian
we see immediately that there can be no non-abelian groups of order n when
n is a prime.
(ii) In a group of prime order any element, other than e, can be taken as a
generator.

Not only are cyclic groups simply defined but their subgroup structure can
be completely described. (Such a description is out of the question for most
groups.)
If G is a finite cyclic group with generator x and if IG1 j k where ke
then it should be reasonably clear that the element x' generates a (cyclic)
subgroup S of G of order k and index j (see exercise 2). Similarly, if G is an
infinite cyclic group with generator x, it is clear that, for E t,
the elements
x' and x generate the same (cyclic) subgroup S of G, that S is an infinite
cyclic group and that 6:51=!.
We amplify these observations in
 Cyclic groups 215

Theorem 5.8.2
(i) Each subgroup S of a finite cyclic group G is a (finite) cyclic group whose
index in G divides IGL Further, given any j such that I IIGI there exists
exactly one subgroup of G with index j.
(ii) Each subgroup S (other than S = {e}) of an infinite cyclic group G is an
infinite cyclic group of finite index in G. Further, given any f t there exists
exactly one subgroup of G with index j.
Proof Let S be a subgroup of the [finite or infinite] cyclic group G = (x). If
S = (e) then S is certainly cyclic, generated by e. Assuming S (e) choose a
to be the smallest positive integer such that €5. [Why does such an a
exist?] Now suppose that (/3 1) also lies in S. By the division algorithm
there exist integers m, r €/ such that /3 = ma + r where 0 r <a. Since x3 €5
and since (x")m €5 we see that f ES. By the choice of a, we see
that r =0 whence /3 = ma and x13 = (x")m. Thus S is the cyclic subgroup of G
generated by x2. Further use of the division algorithm* shows that in either
case G is the disjoint set-theoretic union of cosets 5, xS Thus
JG:SI=a is finite and, in case IGI is finite, G:SIHGI.
To complete the proof we note that the discussion preceding this theorem
shows that in each case there is certainly at least one subgroup of the desired
kind. Suppose now that 5, T are (in case (i) or in case (ii)) subgroups of the
same index in G and let u, r denote respectively the smallest positive integers
such that €5 and xT T. Then, whether G be finite or infinite, a = jG :SI =
G: TI = r follows, as required.
Remark In the infinite cyclic group generated by x the single subgroup of
index 12 can be generated only by x12 or x 12 whereas in the cyclic group of
order 120 the single subgroup of index 12 can be generated by x 12, by x36,
by x84 and by x108 [See exercise 3 if you don't understand why.]
Another way of phrasing part of 5.8.2 is to say: Each subgroup of a group
which can be generated by a single element can itself be generated by a single
element.
The reader may care, after considering a few specific examples, to try and
settle
Problem 4 Let G be a group which can be generated by two elements. Is
it necessarily the case that every subgroup of 0 can be generated by (at most)
two elements? [Exercise 5.6.25 looks helpful.]

Exercises
In these exercises will denote the cyclic group of order n with generator x.
1 Prove that every cyclic group is abelian. Give an example of an abelian
group which isn't cyclic. [Hint: search examples 5.3.4.]
*
Do you recognize this argument? See 3.7.15, 1.4.4, 4.3.4.
216 Basic group theory

2 Given that n = ki where k, 1 e 7 show that has a subgroup of order k.

3 If n = 10 show that is also generated by x3, by x7 and by x9. Deduce
that if S is the subgroup of index 12 in C120 then S can be generated by x 12,
36
byx ,byx 84 andbyx 108
4 Find all elements y E C15 such that C15 = (y).
5 How many elements y E are there such that = (y)? Find all vi for
which there are (i) just 2 such y; (ii) just 3 such y.
6 Let S be the subgroup of C154 generated by x28 and x88. Find vi such that

7 Show that if the group G (e) has no subgroups apart from (e) and G
then G is finite and has prime order. (You are not meant to assume G is
finite. Prove that first!)
8 List all subgroups of C120. (To see relative difficulty now try to list all the
subgroups of
9 Give an example of a finite group G such that all proper subgroups of G
are cyclic but G is not even abelian and, hence, certainly not cyclic. (You do
know of such an example.)
10 Is (0, +) cyclic? Let a, b e 0. Is the subgroup (a, b) of 0 cyclic?
11 Is ({a a, b 0}, +) cyclic?
12 Is ({a +b'12: a, b E0; a, b not both zero}, cyclic?
13 It has been conjectured* that 2 is a primitive root of infinitely many
primes. Show that 2 a primitive root
of 41.
14 Let P={x:xEC and for some nEt,x"=l}. Show that is an
infinite group. Show that P is not cyclic but that every proper subgroup of P
is a finite cyclic group of order ptm for some vn.
15 What can you say about a group with 391 581 x 2216193_I elements?

5.9 Isomorphism. Group tables

In Section 3.10 we made precise the idea of two rings being 'essentially the
same' by introducing the concept of ring isomorphism. Under the heading
'Remarks' we indicated not only ring-theoretic uses for the concept but also
that isomorphism is a basic concept for all of algebra. It is therefore no surprise
that we shall have occasion to call on
*
This conjecture is attributed to Emil Artin (3 March 1898—20 December 1962).
 Isomorphism. Group tables 217

Definition 5.9.1 (cf. 3.10.1)

(i) Let (0, o) and (H, *) be groups. A 1—1 mapping fr: G -÷H from the set
0 onto the set H is called an isomorphism if for all a, b 0 we have
(aob)*=(ai/j)*(bfr)
(ii) If (G, o), (H, *) are groups such that at least one isornorphisrn can be
found between them then we say (loosely) that 0 and H are isomorphic (or
that 0 and H are the same, up to isomorphism) and we write G H.

Elementary consequences of 5.9.1, analogous to those for rings in exercise

3.10.7, are left to 5.10.3 and exercise 5.10.3. Figure 5.6, illustrating this concept
pictorially, is analogous to Fig. 3.2 in Section 3.10.

Fig. 5.6

Examples 5.9.2
(i) The map 0: (1, +) -÷ (2/, +) given by zO 2z is an isomorphism between
these two groups. (Comparison with 3.10.2(vii) is interesting!)
(ii) In the group of all permutations on the set {1, 2,. . , n} the subgroup
.

comprising all permutations fixing '1' is a (sub)group isomorphic to

(iii) The group of 8 symmetries of a square is isomorphic to the subgroup
of S4 generated by (12j(34) and (1234) but not to the subgroup of S6 generated
by (12), (34) and (56), the latter being abelian (of order 8).
(iv) The group of matrices generated by (? and (? with respect
to multiplication is a group of order 8. It is called the group of quaternions
(cf. Section 3.2). It is not isomorphic to the group of symmetries of the square
given in (iii).
(v) For define rO=a0+a1x+
a2x2 + a,x5. Then 0 is an isomorphism.
(vi) (7, +) +); (0, +) +). The former is the case since the equation
nx = a (n e 7) is always soluble within 0 but not always, indeed rarely,
soluble in /. The last two groups are not isomorphic since there is no 1—1
218 Basic group theory

map between C and IL (Recall exercise 2.6.15: 0 is countable whereas R is

not.)
(vii) (la, 0) (see 5.3.4(h)). The maps /jg =3" and =5' both
establish this. The map ito- =2" does not. [Why not?]

We have already exhibited in 5.3.4(m), (n) distinct concrete examples of

groups with four elements, but before attempting to see if any two of them
are essentially the same let us do a little abstract working.
Let us suppose that G is a group with four elements e, a, b, c of unspecified
nature where e denotes the identity element. By Lagrange's Theorem the only
possible orders for elements of G are 1, 2 and 4. If G has an element of
order 4 then the 4 distinct powers of that element already account for all the
elements of G. Hence G is cyclic.
The only other possibility is that each of a, b, c has order exactly 2 [why
not 1?]. Further, from the potential equalities ab = e, ab = a, ab = b, ab = c
only one is possible, namely ab = c [why?]. In a similar manner we find that
we are forced to take ba = c and further [use the symmetry of the situation;
don't laboriously repeat your 'ab = c' proof four more times!] bc = cb a and
ca = ac = b.
As we did for and 17 in Section 2.4 we can summarise this information
in a multiplication table as follows:

e a b C

e e a b c
a a e c b
bbc ea
c c b a e

Our working tells us that, apart from the cyclic group of order 4 there can
exist, up to isomorphism, at most one other group with four elements. Such
a group does exist as we already know (example 5.3.4(m)). Note that, in
particular, every group with four elements must be abelian.
We leave the reader to check (exercise 14) that there is, up to isomorphism,
only one group of order 3 and likewise that there are only two groups of
order 6. (And you already know of a concrete example of each type.)
This sort of analysis can be carried out on groups of higher order: the
reader might care to try to show there are (up to isomorphism) five distinct
groups of order 8, three of which are abelian.
The usefulness of multiplication tables, other than for purposes of exposi-
tion, is just about nil! You can convince yourself of this by writing out the
256 entries in the multiplication table of the group in 5.3.4(1) and then
observing that in fact the group appears to be summarised by the assertions
(i) that group is generated by two elements, a and b; (ii) all relationships
between the various elements can be deduced from three basic ones, viz,
a8=e, b2=e, V'ab=a'. This description of the group, presented in the
 lsomorphism. Group tables 219

form (a, b; a8 =e, = e, b 1ab = a S'), is clearly much more economical than
a 16 x 16 multiplication table. Such 'presentations' have obvious advantages
in the case of infinite groups; furthermore groups tend to make their
appearances in terms of presentations in certain branches of mathematics,
in particular in topology.
We now give three instances of the isomorphism concept in action. Firstly
we confirm the essential uniqueness of cyclic groups of a given order.

Theorem 5.9.3
(i) Every cyclic group of finite order n is isomorphic to the multiplicative
group of all complex nth roots of 1.
(ii) Every infinite cyclic group is isomorphic to the group (/, +).

Proof We shall prove part (i) only (leaving (ii) to exercise 1), using the
additive notation for the given cyclic group.
Thus let (G, +) be a cyclic group of order n with generator a [say, and
consequently elements 0, a, 2a, . $ . , (n — 1)a]. Let (Ca, denote the group of
nth roots of 1 under multiplication. We know the element cos i sin

(which we abbreviate to cis —) generates Clearly the mapping ift: G -*

defined by <n) is a 1-1 map of G onto To ptove

itis a homomorphism we take ua, va E G (0 c u, v <n). By the division
algorithm there exist unique integers m, r such that u + v = mn + r where
Thus we have
(ua +va)ifi = ((u +v)a)Vi = ((mn+r)a)ç& = (ra)Ø
2w' . 21T(mn+r)
= cis — = us —_________
n n
2ir(u+v) 2iru 2iw
=cis—
n fl

thus showing that i/i is a homomorphism, as required.

Questions
(i) Can you supply the reason for each equality asserted above?
(ii) Explain why it is not enough to say (ua +va)çt' = ((u +v)a)fr =
2ir(u+v) 2rnt 2nv
cis
n n vi

It is now a small step to prove that every two cyclic groups of order n (vi
finite or infinite) are isomorphic to each other. Clearly we need

Lemma 5.9.4 If (G, *) and (G, then (H,

220 Basic group theory

Proof Let p, a be the 1—1 mappings from G onto H and from G onto K
which show that G and G aK. Since p is 1—1 and onto there exists the
inverse mapping p -I H -÷ G and it is easy to check is an isomorphism
between H and G (exercise 2).
Now define the map r: H -÷ K by r = p"' °a (where o denotes composition
of functions). We already know that i- is 1—1 and onto K (exercise 2.6.9(u))
and we leave the checking of the remaining property to exercise 2.

The second example of isomorphisms in action establishes a result which

is a surprise—until you see the proof!

Theorem 5.9.5 (li,

Proof One easily checks that the map given by nf,=eX is an

isomorphism.
(Exercise 12 asks the obvious follow-up question!)

Remark The above proof was kept to a minimum in order to impress the
reader with its brevity. You might argue that I ought to show that ui is 1—1
and onto, but these facts together with the third requirement, which is nothing
more than the fact that for all x, y eR we have =eX ?, I feel I can leave
to you solely because the proposed mapping is such a familiar one. In the
case of a less familiar, or more awkwardly defined, mapping between two
groups I would probably feel obliged to go into more specific details in order
to convince you of my claim.

Arthur Cayley (16 August 1821 -26 January 1895)

Arthur Cayley was the second son of a Yorkshire businessman who
lived for some time in St Petersburg. Although born in England he
spent the first eight years of his life in Russia. In 1842 he was elected
a Fellow of Trinity College, Cambridge; but he left Cambridge in
 Isomorphism. Group tables 221

1846 rather than take the Holy Orders necessary for prolonging his
stay. He turned to the Law and during 14 years at the bar produced
almost 300 mathematical papers. Whilst at Lincoln's Inn he became
friendly with J J Sylvester who, like Cayley, was pursuing
mathematics and law simultaneously. Together they essentially
founded the theory of invariants. They were so prolific in this area
that they were given the name 'Invariant Twins'. In 1863 Cayley took
the Sadlerian chair at Cambridge where he remained until his death.
Cayley, who was a good linguist with a keen interest in painting and
mountaineering, wrote 966 papers. Only one book* (on elliptic
functions) bears his name; but he willingly helped in the writing of
several others.
Cayley contributed greatly to the geometry of curves and surfaces.
In 1854 he introduced the concept of abstract group. He also wrote
on matrices, determinants, quaternions, the theory of equations,
dynamics, and astronomy.

* He did write a book on the Principles of Double Entry Bookkeeping!

As an example of the last point let us produce another surprise.

Theorem 5.9.6 (Cayley's Theorem) Let (G, *) be any group. Then (G, *) is
isomorphic to a group of permutations on the set G.

Proof To each element a e G define a map Pa: G -+ G by gpa = g * a for all

g e G. Clearly Pa is a map from G to G. Further, Pa is a permutation on the
set G. For: given any h e G one notes that (h * a 1)pa = h whence Pa is clearly
onto G; further, from gipa = we get g1 * a = * a whence gi = [why?]
so that Pa is 1—1.
Next, the subset S = {Pa a e G} is a subgroup of the group of all permuta-
tions on the set G. For, given Pa, Pa e S we see that for all g C G, ° Pb) =
=(g * a)pa =(g * a) * b —g * (a * b). Thus Pa °Pa Pa*b. Similarly we
find that (pal1 Pa' eS. Thus (5, o)is a group.
Finally we claim that the mapping 6: G -> S given by aO Pa establishes the
isomorphism of G and S. Briefly:
(i) 6 is clearly onto.
(ii) 6 is 1—1. [For: if aO = bO then Pa = Pb. In particular we'd have epa epa,
i.e. a = b.]
(iii) Given a, be G, (a * b)6 =Pa*b Pa °Pa = (aO) ° (be).

Remark This theorem, proved by Cayley in 1854, showed that the new
'abstract' theory of groups was subsumed under the theory of permutation
groups from which the abstract theory was emerging. Since the theorem says
that every finite group of order n is (isomorphically) contained in it is
sometimes remarked that the study of finite group theory is reduced to the
study of the subgroups of every But can one really hope to find out very
much about the five groups of order 8, for example, by looking at the subgroup
222 Basic group theory

structure of the group S8 of order 40 320? Interestingly enough, however,

5.9.6 does give us the idea of attempting to 'represent' a given group by other
groups in which more or easier calculations might be possible and a generalisa-
tion of Cayley's theorem (see 5.10.2(iii)) is useful in 6.6.3 and 6.6.4.

Exercises
1 Show that each infinite cyclic group is isomorphic to (1, +). Deduce that
all infinite cyclic groups are pairwise isomorphic.
2 Show (i) that for each group 0 the identity map i: 0 (see 2.6.5) is
an isomorphism; (ii) that if p: 0 onto H then
H (see 2.6.5) is an isomorphism from H onto 0; (iii) that if a: G -*H
and r: H -* K are isomorphisms then a ° r: 0-* K is an isomorphism. Deduce
that 'is isomorphic to' is an equivalence relation on the class of all groups.
3 Is A6a55? (A proof, or a reason why not, is required.)
4 Show that 56 has an abelian subgroup and also a non-abelian subgroup
each of order 6. Are these subgroups isomorphic?
5 Is the map x -* 3x an isomorphism (i) of with itself; (ii) of (Q, +)
with itself?
6 Is 4, S, ?}, 0) — numbers mod 10 and mod 8 respec-
tively?
7 Is (M(14), ®)?
8 ob =a+b—7?
9 Show that the set of all non-zero matrices of the form ( in M2(R)
forms, under matrix multiplication, a group isomorphic to (CX,.)
10 Show that for every a, b and every n E t there exist x, y e such
y\fl
/ x i a b
that( =(
\—y xl
I
\—b a

11 (a)
Is ({a +b',12: a, bEQ; a, b not both zero},
({a+b'J3:a,beO;a,b
12 Is +)? (Cf. 5.9.2(v).) Is (0, .)? (Cf.
13 Let cv be a complex cube root of unity and the real cube root of 2.
Is (Q(J2)\{O}, (4.3.4 may help.)
14 Prove that there is, up to isomorphism, only one group of order 3 and
just two groups of order 6. Write out their multiplication tables.
*
It's not enough to say n/i = e' is not an isomorphism (since e' may not belong to 0). You must
allow the possibility that a very peculiarly defined map may establish the isomorphism.
 Homomorphisms. Normal subgroups 223

15 Do the following multiplication tables yield groups? If not, why not?

U) — e a b c d f (ii) a b c R S T (iii) MNO P
eeabcdf acabTRS LNPMLO
aaecfbd babcRST MPLOMN
bbcedfa cbcaSTR NMOPNL
ccdfeab RTRSbca OLMNOP
ddfabec SRSTabc PONLPM
ffbdace TSTRcab
(Note that (i), (ii) and (iii) certainly do have identity elements.)
16 Let (G, o) and (H, *) be groups and i/i a 1—i map from G onto H such
that for all E G we have (gi ° = * g1(fi. Show that 0

17 Show that the group of exercise 5.3.1(f) is isomorphic to a subgroup of

index 2 of the group of exercise 5.3.1(g).
18 An isomorphism 0:0-÷ 0 is called an automorphism of G. (Cf. exercise
5.3.7.) Show that the set of all automorphisms of 0 forms a group with respect
to composition of functions. We call this group the automorphism group of
0 and denote it by Aut (G).
For each g e 0 define (fig: 0 -*0 by XLIig = g1xg for all XE 0. Show that (lug
is,for each g, an automorphism of 0. Such an automorphism is called
an inner automorphism of 0. Show that the set Inn (0) of all inner auto-
morphisms forms a subgroup of Aut (0).
19 Try to formulate the concept of 'isomorphism of semigroups' (see exercise
5.3.13). Let S be a semigroup with a multiplicative identity e. For each a €5
define pa: S—÷S by SPa= s*a for each seS. Show that {pa: aeS} forms a
semigroup with respect to composition of mappings and that this semigroup is
isomorphic to S. Show that Pa need not be a permutation on S. What happens if
S has no multiplicative identity?
20 Show that the map, given is even, s(f'=
by s(fi=s if s
s (n +1, n + 2) if s is odd, is an isomorphism betweenand a subgroup
of Deduce that every finite group is isomorphic to a finite group of
even permutations.
21 Show that to each integer n E there exist only a finite number of
pairwise non-isomorphic groups of order n. [The present section suggests two
ways in which this can be proved.]

5.10 Homomorphisms. Normal subgroups

In Section 4.2 we defined the concept of homomorphism from one ring to
another. One reason was to help construct new rings from old (4.2.1, 4.5.1);
another was to investigate given rings by looking at homomorphic images in
which, loosely speaking, calculation is easier (4.2.10). In this section we
introduce the same concept into the theory of groups where the idea is again
224 Basic group theory

of vital importance, as we shall see. Apart from uses within group theory of
the kind just described for the case of rings, many of the applications of group
theory to physics and chemistry involve 'representing' groups by mapping
them homomorphically onto groups of matrices (see, for example, [62], [72]).
With no more motivation we give

Definition 5.10.1 (cf. 4.2.3) Let (G, o) and (H, *) be groups. A mapping
i/i: G -÷ H from the set G into the set H is called a homomorphism itT for all
a,beG wehave(a ob)ilj—_(aifr)*(bil'j).

Remarks
(i) As with 4.2.3 we do not insist that i/i is 1—1, nor that it is onto. Of course,
if it is both 1—i and onto then il' is an isomorphism between G and H.
(ii) The subset Gui = {gufi: g E G} c H is easily shown to be a subgroup of H.
[Show it!] We refer to the group Gil' as a homomorphic image of G. Because
'homomorphic image of' is not a symmetrical binary relation on the class of
all groups we do not use expressions such as 'Gil' is homomorphic to G'.
(iii) You may picture a homomorphism as in Fig. 4.1.

Examples 5.10.2
(i) The map r Z —>74 given by zr = 2 is a rather simple example of a
homomorphism from (Z, +) onto (Zr, r is clearly not 1—1.
(ii) The map r: (R) -> R>< given by Mr = det M is a homomorphism which
is onto and, for n not 1—1.
(iii) The following generalises Cayley's theorem, 5.9.6. Let G be any group
and H any subgroup. Let C0(H) be the set of all right cosets of H in G. For
each a E G define Pa to be the permutation of C0(H) defined by Hga.
Letting S denote the group of all permutations on C0(H), the map -r: G—>S
defined by ar is a homomorphism (exercise 19).
(iv) The map r from the group with 16 elements in 5.3.4(1) to the group
comprising the four matrices

(1 O\(1 —2\(—1 2\(—1 0

iJ'ko —i)'k 0 iJ'k 0 —1

(do these form a group?) given by

0\
xr=ko i) ifxe{I,a ,a ,a }
. 2 4 6

—2\
xrç0 —i)
ifxE{a,a ,a ,a }
- 3 5 7

ifxc{b,ba2,ba4,ba6}
 Homomorphisms. Normal subgroups 225

if xe{ba,ba3,ba5,ba7}
?)

is a homomorphism.

In order to use the homomorphism concept we need some immediate

consequences of the definition. Of course these consequences hold if, in
particular, is an isomorphism between G and H.

Theorem 5.10.3 (cf. 4.2.6) Let i/i be a homomorphism from the group G
into the group H. Then (i) ec$/I=eH; (ii) For all geG, (We
use the notation ec, eH for the identity elements of G and H in order to
distinguish them more readily.)

Proof
(i) e0 = eGea. Hence ea* = (eGeG)Lff = (e0*)(e0i/i). From exercise 5.4.1 it
follows immediately that eQ* = eK.
(ii)
by 5.4.1.

As in the case of rings the concept of homomorphism leads to

Definition 5.10.4 Let /i: G -. H be a homomorphism from 0 into H. The

subset ker i/i of 0 defined by ker i/i = {g: eH} is called the kernel of tfi.

The reader should have little difficulty in proving that ker i/i is a subgroup
of G (exercise 10).
To qualify as a kernel of some homomorphism a subgroup must have special
qualities. We introduce these in

Definition 5.10.5 Let N C 0. If, for all g e 0, we have N = g1Ng (where

g1Ng is defined to be the subset {g1ng: n eN} of 0) then N is called a
normal subgroup* of 0. We write this briefly as N c 0.

Theorem 5.10.6 Let i/i: 0 -*H be a homomorphism. Then ker i/i is a normal
subgroup of 0.

Proof Ker i/i is a subgroup, by exercise 10. Suppose n E ker i/i and let g E 0.
Then (g1ng)tfi=(gtcfr)' (gtfi)=(gt/i)' eH (gcfi)en. Thus g1nge
(n&fi) .

ker tf, for all n e ker q',. Thus for each g E 0 we have g1Ng N. In particular,
for each g €0, (g')'Ng1 c N whence N c g1Ng (see
exercise 11). Thus g - 'Ng = N for each g e 0, as required.
*
This concept was introduced by Galois.
226 Basic group theory

Examples 5.10.7
(i) If is an isomorphism, or more generally if iJi is 1—1, then ker /i = (e0).
Clearly (eG)clG.
(ii) If i/i is defined by gil' = eH for all g E G then ker i/i = G. Clearly G c G.
(iii) The kernels of the homomorphisms described in 5.10.2 are respectively

{zn: z E Z}; the set of all ME such that det M = +1; fl

g€G
g1Hg (which
one can show directly is the largest normal subgroup of G which is contained
in H: see exercise 19); the subgroup comprising I, a4, a6 (see exercise 13).
(iv) Every subgroup of an abelian group is a normal subgroup (see
exercise 12).

We have shown that every homomorphic image of a group G gives rise to

a normal subgroup of G. In the next section we establish the converse assertion
(see 5.11.3). This 1—1 correspondence between normal subgroups and
homomorphic images shows the importance attached to identifying, whenever
possible, the normal subgroups of a given group. We have, in 5.10.7 (iii) and
exercise 14, indicated two ways of picking up normal subgroups in any group.
Another is given by

Theorem 5.10.8 Let G be a group and N < G such that IG : Ni =2. Then
NciG.
Proof Let g be any element of G which is not in N. Since there are just two
left cosets of N in G we conclude that G = NC gN (the dot indicating disjoint
union). But also G = NC'Ng for the same reason. Thus the subsets gN and
Ng of G are identical. That is Ng = gN, whence (by exercise 11) g 'Ng N.
On theotherhandifg EN then gN =N =Ng (exercise 5.7.3)so that g tNg =
N in that case too.

The most commonly quoted instance of 5.10.8 is

Example 5.10.9 For each n E t,

Proof We have already seen in 5.5.11 that Hence

We finish this section by proving a result we shall call upon from time to time,

Theorem 5.10.10 Let be a homomorphism from the group G into the

group H. (i) if ScG then St/icH; (ii) if then* Further,
(iii) if TC3H then cG;Uv) ifScG and i/i is ontoHthen St/i H.
 Homomorphisms. Normal subgroups 227

Remark The conclusion in (iv) is not necessarily true if i/i is not onto H (see
exercise 7).

Proof of 5.10.10
(i) First note that Si/i 0 since S 0. Now let h1, h2 E Si/i. Then there exist
elements g1, g2 ES such that = h1 and g2i/i = h2. Then h1h2 = g1t/i g2(// =
(glg2)t/'ESI/' since gI,g2ES Also hI1 =(giifrYt=gI1i/ieSi/' since
g1ES and by 5.6.5.
(ii) First note that Ti/F1 0 since eG E Ti/F'. Now if E Ti/F' then there
existhi, h2e Tsuchthatgii/i=hiandg2q!i=h2.Butthen(g,g2)&i' . =
h1h2€ T since h1, h2e T and Thus g1g2E Ti/F'. Also =

(iii) Let ge G and let k Ti/F'. Then gi/i€H and ki//€ T. Thus (g'kg)i/i =
(gi/r)'ki/,(gi/i)€ T (since TcH). Thus TI/F' and hence Ti/i 'cG.
(iv) We leave to the reader.

Exercises
1 Show that the maps defined in example 5.10.2(i) and (ii) are
homomorphisms.
2 Let (A, +) be an abelian group. Show that for each m €1 the map
defined by aCm = ma, for all a A, is a homomorphism of A to itself. If m > I
can still map A onto itself?
3 Show that every homomorphic image of an abelian group is abelian and
of a cyclic group is cyclic. Show that under a homomorphism an element of
finite order n maps to an element of order m dividing n. Give a specific
instance where m n.
4 How many pairwise non-isomorphic groups can you find which are
homomorphic images of 53?
S Find, or prove the impossibility of finding, homomorphisms (i) from
®)onto<Z6, ®>;(ii)from<112, e>onto<15, ®);(iii)from<Z, +>
onto <M(14), 0).
6 Let be the complex plane augmented by an extra point 'oo'. Let
a,b,c,d€C Definef:C -*C by:
az+b d

f(x') =
228 Basic group theory

and by
az +b
f(z)=
ifc=0.
f(ao)=ao
Show that (i) under composition these mappings form a group G; and (ii) G
is a homomorphic image of GL2(C). [Hint: -÷ t.] Is this homomorphism
(a
an isomorphism?
7 Find groups G and H, a subgroup S of G and a homomorphism i/i: G -÷ H
such that ScG but SP+H.
8 Give an example of a homomorphism from <S,,, onto <{ — 1, 1 }, >.
9 Let U be a set of generators for a group G. Let /i:G-÷H be a homo-
morphism. Explain why, for each x e G, xg!i is completely determined once
ut/i is known for each u E U.
10 Show that the kernel of any homomorphism is a subgroup.
11 Show that if M c G and N c G and if g e G then (i) gN = Ng itT =
N;(ii)g
12 Show that in an abelian group all subgroups are normal. Find an example
of a non-abelian group G in which all subgroups are normal. [Hint: example
5 .9.2(iv)j
13 (a) Show that the subset I, a2, a4, a6 forms a normal subgroup of the
group of 5.3.4(1).
(b) Show that ((12)) is not a normal subgroup of S3. Thus H <G and
G: Hj = 3 are together not enough to imply H ciG. (Cf. 5.10.8.)
14 Show that every subgroup of C(G) is normal in G.
15 Show that if G has just one subgroup S of order 20 then SciG. What
significance has the integer 20?
16 ShowthatifHciGandKciGthenflnKciG,and(HuK).cG.
17 Show by example that H cK G and K .cG do not necessarily imply
HciG. [Hint: A4j
18 Show that for every group G we have Inn (G)cAut (G). (See
exercise 5.9.18.)
19 Let H c G with G: HI = r and let Hg1, Hg2 Hg. be the distinct right
cosets of H in G. For each g G let be the map sending each Hg to Hg1g.
Show that each is a permutation on the set X = {Hg1 Show that
the map fr: G -+ defined by gØ = is a homomorphism whose kernel is
 Factor groups. The ffrst isomorphism theorem 229

fl g
geG
'Hg. Deduce that fl
gcG
gtHg is the largest normal subgroup of G to
be found in H.
20 Use exercise 19 to show that if a finitç group G with more than 24
elements has a subgroup of index 4 then G has a normal subgroup N such
that (e) <N < G. (We shall use this idea later in the case when IGI = 36.
See 6.6.3.)
21 Let i/i: G—÷M be a homomorphism of the group G onto a (multiplicative)
group of matrices. Given g, he G show that det (hi/i) = det ((g 1hg)i/i) and that
trace (hi/i)=trace ((g1hg)ifr).

5.11 Factor groups. The first isomorphism theorem

In this section we construct to each normal subgroup N of a group G a
homomorphic image of G with kernel N. This construction and consequent
theorems are so important that we do not apologise for essentially repeating
the statements and proofs of 4.2.16 and 4.2.19. Besides its theoretical use in
characterising all homomorphic images of G (see 5.11.6) we also immediately
illustrate the practical use of 5.11.1 and 5.11.3 by answering Problem 2 for
finite abelian groups.

Theorem 5.11.1 Let G be a group and N a subgroup of G. The set of all

left cosets of N in G forms a group with respect to the 'natural' multiplication,
namely aN bN = abN, if and only if N c G.

Proof (&) Suppose N cG and denote the set of (left) cosets of N in G by

the symbol GIN. In attempting to define a binary operation on GIN by the
above formula we first have to check that this 'product' is well defined. (See
Section 2.4 and 4.2.16.) That is, we must check: if aN = cN and if bN = dN
then abN cdN. The given equalities show (see exercise 5.7.4) that a eN
and b 'deN. Since NcG we see that b'(a 'c)beN and hence that
h'(a 'c)b b'd=b 'a abN=cdN.Thustheproposed
method of multiplication on the elements of GIN is a well-defined binary
operation and we only have the group axioms A, N, I left to check.
(A) Given aN, bN, cN e GIN we have (aN . bN) cN = abN cN =

(N) N(=eN) is clearly such that eN aN=aN eN=aN for all aNeG/N;
(1) GiveriaNeG,/Nweseethata 'N=eN=Nsothat
a'N = (aNfl', as required.
For the 'only if' part of this theorem see exercise 1 below.

Definition 5.11.2 The group just constructed is called the factor group* (or
quotient group) of G with respect to N.
*
Sec the footnote to 4.2.17.
230 Basic group theory

Remark The order of GIN is equal to the index of N in G: 1GINI = G:N1.

We thus see the possibility of attempting proofs by mathematical induction.
Theorem 5.11.4 is an example of this. First, however, we show that the
group constructed in 5.1 1.1 fulfills the requirements of the first sentence of
this section.
Theorem 5.11.3 Let G and N ciG be given. Then there exists a
homomorphism 0: G -* GIN such that ker 0 = N exactly.
Proof We define 0 (clearly it is a map) from G to GIN by gO = gN, for all
g E G. For all g, h E G we have (gh )O = ghN = gN hN = gO hO as required.
Clearly gO = eG,N if and only if gN = N, that is, if and only if g E N (exercise
5.7.4). Thus ker 0 N. (0 is called the natural homomorphism from G onto
GIN.)
Problem 5 Will every homomorphism from G onto GIN have kernel exactly
equal to N?
As an application of 5.11.3 we show that, for abelian groups at least,
Lagrange's Theorem has a converse. First we prove
Theorem 5.11.4 Let G be a finite abelian group and let the prime p e
be such that p I IG I. Then G has an element of order p.
Proof If the theorem is false then one can choose a counterexample C, say,
of smallest possible order. Clearly Now let x be any non-identity
element of C and suppose the order of x is n. Then ptn (or else is an
element of C of order p).
Set N=<x>. Then NciC (since C is abelian) and so we can form* Since

plICI=IC:NIINI we see that pllC:N1= By the choice of C we

see that does contain an element yN, say, of order p. Thus (yNY = ec,N = N.
It follows that yP eN (exercise 3) and hence that y" =xt for some r. Con-
sequently y = x = e. Thus y" has order p or order 1. But C has no elements
of order p and so =e. But then yN has order dividing it (exercise 5.10.3)
so that p In after all. This contradiction shows no such counterexample C can
exist and so the theorem is proved.
And now for the converse of Lagrange's Theorem for finite abelian groups.
(See Problem 2.)
Theorem 5.11.5 Let G be a finite abelian group. For each m Ct such that
m IGI, G has a subgroup of order m.

C
* We occasionally write — as an alternative to C/N.
N
 Factor groups. The ffrst isomorphism theorem 231

Proof Suppose that the theorem is false and let C be a counterexample of

smallest possible order. Suppose that k ICI but C has no subgroup of order
k. Let pe7L* be a prime such that p1k. By 5.11.4, C has a (normal) cyclic
C
subgroup N = (x), say, of order p. Now — <ICI and — —. By choice of
kC
N pN
C, has a subgroup M, say, of order It is not difficult to check (see
N p
5.10.1 0(u) and exercise 8) that under the natural map i/i: C -÷ the inverse

image Mi/f1 of M is a subgroup of order — p = k. Thus no such counter-

example C can exist and the theorem is proved.

Actually Problem 2 has, in general, a negative answer. For suppose N were

a subgroup of order 6 in A4. Then A4: Ni = 2 since 1A41 = 12. It follows (see
5.10.8) that Now (123)eA4 hence (123)2eN (since =2). Thus
(123)=(123)4EN. Consequently N contains 1, (123), (123)2=(132) and
similarly (124), (142), (134) and (143) which is already too many elements.
It follows that Lagrange's Theorem has a converse for finite abelian groups
but not for finite groups in general.

Remark We shall improve substantially upon 5.11.4 in the direction of

non-abelian groups in 6.2.8. (See also Theorem H of Section 6.5.)

We have seen in 5.11.3 that every group of the form GIN, where N G,
is a homomorphic image of G. We now show that, conversely, the set of all
homomorphic images of G is essentially restricted to groups of this type.
Applications of this theorem are given in exercise 6 below, and in exercises
6.2.17, 6.3.16 and 6.3.17.

Theorem 5.11.6 (cf. 4.3.1) (The First Isomorphism Theorem) Let G be a

group and i/i a homomorphism of G onto H with kernel K. Then H G/K.

Proof Define a map 6:H—*G/K by: If h =gi/ set h6'=gK. We must first
check that 6 is indeed a map.* Thus suppose h = if' = g21/I. Then (g =
so that g1 e ker i/i = K. Consequently g1K = g2K so that 6 is well defined.
[Do you understand that? Are you sure?]
Clearly B is onto G/K. [If gK G/K then hO = gK if we choose h = gq'i.]
Oisl—1.[IfhiO=h2Owhereh1=g1ifrandh2=g2i/i,thengiK=hiO=h20=
g2K. Thus =ker i/i. Therefore g1ifi=g2Ø; that is h1 =h2.]
a homomorphism. [With notation as above h10 h20 = g1K
o is =
g1g2K = ((gig2)ifi)O = (g1ifr = (h1h2)O, as required.]

*
That is, we must check 0 is well defined.
232 Basic group theory

Exercises
1 Show that if one is to make a group out of the left cosets of the subgroup
N of the group G by defining, for a, b e G, aN bN = abN, then it is necessary
that N be normal in G. [Hint: aN = cN, bN = dN imply a - 1c E N and
Hence if b
a be any element of N, c = 1 and b any element of G.
Cf. exercise 4.2.16.]
2 Prove that V={I,(12)(34),(13)(24),(14)(23)} is a normal subgroup of
54. To which group, with which we are already familiar, is 54/ V isomorphic?
[Hint: What is 154/V1?]
3 Let NcaG. Show that if gN has order n in GIN then gtm eN if! nim.
4 Let T be the set of all elements of finite order in the (possibly infinite)
abelian group A. Show that T is a (normal) subgroup of A and prove that
A/T has no elements of finite order (other than the identity element).

5 Let N CG andH with N cH. Define %to be the subset {hN:h e H}

of Show that and that if! HcG. Show that if then

there exists K such that and S k eK}.

6 Show that, for each group G, Inn (G) G/C(G). [Hint: map G -÷ Inn G
by g (see exercise 5.9.18) and show that the kernel is C(G). Now use
5.11.6.]

7 (a)

If N1 G and N2 G and if is N1 [These may prove

considerably easier when the reader has assimilated Section 6.3.]
8 Let i/i: G -* H be a homomorphism with kernel K of G onto H, both finite
groups. If show that
9 Prove the Third* Isomorphism Theorem (cf 4.3.3). Let N c G, K G
with N K. Show that 6: given by (gN)6 = gK is well defined and

that it is a homomorphism. Show that ker 6 = (cf. exercise 5) and deduce

that
GG/N

*
The Second Isomorphism Theorem is proved in Section 6.5.
 Space groups and plane symmetry groups 233

512 Space groups and plane symmetry groups

According to [132, p. 290], the first direct application of group theory to an
important problem in the natural sciences was in the determining, by the
Russian crystallographer Fedorov,* in 1890, of all the possible symmetry
groups of 3-dimensional repeating patterns. The interest in such patterns
arose from the then widely held view, not confirmed until 1912, that crystals
were assemblages of very small basic units of matter repeated periodically in
3 dimensions. The connection with group theory is that each pattern can be
characterised by its symmetry group. (One must take a little care. See the
exercise at the end of this section.) It turns out that there are only 230 of
these so-called crystallographic space groups amongst which are 22 which
crystallographers prefer to regard as distinct but which, from an abstract point
of view, form 11 pairs of isomorphic groups. Thus the space groups fall into
219 isomorphism classes [70]. The enumeration of these space groups is built
upon the 14 lattices determined by Bravais in 1848 and the 32 crystallographic
point groups (which themselves fall into 18 distinct isomorphism classes [71])
already known to Hessel in 1830. Since this enumeration is quite complicated
([63]) we here look at some of the corresponding ideas involved in the
analogous 2-dimensional problem where 17 groups, no two of which are
isomorphic ([10]), arise. We leave the reader to find the not too difficult details
in, for example, [69] and [73]. (This section is just for fun. You are left to
make what you will of most of the provocative statements in the text. They
are left to you as exercises.)
First recall that an isometry of the plane R2 is a distance preserving mapping
r, say, of R2 onto itself. Amongst such isometries are translations, rotations,
reflections (in lines) and glide reflections. Figure 5.7 adequately describes
these movements, a glide reflection being the result of an ordinary reflection
in some line 1 followed by a translation parallel to I. Translations and rotations
preserve orientation, reflections and glide reflections clearly reverse it; note
the arrowheads on the circles in Fig. 5.7(c), (d).

(a) (6) (c) Id)

Fig. 5.7

* Western authors tend to credit A Schönflies and W Barlow with their independent discovery.
In fact it appears both Fedorov and Schönflies made errors which were eliminated by their
corresponding with one another.
234 Basic group theory

It is not difficult to prove ([69, p. 98]).

Theorem 5.12.1 Every proper (i.e. sense preserving) isometry of is of

type (a) or (b); every improper (i.e. sense reversing) isometry of R2 is of type
(c) or (d).

With an appropriate definition of 'product' the set F of all isometries of

forms a group (5.3.4(1)). As is easily checked, the product of two proper,
or of two improper, isometries is a proper isometry whilst the product, in
either order, of a proper and an improper isometry is an improper isometry
([69, p. 98]).
The 2-dimensional repeating patterns we consider are commonly called
wallpaper patterns (guess why!). By definition these are patterns in which
it is possible to find a bask pattern unit repeated periodically but not
'continuously' in each of two non-parallel directions. Thus we are interested
in patterns such as that in Fig. 5.8(i) and not those in Fig. 5.8(u), (iii).

EFFEF
(i) (ii)

(iii)
Fig. 5.8

In order to accommodate translational symmetries in our discussion we are

obliged to assume that the patterns we consider completely fill the plane.
It is implicit in Fig. 5.8(i) that we are dealing with what are called discrete
groups of isometries of R2. Formally: A group G of isometries of is discrete
if to each point p e R2 there exists a circle C,, with centre p such that for each
g e G either p is not moved by g or it is moved to a point, pg, outside C,,.
Thus the symmetry group of the circle x2 + = 1 is not discrete since it
contains arbitrarily small (rotational) isometries.
Now the complete symmetry group of a wallpaper pattern is a special kind
of discrete group known as a plane symmetry group.* By definition, a plane
*
Also called a planc, or 2-dimensional, crystallographic group.
 Space groups and plane symmetry groups 235

symmetry group is a discrete subgroup of E which contains translations s, t

in non-parallel directions. Let G be such a group and let p e Applying
the isometries s 't' (each such is again a translation) to p produces an infinite
lattice of points (Fig. 5.9).

. •ps2r1. • • •
• • . •pst •pst2 •

•p(1 •p •pt •
S • • S

Fig. 5.9

If we take s, t to be non-parallel translations of smallest possible magnitudes

[how do we know that such 'smallest' translations exist?] it is not difficult to
check that every translation in G is of the form ([69, p. 114]). That is,
the subgroup of all translations of G requires no more than two generators.
Now there do exist (infinite) wallpaper patterns which possess only
translational symmetry (Fig. 5.8(i)). On the other hand some patterns also
exhibit rotational or reflectional symmetry (Fig. 5.13(u), (iii) for example).
Suppose, then, that G contains rotations too. Let t be the smallest translation
in G and let r be a rotation (centre c0) in G of smallest (positive) angle
0= say. [Why does such a smallest exist; why does it take the form

Fig. 5.10

See [132, p. 286].] Suppose t moves c0 to c1 (Fig. 5.10) and that r moves c1
to c'1. Now the isometry t' r - 'tr, which clearly belongs to G, is a translation
which sends c0 to c'1 ([69, p. 102]). Then 1t' is a translation sending c1 to
cL Since t was chosen as small as possible we see that the distance is
2ir
no shorter than c0c1. It follows that angle c1c0c1 is at least 60°. Thus
n 3
whence n 6. The reader is invited to prove that the case n = 5 is also
impossible ([69, p. 111]). We thus have

Theorem 5.12.2 (The Crystallographic Restriction) Let G be a plane

symmetry group. Each rotation of G necessarily has order 1, 2, 3, 4 or 6.
236 Basic group theory

A consequence of 5.12.2 is that there are only 5 basic types of lattice which
can underlie a plane symmetry group. Let us first consider what possible
symmetries, apart from translations, a lattice of points as in Fig. 5.9 may
have. Clearly every lattice has rotational symmetry through angle IT (so-called
diad rotation). Next suppose L is a lattice possessing rotational symmetry of
order 3. There are clearly two cases, Fig. 5.11(i) and (ii), depending on whether

P4
(1) (ii) (iII) (iv)

Fig. 5.11

or not the rotation centre C is or is not a lattice point. In fact the two cases
give rise to identical lattices* comprising equilateral triangles. Clearly this
lattice has 6-fold rotational symmetry about each of its lattice points—and in
fact any 'hexad' lattice must be of this type ([69, p. 112]). A similar analysis
shows that a lattice with 4-fold rotational symmetry must be a square lattice
(Fig. 5.11(iii) and (iv)).
Lattices possessing reflectional symmetry in a line (which need not contain
any lattice point) must be made up of rectangles or rhombuses ([69, p. 112]).
In this latter case the lattice is often thought of as 'centred rectangular' (Fig.
5.12). Finally, if a lattice has glide-reflectional symmetry it is necessarily of
centred rectangular type ([69, p. 113]).
Thus there are only 5 kinds of lattices when we distinguish them by their
symmetries, namely those in Fig. 5.12. These are the 2-dimensional analogues
of the 14 Bravais lattices mentioned earlier.

• . . • S • . S S S

S • • S S • S S • • S

S S • . S • S S S S

Oblique parallelogram Hexagonal Square

• S S S S S S S
• S S
• • • S S S S S
• S S
• S S S S S S

Rectangular centred rectangular

Fig. 5.12
* In (i) F'4, P5, P6 arise from P,, P1 under translations.
 Space groups and plane symmetry groups 237

We now make

Definition 5.12.3 A 2-dimensional crystallographic point group K is a group

of isometries of which fixes a point p and maps a 2-dimensional lattice
containing p into itself.

In any such group there can be neither translations nor glide reflections
(why not?). Consequently either all the elements of K are rotations or one
half of them are rotations and the other half reflections (cf. the proof of
5.5.11). It follows from 5.12.2 that K is (isomorphic to) one of the cyclic
groups or one of the dihedral* groups where n = 1, 2, 3, 4 or 6.
We now show how each plane group G determines a crystallographic point
group as a homomorphic image. We choose some point 0, say, in the plane.
Note that if p is any isometry of which moves the point 0 to the point a
and if t is the translation of R2 moving 0 to a then s = moves 0 to 0,
that is, s is either a rotation about 0 or a reflection in a line through 0.
Further we may write p = st. Moreover the subgroup T of translations of
is a normal subgroup of E (cf. the proof of 5.12.2). It is then immediate that
H = T n G, that is, the subgroup of translations in G is a normal subgroup
of G.
Now suppose that = s1t1 and g2 = s2t2 are elements of G written in the
above form. (Note: it isn't asserted that either or t1 lies in G.) Then
glg2=sls2 t2. Here is a translation (cf. the proof of 5.12.2),
whilst fixes 0 (since both and do). In a similar manner gj' =
may be written siGtsI1, being a rotation about 0, sttisi' being a
translation. We thus see that the set of all s which appear as 'rotation-reflection
components' of elements of G forms a group K fixing the point 0. Further
the map O:G-K given by g10=(sit1)O=s1 is a homomorphism of G onto
K with ker 0 = H. Finally let the elements of H be applied to the point 0 to
produce the lattice L. Suppose a L so that a = Ot for suitable t H. Let
s K. Then there is g G such that g = st1 for suitable t1 T. Then as = Ots =

the lattice into itself. Thus as = Og - 'tg is a point of L. Thus K G/H is a point
group mapping the lattice determined by H and 0 to itself.
The above analysis shows how we should be able to find systematically all
possible plane groups starting from the five lattices and the ten point groups.
The method is indicated in [70] and [120]. Bearing in mind the remark in
[120] that 'the passive contemplation of wallpaper patterns ... is not mathema-
tics' we invite the reader whom we have managed to interest to read [73] or [120]
and look at the 3-dimensional case in [70]. A more geometrical approach is
given in [69] (see also [63]).
There follows a description of the 17 plane groups in pictorial terms and also
in terms of generators and relations (see [10]).
*
D,, is, for each n c the group of 2n symmetries of a regular n-gon (see 5.3.4fl)). The 1-gon
and 2-gon are pictured respectively as '—.—-' and
238 Basic group theory

Ii) p1 (ii) p2
generated by two translations generated by three half turns
<x,y t12=t22 =t32 = = I>

3j, F

F F

(iii) pm (iv) pg
generated by two reflections and a translation generated by two parallel
<r1,r2,y:r12=r22=1,r1y='yr1,r2yyr2> glide reflections
<p,q p2 =q2>

)v) pmm (vi) pmg

generated by four reflections generated by a reflection and
<r1,r2,r3,r4 :r12 =r22 =r3 2 =r4 2=(r1r2)2 two half turns
<Zr, t1, t2;r2 = = t22 1, t1rt1 = t2rt2>
=(r2r3)2=(r3r4)2=(r4r1)2 = 1>

F
E

F
E

(vii) pgg
generated by two perpendicular
glide reflections
<p,q: (pq)2 (p1q)2 = 1>
Fig. 5.13
 Space groups and plane symmetry groups 239

E S E S E
J
94p J
94F
SE dYE

rr
F 9 F [ 9 F
E E S

E. —
E__

(viii) cm (ix) cmm

generated by a reflection and a generated by two perpendicular
parallel glide reflection reflections and a half turn
<r,p = 1,rp2 =p2r> <r1,r2, (r1Lr2t)2 1>

E Fr E rr E rr

9 .1J 9 .Lt 9
E rr E Fr

.u 9 .u -U 9
E E Fr E Fr

-U 9 .u 9 .u

(x) p4 (xi) p4m

generated by a half turn and a generated by reflections in the
quarter turn sides of a (450, 450, 90°) triangle
<s,t ;s4 t2 =(stl4 = 1> <r1,r2,r3 ;r12 'r22
= (r1r2)4 = (r2r3)4 = (r3r1)2 = 1>

E Fr E Fr E Fr

u-F u-F u-F

Sm 5m
-U I I I
E Fr E Fr

u-F £F'\
7[iiy u-F

-U 9
E Fr E Fr E Fr

u-F u-F u-F

5m
-U .11 .U I
(xii) p4g
generated by a reflection and a
quarter turn
<r,s r2 s4 = (f1rsr)2 = 1 >

Fig. 5.13 (conic)

240 Basic group theory

(xiii) p3 (xiv) p3ml

generated by 3 rotations through 120° generated by three reflections ri the
= =1> sides of an equilateral triangle
<81, s3 = =
<r1,r2,r3:r12r22_—r32
:

= (r1r2)3 = (r2r3)3 (r3r1 (3 =1>

(xv) p3lm (xvi) p6

generated by a reflection and a generated by a half turn and a
rotation through 120° rotation through 120°
<r,s : r2 s3 = (s1rsr)3 = 1> <s,t t2 = (st)6 = 1 >

(xvii) p6m
generated by reflections in the sides
of a (30°, 600, 90°) triangle
<r1,r2,r3 : r12 =r22 orj1 = (r1r2)3 = (r2r3)6 = (r3r1)2 1>
Fig. 5.13 (cont.)
 Space groups and plane symmetry groups 241

In each part of Fig. 5.13 the shaded region is a fundamental region in the
sense that when the elements of the group in question are applied to it the
resulting regions completely fill the plane without overlap. No doubt you will
be able to think of different systems of generators and relations to those given.
In view of the remarks at the end of Section 6.1 it is worthwhile reiterating
the earlier remark that the enumeration of the 230 space groups was completed
more than 20 years before the assumption of regularity of crystal structure
on which it was based was confirmed ([64J).

Exercise Try to find all frieze patterns. These are the 2-dimensional repeating
patterns whose symmetry groups are discrete and infinite but also leave a line
in fixed. In such groups the subgroup of translations must be isomorphic
to the infinite cyclic group. There are seven distinct frieze patterns; their
symmetry groups fall into four isomorphism classes. We give you two of the
seven to get you started (see Fig. 5.14). The first has translations and the

TTAT.TT
<t,r <g,r

Fig. 5.14

second has glide reflections, and not vice versa. Although their symmetry
groups are abstractly isomorphic there is (trivially) no isomorphism between
the groups under which isometries in one group map to isometries of the same
kind in the other group. This finer analysis distinguishes the symmetry groups
of these patterns and hence the patterns themselves.
A derivation of these patterns can be found in [69].
6
Structure theorems of group
theory
6.1 Introduction
In much of what we have done so far, applications of the work have been at the
backs of our minds. In Chapters 2 and 3 we gave applications of algebra to
number theory whilst in Chapter 4 we applied our theory to geometrical
construction problems and to polynomial equations over C. In Chapter 5 we
saw the beginnings of a connection between solving equations by radicals and
groups of permutations, a theme we shall greatly enlarge upon in Chapter 7.
In this chapter we offer the reader a different point of view. Putting thoughts
of immediate applications aside we adopt the attitude of many a research
algebraist, for whom the discovery of the deeper properties of rings, fields,
groups, etc. is both the sole aim and the complete reward. This is not to imply
that the results obtained are inapplicable. Far from it. Indeed we shall see how
mere intellectual curiosity leads us, quite naturally, to the concept of soluble
group (Section 6.5)—which, to aid his researches, Galois would surely have
pounced on (see 7.9.3, 7.10.5) if only he hadn't been obliged to invent it for
himself (see 6.5.7) some decades earlier.
We invite the reader to pause from time to time to experience a sense of
wonderment that such deep and beautiful results as are described here are
hidden in the very simple list, 5.3.1, of axioms.

6.2 Normaliser. Centraliser. Sylow's theorems

In this section we penetrate the subgroup structure of finite groups by proving
the three famous theorems of Sylow* (1872). These theorems yield a tremen-
dous amount of information about the subgroups of prime power order in
any finite group. This information will be made great use of in this section
and in Section 6.6.
We begin with
Definition 6.2.1 Let G be a group such that every element of G has p-power
order for some fixed prime p. Then G is called a p-group.
Remarks
(i) 6.2.1 allows for infinite p-groups. Such objects do exist but, on the whole,
they are not especially nice—see the remark following 6.2.9. We shall essen-
tially deal only with finite p-groups.
* Peter I.udvig Mejdelt Sylow (12 December 1832—7 september 1918).
 Normaliser. Centraliser. Sylow's theorems 243

(ii) By Lagrange's Theorem any finite group of order p" is a (finite) p-group.

Example 6.2.2 53 is not a 2-group, nor a 3-group. For each prime p the
multiplicative group of all complex pth power roots of I is an infinite p-group.

Problem 1 Can you show that a finite p-group must have order a power of
p? (This can't be difficult—or can it?)

Definition 6.2.3
(i) Let H be a subset of the group G. The subset g1Hg —{g1hg: h eH}
is called the conjugate of H by g in G. We denote f1Hg briefly by IV.
(ii) If H, K are subsets of G we say that K is conjugate to H in G if there
exists in G an element g such that =K. It then follows that Kr' =H.
Hence we may fairly say that H and K are conjugate in G.
(iii) If = K where H (and hence K) has one element, H = {x}, K = {y}
say, we write y rather than ={y} and say that y is conjugate to x in
G. Copying (ii) we see that x is then conjugate to y in G and say briefly that
x and y are conjugate in G.
(iv) The subset N0(H)={g: g G and =H} is called the normaliser of
H in G. If H comprises the single element x we have ({x}) = {g: g E G and
g1xg—x}={g:geG and xg=gx}. We call NG({x}) the centraliser* of x in
G and denote it by C0(x).

Remarks
(i) Conjugacy, as defined in 6.2.3, determines an equivalence relation on
the set of all subsets of G and also on the set of all subgroups of G.
6.2.3 (iii) determines an equivalence relation on G (exercise 3). In each case
the corresponding equivalence classes are called conjugacy classes.
(ii) It follows easily that Nc(H)EG. If then NG(H) is the unique
largest subgroup of G in which H is a normal subgroup (exercise 6). C0(x)
is also a subgroup of G.

Examples 6.2.4
(i) In 53 the subsets {(12)}, {(23)}, {(31)} (equivalently, the elements (12),
(23), (31)) are conjugate since (1311(12)(13)=(23) and (12)_l(23)(12)
(31). [What about the pair (12), (31)?]
(ii) In 54 the normaliser of V = {I, (12), (34), (12)(34)} is ((12), (34), (1324)),
a subgroup of order 8.
(iii) In 54 the centraliser of (123) is ((123)), a subgroup of order 3.
(iv) Any normal subgroup is conjugate only to itself.

One result we shall.use several times below is

Theorem 6.2.5 The number of distinct conjugates of the subset H in the

group G is equal to G : and hence divides G(.
*
Cf. exercise 17,
244 Structure theorems of group theory

Proof Foranyx, yeG weseethatx1Hxy1Hy if!

is, ill EN0(H), that is, if! NG(H)X —N0(H)y. This is enough.

We obtain immediately (cf. 5.6.1O(iii))

Theorem 6.2.6 A group of order (>1) has non-trivial centre.

Proof Let C1, C2,.. ., be the equivalence classes of conjugate elements

(i.e. conjugacy classes) of G where C1 is the class containing just the identity
elements of G. Let C2,..., r) denote the remaining equivalence classes,
if any, which contain just one element. By exercise II, U C1 = ((G), the
centre of G. Since G = U C1 and the C1 are pairwise disjoint we see that*

GI=EICkI+ IdlI
k=I
(This equality is called the class equation of G.) But each of the C1j divides
(by 6.2.5) and hence is a power of p, possibly p°. Now and pllCd

for l=t+l,...,r. It follows that t>1 and hence that G

has non-trivial centre.

Remark A group of order p and q distinct primes, may easily have

trivial centre—S3 for example (see S.6.1O(iii)).

Corollary 6.2.7 A group of order p2 (where p is a prime) is necessarily

abelian.

Proof By the previous theorem Thus 1(G)l=p or 11(G)I=p2.

In the latter case 1(G) = G and G is abelian immediately. We show that the
other possibility cannot occur. Suppose then that C(G)1 =p. Then =
p,which implies that G/1(G) is a cyclic group, by Lagrange's Theorem. Taking
the elements of G/1(G) to be 1(G),
= gP G is the disjoint union of these cosets.
Suppose x, y G. Then for some i, I and for suitable a, b 1(G) we
have x = g'a, y = g'b. But then xy = gag'b = g'g'ab = g'g1ba = glbgLa = yx.
Consequently G is abelian, whence G = 1(G). Thus I1(G)I = Gi >p, a
contradiction.
Remark Examples prove there is no hope of extending 6.2.7 to groups of
order p3. (See exercise 16.)
Theorems 6.2.8 and 6.2.12 are described in [31] as constituting the second
most important result in finite group theory. The first of these extends 5.11.4
*
1C,l denotes the number of elements in C1.
 Normal/set. Centraliser. Sylow's theorems 245

as far as possible in the case of non-abelian groups (cf. the example of A4

given after 5.11.5).

Theorem 6.2.8 (Sylow's First Theorem, 1872) Let G be a finite group* of

order pt's where p is a prime and (p, s) = 1. Then for each j3 E 1 such that
G hasasubgroupoforderp'3.
Proof We prove the result by induction on 1G. Clearly the result is true if
= 1 so we assume that IG = p"s and that the result holds for all groups
of smaller order. We use the class equation I I C,j + IC,! again.
k1 i=t+1
If, for some 1, we have p t lCd, where C, is the set of conjugates of g, say,
thenptlG
the induction hypothesis we can infer that C0(g), and hence G, contains a
subgroup of order p".
If, for each 1, we have lcd then, as in the proof of 6.2.6, we find p1
By 5.11.4 has a subgroup Y, say, of order p. Now Y -C G (exercise
5.10.14). Look at the factor group G/Y. Clearly 1G/Yl =p°1s and so, by
the induction hypothesis, G/ Y has a subgroup of every order p7 for 0 y
a —1. It then follows from exercise 5.11.8 that 0 has a subgroup of order
us required.

Remark 0 can have no subgroup of order with a <fi by Lagrange's

Theorem.

We can now answer Problem 1. It is interesting to see how much machinery

has been necessary in order to solve what might have seemed, at first glance,
a rather simple problem.

Theorem 6.2.9 Let 0 be a finite group. Then 101 for some a if and
only if the order of every element of 0 is a power of p.

Proof One way round is easy (given Lagrange's Theorem!): If 0, of order

contains an element of order n then n pa by Lagrange's Theorem.
The other way round is also easy (now!): If 0 contains only elements of
p-power order then 101 =p" for some a; otherwise there would be a prime
q such that qllGI and then by 6.2.8 a subgroup and hence an element of
order q in 0.

Remark From 6.2.6 and 6.2.9 every finite p-group (other than the trivial
group) has non-trivial centre. There do exist infinite p-groups with trivial
centre. The first example was constructed by A G Kurosht in 1939.
*
We may assume G non-abelian—if it is abetian use 511.5.
Aleksandr Gennadievich Kurosh (19 January 1908 — 18 May 1971).
246 Structure theorems of group theory

Sylow's investigations led to certain subgroups in any group 0 being named

after him.

Definition 6.2.10 Let G be a finite group and let where p is a

prime and (p, s) = 1. Each subgroup of order in 0 is called a Sylow
p-subgroup of 0.

Examples 6.2.11
(i) The Sylow 2-subgroups of 53 are {I, (12)}, {I, (23)}, and {I, (31)}. There
is just one Sylow 3-subgroup, namely that generated by (123). For each other
prime there is only one Sylow p-subgroup, namely the trivial subgroup.
(ii) The Sylow 2-subgroups of all have order 8. One such is the subgroup
in 6.2.4(u). Another is the conjugate of this by the element (13). In fact, as
the following theorem shows, 54 has precisely 3 Sylow 2-subgroups which are
moreover conjugate to one another.

Remark The important part of Sylow's First Theorem says that, in every
group 0, Sylow p-subgroups do exist. Lagrange's Theorem, on the other
hand, assures us that in a given group certain types of subgroup don't exist!
Thus Lagrange's Theorem gives necessary conditions and Sylow's First
Theorem sufficient conditions for the existence in a group of subgroups of
certain types.

We put Sylow's Second and Third Theorems together in

Theorem 6.2.12 Let 0 be a finite group of order pas where (p, s) = 1 and
let P be a Sylow p-subgroup. Then each Sylow p-subgroup of 0 is conjugate
(and hence isomorphic) to P. Further, the number of such conjugates divides
blip" and is congruent to 1 (modp).

We shall need two preliminary lemmas.

Lemma 6.2.13 Let P be a Sylow p-subgroup of 0 and let g e 0 have p-power

order. If p& = P, then g P.

Proof Since P c NG(P) and since g C N0(P) we have gP C NG(P)/P. Further

(gP) is a subgroup of some p-power order since g and hence gP each have
p-power order. Let S be the inverse image of (gP) under the homomorphism
from N0(P) onto N0(P)/P. Then since (gP) and P have p-power order it
follows from exercise 5.11.8 that S has p-power order. Since P is a Sylow
p-subgroup of G we conclude that S = P. Thus (gP) is the trivial group; that
is, gEEP.

Lemma 6.2.14 Let T G and H 0. Then the number of distinct conju-

gates of the form H' where t e T is equal to IT: TnN0(H)b.
 Normal/set. Ceritraliser. Sylow's theorems 247

The proof of this generalisation of 6.2.5 is left to exercise 7.

We can now give

Proof of Theorem 6.2.12 Let P be as given and let K ={P = P0, P1 Pr}
denote the set of distinct conjugates of P in G. Clearly each is a Sylow
p-subgroup of G. We place an equivalence relation on K by writing —P1
for some aeP. Then {P} is an equivalence class. However, no other
is: otherwise we should have for each heP and hence (a
contradiction) by 6.2.13.
Next the number of conjugates of Pk (k 0) under P is, by 6.2.14,
JP: PnNC(Pk)I and this is a power of p bigger than p° [why?]. Thus we see
immediately that the conjugates P1,. , Pr split into classes each with a
. .

multiple of p elements. Consequently the number of conjugates of P in G

(including P itself) is of the form 1+ mp.
Suppose Q is any Sylow p-subgroup of G not listed in K. We apply the
same technique as that just used except that we split K into classes by putting
— P1 if! = P1 for some c e 0. An identical argument to that in the second
paragraph above shows that this time r + 1 is a multiple of p which contradicts
the formula above. This shows that all the Sylow p-subgroups are listed in
K; that is each is conjugate, in G, to P.
Finally we observe that conjugate subgroups are clearly isomorphic (exercise
4) and that 1+mp=r+1=IG:NG(P)I. That follows
immediately.

Remarks Sylow's Theorems can be generalised in several ways, on the one

hand to infinite groups under certain extra hypotheses [36]. (One then defines
a Sylow p-subgroup to be not contained in any bigger p-subgroup,
rather than as one of 'maximum p-power order'.) On the other hand one can
define the concept of Sylow ir-subgroup where ir is a set of primes (see
exercise 6.5.20 and the concept of Hall ir-subgroup in Section 6.5).

We now proceed to some applications. Others will be given later.

We begin with a definition whose importance will be revealed in Sections

6.5 and 6.6. See also 7.10.5 and exercise 7.9.2.

Definition 6.2.15 A group G is called simple if! it has no normal subgroups

other than (e) and G.

One of the main uses of the Sylow Theorems is to prove, for certain special
n, that no group of order n is simple. We give three examples.

Examples 6.2.16 (i) Let =42. Then G is not simple.

*
A p-subgroup: a subgroup which is itself a p-group.
248 Structure theorems of group theory

Proof By 6.2.12, the number of Sylow 7-subgroups of G is of the form

1 + 7k and divides 42/7 =6. Thus k =0 and G has just one Sylow 7-subgroup
P, say. But then all conjugates of P, being themselves Sylow 7-subgroups of
G, must coincide with P. Thus P is normal in G.

(ii) Let 1GI = 56. Then G is not simple.

Proof The number of Sylow 7-subgroups of G is of the form 1+7k and

divides 8. Thus there are 1 or 8 of them. The number of Sylow 2-subgroups
of G is of the form 1 +2k and divides 7. Thus there are 1 or 7 of them. Thus
neither prime dividing 56 gives us a normal subgroup as immediately as in
the last example. Let us do a finer analysis. Suppose G had 8 Sylow 7-
subgroups. Since 7 is a prime each pair of these subgroups can only have (e)
in common. [Are you convinced?] Thus these 8 subgroups yield 8(7—1) = 48
non-identity elements of G. Now G has a subgroup of order 8 (by 6.2.8), the
order of each element in this subgroup being a power of 2. Thus this subgroup
of 8 elements is totally disjoint from the subset of 48 non-identity elements
of G already listed. It follows that if G has 8 Sylow 7-subgroups then G has
room for only one Sylow 2-subgroup which would have to be normal.
Of course if G has only one Sylow 7-subgroup we proceed as in case (i).

(iii) If GI = 48 then G is not simple.

Proof Let H be a Sylow 2-subgroup. Then HI = 16. If H is the only Sylow

2-subgroup of G there is nothing more to prove. Otherwise there will be 3
such. Let K be one of these. Consider HnK. If HnK has order 1
or 2 or 4 (and hence index 16 or 8 or 4) in H, it is not difficult to check that
there are 162 or 16 or 16 distinct elements of the form hk (where h H
and k €K) in G (exercise 25). None of these is a possibility since IGI =48.
Thus H n K has index 2 in H and in K. Consequently H and K are subgroups
of Since we see that H<NG(HnK). Thus
16IING(HnK)II1GI and IN0(HnK)I>16. Thus ING(HnK)1=48, whence
H rm K G, as required.

Exercises
1 Show that a subgroup and a factor group of a p-group is again a p-group.
Show that if H c G and if H and G/H are p-groups then so is G.
2 Give an example of a finite group G generated by two elements of order
2 but where G is not a 2-group.
3 Show that conjugacy determines an equivalence relation on (a) the set of
all subsets of a group G; (b) the set of all subgroups of G; (c) the set of
elements of G.
 Normaliser. Centraliser. Sylow's theorems 249

4 Show that conjugate subgroups are isomorphic as groups.

S Let H C G. Show that N0(H) G and that H c NG(H). Find examples,
where G is a group, H a subgroup and K a subset but not a subgroup of G,
such that (a) N0(H)=0; (b) NG(H)H; (c) N0(K)=G; (d) NG(K)=(e).
6 Let H 0. Show that H c N0(H); indeed NG(H) is the unique largest
subgroup of 0 in which H is normal.
7 Prove Lemma 6.2.14.
8 Let 0 be a finite group and H <0. Show that U W 0. [Hint: Use
gEG
the fact that H 6.2.5 and IHI Now count!]
9 Find the centraliser of (1234) in 54. How many conjugates has (1234) in 54?
10 Show that, in every group 0, C(G)= fl C0(x).
xeG
11 ShowthatxeC(G)iffIG:CG(x)1=1.

12 Let 0 be a group of order p". Show that 0 has a series 0 =

of subgroups such that, for each i and
[Hint: Induction on n. Given 0, find On_i ciG such that
=p. Use the induction hypothesis to find in G/Gn1 a series
Gi 0, 0
—>...>------whereeach--—-c———andwhere
0,
=p.Use
0n—1 0n_i tin_i 0n-i tin-i -n—i
exercise 5.11.9 and 5.1O.1O(iii) to show G >Gn is a series
as required.]
13 Verify the class equation for 53, 54 and D4. (D4 is the group of 8
symmetries of the square.)
14 Use the class equation to show that if H c 0 (H (e)) and if 0 is a
p-groupthen [Hint:

IS Let 0 be a p-group of order p". We know (from 6.2.8) that G has at

least one subgroup of order Show that every such subgroup S is normal
in ci. [1-lint: If x then (5, x) =0 and 5 -c (5,4. Otherwise C(G)
Pass to — Doesn't exercise 12 already prove this?
—.]

16 Exhibit, for some prime p, a non-abelian group of order p3. [Hint: try
p =2.]
17 Let 0 be a group and S a subgroup. Define C0(S), the centraliser of S
in 0 by C0(S)={g:geG and gx=xg for all xeS}. Show that C0(S) is a
subgroup of 0 and that C0(S) -c N0(S). Prove that NG(S)/CG(S) is isomorphic
to a subgroup of Aut (5). (See exercise 5.9.18.)
250 Structure theorems of group theory

18 Show that any conjugate of a Sylow p-subgroup is a Sylow p-subgroup.

19 Let P be a Sylow p-subgroup and H a subgroup of the group G. Is it

true that P n H is a Sylow p-subgroup of H? [Hint: try 53.]
20 Show that a Sylow p-subgroup P of a group G is a Sylow p-subgroup of
its normaliser. Show further that P is the only Sylow p-subgroup of its
normaliser.

21 Let S be a p-subgroup and P a Sylow p-subgroup of G. Show that S is

contained in some conjugate of P. Deduce that one could equally well define
a Sylow p-subgroup to be 'a p-subgroup of G not contained in any larger
p-subgroup of G'.
22 Show that NG(No(P)) = NG(P), P being a Sylow p-subgroup of G.
23 Show that no group with order 135, 30 or 96 can be simple.
24 Find two subgroups of order 4 which are not conjugate in (This
contradicts 6.2.12, surely?)
25 Let H, K be subgroups of a finite group G. Show that the set
{hk: h e H, k E K} contains exactly elements.

6.3 Direct products

In this section we provide a tool which can, on the one hand, be used to
dissect given groups in to more easily managed pieces but which, used the
other way round to build up 'big' groups out of little ones, can be useful for
the purpose of constructing counterexamples. (See exercises 21 and 22 below,
exercise 5.11.7 and exercise 6.5.11.) There are many such tools in existence
in group theory, this one being just about the simplest of its kind to describe.
Despite this simplicity, it is far from useless, being exactly the right tool to
penetrate the structure of all finitely generated abelian groups (see 6.4.3 and
6.4.4 and the Remarks at the end of Section 6.4).
To keep things straightforward we shall deal only with the case of a finite
number of groups. The infinite case presents two faces (which coincide in the
finite case) but not, at this stage, any extra difficulty (exercise 20).

Definition 6.3.1 Let G1, G2 be given groups and set G =

G1 x G2 x the cartesian product of the sets
G a binary operation by putting g2 (h1, h2 =
(gihi, g2h2, . . ,
. (G, .), which is clearly a group (see exercise 1), is called
the direct product of the groups G, (1 c i c n).
 Direct products 251

We easily prove

Theorem 6.3.2 The group G contains subgroups H1 (1 c i c n) such that (i)

for each i, (ii) for each i, H1 c G; (iii) G —(H1,H2,. (see . .

5.6.11); (iv) for each i, H1 n(H1,.. , H1+1, .. , . .

Proof We define H1 to be the set {(e,.. , e, h1, e,.. , e): h1 . . G1}. It is easy to
check that H1 is a subgroup of G. The map (e,.. , e, h,, e, . . . . , e)—.h1 is clearly
an isomorphism from H1 onto G1. The equality (gi . . . , g1,. . . ,

(e, . . . , e, h,, e, . . . , e) (g1, . . . , . . . , = (e, . . . , e, g[1h1g1, e, . . . , e)

shows that H G. The equality (gi, , ge,.. , gj = (gi, e, e,. . . .

(e, g2, e,...)' ... (e, e,.. . , e, shows that G = (H1, H2,. Ha). Finally . . ,

(iv) is a simple set-theoretic matter (see exercise 2).

Remarks on notation and terminology

(i) When each G1 in 6.3.1 is abelian and when we are using + for the binary
operation on each G1 we usually replace by ®, so that (gi, ga,. . , .

and refer to G as the direct

sum of the G = €9 . .$ G,,. .

(ii) In 6.3.1 we began with a set of groups G and formed a new group G.
The G are themselves not subgroups of G; they are only isomorphic to the
groups H1 which are actually inside G. For this reason the group G of 6.3.1
is sometimes called the external direct product of the G1. By way of contrast,
any group G satisfying the properties (ii), (iii), (iv) of 6.3.2 is called the internal
direct product of the H,. Clearly the word 'internal' is applicable; to justify
the use of the expression 'direct product' requires the following

Theorem 6.3.3 Let G possess subgroups H, with the properties (ii), (iii), (iv)
as in 6.3.2. Then G is isomorphic to the direct product of the H1, considered
as groups in their own right.

Proof
h1eH,cG where Clearly h71h71h1h1=hHhI'h1h1)eH since H cG.
But also [why?]. Thus hI1h[1h1h1e
H1 n H1 = (e) [why?]. It follows that, for all h1 e H1, h1 H1 with i f, h1h1 = h11i1.
Since G = (H1,. , we can clearly write every element g of G in the form
. .

of a product g = x11x12. say, where each x1, is in one of the H1. Using
. .

the fact, just proved, that each element of H1 commutes with every element
of H2, H3, .. , we can, in the expression for g, simply move all the x1
.

which belong to H1 to the left and then take their product to reduce the
number of elements from H1 in the product for g to one, h1 say. Next we
similarly collect all elements of H2 in this new product into one element, h2,
lying immediately to the right of h1. Repeating this for the elements from
H3, H4,. . , shows that g can be expressed in the form
.

g=h1h2. (*)
252 Structure theorems of group theory

where each h1 e H (h1 may or may not be the identity of H1). The final step
in this initial stage of the proof is to show that the representation (*) of g as
a product is unique. That is, if also g = k1k2. k,, with k1 H1 (1 i c n), we
. .

wish to show that h1 k1, h2—k2,. , . . We leave this to exercise 9.

The alleged isomorphism is now easy to find. One defines 0: x I-In -,

The above remarks show in turn that C is onto and 1—1. Thus only the
homomorphism property remains. This again we leave to exercise 9.

Remark 6.3.2 and 6.3.3 demonstrate the intimate connection between the
concept of internal and external direct product and the reader may ask why
we bother with both. The answer, as the reader will see if he tries to do
without one or the other concept in what follows, is simply 'for ease of
exposition': we can (and will) be careless in talking of groups as 'being direct
products' omitting the words 'isomorphic to' or 'of its subgroups', and we
shall occasionally write = when might be more accurate.

Exercises
1 Show that the pair (G, in 6.3.1 is a group.
2 Prove 6.3.2(iv).
3 Show that the direct product of any (finite) number of abelian groups is
abelian. Does this result remain valid if the word 'abelian' is replaced by (a)
'finite'; (b) 'cyclic'; (c) 'matrix'?
4 (a) Is G C6®C10 cyclic? If not, what is the largest order of any of the
elements of G? (Clearly the answer is 60 if G is cyclic.)
(b) Is A5 S3 x D5 (D5 being the group of symmetries of a regular
pentagon)?
5 A group G is a proper direct product if G A x B with neither A nor B
the trivial group. Are the following groups proper direct products (or sums)?
(a) S4; (b) D6; (c) (for each n 3); (d) (C, +); (e) C,,', p a prime.
6 ShowthatCrxC,=Crsif(r,s)=1.
7

Show that
(a)

the proof of 6.3.3 where indicated.

10 Write $) in as many different ways as you can as a direct sum of two

or more groups.
 Direct products 253

11 How many subgroups has 12$12$Z2? [Hint: try z2ez2 first where the
answer is not 4.]
12 Is it true that every subgroup of A x B has the form K x Y where K A
and YEB?
7L$Z
13 Write as a direct sum of two cyclic groups (if possible!) when N
N
is (a) the subgroup of Z$Z generated by (2, 2); (b) the subgroup of Z$Z
generated by (2, 3).
14 Show that where R/Z is defined in the same
manner as the group of exercise 5.6.16.
15 Show that the map fi:01x02-÷01 given by (gi,g2)ifr=g1 is a
homomorphism of x 02 onto 01. What is ker ii?
16 Let N1<G1,N2CG2 and N=<(nl,eG2), (e61,n2):n1eN1,n2eN2)
_cG1 x G2. Show that N=N1 xN2, NC xG2 and that

17 Let N1, N2 c 0 and let N =N1 nN2. Show that is isomorphic to a

subgroup of x $-. [Hint: try mapping 0 -* x by g -. (gN1, gN2). Show

that this is a homomorphism with kernel N.] Deduce that if N1 nN2 = (e)
then x contains a subgroup isomorphic to 0.

IS Find the centre of 53 xD8. More generally relate and C(B) to

C(A xB).
19 Let 0 be a group of order p2q2 where p, q are distinct primes and where
qkp2— I and flq2— 1. Show that 0 is a direct product of a group of order
p2 and one of order q2. Deduce that 0 is abelian.
20 Let 02,... be an infinite set of groups. Their cartesian product set
O = 01 x 02 comprises all infinite vectors (gi, g e 0. Show
that with multiplication defined by (gi, g2,. . (h1, h2,. .) = (g1h1, g2h2,.. .),
.

O becomes a group. Show that the subset 0 of 0 comprising those infinite

vectors which contain only finitely many non-identity elements is a subgroup
of 0. (0 is called the restricted direct product and 0 the unrestricted direct
product of the 0.)
21 Show that there exist groups A, B such that each is isomorphic to a
subgroup of the other and yet A B. [Hint: Use infinite direct products
involving C2 and C4.]
22 Show that there exist groups A, B such that each is a homomorphic
image of the other and yet [Hint: Same as for exercise 21.]
254 Structure theorems of group theory

6.4 Finite abelian groups

We now show how the relatively simple concepts of direct sum and finite
cyclic group enable us to characterise and then classify finite abelian groups.
Thus for these groups, as for finite fields, the algebraist's dream is realised.
(See the end of Section 4.5 and Remarks in Section 3.10.) Actually it is
possible similarly to characterise and classify all finitely generated abelian
groups but we shall stick to the finite case mainly because it is particularly
easy and the main aim of this section is to introduce the reader to the spirit
of the development as much as to the theorems themselves.
We begin, using the additive notation, with

Theorem 6.4.1 Let A be a finite abelian group. If, for each prime p dividing
we let denote the set of all elements of p-power order in A (including
0) then each is a subgroup of A and A is their direct sum.

Proof That each 5,. is a subgroup (indeed a normal subgroup) of A we leave

to exercise 1. Next let x eA (x #0). Then x has order n where nilAl. Suppose
that n . where the are distinct primes and each
. 1. Write,
for each i (1 i r), q1 Then the positive gcd of the qj is 1 and so
there exist (exercise 1.4.6) integers k1,. .. , such that k iq1 + k,q7 = 1.
It follows that x = (k +. + k,qr)x = k iX + k4rX. Now q1x (and hence
has order dividing [even equal to ?] Pr. Thus x has been expressed as
a sum of elements with orders respectively. It follows
that XE <S,,, S,,2, ., Si,) But x was any element ofA.Thus
. . Al> =
A.
Finally A will be the direct sum of the S,. if only we can show that, for
n(S,.: plIA! andp q) =(0). Butthisiseasybecause anyelement
of Sq has order a power of q whilst any element of (5,.: pjjAl and p has
order coprime to q. Thus the only element common to both subgroups is 0.

We now show how to split each of these p-primary components 5,. into
cyclic components.

Theorem 6.4.2 Let 5,. be a finite abelian p-group. Then 5,. is a direct sum
of cyclic groups (of p-power order).

Proof Let s be any of the elements of maximal order p°, say, in 5,. and let T
be as large a subgroup of 5,. as possible satisfying (s) n T = (0). Then (s, T) =
(s)$T. (i) (s) and T generate (s, T)!; (ii) (s) and T are normal in
For:
(s, T) (iii) (s)nT=(0) (by choice!). If (s)EDT<S,. we can find an
and
such that Since = 0 [why?] E(5)$T. There then exists
Now py€(s)$T. Hence py=ls+t (1E7L,tET). Then O=p5=
p"11s +pa_lt. Thus (s) n T = (0). It follows that pal, pk = 1, say, and
 Finite abelian groups 255

hence that p(y—ks)——py--ls=teT. However, [Why not?] Thus

(T, y —ks)> T. Consequently (7', y —ks)n(s)>(O). [Why?] That is, for some
in, n E/ and VET we have =v+n(y—ks). Here p4'n (or else mse
(s) n T = (0), a contradiction). Thus (n, p) = 1. Also ny = ms — v + nks (s)$
T and [by choice] py E (s)$ T. Thus y (s)$ T [why?], a contradiction. Thus
(s)$T = 5,, after all. An argument using induction now completes the proof
of the theorem. < IS,,I so T is a direct sum of p-power cycles. Now use
exercise 6.3.8(a).)

Combining 6.4.1 and 6.4.2 we find (exercise 6.3.8(a) again)

Theorem 6.4.3 Let A be a finite abelian group. Then A is the direct sum
of cyclic subgroups of various prime power orders, the primes involved being
those which divide Al.

Remarks
(i) If A is cyclic of prime power order it is not decomposable into a direct
sum of two or more non-trivial groups (exercise 6.3.5(e)).
(ii) Since every finite direct sum of (prime power) cycles is a finite abelian
group we have characterised finite abelian groups as precisely those expressible
as direct sums of (prime power) cyclic groups. In order to effect a classification
of these groups we need to explain how to tell them apart. This follows from
considering the question: Given an abelian group A, is its decomposition
into a direct sum of prime power cyclic groups unique? The immediate answer
is no; from A =(s)$(t)=(s)$(u) where (s), (t), (u) are cycles one cannot
infer (t) = (u) (exercise 2). Recalling from Section 3.10 that an algebraist is
a person who cannot see any difference between isomorphic systems, never
mind equal ones, we see that the next theorem yields as much as the algebraist
could ask for.

Theorem 6.4.4 (The Fundamental Theorem of Finite Abelian Groups*) Let

A be a finite abelian group. Then any two decompositions of A into direct
sums of cyclic groups of prime power order contain the same number of
summands of each order.

Remark It is easy to check that if A=S,,,EB" are

decompositions of the type provided by 6.4.1 then S,,2 = . . . , = 5',,,.

Thus we can concentrate on the case where A is an (abelian) p-group.

Perhaps the pictorial representation in Fig. 6.1 of an abelian p-group which
decomposes as a direct sum of 2 p-cycles, 3 p2-cycles, 2 p4-cycles and one
p 5-cycle will help the reader through the following proof.

*
First proved in 1878.
256 Structure theorems of group theory

81 84 85

This
'slice'is{]hedirectsumof3p-cvcles ® 3 A
pA
p2A - I t
I
J

Fig. 6.1

Proof of Theorem 6.4.4 Let a decomposition of A into a direct sum of

p-power cycles be given. For each i define to be the direct sum of all cyclic
summands of order p in this decomposition. Then A B1 $B2®
(say) (exercise 6.3.8(b)) and, for eachjEl*, p'A = .

(Herep'A ={p'a: a eA}. It is clear thatp'A isasubgroup

of A.) Now consider the factor group* This group is isomorphic to

pA .

(exercise 6.3.16), a direct sum of (a direct

p B1÷2 p
sum of) p-cycles. Now the number of summands in this direct sum of cycles
is b1±3 +b142+ . . +b, where b, is the number of cyclic summands in Thus
the total number of cyclic summands in is equal to the number of cyclic
summands in A which have order at least Thus the number of cycles of
each order in a direct decomposition of A into p-power cycles depends
only on A.

Remarks
(i) In view of our stressing, in Chapters 1 and 3, the concept of unique
factorisation in 7 and in various number rings, it is appropriate to observe
that 6.4.3 and 6.4.4 together yield yet another result of this type. It follows
that two finite abelian groups are equal in the algebraist's eyes (i.e. isomorphic)
if they decompose into the same number of indecomposable pieces (i.e. prime
power cycles) of each kind. Another unique factorisation theorem for finite
abelian groups is given in exercise 12.
(ii) The structure of an infinite abelian group A can be very involved. But
if A is finitely generated we have: A is the direct sum of a finite number of
infinite cyclic groups and a finite number of cyclic p-groups for various primes
p. The number of summands of each type completely characterises A. ([27, p. 96];
[36, p. 106].)

*
We keep to the fractional notation even in the case of abelian groups. The difference notation,
p'B risks confusion with
 Finite abe/ian groups 257

Example 6.4.5 There are 10 different types (i.e. isomorphism classes) of

abelian groups of order 1008 32.
7), namely

C2$C2®C2®C2$C3®C3®C7 C2®C2EBC2®C2®C9®C7
C2$C2®C4
c2®c8 sc3®c3®c7 c2®c8
C4®C4 ®C3®C3®C7 C4®
®C3®C3®C7 C16

The largest order of an element in each of these groups is respectively

42, 84, 168, 84, 336 and 126, 252, 504, 252, 1008.

Exercises
1 Prove that the subsets of 6.4.1 are subgroups, as claimed.
2 Show by means of an example that if A is the internal direct sum of
subgroups X and Y and also the internal direct sum of X and Z then it is
possible that Y [Hint: C2®C2.]
3 Find the number of (isomorphism classes of) abelian groups of order (a)
360; (b) 218; (c) p5 where p is a prime. In each case give the maximum of
the orders of the elements in the respective groups.
4 Are there more (isomorphism classes of) abelian groups of order i13 than
of order 2232?

5 How many different abelian groups of order 720 have (a) exactly 3, and
(b) exactly 4, subgroups of order 2? (Careful with (b).)

6 Write (M(16), 0) as a proper direct product of cycles—or prove such a

decomposition impossible.

7 Let A be a finite abelian group. Show that A has a finite series of subgroups
A =A0>A1 >.. . >(0) such that each is a (cyclic) group of prime
order.

8 Two finite abelian groups A, B have exactly the same number of elements
of order m for each m E zt Show that A B.
9 Show that if a finite abelian group A has, for each m 1A1, exactly m
elements such that i-na = 0 then A is cyclic. [Hint: consider C2 x C2.J
10 Show that, in an infinite abelian group, the elements of infinite order,
together with 0, do not necessarily form a subgroup. [Hint: C2 x Give
an example other than (/, +) in which they do.
258 Structure theorems of group theory

11 A finite abelian group A is generated by two elements a, b such that

4a = 4b = 0 and 2a = 2b. Does this mean that A cannot be expressed as a
proper direct sum?
12 Let A be a finite abelian group. Show that A can be expressed as a direct
sum of cyclic groups of orders n1, n2,.. . ,n, (say) where . It'r. [Hint:
. .

Try a specific example, say C2®C2®C3, first and see if you can generalise
the argument. Recall that Cr$C, = Crs if (r, s) = 1.] This is yet another kind
of unique factorisation theorem since A determines the n1 uniquely. (Proof?)
13 (Cf. exercise 4.5.5 and exercise 9.) Let (F, +, .) be a finite field. Prove
that (Fx, is a cyclic group. [Hint: assume not and deduce that some equation
of the form f' — I =0 has more than p solutions in F.]
14 Let G, H, K be finite abelian groups. Show that if GEF3G asHE9H then
and that if then HaK.

6.5 Soluble* groups. Composition series

After Ruffini (1799, 1813) and Abel (1824) had shown the impossibility of
finding a universal formula in terms of +, —, ± and which would
yield the roots of any given quintic equation (see Section 7.9) the question
of exactly which particular equations were soluble* by radicalst was answered
by Galois in 1832. The answer is: those equations whose associated group
satisfies an easily stated condition (see 6.5.7). Groups having this property
were christened 'soluble' by Jordan in 1867. However, even had it been of
no significance for the theory of equations the solubility property would
undoubtedly have attracted the attention of research workers in group theory
because of the following philosophy. We have just seen what nice groups
finite abelian groups are in that their structure can be explicitly written down.
Given a quite arbitrary (finite) group G one can investigate it by looking for
special (e.g. Sylow) subgroups or by mapping it homomorphically onto a
member of some class of groups about which we are fairly knowledgeable.
Into this latter class fall the finite abelian groups, so a good ploy might be to
map G onto as 'large' a finite abelian group A, say, as possible. What A tells
us about G may not be much but it may well be better than nothing. Now
this homomorphism produces a kernel K, say, and we can apply the same
procedure to K that we have just applied to G. That is, we find a subgroup
N of K such that N is normal in K and K/N is as big an abelian homomorphic
image of K as possible. Continuing in this way we may (or may not!) reach
the identity subgroup of G. If we do, G is called a soluble group (Definition
6.5.5). It may be thought of as being made up of 'slices' (like a sliced loaf!),

*
Solvable, in USA.
Definition 7.7.2.
 Soluble groups. Composition series 259

each of which is abelian. The prospect thus generated would surely be enough
to set one about the study of such groups. (Presumably, if the theory had
developed this way, the groups considered would not be called soluble but
maybe multi- or poly-abelian?)
If G is a group, if N G and if G/N is abelian then, for every pair of
elements a, b E G, we have aNbN = bNaN, that is a 1b 1abN = N, which
means that a 'b 1ab EN. This helps motivate the following

Definition 6.5.1 Let G be a group and a, bEG. The element a1btab,

denoted briefly by [a, b], is called the commutator of a and b.

Remarks
(i) 'Commutator' is a good word to use: for [a, b] is a kind of measure as
to how near a and b come to commuting, [a, b] being the identity element
of G if and only if ab = ba.
(ii) It is perfectly natural to look at the set of all commutators in a group
G. The sad fact is that this subset may not form &subgroup of 0 (see [26,
p. 38]) so we move to the next best thing.

Definition 6.5.2 Let G be a group. The subgroup generated by the set

{[a, b]: a, b E G} is called the commutator subgroup or derived subgroup of
G. It is denoted by G' or or [0, 0].
More generally, for subgroups A, B of a group 0 we denote by [A, B] the
subgroup of 0 generated by the set {[a, b]: a EA, b €B}.

Examples 6.5.3
(i) Since [a, a] = e the identity element of each group is a commutator. In
an abelian group (and only in an abelian group) is e the only commutator.
Hence 0 is abelian if 0' = (e).
(ii) [a, b] = [b, aj1; [a,
= bfl for all a, b, g EQ.
(iii) In A5 [(l2)(35),(234)]=(l2)(35)(432)(l2)(35)(234)=(l2345) is a com-
mutator.
(iv) The derived group of the 8-gon group in 5.3.4(1) comprises I, a2, a4, a6.
(v) for all

n (See exercise 4 and 6.6.1.)

To show that G' does what we are asking of it we prove

Theorem 6.5.4 For each group 0, 0' c G. Further, if N c G then GIN is

abelian if N 0'. (Thus Q,/Q' is the largest abelian homomorphic image of
0.)

Proof 0' is, by definition, a subgroup. Each element x of 0' can be expressed
as a product x = [ga, g2][g3, g4]. . .g2j of commutators of 0. Now for
260 Structure theorems of group theory

each geG we have (see 6.5.3(u))

Thus G' G.
We saw earlier that if GIN is abelian then G' N. Conversely, if G' N
then GIN is abelian: for, given a, b and G we have a1b1ab E
Consequently aNbN = bNaN, as required.

Proceeding further, we make the formal

Definition 6.5.5 For each non-negative integer n define G(n±l) to be

G°'1, the derived subgroup of where G101 = G. The non-increasing
sequence G
of G. If, for some n, we have G is said to be a
soluble group and if in is the least such integer G is said to be soluble of
length m.

Examples 6.5.6
(i) = (I); 52 is soluble of length 1 (that is, 52 is abelian!); 53 is soluble of
length 2; 54 is soluble of length 3. The remaining 5,, (n 5) are not soluble
at all (see example 6.5.3(v)).
(ii) A2 = (I); A3, A4 are soluble of lengths 1 and 2 respectively. An is not
soluble if it
(iii) The 8-gon group of example 5.3.4(1) is soluble of length 2.
(iv) Let U,, (li) denote the set of all it x n matrices of the form

1 a12 a13 . . .

o 1 a23 . . .

A= 0 0 1 ... a3n

00 0 ... 1

where the a11 are real numbers.

It is not difficult to check that A' exists and that if
1 b12 . . .

0 1 1j23 ... b2,,

0 0 1 ... b3,,

00 0 ... 1

then b11+1 = It follows that the derived group comprises all

matrices whose form is that of A except that the elements 12, a23,.. . , a,,_1,,,
on the first 'superdiagonal' are all 0. In a similar manner one finds that
comprises matrices in which the first — I superdiagonals are full of Os.
It then follows easily that the group is soluble of length — [—log2 nJ.
 Soluble groups. Composition series 261

Note 6.5.7 6.5.5 is a more recent definition of the term 'soluble'. 6.5.8 below
gives an equivalent definition (for finite groups): A (finite) group G is soluble ¶
it has a series
(*)
of subgroups such that each + is a proper normal subgroup of with the
property that each quotient +11 is a prime. Since Galois had shown that an
equationf(x) = 0 is soluble by radicals if its (Galois) group had a series as in (*)
it was a natural step for Jordan* to call such groups soluble.
For yet another variant lying midway between (*) and 6.5.5, see exercise 9.

Remark We cannot infer from the facts that G1 c G and G2 G1 that

ci G. That is, normality is not transitive (exercise 5.10.17).

That the two definitions of soluble are equivalent for finite groups is given
by

Theorem 6.5.8 Let G have a series of type (*). Then = (e). Conversely,
if G has a series of type (*).

Proof To prove the first part we use induction. Clearly G0. Suppose
Gk. Then {G(k) Gld]c[Gk, Gk by 6.5.4, since
1

Gk/Gk+l is abelian.
For the converse, let G = G°> .> = (e) be the derived series
.

of G. For each i, is a finite abelian group and hence, by exercise

6.4.7, has a series of (normal) subgroups . >(e) .

such that each factor group A1/A1+1 is prime cyclic.

Using methods like those called for in exercise 6.2.12 we see that the
inverse images of these A under the natural homomorphism from onto
form a series of normal subgroups of with corresponding
factor groups again of prime order. Inserting these inverse images between
and 1) for each i refines the derived series of G into a series of type
(*) for G, as required.

Remarks
(i) From this characterisation and from exercise 6.2.12 we see immediately
that every finite p-group is soluble.
(ii) Subgroups and homomorphic images of soluble groups are again soluble
(exercise 7). Further, given a group G and a normal subgroup N such that
N and GIN are soluble one can show that G is necessarily soluble (exercise
17).
*
Camille Jordan (5 January 1838—22 January 1921).
262 Structure theorems of group theory

In (*) we see that the subgroups G, form a sort of maximal series—the G1

are packed together as closely as possible; since the factor groups are of prime
order there is, by Lagrange's Theorem, no room for any more insertions.
Such a series is, according to the following definition, an example of a
composition series.

Definition 6.5.9 Let G be a group and a

decreasing series of subgroups such that, for each i (0 c i n — 1) G1+1 ci
and is a simple group. Such a series is called a composition series for
G and the groups are called composition factors of G. n is called
the length of the series.

Remarks
(i) By 5.10. lO(iv) one can equally describe each subgroup in 6.5.9 as
being a maximal normal subgroup of maximal in the sense that no bigger
normal subgroup of can be placed strictly between G1 and
(ii) Clearly every finite group has a composition series; an infinite group
may not have (exercise 12).
(iii) Series (*) is a composition series for G; clearly finite groups of prime
orders must be simple.

Examples 6.5.10
(i) 53 >A3 > (e) is a composition series for S3. The composition factors here
are isomorphic to C2 and C3.
(ii) where C2={I, (12)(34)} is a composition series
for S4. Composition factors are C2, C3, C2, C2.
(iii) G = (a, b)>(a)>(a2)>(a4)4e)isone composition seriesforthe 8-gon
group of 5.3.4(1). (a, b)>(a2, b)4a4, b)>(a4)>(e) is another. In each series
the composition factors are C2, C2, C2, C2.
(iv) If S is a simple group then S > (e) is the only composition series for S.

In the above terminology Galois' theory says that an equation is soluble

by radicals if and only if the Galois group of the equation has a composition
series with prime cyclic factors. Now there will, in general, be many ways of
picking a composition series for a given finite group G. Beginning with G0 = G
one simply takes, for each i, to be any one of the (possibly many) maximal
proper normal subgroups of continuing in this way until one reaches (as
one must, since G is finite) the trivial subgroup. Suppose that one of the
composition factors in a series so obtained is not prime cyclic. Does this mean
that the corresponding equation is not soluble by radicals? Surely not. Maybe
if we had chosen the subgroups G, in a different manner we would have
obtained a composition series in which each of the factors is prime cyclic.
 Soluble groups. Composition series 263

Thus it looks as if we have to search through the various composition series

of G until either we find a series with all factors cyclic, or else we prove that
there is no such series—by examining each and every one!
Fortunately, Jordan proved in 1869 that the orders of the various composi-
tion factors of a given group G are invariant from one composition series to
another. Thus G will have one composition series with all its factors prime
cyclic itT this is the case with all composition series of G. That is, the first
series one examines completely answers the problem regarding solubility.
Jordan's result was strengthened by Holder* in 1889 to give, for any finite
group, the

Theorem 6.5.11 (The Jordan—Wilder Theorem) Let G be a finite group with

composition series

and

Then r s and the r factor groups can be put in 1—1 correspondence

with the r factor groups K3/K1÷1 in such a manner that corresponding groups
are isomorphic.
In order to prove this we make first the

Definition 6.5.12 Let H, K be subsets of the group G. We denote by HK

the set {hk: h EH, k EEK}.

Next we note the

Lemma 6.5.13 If H G and K -cG then HK is a subgroup of G.

Proof We leave this to exercise 13.

The tool required to prove 6.5.11 is

Theorem 6.5.14 (The Second Isomorphism Theorem) (cf. 4.3.2) Let

Proof ClearlyKciHK sinceKcG andKcHKcG. WedefineO:H-*

HK/K by hO = hK. Clearly 0 is a mapping. Rather easily we see that for
g, heH we have Also 0 maps H onto
HK/K since every element of HK/K is of the form hkK where h e H and
keK. But hkK=hK since keK.
* Otto Ludwig Holder (22 December 1859- 29 August 1937).
264 Structure theorems of group theory

Thus C is a homomorphism from H onto HK/K. According to 5.11.6 we

see that Now if hekerO then hK=eK =K. Thus heK,
that is ker C cH nK.
The reverse inclusion is easy.

Now for the

Proof of Theorem 6.5.11 We proceed by induction on K?. Thus we suppose

(i)

and
(ii)

are two composition series for G.

Clearly we may suppose H1 K1. (If H1 = K1 an application of the induction
hypothesis to H1 = K1 soon finishes the task.) From this we infer that H1K1 = G
GH1K1 H1 G K1
[why?] and hence that proof that is
K1H, The
nH1
similar.
Consider now the series
(iii)
and
(iv)

where

H1 n
series H1 series
H1; similarly with those parts of (ii) and (iv) in relation to By induction
we may infer that the factors
fG\ H1
—H2
—. .
. from(i) (A)
\H1/ H2' H3 H,
and the factors
/G\
I—I —
H1 H1nK1
. . . from (iii) (B)
\H11 H1nK1' L2 ' '

can be put into 1—1 correspondence in such a way that corresponding factors
are isomorphic groups.
Similar remarks apply to the factors

from (ii) (C)

 Soluble groups. Composition series 265

and the factors

fG\ K1 K1nH1
from(iv) (D)
L2

Finally, since the groups in (B) and (D) can, by the above isomorphisms
G H1 G K1
and clearly be put in 1—1 correspondence in the
K1H1 nK1 rmH1'
above manner, so can the factors of (A) and (C). This proves the theorem.

Remarks
(i) The Jordan—Holder theorem is yet another sort of unique factorisation
theorem (see Remarks at the end of Section 6.4). Notice that we do not claim
G GH1K1 H, K1
H1 —, --- -p-, . . . , etc., only that the factors
K1
and are isomorphic
111±1
in pairs, in some order (see exercise 18(b)). Comparing this situation with that
in Z we see that we can regard G as a sort of (in general non-commutative)
'product of primes', namely the simple composition factors. If, in particular,
the Jordan—HOlder theorem is applied to cyclic groups of finite order we
obtain a splendidly long-winded proof of the uniqueness part of the funda-
mental theorem of arithmetic!
(ii) As two finite groups G1 and G2 may have isomorphic composition factors
and yet not be isomorphic (see exercise 18) the Jordan—HOlder theorem is
mainly of use in a negative sense: if H1 and H2 have distinct sets of composition
factors then H1 and H2 cannot possibly be isomorphic.
We close this section by mentioning two of the more recent developments
in the theory of finite soluble groups. First we indicate how Sylow's theorems
can be generalised in the case of soluble groups.
Let n be a finite set of (positive) primes and n' denote the complementary
set of (positive) primes not in it. A it-number (respectively n'-number) is any
positive integer (including 1) whose prime factors belong to the set it (respect-
ively, to iv'). A finite group K is called a it-group if IKI is a it-number. A
subgroup H of a finite group G is called a Hall n-subgroup of G if (i) H is a iv-
group and (ii) G : Hj is a ir'-number. In particular if it = {p} any finite group G
I

possesses Hall it-subgroups, namely its Sylow subgroups. However in general a

group need not have a Hall it-subgroup. For instance the group A5 of order
60 = 22.3.5 has no Hall {3, 5}-subgroup (exercise 20). However theorems like
Sylow's can be proved for it-subgroups of soluble groups. In fact in 1928 Philip
Hall* proved

Theorem H Let G be a finite soluble group and n be a finite set of primes.

Let A be any n-subgroup (including (e)) of G. Then
(i) G has Hall n-subgroups: there is even one containing A;
(ii) any two Hall n-subgroups of G are conjugate.
* Philip Hall (11 April 1904—30 December 1982).
266 Structure theorems of group theory

The proof of this theorem appears in most modern introductory texts on

group theory. (We refer the reader to [34] or [36] to name just two sources.)
On the other hand the othçr result we wish to mention, proved as recently
as 1963, possibly never will. We first reveal the intimate connection with
finite simple groups.
By 1900, several infinite 'families', together with five 'sporadic' examples,
of non-abelian finite simple groups were known (see next section). All were
of even order. Burnside* (1897) suggested an investigation into the existence
of non-abelian simple groups of odd order be carried out and in 1911
he conjectured that there is no such group. Assuming Burnside to be correct
this implies immediately that every group of odd order is soluble. For if G
is any group of odd order then G has a composition series G =
Go> Q1> . = (e) with simple, odd-order, composition factors. By
Burnside's conjecture each factor must be simple abelian and hence prime
cyclic (exercise 6.6.1). Thus G is soluble. Conversely if a non-abelian group
G (whatever its order) is soluble, then it cannot be simple since G' is a normal
proper and non-trivial subgroup of G. The theorem that all non-abelian simple
groups must have even order was proved by Feit and Thompson, two alge-
braists from the USA (and then in their early thirties) in a mammoth 254-page
paper, 'Solvability of groups of odd order' which occupied an entire edition
of the Pacific Journal of Mathematics in 1963. (For his subsequent work in
determining all minimal finite simple groups, that is all finite simple groups
in which every proper subgroup is soluble, John Thompson was awarded the
Fields Medalt at the 1970 International Congress of Mathematicians.)

Exercises
1 Let G be a group and letx e G. Show that x C C(G) if [x, g]= e for alIg e G.
2 Show that for all a,b,ceG,
[a, c] = [a, c], b][b, c].
3 Let a, b, c C G. Is it necessarily true that [a, [b, c]]? [Hint:
Look at S3.]
4 Show that there are (even) permutations a, b, c, d in A5 such that [a, b]
is (123) and [c, d] is (12)(34). Deduce that every element of A5 is a commu-
tator. Deduce that A5 is not soluble.
S Show that if A c G then [A, B] c G and [A, B]si AnB.
6 Let 0: G—*G0 be a homomorphism. Show that [a, b]0 =[aO, bO] for all
a, be G. Deduce that (G')O = (GO)'.
*
William Burnside (2 July 1852 —21 August 1927).
±For details of this 'Nobel Prize' of Mathematics see [99]. There is a photograph in Bulletin of the
American Math. Soc., Vol. 40, 1934, p. 189. See also [96].
 Soluble groups. Composition series 267

7 Show that subgroups and homomorphic images of soluble groups are again
soluble. [Hint: for homomorphisms extend exercise 6.]
8 (a) Show that the direct product of two soluble groups is again soluble.
Find the solubility length of the direct product in terms of the lengths of its
factors.
Let H c 0, K c 0 be such that — and — are soluble. Show that 0
G G
(b)
H K I-InK
is soluble. [Hint: use exercise 6.3.17.]
9 In 1864 Jordan, in effect, stated: The group 0 is soluble if 0 has a series
and
is abelian. Prove this. [Hint: For the given series prove by induction
that 0. The other half of the proof is immediate.]
10 Show that the solubility length of 54 is 3 by finding the derived series of
54. Do the same for the group D5 of symmetries of the regular pentagon.
11 Show that a direct product of infinitely many soluble groups need not
be soluble. [Hint: find groups 0 with solubility length i, for each i e
12 Show that the infinite cyclic group has no (finite) composition series.
Show that if one allows infinite series with simple factors then (1, +) has
composition series but no analogue of the Jordan--Holder theorem can hold.
[Hint: Look at the series and />31>97>...]
13 Let and Show that HK need not be a subgroup of 0.
Show that HK is a subgroup of G if one of H and K is normal in G and
that HK -c 0 if both H, K are normal in 0.
14 Write down as many distinct composition series as you can for the cyclic
group of order 180. Find the composition factors in each case. Hence verify
the Jordan—HOlder theorem directly for this group. Do the same for the group
53xD5.
15 Let 0 be a finite group and let H c 0. Show that 0 has a composition
series in which H appears as a term.
16 Does 54 have a sequence 54=H0>H1>H2>H3>(e) in which each
H1/H1+1 is prime cyclic and each H1 is normal in 54?
=
17 Let N c G. Show that Deduce that if is soluble of
length m and if N is soluble of length n then G is soluble of length rn + n.

18 (a) Give an example of two finite groups 0, H which have the same
composition factors and yet are not isomorphic. [Try order 4.]
(b) Show that C6 has two composition series C6 = >02 = (e) and
G1 H3
C6 = G0>H1 =(e) such that
03H1 and 02112
268 Structure theorems of group theory

19 Show that in a soluble group G the length of each composition series

for each proper subgroup H of G is shorter than the length of each composition
series of G (cf. exercise 6.6.10).
20 Show that if one defines a Sylow n-subgroup to be a maximal n-subgroup
(see the remarks preceding 6.2.15) then every finite group has Sylow ir-
subgroups. By considering Sylow {3, 5}-subgroups of A5 show that such Sylow
n-subgroups may not be pairwise isomorphic. Show that A5 has no Hall
{3, 5}-subgroup.

66 Some simple groups

Because of the intimate connection between soluble groups and simple groups
we complement the previous section with just a few remarks on simple groups.
Clearly every (cyclic) group of prime order is simple since, by Lagrange's
Theorem, it can have no subgroups other than itself and the trivial subgroup.
Conversely, if an abelian group is simple then it must be (cyclic) of prime
order (see exercise 1). We establish the existence of infinitely many non-
abelian finite simple groups by proving

Theorem 6.6.1 For each integer n 5 the alternating group is simple.*

The strategy is to prove that

(i) If H (e) and if H then H contains a 3-cycle;
(ii) H contains all 3-cycles;
(iii) is generated by 3-cycles.
It follows immediately that H = whence is simple.

Proof
(i) Let H c and let h be an element of prime order. Write h as a product
of disjoint cycles. For ease of reading we use letters rather than integers in
expressing the elements of
(a) If and if h=(a1a2. . we have
h' = so that H contains a 3-cycle.
(b) If h = (abc)(def)... we have =
(bdecf) H whence a 3-cycle can be found in H by using step (a).
(c) If Ihi=2 then either h=(ab)(cd) or using
an even number of transpositions. In the first case (bde )h h=
(aebdc) H whence a 3-cycle can be found in H by using step (a).
In the second case (bde)h(edb) . h =(afc)(bde) whence a 3-cycle can
be found in H by using step (b).
Thus whatever order h has, it leads us to a 3-cycle in H.

*Galois knew that A5 is simple.

 Some simple groups 269

(ii) Let a = (xyz) be a 3-cycle in 5,,. We know (exercise 5.5.6(u)) that all
n(n—1)(n--2)
3-cycles in Sn are conjugate and, by counting, that there ate
n(n— l)(n—2)
of them. Thus* whence I = 3(n — 3)! Now
=
(a), being a group of permutations, comprises either all even permutations
or half even and half odd permutations (see exercise 2). Since n 5, there
exist letters 1, m distinct from x, y, z. Then (xyz)(lm) is an odd permutation
which clearly lies in (a). Hence (a) comprises permutations half of
which are even and half of which are odd. Thus there are exactly j. 3(n —3)!
permutations in An which commute with (xyz). That is, = 3(n —3)!

and so An: CA,ja )I . Thus conjugating (xyz) only by elements

=
n(n—1)(n—2)
of An already yields all the 3-cycles which exist in Sn, and so
starting with one 3-cycle in H we find, since H <An, that all the 3-cycles of
are in H.
(iii) Every element of An is expressible as a product of an even number of
transpositions. Now (ab)(bc) = (acb) whereas (ab)(cd) = (ab)(bc)(bc)(cd) =
(acb)(bdc) so that every element of An is a product of 3-cycles. Thus An is
generated by the set of all 3-cycles.
This completes the proof of 6.6.1.

Of the groups mentioned in 6.6.1 A5, with =60 elements, is the smallest.
Are there any groups with order less than 60 which are also simple? Certainly
all groups of prime order are. But what about the remainder? No group of
order p n with n > 1 can be simple since every such group is either abelian
(when trivially it is not simple) or has a proper non-trivial centre (see Section
6.2). 17 of the remaining groups have order pq where p, q are distinct primes.
We prove

Theorem 6.6.2 No group G of order pq, where p, q are distinct primes, can
be simple.

Proof Suppose WLOG that p > q. The number of Sylow p-subgroups is of

the form 1+ kp and divides q (see 6.2.12). This is impossible unless k = 0.
Thus there is just one Sylow p-subgroup of G. Consequently it is normal in
G and we are done.
tTheorem 6.2.5.
270 Structure theorems of group theory

Remark One can take this analysis further. It follows easily that G is soluble.
In the case that q tp — 1 we can even prove that '7 is a cyclic group. (See
exercise 3.)

We leave it to the reader to check that all groups of order p2q have a
proper non-trivial normal subgroup and hence are soluble. We go one stage
further.

Theorem 6.6.3 No group'7 of order p2q2 is simple.

Proof Suppose WLOG that p > q. The number of Sylow p-subgroups is of

the form 1 +kp and divides q2 (by 6.2.12). Thus 1 +kp = 1 or q or q2. In the
first two cases the only possibility is k = 0 whence G has a unique, and hence
normal, Sylow p-subgroup of order p2. If 1 + kp = q2 then kp = —1 whence
p lq — 1 or p lq + 1. Since p > q the former is impossible whilst in the latter case
p = 3 and q = 2 is the only possible solution. Thus '71 = 36, and, if'7 is simple,
the number of Sylow 3-subgroups must be 4. These 4 Sylow 3-subgroups are
all conjugate (by 6.2.12) and so, if P is one of them, IG :NG(P)I =4 (6.2.5).
Let us denote briefly by H, and write '7= H u Hg1 u Hg2 u Hg3 as a
disjoint union of the four cosets of H in '7. We now define a homomorphism
o (see 5.10.2(iii)) from '7 into the group of 24 permutations on the set
{H, Hg1, Hg2, Hg3} by setting, for each g E'7,

g
0(H Hg1 Hg2 Hg3
\Hg Hg1g Hg2g Hg3g
That 0 is indeed a homomorphism was noted in exercise 5.10.19. In particular
o is mapping the group'7 of order 36 into a group of order 24. Hence the
kernel of 0 is non-trivial. Also ker 0 c H. Thus '7 has a non-trivial proper
normal subgroup, as required.

We leave it to the reader to prove (exercise 5) that, apart from the prime
cyclic groups and the trivial group (!) there are no simple groups of order less
than 60 and to deduce that all groups of order less than 60 are soluble.
There also exist ([36, p. 292]) infinite classes of (finite) matrix groups over
finite fields which are simple. Some of these are isomorphic to some of the
alternating groups introduced above. On the other hand there do exist pairs
of non-isomorphic simple groups which have the same order. In particular
there is such a pair of order 20 160, one of the two groups being A8. Let us
now show that amongst all groups of order 60 there is, up to isomorphism,
only one simple group.

Theorem 6.6.4 Let'7 be simple and of order 60. Then

Proof By 6.2.12 the number of Sylow 5-subgroups is 1 or 6. Since '7 is

simple there must be 6. Hence we have already 24 non-identity elements of
 Some simple groups 271

G. The number of Sylow 2-subgroups is either 3, 5, or 15. There cannot be

3—or else G would have (cf. the proof of 6.6.3) a subgroup of index 3, and
hence there would be a homomorphism of G into S3 with kernel not equal
to G. Suppose there are 15 Sylow 2-subgroups-—each of order 4. If no two
intersect in more than (e) we pick up 45 non-identity elements of G disjoint
from the 24 we already have. This is impossible. Thus, if G has 15 Sylow
2-subgroups then at least one pair must intersect non-trivially—in a subgroup
of order 2. Calling this pair of subgroups X, Y we see that if G = (X, Y) then
K n Y is central and hence normal in G, which is impossible. Thus (K, Y),
which contains at least 4 elements (exercise 6.2.25), must have index 3 or
5. The former is impossible, as shown above. Thus the latter holds. Note also
that if G has 5 Sylow 2-subgroups we have immediately a subgroup of index 5.
Thus we have reduced the problem to showing that if G is a simple group
of order 60 and with a subgroup of index 5 then G
The by-now usual argument shows that G is (isomorphic to) a subgroup
of S5. Clearly ISs: = 2 SO G is normal in S5. It follows that G nA5 A5
and hence that G nA5 = A5 or (e). This latter is easily seen to be impossible
(exercise 6) whence G =A5 follows.

Having given in 6.6.1 one infinite family of non-abelian finite simple groups
we close with a few words on some of the other non-abelian finite groups
known to be simple. Matrix groups, with their entries coming from finite
fields, yield several infinite families of finite simple groups. One such family
is obtained as follows. Let n be any integer greater than 1 and let F be any
field, finite or infinite. The set of all n x n matrices with entries in F and with
determinant 1 forms a group denoted by (F)—the special linear group of
degree n over F. This group is in general not simple since it might have
non-trivial centre Z, say. (In fact Z comprises all matrices of the form a!
where I is the n x n identity matrix and a EF is such that = 1, because of
the determinant having to be 1.) However, the factor group can, except
when n=2, F—i2 and n=2, F=73, be shown to be simple ([36, p. 294]).
In case F is an infinite field we thus have an example of an infinite simple
group; if F is a finite field we obtain a finite simple group. (These groups
were introduced in Jordan's Traité in 1870.) There are other families of simple
matrix groups which you can find by looking up the terms orthogonal, symplec-
tic and unitary groups. Besides these infinite families there were also known
five groups, the smallest being of order 7920 and the largest of order
244 823 040, discovered by E Mathieu* (in 1861 and 1873) which did not
appear to be members of any infinite family. Dicksont discovered some further
families of finite simple groups around 1905 and then not a single one was
*
Emile Leonard Mathieu (15 May 1835—19 October 1890).
± Leonard Eugene Dickson (22 January 1874—17 January 1954). Author of 18 books including
the monumental History of the Theory of Numbers.
272 Structure theorems of group theory

found until 1955 when Chevalley found yet more families. Variants of these
were soon found and the list of finite simple groups was then presumed, by
some, to be complete (all that was lacking was a proof!) when Janko found
a group of 7 x 7 matrices with entries in the field Z11 and of order 175 560.
Was this a member of another infinite family? At first it was thought that it
might be. Now we know otherwise, (see for example [90])—there are no more
finite simple groups to be found. The proof that besides the various infinite
families of finite simple groups mentioned above there are just 26 sporadic ones,
the largest being of order approximately is due to the efforts of many
mathematicians working in concert. Their combined contributions to the proof
amount to several thousand pages. For an introduction to this proof see [16].
A very readable article, written just before the completion of the classifica-
tion theorem, is in volume 84, number 9 of the American Mathematical
Monthly.
As further study of simple groups is far from simple we conclude our chapter
on 'raw' group theory here and look, instead, at its most famous application.
Exercises
1 Let G be a simple abelian group. Prove that G is finite of prime order
(and hence cyclic).
2 Show that if H is a subgroup of then either (i) all the elements of H
are even permutations or (ii) exactly half are even and the other half are odd
permutations. [Hint: recall the proof of 5.5.1 1.]
3 Show that if p. q are primes, if p >q and if q 2'p —1 then a group G of
order pq has a normal subgroup of order p and one of order q. Show that G
is the direct product of these two subgroups and is, in particular, cyclic. Show
that whether qip — 1 or not, G is soluble of length at most 2.
4 Show that any group G of order p2q has a proper and non-trivial normal
subgroup. Deduce that G is soluble. [Note: you may not 'assume WLOG
that p >q'. Why not?]
5 Show that the only finite simple groups amongst the non-trivial groups
with orders less than 60 are the cyclic groups of prime order. Deduce that
every group with order less than 60 is soluble.
6 Show that S5 has one subgroup of index 2. Is this true of 54; 53; D4?
7 Consider the set P of all even permutations on the set t of positive
integers. Thus P is by definition the set of all those permutations on jF each
of which can be expressed as a product of a (finite) even number of transposi-
tions. In particular, any element of P moves only finitely many elements of
Show that P is a group and that it is an infinite simple group. Show that
P is not a finitely generated gro.up. [Hint: Think of P as the set-theoretic
union of the increasing sequence A1 <A2 <A3 < of alternating groups.
Note that if N then for each n.]
 Some simple groups 273

There do exist infinite simple groups which are finitely generated. The first
was discovered as recently as 1951 by Graham Higman.
8 Using 5.9.6 and exercise 5.9.20 prove that to every finite group G there
exists a finite simple group S such that S contains a subgroup isomorphic to
G. (We say that S contains an isomorphic copy of G or (cf. 3.10.2(v)) that S
embeds G.)
Prove that there is an infinite simple group which contains an isomorphic
copy of every finite group. [Exercise 7 helps.]
9 Let G be a finite group with all its Sylow subgroups abelian. Need G be
soluble? [Hint: which is the smallest you know?]
10 Exhibit a group G with subgroup H such that all composition series for
G have length 1 whereas all composition series for H have length 35. (The
number 35 has no special significance!)
11 Show that if n <60 then any two groups of order n have identical sets
of composition factors. What about the case n = 60?
*
Or, insoluble.
7
A brief excursion into Galois
Theory

7.1 Introduction
At the start of Chapter 5 we indicated Lagrange's approach to the problem of
finding 'algebraic' (or 'radical') formulae—that is, formulae involving (only) the
operations +, —, x, + and for various n—which would yield the zeros* of
any given polynomial. His investigations in the case of degree 5 proved
inconclusive and it was left to Ruffinit (1799, 1813) to indicate and then Abel
(1824) to demonstrate the non-existence of such a formula for the zeros of the
general quintic x5 + ax4 + bx3 + cx2 + dx + e in terms of the (literal) coefficients
a, h, c, d, e. (See 7.9.5.)
Nevertheless, there certainly exist spec jfic polynomials of degree 5 (and
higher) which are irreducible over 0 and for which all the zeros can be expressed
in radical form (see exercises 4.6.6, 4.6.7 and Section 7.6). In particular Gauss
showed, in the Disquisitiones, how, for each positive integer n, the zeros of the
polynomial x"— 1 could be so expressed (see 7.8.2. and 7.10.1). But it fell to
Evariste Galois to discover a criterion—in terms of certain groups of
permutations—by which one can decide (in theory, if not always easily in
practice) whether or not the zeros of a given polynomial with numerical
coefficients are expressible in radical form.
We shall not try to follow Galois' presentation. For one thing the details (as
distinct from the ideas, which are fairly easily described—see Section 7.12) of his
original memoire were not readily comprehensible—a fact Galois partly
ascribed to the novelty and nature of the material—even to mathematicians of
the calibre of Poisson, Lacroix and others. (Those wishing to see a detailed
discussion of Galois' method should consult [83].)
Accordingly, the approach taken here follows a different path from that of
Galois, being essentially that inspired by Dedekind (1894) and Webert (1893,
1895) and coming to full fruition under Emil Artin (1938 and 1942). (It is
interesting to note that the 'old' methods took a long time to die out; an account

* In this chapter we shall talk of zeros (rather than roots) of polynomials to avoid possible confusion
with the various (including roots of unity) which occur.
tPaolo Ruffini (22 September 1765—10 May 1822).
Weber (5 May 1842-17 May 1913).
 Introduction 275

Evariste Galois (25 October 1811—31 May 1832)

Galois was born at Bourg-la-Reine where his father was at one time mayor.
Whilst still at school he wrote his first paper (on continued fractions). In July
1829 Galois' father committed suicide and in August Galois failed in an attempt
to enter the Ecole Polytechnique, because of his inability, apparent even in his
early school life, to submit to the generally required methods of procedure. Later
that year he entered the Ecole Normale Supérieur. In 1829/30 Galois had
published several research papers including a note on the solubility by radicals of
equations of prime degree and one on finite fields. But others, unfortunately, were
'lost' and Galois felt he was being persecuted. He then joined the revolution of
1830 and in December of that year was expelled from the ENS. In 1831, at
Poisson's request Galois submitted a third revised edition of his memoire on the
solubility of equations by radicals to the Academy of Sciences. When he learnt of
its rejection—since Poisson and others couldn't understand it as written—Galois
was in prison. After his release he was persuaded into a duel, some say over a
love affair. He spent the night before the fatal duel writing many letters, one
containing his mathematical ideas. His work, described in modern terms in this
chapter, was published by Liouville in 1846 but not really appreciated for another
25 years.
Galois also contributed to the theory of numbers and sketched out his ideas
relating to elliptic functions and abelian integrals.

[1!] of Lagrange's and Galois' work wholly in the 'old' style was published as
late as 1930.)
However, rather than follow, unerringly, the route described by Artin in
[3],we shall, in the hope of maintaining some sense of discovery and involvement
on the part of the reader, adopt a slightly more 'experimental' approach in
which the results we aim for will be determined more by perceived need—
although, if we see a result or idea which looks interesting or of possible future
use, we shall record it. One consequence of this is that not all results will be
stated as sharply as possible.
276 A brief excursion into Galois Theory

7.2 Radical towers and splitting fields

We have already seen (4.2.20, 4.2.2 1), on being given the polynomial*f(x) in the
polynomial ring F[x], F being a field, how we can construct a larger
fieldt off(x) over F, which contains F and in whichf(x) splits into a
product of linear factors. To say that all the zeros of f(x) are radically
expressible over F is, then, simply to say that S1 can be chosen inside some
suitable radical tower over F according to the following definition.

Definition 7.2.1 The field R is a radical tower over the field F if there is a
sequence F=E0c E1c ... of subfields** of R where, for each i
(1 i s), E1 = - 1(r1), r1 being a zero of an equation of the form — a1 =0,
where a1 e E1 (Clearly there is no loss of generality in assuming that each n, is
a prime. Prove this!) We shall call each a radical extension of E1.

Thus, formally, we arrive at

Definition 7.2.2 The polynomialf(x) E F(x) is soluble by radicals (over F) if its

splitting field (over F) is contained in some radical tower R over F.

In this format, Galois' idea, following that of Lagrange (but from a different
perspective), of (essentially) looking at certain permutations of the zeros off(x)
in (see exercise 11) is, for the most part, reinterpreted in terms of those
automorphisms of which fix F elementwise. (See Section 7.4.) That is, the
new idea is to study the Galois group Gal(Sf/F)—see exercise 5.3.7. (One
advantage of this change is that one is then studying the whole of some group,
namely Gal(S1/F), rather than some obscure (?) subgroup of some group of
permutations. See the Remark on p. 221.) Later we shall refer to Gal(S1/F) as
the Galois group of 1(x) over F.
Earlier (4.2.21) we asserted the uniqueness, up to isomorphism, of As we
are now calling on this uniqueness to define Gal(S1/F) unambiguously, we really
ought to offer a proof of it.
How could one set about such a proof? Since, from 4.2.21, S1 can be
expressed in the form . the ; being the roots off(x) in SO,.,
. .

an induction argument, taking the ; one at a time, suggests itself. So surely, we

must start with

Lemma 7.2.3 Let be splitting fields forf(x) over F and let Yi e

be zeros of the same factor g(x) (off(x)) in F[x]. Then the identity
* It
seems preferable to use the notation f(x), rather than here.
§A more accurate notation would be 51(F), or similar.
t There should be little chance of confusing this notation with that for symmetric groups.
Also called a splitting extension of F (byf(x)).
** We shall use c, ç etc.
to denote subfields and, later, <, etc. to denote subgroups.
condition is forced on us by the desired conclusion. See exercise 13.
 Radical towers and splitting fields 277

map i: F—>F can be extended to an isomorphism K: F(y such that

K(71)=72 and K(t)=t for all teF.

At this point it is best to admit to a little hindsight. It turns out that 7.2.3 is
not general enough for us successfully to apply an induction argument to extend
the map i to an isomorphism between and S2. (Exercise 16 asks you to
determine why not. When you see the reason, the appropriate modification,
7.2.3', immediately suggests itself.)
Recalling that our approach is 'experimental', and that we are not, initially,
attemping to make our lemmas and theorems as 'tight' as possible, it will be no
surprise to find that 7.2.3' contains a large slice of redundant hypothesis. Can
you spot it? In addition yet more hindsight would allow a yet smoother and
cleaner presentation. However, in the interests of not losing sight of our
immediate target, we shall refrain from such generalisation.

Lemma 7.2.3' Let F,, F2 be fields and let 2: F,—÷F2 be an isomorphism

between them. Let and let f2(x)=
2(a0) + 2(a,)x + ... + be the corresponding polynomial in F2[x]. Let
g,(x) be an irreducible factor of f,(x) in F1[x] and g2(x) the corresponding
(irreducible—exercise 17(b)) factor of f2(x) in F2[x]. Finally, suppose, for
i= 1,2, that y, is a zero of g,(x) (and hence off1(x)) in the splitting field 5, off,(x)
over F,.
Then 2 can be extended to an isomorphism p between F1(y1) and F2(y2) such
that 4u(y1)=y2 and j4t)=2(t) for all teF,.

Proof We first note that the isomorphism 2 easily extends*, as implied in the
statement of the lemma, to an obvious isomorphism 2:F,[x]—*F2{x] which, in

turn, gives rise to an isomorphism A: (exercise 17(a)).

Since, for i = 1, 2, each g,(x) is irreducible in F,[x], g,(x) is the minimum

polynomial of y, over F,. (Prove this.) Hence, by 4.3.4, we have for i= 1,2

an isomorphism a,: We now define u to be the composition

cc1 F,[x] A F2[x] c2

F, ('i'

of the isomorphisms aj', A and a2. Noting that a,(x + [g,(x)}) = y, (i = 1,2), we
readily see that p(yj=y2, as claimed.

prove this is a tedious chore which 1 happily delegate to you!

278 A brief excursion into Gab/s Theory

Did you spot the redundancy? It is, of course, that no use was made of the
fact that and are splitting fields. (Can you write down the exact hypotheses
used in the proof of 7.2.3'?)
As an example of 7.2.3' in action, consider the isomorphism A: O(i)—+Q(i)
given by complex conjugation. Take f1(x)=g1(x)=x2—i, so that f2(x)=
g2(x) = x2 + i. Let 71 = = be a primitive 8th root of unity. The zeros of
x2 + / are then a3 and According to 7.2.3' A can be extended to an
isomorphism p of C(s) with itself in which p(a) = (or a7).
To extend the isomorphism p above to one between and we proceed by
induction—on the dimension :F1} of S1 over F1 (see 4.5.6). We shall need to
note that [S1 :F1] is finite (exercise 12). Our aim is then to prove

Theorem 7.2.4Continuing with the notation of 7.2.3', there is an isomorphism

v:51—÷52 such that v(t)=A(t) for all teF1.

Proof If :F1]= 1 then f1(x) factorises completely into linear factors in

F1[x]. Hence, trivially, the same is true forf2(x) in F2[x]. Consequently = F2,
so that we can take v = A in this case.
Now suppose :F m> 1 and that the theorem is true for all splitting
extensions L K for which [L: KJ <m. Since : F1]> 1, f1(x) has at least one
irreducible factor h1(x), say, of degree 2 in F1[x]. Let h2(x){ =2(h1(x))} be the
corresponding (irreducible) polynomial in F2[x].
To make use of the induction hypothesis we note that, for / = 1,2, 5, is a
splitting over F,(7,) and that <[5,: F1]. (See exercise 10.)
Thus by the induction assumption, taking F1(71), F2(72) rather than F1, F2 as
the 'base' fields, we can extend the isomorphism p:F1(71)—*F2(72) of 7.2.3' to an
isomorphism v: S such that v(u) = ,u(u) for each u e F1 (71)—hence, in
particular, for each t e F.
Taking F1 = F2 F and A to be the identity map ,:F1—*F2 we achieve our aim,
namely

Theorem 7.2.5 Let F be a field,f(x)e F[x] and be splitting fields forf(x)

over F. Then under some isomorphism v. Furthermore v can be chosen
so that it acts as the identity map when restricted to F.
Having tidied up that point, let us think how we might find something out
about Gal(Sf/F), given that is itself contained in some radical tower R over
F. Although there are infinitely many different possible choices for R(254,
each such R is constructed in the same general way from a succession of simple-
looking radical extensions starting with F. We might hope that the existence in
each tower R of these simple 'slices' is somehow reflected in the structure of
Gal(Sf/F). All this suggests one line of attack on Gal (Sf/F): first find out how
Gal(R/F) is related to its component parts—namely the various
1)—and then how the structure of Gal(R/F) is inherited, if at all, by
its smaller brother Gal(5f/F). (An immediate, but rather trifling, relationship
between certain Galois groups is given in exercise 21.)
 Radical towers and splitting fields 279

Exercises
1 Use the method of exercise 4.6.6 to try to show that the zeros of — I are
expressible in radical form over 0. What is the main problem you come across?
2 Let Fz=0, E1 =0(12), E2=E1(1(1 +12)). Find the minimum polynomial
over F of 1(1 +12). Is E2 its splitting field?
3 Find a polynomialf(x) in 0(x) having as a zero. Exhibit a
radical tower over 0 which contains the splitting field of f(x).
4 Find a splitting field S1 and a radical tower containing S,- for each of the
following polynomials over 0: (i) x4 + 5x2 + 6; (ii) x4 — Ox2 + 1; (iii) x5 — 1;
1

(iv)x7—l;(v)x9-—l;(vi)x6+l;(vii)x3+x+l.
5 Confirm that is a splitting field for x4 — lOx2 + 1 over 0. Find
(i) a polynomial of degree 6; (ii) a0+a1x+a2x2+a3x3+x4, with
none of the a splitting field.
6 Show that if KcLcM with M a radical tower over K, then M is a radical
tower over L. Do you think L is necessarily a radical tower over K? (See exercise
7.10.3.)

7 Show that each polynomial f(x)= a0+a1x+ ... +a,,f in l1[x] is soluble by
radicals over It Do you think it might also be soluble by radicals oyer
0(a0, a1,.. . , aj? Is x2 + (e + lm)x + soluble by radicals over 0(le, \/7r)?
8 Let F be a field of characteristic #2. Show that each polynomial ax4 + bx2
+ c (a #0) is soluble by radicals over F.
9 (a) Let K be a finite field. Show that K is a splitting extension and a radical
tower over its prime subfield.
(b) Find splitting fields for over /7 and for x3—S over
10 Let M be a field such that Fc_M_cS1. Show that SO,- is a splitting field of
some polynomial over M. [Hint: What polynomial springs to mind?!}
Ii Let 5= FOx1, cx2,. . be a field extension ofF. Show that each automor-
.

phism c of S which is the identity on F is completely determined once is

known for each i(l i n). Show further that each element of Gal(51/F) gives
rise to a permutation on the zeros of 1(x) in Finally show that distinct
elements of Gal(51/F) give rise to distinct permutations.
12 (a) Prove, with F[x], f(x), S. and R as in 7.2.2., that [S1:F],[R:F],
I
Gal(51/F) and Gal(R/F) I are finite. (b) Show that, in fact, [Si: F] n! where
I

n = deg(f(x)).
13 Two of the (complex) zeros of f(x) = x4 — 2x3 + 5x2 — 4x + 6 are isJ2 and
I + (12. Is there an element of Gal(Sf/0) which maps one onto the other?
Answer the same question for the zeros 12 and kJ2 of the polynomial x4 —4.
280 A brief excursion into Galois Theory

14 Let f(x) frCt) be a product of distinct irreducibles in F[x].

Show that under any element of Gal(S1/F), each zero off1(x) can only map to a
zero off1(x).
15 Show that the isomorphism ic of 7.2.3 is the only isomorphism between
F(y1) and F(y2) which is the identity on F and for which ic(y1)=y2.

16 Explain why 7.2.3 has to be modified (to 7.2.3').

17 (a) Establish the isomorphisms 2, A of 7.2.3'. (b) Prove (see 7.2.3') that if
g1(x) is irreducible in F1[x], then g2(x) is irreducible in F2[x].
18 In 7.2.3' put F1 =F2=0W2) and let 2:F1—+F2 be defined by 2(12)
—12. Let S1 be the splitting field over F1. Determine the
splitting field of g2(x) (as in 7.2.3') over F2 and describe the isomorphism p.
19 (a) Does the automorphism a of 0(12) given by a(\/2)= —12 extend to
an automorphism of 0(\/(l +12))? Tn how many ways?
(b) Does the automorphism t of given by = —16 extend to an
automorphism of 0(12, 13)? In how many ways?
20 Let be fields with [L:K]<cc. Show that if a:L—+L is a *l_l
homomorphism which acts as the identity on K, then a maps L onto L. [Hint:
Let 26 L and let g(x) be the minimum polynomial of 2. Let Z be the set of zeros
of g(x) in L. Show that Since a is 1—I, a maps Z onto Z. Or: show that
a basis for L over K maps to a set of elements (of a(L)) which is linearly
independent over K.]
21 Show that if K M are fields, then Gal(M/L) Gal(M/K).

7.3 Examples
As it is always a good idea when beginning a mathematical investigation, let us
start by looking at some particular examples. All are radical extensions or
towers—some, but not all, are splitting extensions.

Example 7.3.1 (i) Let F=0 and n=12, the positive square root of 2. Set
a0,a1 eO}clt Each automorphism a of R automati-
cally 'fixes' F elementwise (i.e. a(t) = t for all t e F—exercise 3.10.8). On the
other hand {a(r)}2 = a(r2) = a(2) 2, 50 that a(r) = ± 12. It follows that
a(a0 + a112) = a(a0) + a(a1)a(12) = a0 + a1( ± ]2).
So, evidently, there are just two automorphisms of 0(]2) (each leaving C
fixed). [See exercise 3.10.1(i).]
(ii) Let F=0 and the real cube root of 2. Set R=0(r)={a0-l-a1r
+a2r2:a0,a1,02 eQ}cR—see 4.3.5.
* Usually called a tnonornorphism. See 7.4.3.
 Examples 281

Again each automorphism a of R automatically fixes 0 elementwise. But

that isn't all that a fixes. Indeed, since r3=2, we find {a(r)}3=a(r3)=a(2)=2.
Since R contains only real numbers, a(r) must be the (unique) real cube root of
2. That is, a(r) = r. But then a fixes all elements in R. In other words, a is the
identity automorphism of R.
On the other hand, roots of unity yield nice results. As an example:
(iii) Let F = 0 and and R =0(r) where r( I) is a (primitive) complex pth root of
unity, p being a prime. Then r = e2"11" (for some k, I k <p) and r has minimum
polynomial +x+l (cf.: 1.9.17(u)). Yet again, if a is an
automorphism of R, then a fixes F elementwise, and so a(r) also satisfies
1
+x"2+ +x+l =0. (Cf. exercise 7.2.14.) Thus for some
j(l jcp), that is, a(r)=rU for some u(l Furthermore each such u
produces a d(fferent automorphism of R (exercise 2).
As 'two-step' examples we offer:
(iv) Let F=0, E2=E1(13)=R. Then
R={a0+a1\/2+a2\/3+a3\/6:a0,a1,a2,a360}.
Here there are at most 4 automorphisms of R. For: each such automorphism a
fixes F; or and a(%J3)=13 or —%J3, and this information
suffices to determine the action of a on each element of R. We leave you
(exercise 6(a)) to check that each of the potential automorphisms is indeed one.
(In due course see 7.4.7.)
(v) Let F=Q, E1=F(w) and E2=E1(r)=R, where w=(—1+i\/3)/2 is a
complex cube root of 1 and r is the real cube root of 2. Here
0(co, r) = {a0 +a1r+ a2r2 + a3w + a4wr
+a5wr2:a0, a1, a2, a3, a4, a5eQ}.
Once again each automorphism a of R fixes F. Also, as in (ii), (a(r))3 = 2, but,
this time, a(r)=r or cur or w2r. Similarly a(w)=w or a?. As in (iv) these
observations determine the action of a on every element of R. It follows that
there are at most 6 automorphisms of R (all fixing 0). Again we leave the
checking (fairly lengthy if done naïvely!) to you that there are exactly 6. (7.4.7,
together with 7.5.1, gives this result quickly.)
Other interesting examples include:
(vi) Let F = E1 = E2 = E1('2) = R. Put y = the positive (real)
eighth root of 2. Then y has minimum polynomial x4 — \/2 over F and so the
only automorphisms of R fixing F are those determined by y—*y and y—÷ —
(vii) Let F = and let R be the field of 4 elements. Since each non-zero element
of R satisfies the equation x3 —1=0 we see that R is both a radical extension and
a splitting extension of F. Denoting the elements of R by 0, 1, and I + allows
us to check easily that the map given by z—÷ I + x determines an automorphism
of R. Thus there are exactly two automorphisms of R (each 'fixing' 4).
(viii) Let F=0, E1 Then has
minimum polynomial x3—(3 + over E1 and x6 — 6x3 +7 (why?) over F. The
zeros of this sextic are + — and four complex (i.e. non-real)
282 A brief excursion into Galois Theory

zeros. One can show (exercise 16) that — 12) As a consequence (as in
(ii)) the only automorphism of E2 is the identity map.

Exercises
1 Let n e V. Find the number of automorphisms of if (i) n is odd;
(ii) n is even.
2 Show, using 7.2.3', that, in 7.3.l(iii), for each u such that I I, there
is an automorphism a of 0(r) for which a(r) =
3 Let w 1 be a cube root of unity. Does 0(w) have an automorphism a for
which a(w) = —
4 Let cx be a non-real cube root of 2. Is the subfield of C: (i) a radical
tower over 0; (ii) a splitting extension of x3 —2 over 0?
5 Describe informally as in 7.3.1: the automorphisms (all fixing 0) of
(i) 0(i'\13); (ii) 0(i, 13); (iii) 0(i, 0(12, and of(v) fixing
0(13); (vi) 0(w,I—3) fixing 0(1—3); (vii) fixing 0(i13).
6 (a) Show that there are exactly 4 automorphisms of 0(12,13) and exactly 6
automorphisms of 0(w, [Hint: Use the methods of 7.2.3' and 7.2.4.] (b)
Informally, how many automorphisms are there of 0(12,13,
7 How many automorphisms are there of w) (i) over 0; (ii) over
0(12); (iii) over 0(w)?
8 Write down in full the permutations of the zeros corresponding to the
automorphisms of R in 7.3.1(v).
9 Indicate informally how many automorphisms there are of the splitting field
over 0 of(i) x2—3; (ii) x3—3; (iii) x4—3; (iv) x5—3?
10 How many elements are there in Gal(S1/0) if f(x) =: (i) x4 + I;
(ii)x4+x2+l;(iii)x3-F2x—l;(iv)(x2-F2x—l)2(x3—2)?
11 If F c E are fields with [E: F] =2 show that, if the characteristic of F 2,
there exists a non-trivial automorphism of E fixing F.
12 Let p be an odd prime. Write the splitting field S1 of x"—2 over 0 as a
sequence of radical extensions. What is
13 Find a radical tower R of least dimension over 0 containingwhere
cc = + How many automorphisms does R have? Letf(x) be the minimum
polynomial of cc over 0. How many elements are there in: (i) Gal(Sf/0(12));
(ii)
14 Let f(x) be the miniriium polynomial of cc = 1(1 + 12) in 0[x]. Find the
orders of Gal(Sf/0) and Gal(51/0(12)).
 Some Galois groups: their orders and fixed fields 283

15 What is the order of Gal(0(cos 72°)/0)?

16 Show that the six zeros, in C, ofx6—6x3+7 are as stated in 7.3.I(viii) and
that — + [Hint: Assume z = — e +
=0(t)=K, say. Then 71/3 =al +bt+c12 where Cube up to get
7=Al+Bt+Ct2 where B
=C=O. Deduce that if a$O then b=c=O, whilst if a=O then b=O or c=O.
Hence 7U3=a(if a$O) or hz or cz2 (if a=O). If 7'13=cz2 then c3=
(3+ — Setting c = (a + (where a, b, d e 7 and (a, b, d) = 1), try
to show that this equality is impossible in

7.4 Some Galois groups: their orders and fixed fields

We can make a number of interesting observations about the above examples.
For one of these it is helpful if we first make (cf. exercise 5.3.7).

Definition 7.4.1 Let L be a field and let E be a (non-empty) set of automor-

phisms of L. The fixed field, *Fix(E), of E is the set {teL:cx(t)=t for

a subfield of L. Exercise 2 invites you to

confirm this.
Let us reconsider 7.3.1.

Example 7.4.2 In (i), R=O(r) is a splitting field off(x)=x2—2 over F=0,

Gal(R/F) j = 2 = [R : F] and Fix(Gal(R/F)) = F;
In (ii), R=0(r) is not a splitting field of f(x)=x3—2 over F=0,
Gal(R/F)I = <[R : F] =3 and Fix(Gal(R/F)) =
1

In (iii), R =0(r) is a splitting field of f(x) = x" — I over F = 0, Gal(R/F) 1 I

=p—l=4R:F] and Fix(Gal(R/F))=F;

In (iv), is a splitting field off(x)=x4— lOx2+ I over F=0,
IGal(R/F)I =4=[R :F} and Fix(GaI(R/F))=F;
In (v), R—0(w,r) is a splitting field of f(x)=x3—2 over F=0,
lGal(R/F)} = 6=JR : F] and Fix(GaI(R/F)) = F; —

In (vi), R = 0e.J2) is not a splitting field of f(x) = x8 —2 over F =

IGa1(R/F)j = 2< [R : F] = 4 and Fix(Gal(R/F)) = E1.
In (vii), R is a splitting field off(x)=x2+x+ over F=7L2, Gal(RIF)I—2
1

=[R:F] and FixçGal(R/FD)=F.

In (viii), R = + is not a splitting field of f(x) = x6 — 6x3 +7 over
F=0, IGal(R/F)I= 1 .c[R:F]=6 and Fix(Gal(R/F))=R.
The assertions in (i), (iii), (iv), (v) and (vii) regarding fixed fields and the orders
of the Galois groups look attractive. Have we happened to choose particularly
nice examples or is there something 'deeper' at work here? No prizes for
* Again, a better notation would be Fix(E, L), or similar.
284 A brief excursion into Galois Theory

guessing the answer! Note, too, that we have given no example of a splitting
extension which is not itself also a radical tower. Is this another universal truth?
We leave the reader to ponder this for a while. (A plausible (?) proof there there
is no such example is: Surely each splitting field S1 F which is contained in
some radical tower F C c C E, is itself radical over F via the sequence
FccE1rtS1c . .. In due course see exercise 7.10.3.)
Let us begin by looking at the apparent bounds placed on the orders of the
Galois group. Since not all the parts of 7.4.2 concern splitting extensions we
shall learn from our experience with 7.2.3' and 7.2.4 and keep our options open
by working as generally as seems prudent. Although we are going to count
automorphisms rather than merely show their existence, it is not too surprising
to find that the same line of argument used earlier will prove useful. Accordingly
we shall only sketch the proof of 7.4.3. We hope the brevity here will help you
more readily see what makes the earlier proof tick.

Lemma 7.4.3 Suppose K5cL1 and K2c_L2 are fields with [L1:K1}<cx.
Further suppose that 0 is an isomorphism from K1 onto K2. Then there are at
most [L1 :K1] ways of extending 0 to a monomorphism* of L1 into L2.

Proof We first check that the induction step will work.

Choose in L1\K1 and let g1(x) e K 1[x] be the minimum polynomial of m.
Suppose that in L2 there exists an element fi whose minimum polynomial g2(x)
(over K2) is 0(g1(x)) (ci the map 2 in 7.2.3' — and note that, this time, there is no
guarantee that such a ,0 exists — exercise 7.2.19). Then, as in 7.2.3' we can find an
isomorphism p:K1(71}—.K2(J3) extending 0 and such that p(yj=/?. Since each
root of g2(x) which lies in L2 can give rise to such an extension of 0 and since, for
each such extension, j4y1) is necessarily a zero ofg2(x), we see that no more than
degg2(x) (—degg1(x)——[K5(y):K1]) such extensions are possible. [As an
obvious, but important, aside let us note here that there will be exactly
[K1(y1):K1] such extensions if g2(x) has degg2(x) distinct zeros in L2.]
We now begin the induction—cf. 7.2.4.
If [L1:K1]=l there is clearly only one 'extension' of 0—itself! If [L5:K1]
=m> 1 choose 6 L1\K1 as above. Then O:K1—'K2 can be extended to a map
p:K1(75)—>K2(/3) in at most [K 1(y1):K1] ways. By induction, replacing K1cL1
and K2c_L2 by K1(y1)cL1 and K2(f3)cL2, each such p can be extended to a
map v:L1—÷L2 in at most [L1 :K1(y1)] ways. It follows immediately that there
are at most [K1(75):K1}[L1:K1(y1)] (=[L1:K1]) ways of extending the isomor-
phism 0:K1—÷K2 to a monomorphism v:L1—*L2. (Incidentally all monomor-
phisms from L1 to L2 must be obtainable by this 'climbing up' method. Can you
see why?)
An immediate and interesting corollary is:

Corollary 7.4.4 Let K c L be fields with [L: K] < oo. Then IGal(L/K)I [L : K].
*
We prefer to use this (standard) terminology for a homomorphism which is 1 1 but not necessarily
onto, to the word 'embedding' introduced in 3.10.2.
 Some Galois groups: their orders and fixed fields 285

Proof Exercise 15.

This result helps to explain the relationships between I Gal(R/F) and [R F] in
I

all sections (splitting field or not) of 7.4.2. Can we go further and show that
equality always holds whenever R is a splitting field over F?
From the aside in the proof of 7.4.3 it is clear that to turn the inequalities in
the statements of 7.4.3 (resp. 7.4.4) into equalities we only need to assume that
the polynomial g2(x}—and its equivalents which are implicitly present at the
induction step—has as many distinct zeros in L2 (resp. L) as its degree. This
requirement of distinctness strongly suggests that we should introduce

Definition 7.4.5 Let F be a field,f(x) e F[x] and be its splitting field. We say
thatf(x) is separable over F if each irreducible factor of f(x) in F[x] has no
repeated zeros in
Simply declaring that each of the g(x) appearing in the proof of 7.4.3 be
separable looks like a distinct 'fiddle', especially if it is a bit difficult to identify
exactly which g(x) are involved. (Perhaps we could just assume that 'all
polynomials arising' are separable.) However, in the case in which we are
presently interested, namely the splitting field of a given polynomial, the g(x)
which arise can be easily identified—as being factors of f(x) which are
irreducible over various fields containing F. And, of course, their separability is
then subsumed by that off(x) itself.
The following result is an immediate consequence of these remarks.
Theorem 7.4.6 Let 2:F1—÷F2 be an isomorphism, let f1(x) and f2(x) corre-
spond under the natural extension 2:F1[x]—.F2[x] of A and let and denote,
as usual, splitting fields forf1(x) andf'2(x) over F1 and F2, respectively. Iff1(x) is
separable over F1 then f2(x) is separable over F2 (exercise 20) and A can be
extended to an isomorphism of onto in exactly [S1 :F1] distinct ways.
The result we are looking for then follows if we specialise 7.4.6 to the case
where F1=F2=F (say),f1(x)=f2(x)=f(x) (say), S1=S2=S1 and take). to be
the identity map z : F—*F. For then we are counting the number of auto-
morphisms of which leave F fixed. That is, we have

Theorem 7.4.7 Let S1 be the splitting field of the separable polynomial f(x)
over the field F. Then I Gal(51/F) = F].
I

7.4.4 and 7.4.7 explain fully the relationships between the IGal(R/F)I and the
[R:F] in 7.4.2. In addition, 7.4.7 confirms, without the need for extensive
calculation, that the p — I (respectively 4, 6) 'maps' c of 7.3.1 (iii) (respectively
(iv), (v)) are indeed automorphisms of R.
What about fixed fields? The various parts of 7.4.2 suggest that, if
I
Gal(S1/F) = [Si: F], then Fix(Gal(S1/F)) = F. That this is true generally is the
content of

Theorem 7.4.8 Let F, f(x) and be as in 7.4.7. Then Fix(Gal(51/F)) = F.

286 A brief excursion into Galois Theory

Proof Suppose Fix(Gal(Sf/F)) = H F. Then Gal(51/F) = Gal(S1/H) -

exercise Consequently, by 7.4.7, [Sf:F]=[Sf:H]. But [S1:F]
21.
=[Sf:H][H:F]>[Sf:H] (see 4.6.1) — a contradiction, since [H:F]>l, by
assumption.

The information given in 7.4.2 (ii), (vi), (viii) regarding the fixed fields of these
non-splitting (over F) extensions is at least partly explained by 7.4.4.

Exercises
1 Let S1 be the splitting field over R of the minimum polynomial over l1 of
e+irr. Write down this polynomial and find Gal(S1/lk).
2 (i) Prove that Fix E (see 7.4.1) is a subfield of L. (ii) Show that Fix E = Fix G,
where G is the subgroup of Aut L generated by E.
3 Let G be a group of automorphisms of the field L. Prove that the prime
subfield P of L is contained in Fix G and that Aut L = Gal(L/P).
4 Find the fixed fields over 0 of: (i) (ii)
(iii) i)).

5 Find the fixed fields of s/3, i)) over (i) 0(i); (ii)
6 How many subfields has F = 0G12, \/3)? How many can act as Fix a for
some automorphism a of F?
7 What are the fixed fields of the automorphisms p, a, t: 0(x)—*0(x) given by:
(i) p(x)= —x; (ii) a(x)=x/2; (iii) r(x)= 1 +x? [Hint: for (ii) consider
f(x)/g(x) = h(x). If degf(x) = deg g(x), look at h(x) — r for suitable r e 0.]
8 Let (1 i n) be subgroups of Gal(L/K). Prove* Fix<uG,> = n(Fix Ge).
Is it necessarily true that = <uFix G1)?
9 Let SO,- be the splitting field of x8 — over 0. What are the fixed fields of:
1

(i) a; (ii) a2; where a is the automorphism determined by = where

10 Let regarded as the group of all permutations on x1, x2, ..., give
rise to automorphisms of 0(x1, x2, . . ., in the obvious way (c.f. 5.5.9). Take a
guess at what Fix(Sn) is.
11 Prove the assertions in 7.4.2 (i), (iii), (iv) regarding their fixed fields.
12 Prove the assertions in 7.4.2 (vi), (viii) that R is not a splitting field and that
the fixed field is larger than F.
*In what follows (Xu Y>((uX1)) etc. will denote the smallest subgroup or subfield (as appro-
priate) which contains X and V (all the X1).
 Separability and normality 287

13 Prove the assertions in 7.4.2 (ii), (iii), (v), (viii) concerning Gal(R/F) and
I

{R:F].
14 Determine whether or not [R :0] = [Qal(R/0) if: (i) R =
I
I+
(ii) (iii)

15 Prove 7.4.4.
16 Confirm, using 7.4.7, the orders of the Galois groups given in 7.4.2 (iii),
(iv), (v).
17 Given fields K L c M do we always have IGal(M/K)I =
Gal(M/L)I . IGal(L/K)I?
18 Show that for each keV the polynomial + +x+ lis separ-
able over 0.
19 Prove that if f(x) e F[x] is separable over F and if F cE, then f(x) is
separable over E.
20 Let be an isomorphism of fields. Show that if f1(x) e F1[x] and
f2(x)—Mf1(x)) e (see 7.2.3') then f1(x) is separable over F1 ifff2(x) is
separable over F2.
21 Show, as asserted in 7.4.8 that, if Fix(Gal(S1/F)) = H, then Gal(Sf/F)
= Gal(S1/H).
22 Let F = 4(t) (cf. exercise 4.5.13) where t is an 'indeterminate'. Let
f(x) = — t e F[xJ and let S1 be a splitting field forf(x) over F. Show that if cr is a
zero off(x) in Si., then (x—cx)"=x"-—t in 51[x].

7.5 Separability and normality

In Section 7.4 we were virtually forced to introduce the idea of separability for
polynomials. Before continuing we ought to determine if this imposes any
serious restrictions upon us, there being little point in developing a theory which
is applicable to only a small number of examples. Fortunately separability is not
a very strong restriction. There is, as they say, 'a lot of it about' (cf. exercises 2—4
in Section 1.11).
We have

Theorem 7.5.1 Let F be any field of characteristic 0. Then each polynomialf(x)

in F[x] is separable over F.

Proof Let g(x) be an irreducible factor in F[x] of f(x) (and so of degree 1)

and let g'(x) be its (usual) derivative. (See exercise 1.11.2.) Since
deg g'(x) <degg(x), the irreducibility of g(x) forces the gcd (g(x), g'(x)) to be 1.
Consequently (cf. exercises 1.9.17, 1.10.3) there exist, in F[x], polynomials s(x),
t(x) such that 1=s(x)g(x)+t(x)g'(x) in F[x]. Regarding this as an equality in
288 A brief excursion into Galois Theory

we obtain I = s(c4g(cx) + t(cx)g'(ct) for each i in If, in particular, x — is a

repeated factor of g(x) in S1[x], so that is a factor of g'(x) in S1[x] (exercise
1.11.2), we obtain 1 =s(jx) 0+ 0 in S1, a manifest contradiction.
A similar result can be obtained if F is a finite field (see exercise 13(c)). Note,
however, that there do exist examples of infinite fields F of prime characteristic
such that not all polynomials in F[x] are separable over F—see exercises 4.5.13.
and 7.4.22.
Although Galois, in a different piece of work, was responsible for the
introduction of fields of finite characteristic he appears to have ignored any
discussion of separability in his work on (zeros of) polynomial equations, by
assuming that the zeros of his polynomials, which he supposed to exist
'somewhere', were always pairwise distinct.
7.5.1 (and exercise 13(c)) tells us that, more often than not, an irreducible
polynomial f(x) e F[x] will factor into a product of distinct linear factors in
S1[x]. Of course, S1{x] depends heavily upon F and f(x): it is specially
constructed to meet the needs of f(x). It is, therefore, mildly surprising that,
given half a chance, an irreducible polynomial in F[x] will also try to split in
S1[x]. Formally this becomes

Theorem 7.5.2 Let Sf be the splitting field of the separable polynomial 1(x)
over F and let g(x) e F[x]. If g(x) is irreducible in F[x] and has at least one zero in
S1\F, then g(x) splits (completely) into a product of distinct linear factors in
S1[x].

Proof Let e S1\F be a zero of g(x). Denote by 132,. , /3, the distinct . .

values of v(J11) as v ranges over the (finitely many—why?) elements of

Gal(S1/F). Setting h(x)= ]JJ
h0 +h1x+ +h,x', say, each v permutes
j= 1

these factors and so maps each to itself. Consequently, by 7.4.8, each

coefficient lies in F; i.e. h(x) e F[x]. If g(x)%h(x) in F[x] the irreducibility of g(x)
leads (as in the proof of 7.5.1) to 1 = s(x)g(x) + t(x)h(x) for suitable
s(x),t(x)eF[x] and hence, as in 7.5.1, again to l==s(131)g(131)+t(/31)h(/31)=0.
Thus g(x) h(x) in F[x] and hence in Sf[x]. Since h(x) splits into distinct linear

factors in 51[x}—by definition!—so too must g(x). (In fact g(x) = h(x). Can you
see why?)
This property surely deserves special attention.* So we make

Definition 7.5.3 Let K c L be fields with [L K] < cc. Suppose that each
irreducible polynomial g(x) e K[x] either has no zeros in L or factorises into a
product of linear factors in L[x]. Then L is said to be a normal extension of K (or
to be normal over K).
* In fact we've already given it very special attention. See exercise 7.5.9 for a pleasant surprise!
 Separability and normality 289

(We shall see in exercise 7.10.1 that this name is not idly chosen!)
We can succinctly summarise where we've got to so far if we extend 7.4.5 to

Definition 7.5.4 Let K c L be fields with [L: K] c cc. If, for a e L, the minimum
polynomial Mjx) of a over K is separable over K, then we say that a is
separable over K. If, for all a e L, a is separable over K, then we say that L is a
separable extension of K (or is separable over K.)
Our summary is then given by

Theorem 7.5.5 Let F,f(x) and S,- be as in 7.5.2. Then S1 is a (finite*) normal
and separable extension of F.

Proof Bring together 7.5.2 and exercise 7.2.12.

In fact, after you have established the result in exercise 9, you will be able to
accept

Theorem 7.5.6 Let K L be fields. Then (i) L is the splitting field of a separable
polynomial over K if (ii) L is a (finite) normal and separable extension of K.

Comments 7.5.7 (a) Further conditions, equivalent to 7.5.6(i), (ii) are noted in
exercise 8.
(b) Although, in 7.5.6, (i) is equivalent to (ii), it is interesting to see (in (ii)) how
we can get away from a lopsided dependence on a particular polynomial f(x)
and its splitting field to a more symmetrical setting in which no polynomial
dominates.
(c) Noticing the word 'separable' in each part of 7.5.6, one is naturally led to ask
if (i) and (ii) remain equivalent if the word 'separable' is removed. Exercise 9
gives the answer.
It turns out that extensions which are finite, normal and separable will play a
major role in what is still to come. Accordingly it is worthwhile making

Definition 7.5.8 If K L are fields satisfying (either of) the conditions of 7.5.6,
we shall describe L as a Galois extension of K. Using this terminology we can
rewrite 7.5.5 as

Theorem 7.5.5' Let 5,- be a splitting field of the separable polynomialf(x) over
the field F. Then S,- is a Galois extension of F.

Exercises
1 Let f(x) e F[x] where the characteristic of F is 0 and degf(x) 1. Let d(x)
denote the gcd of f(x) andf'(x) in F[x] (see exercises 1.10.3 and 1.1 1.2). Show
* i.e. finite-dimensional (see 4.5.6).
290 A brief excursion into Galois Theory

that f(x)/d(x) (trivially separable over F) has the same splitting field over F as
f(x). Test this on exercise 7.3.10 (iv).
2 Let KcL be fields with [L:K]<cc. Show that L is normal over K iffL
contains (an isomorphic copy of) Sf for each irreduciblef(x) e K[x] which has a
zero in L.
3 Let be fields. Show that if M is normal over K then Mis normal
over L. Give an example to show that L need not be normal over K. Show that
L is normal over K if, in addition, a(L) L for every automorphism a of M
which fixes K.

4 Use the radical tower to show that ifKcLcM and if L

is normal over K and M is normal over L, then M need not be normal over K.
5 Let K L with [L: K] < cc. Let F = Fix(Gal(L/K)). Show that L is a normal
extension of F.
6 Let K L M be fields with M separable over K. Show that M is separable
over L and L is separable over K.
7 Artin [3] defines L to be a normal extension of K if [L: K] < cc and if
Fix(Gal(L/K))=K. Show that if L is normal in Artin's sense then it is in our
sense—but not necessarily conversely—unless the separability of L over K is
also assumed.
8 Show that if [L:K]< cc then conditions (i) and (ii) of 7.5.6 are also
equivalent to: (iii) Gal(L/K) = [L: K]; (iv) the fixed field of Gal(L/K) = K.
I I

9 Let K L. Show that L is a normal extension of K if L is a splitting field

over K of some g(x) e K[x]. [Hint: 'Only if': Since, by definition, we have
[L:K]< cc, we also have L=K(cx1 ;) for suitable ; in L. L is the splitting
field of . . MJx) where M21(x) is the minimum polynomial over K of cx1.
'If': Let h(x) be irreducible in K[x] with a zero in L. Let M( L) be a splitting
field for h(x) regarded as an element of L[x]. Let /3 be any zero of h(x) in M\L.
Use 7.2.3' to show K(f3) by an isomorphism which acts identically on K
and sends cx to /3. L is a splitting field for g(x) over KUx) and L(/3) is a splitting
field of g(x) over K(fl). Show L L(/3). Deduce that /3 e L.]
10 Let L1,L2 be normal extensions of K with KcL1,L2CM. Show that
L1nL2 and <L1uL2> (see footnote on page 286) are both normal over K.
Assuming only that L1 is normal over K prove that <L1 uL2) is normal over L2.
11 Let ;) where [L:K].ccc. Show that if the minimum
polynomial of each ; splits in L[x] then L is a normal extension of K.
12 Give an example of a separable extension which is not normal and an
example of a normal extension which is not separable.
In the following assume that each field has characteristic p 0.
 Sublields and subgroups 291

13 (a) Letf(x) e K be irreducible. Show thatf(x) is not separable in K[x] ifff(x)

is of the form a0+a1f+ where the
a finite field. Prove that the derivative off(x) e K[x] is 0 if
f(x) = (g(x))" for some g(x) e K[x}.
(c) Prove that 7.5.1 holds if F is a finite field (of characteristic p).
14 Define the map 4):K—÷K by 4)(a)=o" for each ae K. Show that 4) isa 1—1
homomorphism of K into K—called the Frobenius monomorphism of K. Show
that the fixed field of 4) is the prime subfield P of K and that if K is a finite field
then 4) is an automorphism of K. [Hint: To show that Fix 0—P note that if
c"=c then c is a zero of x"—x, which has at most p zeros in a field.]

7.6 Subfields and subgroups

7.4.7 and 7.4.8 told us something about the order and fixed field of Gal(Sf/F).
To find out more about the structure of this group we are led, naturally, to
investigate its subgroups (and their corresponding fixed fields in S1).
Having just indicated (7.5.7) how we can free ourselves from direct consider-
ation of a specific polynomial and its splitting field let us reinterpret 7.4.7 and
7.4.8—and even extend them somewhat—in these terms.
With very little effort we obtain

Theorem 7.6.1 Let K c L be a Galois extension and suppose that Kc Cc L, C

being a field. Then
(i) IGal(L/C)I =[L:C]; (ii) Fix(Gal(L/C))=C;
(iii) The map given by 'V(C)=Gal(L/C) is a 1—1 inclusion reversing
map from the set F = {C: K c Cc L} of all fields between K and L into the set
of all subgroups of Gal (L/K).

For a pictorial interpretation of cb see 7.6.5 below.

Proof (i) and (ii). These are easy: We know, by 7.5.6, that L is a splitting field
over K of some separable polynomialf(x) e K[x}. It follows trivially that L is a
splitting extension of f(x) over C and that f(x) is separable over C (exercise
7.4.19). Hence (i) and (ii) follow from 7.4.7 and 7.4.8.
(iii) Given KcC1, C2cL and Gal(L/Cj=Gal(L/C2), 7.6.1(u) tells us at once
that C1 = Fix(Gal(L/C1))= Fix(Gal(L/C2))== C2.

Maintaining our 'experimental' approach may lead us to some dead ends, but
surely we must follow up the obvious question suggested by 7.6. l(iii): Is there a
'corresponding' I—I mapping from to F? Put another way:
Given H Gal(L/K) does there exist a field D such that K c Dc L and
Gal(L/D) = H?
In view of 7.6.l(iii) and the finiteness (why?) we need only check whether
or not D onto
292 A brief excursion into Galois Theory

Now there were such a D then, by 7.6. 1(u), we'd have

Fix H = Fix(Gal(L/D)) D. So let's look at Gal(L/Fix H)). Surely this must be
equal to H?
Now, by 7.6. l(iii), 0 sends Fix H to Gal(L/Fix H)), a group all elements of
which surely fix Fix H elementwise! Because H is a group of automorphisms of
L fixing Fix H and Gal(L/Fix H)) is the set of all automorphisms of L fixing
Fix H, there seems no obvious reason why H Gal(L/Fix H)) should not be
possible. Fortunately the next result, called Artin's Lemma but known, in
essence, to Dedekind, does just what we want. [Warning!! The proof is so
beautiful you may feel impelled to tell it to your friends!]

Lemma 7.6.2 Let L be a field and let H be a finite group of automorphisms of

L. Then [L:FixH]IHI.

Proof Let H=={01=e,02,.. so that and suppose that

[L:FixH]>IHI. Then there will exist a set ofn+ 1 elements of
L which are linearly independent (see 4.5.2 and 4.5.6) over Fix H. Consider the
following system of n homogeneous linear equations in n + I unknowns
x1,x2,. . with coefficients in L:

By a well-known theorem of linear algebra, (see, for example, [18]) the system
(*) has a solution (a1, a2,. . , say, where a1 e L(i= 1, 2, . , n + I) and not
. . .

all the are equal to zero.

From amongst all these non-trivial solutions we choose one having the least
number r, say, of non-zero 'components' a1. By reordering the if necessary,
we may assume that this non-trivial solution is (a1, a2,.. . , ar, 0,. , 0) where. .

r 1. First note that r> 1 (since ct1a1 0 is impossible—why?). Next observe

that (l,b2,.. .,br,0,. ..,0), where is also a solution. (Why?) That is,
for each i(1 we have
... +Oi(2r)br0 (**)

Now all the bk belong to Fix H. (Why not? Concentrate on the first equality
not
in (**)).
Suppose
that is one such element. It follows (why?) that there exists
Be H such that Applying U to the equalities (**) we obtain the new
system of equalities in L:
O(01(cx1))O(1) +O(01(z2))O(b2)+ ... + cxr))O(br) O(O)0 (1 i n) (***)

Now, H={01,92,... '0n}={801'°82 5.4.5. So, toeach

i(1 i n), there exists j(i)(l n) such that= Consequently, for
each i(1 i n) the ith equality of(**) and the j(i)th equality of(***) are:
 Subfields and subgroups 293

0
and
+ + =0,
respectively. Subtracting gives:
01(cx2)[b2 — 0(b2)] + ... + 01(2j[b,. — 0(br)] = 0 (1 i n).
That is (0,b2 — 0(b2),.
. , b, — 0(br), 0, . . .0) is a non-zero (why?) solution of the
.

system of equations (*) and yet has fewer non-zero components.

This contradiction to our choice of r shows that the assumption
[L: Fix H]> Hi is untenable. Consequently, [L: Fix H] HI.

As an immediate corollary we get

Corollary 7.6.3 Let L and H be as in 7.6.2. Then [L: Fix H] = PHI and
H = Gal(L/Fix H). In particular L is a (finite) normal and separable (hence
Galois) extension of Fix H.

Proof Trivially H Gal(L/Fix H). But, by 7.4.4, IGal(L/Fix H)j [L: Fix H].
(Why is 7.4.4 applicable?) Consequently PHI iGal(L/FixH)I [L: Fix H] HI,
the last inequality coming from 7.6.2. The final remark follows from exercise
7.5.8.

Specialising to Galois extensions we get

Theorem 7.6.4 Let Kc L be a Galois extension. Then there exists a 1—1 and
onto mapping 'b : given by: 0(C) = Gal(L/C), with D '(H) = Fix H.

Proof If H Gal(L/K) then H = Gal(L/Fix H) (by 7.6.3) = G(Fix H) (see

7.6.1 (iii)).
This result is a part of the so-called Fundamental Theorem of Galois Theory
as normally presented—see 7.10.6.
At this point we offer just two specific examples of the above correspondence
in action. Wanting one which is neither too trivial nor too involved we arrive at
what is probably the most used illustration.

Example 7.6.5 The field S = Q(r, i), where r = is the real positive 4th root of
2 and i2 = — 1, is a splitting field over 0 of the polynomial x4 —2 (cf. 7.4.2(v))
and so is a Galois extension of 0 (7.5.5').

To show you that converting problems concerning subfields of fields into

problems about subgroups of groups is a worthwhile activity, let us ask you now
294 A brief excursion into Galois Theory

to describe all the subfields of O(r, i) which contain 0. When you have found a
few how confident will you be that you then have them all?
To determine G = Gal(S/Q), note that each automorphism a of S will
automatically fix 0 and will map i to i or — i and r to r or ir or — r or — ir (cf.
7.3.1(v)). Thus JGal(S/Q)I But, clearly, [S:Q]=8 and so, by 7.4.7, we must
have IGal(S/0)I =8. We can describe these automorphisms succinctly by their
actions on i and r as follows:
_[a2 (73 (74 (16J (77

r r r 7 ir —r
i
—ir r
—i

Setting b = a2 we find b2 = C3, b3 = ff4, b' = a1 = e, the identity element. Setting

a = (75 we find a2 = e. Further (ab)(r) = a(b(r)) = a(ir) = — ir = (b3a)(r) whilst
(ab)(i) = a(b(i)) = a(i) = — i =(b3a)(i). Consequently ab = Ma. Thus G is seen to
be the dihedral group of order 8, its three subgroups of order 4 and five
subgroups of order 2 corresponding to subfields of 0(r, i) which have dimen-
sions 2 and 4 respectively over 0. This correspondence, determined by the
mapping is best understood via the following diagrams in which fields and
groups in the same 'place' in each diagram correspond under the map b of 7.6.4.
(Notice that we place 'larger' fields above 'smaller' but reverse this principle
for ¶L)

GagS/O)Z
You are asked, in the exercises, to fill in the 7?
How can we confirm this correspondence? Naïvely we take a typical element
a01+a1r+a2r2+a3r3+a4i+a5ir+a6ir2+a,ir3 (a,e0) (1
using the obvious basis {l, r,. ir3} of 0(r, i) over 0. Applying
. . ,
6 and
equating the result to (*), we find that if (*) is not changed by 6 then (*) takes
the special form a0 I + a1(r + ir) + a3(r3 — ir3) + cz6(ir2).
Now you might just recognise that ir2 e 0(r + ir) — since (r + ir)2 = r2(l + i)2 =
2ir2. Further (I +i)3=2(i— l)= —2(1—i). Hence r3—ir3eO(r+ir), too. So, the
fixed field of °6} is 0(r+ir).
Alternatively: If, in general, a e Gal(L/K) has order n and if t e L then a maps
+a" '(t) to itself. (In the above example, 6 has order 2 and so
maps r+ir to itself.) Consequently 0(r-l--ir)cFix{a1, a6}. Hence, if
then the subgroup of Gal(S/0) fixing 0(r+ir) is strictly
bigger than Fix{a1, a6). But you can easily check that no other fixes r+ir.
Hence 0(r+ir)=Fix{a1,a6}.
The next example is even more impressive—at least in what it tells us.
 Subfields and subgroups 295

Example 7.6.6 Let S be the splitting field over 0 of the polynomial —1,
where p is a prime. We know (7.3.l(iii) and exercise 7.3.2) that
Gal(S/O)I=[S:O]=p— 1. Put (=cos(2ir/p)-i-isin(2n/p) and set a=(+ l/C=
2 cos(2ir/p). Then a e P and 0 c 0(a)c 0(0=5. Suppose we now choose pto be
of the form 2" + 1. (Then, necessarily, k = for some integer n—exercise 4.6.5.)
Consequently [5:0] = 2k and so [0(a): 01=2' for some integer i (why?). Since
Gal(S/0) is abelian [in fact it's cyclic! Try to prove it abelian but, if stuck look at
exercise 7.8.6] we deduce from exercise 6.4.7 that Gal(S/Q) has a decreasing
sequence of subgroups

each of index 2 in its predecessor. Hence, by 7.6.4, we see that there is a

corresponding sequence
cQ,=Q(a) with foralli.
Hence, by exercise 4.6.3, we can deduce that the point (a, 0) is obtainable by
straightedge and compass construction. Thus we have proved the 'if' part of
Gauss's assertion concerning constructible polygons. (See Section 4.6.)

Exercises
I Suppose that L2M2K and G H <e) where G=Gal(L/K). Write G(M)
for Gal(L/M) and F(H) for the subfield Fix(H) of L. Prove that G(F(H)) H,
F(G(M)) 2 M and (omitting brackets) that FGF(H) = F(H), GFG(M) = M.
2 Suppose K L with [L: K] c cc and let Gal(L/K) have fixed field F(2 K).
Show that there is a 1—1 correspondence between the set F of subfields of L
containing F and the set cc of subgroups of Gal(L/K). Deduce that the mapping
'V of 7.6.1 is always onto cc.
3 (Dedekind's Independence Theorem: DIT) Let H and K be fields and let
0,. be distinct monomorphisms* of H into K. Prove that . ., qS,. are
linearly independent over K—i.e. (cf. 4.5.2(u)): If a1,.. . , a,e K are such that
+a4,(h)=0,, for all heH, then all the a,=OK.
[Hint: Suppose false and let . . , 0,, (n r) be a smallest non-empty set such
that a1Ø1 + is the zero map H—÷K with no a,=0. Since q5,(l,,)= 1K'
n 2. Since 0,, there is some he H such that 0,,(h). For all g e H,
a1q51(hg)+ ... + 0.

But also
+ a2qS,,(h)Ø2(g) + ... + =0
(why?). Subtract this from the previous equality to get a1(Ø1(h)— Ø,,(h))Ø1(g)
+ +Th=0. But 0i(h)5e&(h).]
*
See 7.4.3.
296 A brief excursion into Galois Theoiy

4 (1St application of DIT) Let k1,. .. , be non-zero elements of a field K and

let t1,.. , be distinct automorphisms of K. Then there exists k e K such that
+kfltfl(k)#OK.
Check this theorem with the four automorphisms: Identity,
t(12 2,%J3—>---\/3) and at of given
that k2t—3—l\/2+2\/3+0\/6,
—213— 116 and k4 =3 + 112 + 213+016. [This result is helpful in giving an
alternative proof of part of 7.10.4.]

5 (2nd application of DIT) Prove directly, without using 7.6.3, that if G is a

group of automorphisms of the field K and if F0 denotes Fix(G), then
finite

GI[K:F0].[Hint:Supposem=[K:F0]<IGI=nandthatx1 ,xmisabasis
for K over F0. Let G={r1,.. Note that there are not all
equal to such that
(/=l,...,m).
Let aeK. Then a=oc1x1+ for suitable ;eF0. Multiply the above
equations by cxi,. . , a,,, respectively in order. Using
. = (why
equality?) add the m resulting equalities to get t1(a)k1 + ... =0,
contradicting exercise 3.]
6 Let K be a Galois extension of F with Galois group G. Given fields E1, E2
such that F E1, E2 K with = 1r11 G(i = 1,2), show that
Gal(K/<E1uE2>)=H1nH2, Gal(K/E1nE2)= <H1uH2).
7 Suppose G= ., is a (finite) group of automorphisms of the field L
with fixed field K. Suppose that ç& e Aut(L) and that i/i fixes K. Show that çfr e G.
8 Suppose K L M are fields with [L: K] < cc. Show that
[Gal(M/K): Gal(M/L)] [L : K] and that equality holds if M is a Galois exten-
sion of K. [Hint: Show that elements in the same coset of Gal(M/L) in
Gal(M/K) give rise to the same mapping of L into M. Now use 7.4.3 with
K1=K2=K etc.]
9 Determine the subgroups and subfields omitted from the diagram of 7.6.5.
10 Determine Gal(S1/Q) where: (i) f(x)=x3—2; (ii) f(x)=x5—2;
(iii) f(x)=x6— 2. Draw the subgroup and subfield 'lattices' as in 7.6.5.
11 Determine Gal(S1/cP) where: (i) f(x) = —3; (ii) f(x) = —4. Draw the
subgroup and subfield lattices.
12 Find Gal(S1/Q) iff(x)=(x2+2x— l)4(x3—2). (Cf. exercise 7.3.10)
13 Determine Gal(S1/F) wheref(x)=x4+2 and F is 74.
14 Draw the subfield lattice for the splitting field of x3 + x + 1 of exercise
7.2.4(vii). (Its Galois group is S3.)

15 How many subfields has 0(12,13, \/5) got?

 The groups GaI(R/F) and Gal (S1/F) 297

16 Let f(x)==x8— O[x]. Check that GaI(S1/Q) is C2 x C2. Show 0(i),

1

Qe.J2) and 0(i, 12) are subfields of Sf. What are the others?
17 Let L be a Galois extension of K and suppose that Gal(L/K) is cyclic. Show
that for each dI[L: K] there exists just one field T such that K T c L and
[L: TI = d.

7.7 The groups GaI(R/F) and GaI(Sf/F)

Whilst our investigations so far have followed a fairly natural path and
produced at least one very powerful looking by-product, (7.6.4), it is surely time
to return to our motivating problem, the solubility of equations by radicals.
To keep our main aim firmly in view, we work, once again, in terms of
splitting fields of separable polynomials rather than talk of Galois extensions.
We consider, first, the groups of the title where R, are as in 7.2.2. The
following result is both readily spotted—and proved!

Lemma 7.7.1 Suppose that F R are as in 7.2.2 and suppose that

OeGal(R/F). Then O(S4=S1.

Proof LetseS1=F(;,z2,. . ,;)where ;(l

. the zeros off(x)in S1.
Since t9 maps F identically we see that, for each oct, O(;) is a zero of f(x), i.e.
O(;) = 2j(i) for some j(i). O(Sf) is then immediate. [What about equality?]

Thus, in order to relate Gal(R/F) to Gal(S1/F) it seems reasonable to

associate, with 0€ Gal(R/F), the restriction O (say) of U to Gal(S1/F). (O is
certainly meaningful—by 7.7.1.) This suggests that there should be a (natural)
homomorphism t, say, from GaI(R/F) to GaI(S1/F). Let's confirm this.

Lemma 7.7.2 Let be as in 7.2.2. The map ftGal(R/F)—*Gal(S1/F)

given by Q(O) = O is a homomorphism.

Proof Let 4, e Gal(R/F) so that

i/i 7e Gal(S1/F). We find, for all s e
= (4n/i)(s) = = q5(t7(s)) = = [You should try to
answer the question 'why?' at each equality—especially the one indicated!] Thus
the map Q defined by Q(O)=? is a homomorphism.

7.7.2 may look nice but, as it stands, it is not all that much use,-for there seems
little likelihood of being able to determine much about the structure of
Gal(S1/F) from that of Gal(R/F) unless Q above maps Gal(R/F) onto
Gal(S1/F). (In fact Gal(R/F) = <e) Gal(51 /F) is possible—exercise 4.)
Let us see what it might cost us (in terms of hypotheses) to ensure that L maps
Gal(R/F) onto Gal(S1/F). One criterion is immediate: If F) is the fixed field
of Gal(R/F), then each 0 is required to fix all of V. But, by 7.4.8,
Fix(Gal(S1/F)) = F. Consequently we are forced to assume that VnS1 = F.
298 A brief excursion into Galois Theory

Is V F? More than likely! (See exercises 3, 4.) But let us press on and see
what turns up. Certainly Gal(R/F)=Gal(R/V). Also R is a normal extension of
V (by 7.6.3). If we denote by T the smallest subfield of R containing V and S1,
we see that T is a splitting extension off(x) over V—so that V is also the fixed
field of Gal(T/V). As we shall see below (cf. 7.7.3) this is enough to ensure that
the natural map (cf. 7.7.2) from Gal(R/V) (=Gal(R/F)) to Gal(T/V) is onto.
This looks promising! Can we now map from Gal(T/V) onto Gal(S1/F)?
Since there is a homomorphism, as in 7.7.2, from Gal(T/V) into
Gal(Sf IF). Is it onto? Sadly ... in general, no! [Since T = V(ct1, ;), each
element of Gal(T/V) determines a permutation on the set of which, itself,
determines a corresponding element of Gal(S1/F). Unfortunately there is no
guarantee that these permutations on the ; yield every element of Gal(Sf/F).]
So, what can we do? Since our choice of R is unlimited (whereas we are stuck
with F,f(x) and S1) perhaps we can choose a new R for which V = F? It may be
a long shot, but it's worth a try! First, to save time if we are wrong, let us check
if indeed Q (of 7.7.2) will be onto if Fix(Gal(R/F)) is assumed to be no bigger
than F.
In fact we obtain

Theorem 7.7.3 Assume* that Fix(Gal(R/F)) = F (so that R is a Galois exten-

sion of F, by 7.6.3). Then the map Q is a homomorphism from Gal(R/F) onto
Gal(S1/F). Further ker Q = Gal(R/S4. Consequently Gal(R/S1) < Gal(R/F)
and Gal(Sf/F) Gal(R/F)/Gal(R/S1).

Proof Let i/i e Gal(S1/F). Since R is a Galois extension of F (by 7.6.3 with
L=R and H=Gal(R/F)), R is, by 7.5.6, a splitting field of a suitable separable
polynomial h(x), say, over F and, hence, of h(x) over Sf (trivially).
On taking F1=F2=S1 and 2=i/i, 7.2.4 tells us that we may extend the
automorphism i/i of S1 to an automorphism v of R. Since cli acts identically on F
so does v. That is, v e Gal(R/F)—and, by construction, = i/i. Next:
ker 11 = Gal(R/S1). For, if v e Gal(R/F), then v e ker Q if is the identity map on
5,-; that is if veGal(R/S/).
Finally: the isomorphism then follows from 5.11.6.

So, our immediate task is to show that the given radical tower R can be
replaced, when necessary, by a radical tower which is also normal over F. Can we
take a hint from 7.3.1(ü) and (v) where adjoining a cube root of unity changed
the rather sterile into the splitting field (and hence normal extension)
of x3—2 over 0?
Perhaps it is sufficient to adjoin to R, a primitive nth root of unity, where
(See 7.2.1) Unfortunately this won't do. As exercise 8 shows, there is

* In fact R can he any Galois extension of F which contains s1•

 The groups Ga/(R/F) and GaI(S,/F) 299

still no guarantee that the new field will be normal (i.e. a splitting field) over F.
Of course normality (perhaps without radicality) is easily obtained—given
exercise 7.5.9. (See also 7.5.7(c).)
To get the desired type of extension note first that (in 7.2.1) R
= F(r1, r2, . . . For each i let Mr(X) denote the minimum polynomial of
over F (why is there such a polynomial?) and put m(x) = fi Let
Sm = R(u1, u2,. . . ,u,) be the splitting field of m(x) over R. It follows (exercise 10)
that is a splitting field—hence a normal extension (see 7.5.7(c))—for tn(x)
over F. But there's an obvious question. Have we just destroyed the required
radical property? In fact we have

Theorem 7.7.4 5m is a radical tower over F.

Proof We have F_cSf_CF(rl,r2, . . . 7.2.3' and 7.2.4 show that if u is

any zero (in Sm) of the minimum polynomial Mr1(X) of over F then an element
of Gal(Sm/F) exists which maps r1 to u.
Consequently, Sm=F(ii(ri),. . q1(rj, . . , . , (r1),. . . . . . ,

where tj1 = e,. are the elements of Gal(Sm/F). But this implies that
. . is a
radical tower over F. For, Sm contains the increasing sequence
FcF1,1cF12c ... _cF1 (*)

of fields where, . . so that* Fk,

= Fk,, where ndr1) = e Fk, —

(Note that the n1 arising here are precisely those appearing in the originally
given tower in 7.2.1.)

Exercises
I Establish the equality in 7.7.1.
2 Show that there are fields QcKcLcC with Gal(L/Q)=Gal(K/O)=<e).
[Hint: n=?]
3 Let K = to), L = to). Find Gal(K/Q) and Gal(L/Q) and
show that Fix(Gal(L/Q)) D Fix(Gal(K/Q)).
4 Noting that 0 c 0(\/2) c 0(/(3 + is a radical tower containing
the splitting field over 0 of x2 —2, compare with
(See exercise 7.3.16.)
5 Check, in the discussion following 7.7.2, that T is indeed a splitting
extension off(x) over V and that Fix(Gal(T/V)) = V.

* Read F10 as F1
300 A brief excursion into Galois Theory

6 Find three radical towers over 0, with distinct Galois groups, all containing
the splitting field Q(\/2).
7 Prove that if K c L M, if M is normal over K and if a is an isomorphism of
L into Al which fixes K, then a is the restriction i/i of some automorphism of M
which also fixes K.
8 Let 0 c c Q(\/(l where de 1, be given. Show that if d= 2
then the extension + \/d'), i) is a normal extension of 0, whereas if d= 3 it
is not. [Note: we are adjoining i, a primitive 4th root of unity, since the given
radical tower comprises two quadratic extensions.]
9 Given that where L is normal over K, show that Gal(M/L) is a
normal subgroup of Gal(M/K).
10 Confirm the assertion in the paragraph preceeding 7.7.4 that the splitting
field of m(x) over R is actually a splitting field for m(x) over F.

78 The groups GaI(F1/F11)

Recall that our main aim is to express the 'big' group (which is now
Gal(Sm/F))—and hence the group of real interest, Gal(Sf/F)—in terms of the
(we hope) 'simpler' groups* Gal (F1 3. Unfortunately, as 7.3.1 (ii) shows,
without roots of unity present, the groups 1) may be too simple,
namely they may be of order 1!
There seems nothing for itt, therefore, but to adjoin to Sm the roots of I for
all after all. Does this destroy the radicality and normality we've worked
hard to get?
Fortunately not! Indeed if S = Sm(() is obtained from by adjoining a
primitive nth root ( of unity (where n is the 1cm of the n1 occurring in 7.2.1—see
the remark at the end of Section 7.7) we see that S is certainly a radical tower
over and a splitting field over F of the polynomial m(x)(x"— 1). (Prove this!)
There's one last rearrangement required which it is easy to spot as being helpful.
In order that, when considering 3, we already have an root of
1 available in we place at the 'beginning' rather than at the 'end' of the
radical chain. That is, we consider the sequence—-it is (still) a radical tower over
F—
FcF(QcF1, ... (7.8.1)
Note that, in 7.8.1, is a splitting (and radical) extension of F. Further
each is a splitting (and radical) extension of 1(c) since
= where e and contains a primitive
1 root (
of unity.
What, then, are the Galois groups of these extensions? To deal with the
extension ofF we prove (cf. 7.3.l(iii)):
* See footnote on p 299.
tOut see exercise 1!
 The groups GaI(ñ1/ñ1_ 301

Theorem 7.8.2 Lett L be a splitting field of xn — I over a subfield K. Then

Gal(L/K) is abelian.

Proof Let 1, s,. . , -' be the zeros off — 1 in L, so that e is a primitive nth
.

root of 1. Then L = K(s). Let be the minimum polynomial of s over K.

Then each zero of is one of the Now, for general n, not all the will be
zeros of M6(x)—exercise 2. Nevertheless, if 4' e Gal(L/K), then 0(s) = & for
somej (coprime to n—why?). [Note that 4)(s") is then determined for all k—viz.
= 5kj1 Suppose that Ok' e Gal(L/K) are given by Ok(s) = 5k and Ø,(s) =
Then = 51k = = Ø,(5k) This is enough to prove that
= =
Gal(L/K) is abelian, as claimed. (Why?)
For dealing with all the other extensions, we obviously need (c.f. 7.3.1(v)):

Theorem 7.8.3 Let L be a splitting field off — a over a subfield K where a e K

and where we assume that K contains n distinct nth roots of unity. Then
Gal(L/K) is abelian.

Proof (Briefly) If s is a primitive nth root of I and if is one of the zeros of

— a in L, then the zeros of x" — a in L are precisely the elements
, sn - In particular, L = K(cc). (Why?) The minimum polynomial of
has some of these as its zeros. Thus, again, if e Gal(L/K) is such that
&x, then OJ4)k(ct) = = = OkOj@). Hence Gal(L/K) is abelian.

Exercises
1 Let F=E0cE1c ... CE5 where and where
being some prime in /. Let M(x) = HM,Ax), where M,1(x) is the minimum
polynomial of over F. Let SM be the splitting field of M(x) over E5. Prove that,
for each 5M contains all complex p1th roots of unity. [Hint: What are the
zeros in 5M of the minimum polynomial Nri(X) of over E1_ i?]
2 Find the minimum polynomial (notation Gjx)) for the primitive nth roots
of unity for each n such that 1 n 12. (The polynomials you are dealing with
are called cyclotomic polynomials. C.f. [15], Theorem 15.3.) [Hint:
=(x4—l)/(x2—l)=x2-Fl. Infactf—1=flGd(x).]
din

3 Find Gal(51/Q) where f(x) = — 1. Determine all the subgroups of

Gal(51/Q) and all fields between and 0.
4 Let be the splitting field of x"— I over 0. Prove that Gal(51/0) is
isomorphic to the multiplicative group 4, say, of invertible elements of 4.
5 Find the Galois group over 0 of: (i) x4 + 1; (ii) x5 + 1.

tTo avoid problems concerning separability we need only assume characteristic KJn. (See [15].) For
simplicity we here assume characteristic K = 0.
302 A brief excursion into Galois Theory

6 Prove that the groups in 7.8.3 are cyclic for all n and that those in 7.8.2 arc
cyclic if n is primc. [Ci exercise 4.]
7 Letf(x) = x" — a be irreducible in Q[x}, p being a prime. Show that the Galois
group of f—a over Q, that is, Gal(Sf/O), is isomorphic to the group of
transformations on 1,, given by y—+iy +j(i,j i 0).
8 Show that the Galois group of f —2 over 0 is not abelian if n> 2.
9 Let F be any field of characteristic 0. Show that the Galois group of
1(x) = x'1 — a over F has an abelian normal subgroup with abelian factor group.
10 Let with (m,n)=l. Put f(x)=xm—l, g(x)=f—l and h(x)
= x

7.9 A necessary condition for the solubility of a

polynomial equation by radicals
We now come to the first half of Galois' result concerning the existence of
radically expressible zeros of polynomials. We shall suppose, as Galois did
implicitly, that all fields are of characteristic 0 so that all polynomials considered
are automatically separable (by 7.5.1).
Suppose, then, that F, 1(x) and R are as given in 7.2.2 and that the normal
radical tower S over F of 7.8.1 has been built around R. For convenience let us
relabel the terms of 7.8.1 as
(7.9.1)
(From now on these new replace those given in 7.2.1.) Since S is a Galois
extension of F we know (by 7.6.4) that there exists a 1—I correspondence
between the subgroups of Gal(S/F) and the subfields of S which contain F. In
particular the (increasing) sequence of fields in (7.9.1) corresponds to a
(decreasing) sequence of subgroups
Gal(S/F) Gal(S/E1) Gal(S/E2) ... Gal(S/S)= <e>. (7.9.2)
Since S is a splitting extension of F, S is also a splitting extension of each field
V for which In particular, it is a splitting extension of each
—1(1 i w). Furthermore, as already observed, each is a splitting exten-
sion of its predecessor E1_1. Consequently, as in the proof of 7.7.3 (see also
exercise 7.10.1!) each subgroup in 7.9.2 is normal in its predecessor. Since each
of the corresponding factor groups is, by 7.8.2 and 7.8.3, abelian, we deduce
that Gal(S/F) is itself soluble. (See exercise 6.5.9.)
Since we also have F c c 5, where is a splitting extension of F, the same
argument (which we have already given in 7.7.3!) shows that Gal(51/F) is a
homomorphic image of Gal(S/F). Thus, from exercise 6.5.7, we obtain our
criterion.

Theorem 7.9.3 Let F,J(x) and be as usual with characteristic F = 0. If is

contained in some radical tower over F, then Gal(51/F) is a soluble group.
 A necessary condition for the solubilityofapolyn omial equation by radicals 303

Comment 7.9.4 (See exercise 3) This part of Galois' result (7.10.5) is already
sufficient to show the existence of polynomials of degree 5 whose zeros are not
(all*) expressible in terms of radicals. We shall leave examples of specific
polynomials until Section 7.11, and content ourselves with merely stating here
(but see exercise 2) the famous result of Abel and Ruffini, namely:

Niels Henrik Abel (5 August 1802-6 April 1829)

Niels Abel, born in Findö, Norway, was the second child of a poor
but gifted churchman. Abel's talents, at first dormant, blossomed
under the care of a schoolteacher, Berndt Michael Holmbäe, seven
years Abel's senior. Together they studied the works of Euler,
Lagrange and Laplace. Whilst still at school Abel believed he had
discovered how to solve the general quintic equation by radicals. He
himself later found the error and in 1824 he published, at his own
expense, his proof of impossibility.
After studying at the University of Christiania (Oslo) Abel travelled
to Paris where he was ignored by the leading mathematicians, and
then to Berlin where he was befriended by Crelle who published
Abel's papers in his newly founded journal. Crelle tried to get Abel a
professorship at the University of Berlin but because of illness and
poverty Abel had to return to Christiania. A few days after he died of
tuberculosis a letter from Crelle arrived telling Abel of his
appointment to a position in Berlin.
Besides his work on equations Abel did fundamental work on the
theory of convergence of power series. He is also regarded as a
cofounder, along with Jacobi, of the theory of elliptic functions
(although Gauss, in work unpublished in his lifetime, anticipated
many of their results).

Theorem 7.9.5 + ... -l-a1x-*a0eF[x], where F is a

field of characteristic 0 and the a1 are 'general' coefficients. [In formal terms they

*See exercise 7.10.8.

304 A brief excursion into Galois Theory

are elements of the field F(a0, a1 an) of fractions of the polynomial ring
F[a0, a1,.. , .in the distinct letters a0, a1 an.]
For each integer n 5 the equation f(x) = 0 is not soluble by radicals over
F(a0,. . ,aj. That is, there exists no universal 'formula' in terms of
. —' ±,
the various and the coefficients an, an_ 1' . ,a0 which will always yield the
solutions off(x) = 0 on substituting specific numbers for the a1.

Note 7.9.6 Despite the content of 7.9.5 and its interesting corollary, exercise
7.9.5, it is still possible that each individual equation f(x) = 0 over 0 might
succumb to a solution by radicals via a 'trick' peculiar to itself. However, in
Section 7.11, we'll see how to construct any number of specific quintic
polynomials in 0[x] which are not soluble by radicals over 0. Of course the
existence of even one such is sufficient to prove 7.9.5 (at least for n = 5), but you
are invited to prove 7.9.5 directly in exercise 2.

Exercises
1 Let F be any field and let L = F(x1 xj, where the x1 are 'indeterminates'.
Let Sn' regarded as the group of all permutations on x1,.. . , give rise to
automorphisms of L in the obvious way (cf. 5.5.9). Let be the jth
elementary symmetrical polynomial on the x1 (see Section 4.7). Show that
K =F(s1,. . ,s,j is the fixed field of as follows:
.

(i) Show that L is a splitting field for the polynomial

yn_syn_l+
and deduce that L is a Galois extension of K.
(ii) Use exercise 7.2.12(b) to deduce that [L:K]n!
(iii) n! = l5nH [L : Fix Sn] [L: K]
Deduce the following extension of 4.7.1: Every symmetric rational function in
n indeterminates with coefficients in a field F is expressible as a rational
function, with coefficients in F, of the j n).

2 Let F be any field. Use exercise I to prove the insolubility by radicals of the
'general' quintic equation
f(x)=a0+a1x+a2x2+a3x3+a4x4+x5 eF(a0, a1,.. .a4)[x],
the a1 being indeterminates over F.
3 The first sentence of 7.9.4 is so vague as to be almost worthless. Why?
4 Deduce from exercise 1 that, given any finite group G, there is a pair F c E of
fields such that GaI(E/F) G.
S Find specific real numbers a0, a1 a4 such that the polynomial
x5 +a4x +a3x +a2x +a1x+a0
 A sufficient condition for the solubility of a polynomial equation by radicals 305

is not soluble by radicals over the subfield Q(a0,. . . , a4) of It

[This is probably rather hard. Let's just say that the uncountability of R
(exercise 2.6.15) might help! A starting question might be: Can you find a real
number which is not a zero of any polynomial in Q(ir)[x]? Does this generalise?]

7.10 A sufficient condition for the solubility of a

polynomial equation by radicals
One naturally asks if the necessary condition stated in 7.9.3 is also sufficient. It
would certainly be very satisfying to be able to prove:

Theorem 7.10.1 Let S1 be the splitting field of f(x) e F[x], where F has
characteristic 0. If Gal(S1/F) is a soluble group, then there exists a radical tower
over F containing 5,,..

Taking a hint from Section 7.9, we first adjoin to some roots of 1 by taking
the splitting field over 5,,. of the polynomial x" — for some n. But which n shall
1

we choose? Let us postpone a decision for the moment. In fact, in our first
result, the value of n seems to be immaterial!

Lemma 7.10.2 Let F,f(x), be as indicated and (be a primitive nth root of
unity. If we let F(() be a splitting field of x" — I over F and T be a splitting field
of f(x) over then and is (isomorphic to) a
subgroup of Gal(51/F).

Proof By definition T= , ;) = F((, ocr,. ..,;) where ;,

; are the zeros off(x) in T. Hence T= F(;, SO,- being
the splitting field of f(x) over F. If Be Gal(51(O/F(O), then B fixes F and
permutes the ; amongst themselves (exercise 7.2.11). Hence each B maps 5,,. to
itself and so gives rise to an element O of Gal(51/F). The map is then a
homomorphism of Gal(51(O/F(43) into Gal(51/F) (as in 7.7.2). Finally, if O fixes
(F and) all the;, then B fixes F, the ; and (. Hence B=e. That is, the map
is I—I (and into).
We deduce immediately that: If Gal(51/F) is a soluble group then so is
Gal(51(43/F((fl. Consequently there is a chain of subgroups
(*)
Gal(51(Q/F(O) = G0 > G1> > = <e>,

each being normal in its predecessor with 1/Gd being cyclic of prime
order. (See 6.5.8.)
We know from 7.6.4 that (*) gives rise to a corresponding (increasing)
sequence of subfields
= F0 c F1c F2 c
306 A brief excursion into Galois Theory

so that G1=Gal(S1(O/FJ, Fix(GJ=F1 and with each [F1:F1_ prime. Does the
normality of each G1 in G1_ and the primality of each G1.
1
imply anything
special about the field extension F1DF1_ We shall see that it does!
Let us first see what the relation of normality implies. Since is a splitting
field (forf(x)) over F(Q, it is also a splitting field over (F1 and) F1_ for each 1

i( I in). There then follows

Lemma 7.10.3 For each i(l and for each 4) e we have:

(i) 4)(F1) = F1; (ii) F1 is a splitting extension of F1—

Proof (i) Let 0€ G1 and 4) e G1 - 1 = Gal(S1g)/F1 - 1). Then,

since G1<G1_ 1' we have G1. Hence, for v1e F1, '044(vJ=v1. Conse-
quently O4)(v1) = Ø(vJ—for all 0€ G1. Hence 0(v1) e Fix(G1) F1 by 7.6.4. Thus
Ø(F1)cF1. To prove equality, note that [F1:F1_1]=[Ø(F1):4)(F1_1)]
= [0(F1): F1 - (why?) c [F1: F1 - That 0(F1) =F1 follows immediately.
(ii) I
Gal(F1/F1 — 1) = I Gal(S1g)/F1 1)111 Gal(S1g)/F1)
— I
(exercise 7.6.8)
=[S1(fl:F1_ l]/[Sf(():F1] =[F1:F1_ by 4.6.1. The result claimed then follows
immediately from exercise 7.5.8.
Thus 'all' we need to do is to show, for each i(1 n), that F1
i F1 -
where & F1_1 for some prime p. Surely we won't be so lucky? We are!!

Theorem 7.10.4 Letp be a prime and let K be any field containing thep distinct
pth roots of 1. Let L be a splitting field over K of a separable polynomial and
suppose that [L:K]=p. Then L=K(ó) where &€ K (so that L is the splitting
field of x"—& over K).

Proof Since [L: K] =p we have Gal(L/K)I = p and so Gal(L/K)

is cyclic of prime order. Let ceL\K. Define c0=c,
= 0(c), c2 = 02(c), etc. so that &(c) = c. Let a( I) be one of the pth roots of
unity in K and put* 3=c0+cc1+e2c2+ +ur'c,,_1. Then 0(3)=e_13.
Consequently 0(&) = & so that & Fix(Gal(L/K)) K (since Fix(Gal(L/K))
K whilst [L: =p). But K since 0(3) 3. Hence L = K(3) since [L: K]
as required. (Isn't that nice? Well it would be, except for one thing. Why is 3
So the big question is: can we assume that 3 0? Exercise 10 tells you
'essentially yes'.
The main content of sections 7.9 and 7.10 is thus

Theorem 7.10.5 (Galois) Let f(x) e F[x], where char F = 0 and let Sf be the
splitting extension off(x) over F. Thenf(x) is soluble by radicals over F if and
only GaI(Sf/F) is a soluble group.

* Elements of this type were used by Lagrange and Galois and go under the name 'resolvent'. See
and 7.12.
 A sufficient condition for the solubility of a polynomial equation by radicals 307

This result provides the unification which Lagrange was seeking. It tells us that,
because the symmetric groups S2, 53, S4 are all soluble, there are general
formulae (that is, involving literal coefficients) for obtaining the zeros of
polynomial equations of degrees 2, 3 and 4.
The above is just one consequence of what is now called the Fundamental
Theorem of Galois Theory, namely:

Theorem 7.10.6 Let L be a finite normal and separable (i.e. Galois) extension
of a subfield K. The map from the set F of fields lying between L and
K to the set cc of subgroups of Gal(L/K) given by cD(M) = Gal(L/M) is a 1—I
correspondence. In addition M is a normal extension of K if Gal(L/M) is a
normal subgroup of Gal(L/K). In this case Gal(M/K) Gal(L/K)/Gal(L/M).

Proof Put together 7.6.4 and exercise 7.10.1.

Exercises
1 Let F c K c F with F a Galois extension of F. Prove, in full, that K is a
normal extension of F if Gal(E/K) is a normal subgroup of Gal(E/F).
2 Find a normal extension N of €1 such that N a radical
tower over 0?
3 By considering the subgroups of the Galois group of f(x) = — 1 over 0,
show that there exist normal extensions which are not radical towers over 0.
(Hence, even iff(x) is soluble by radicals over 0, need not itself be a radical
tower.)
4 Show that if K is a radical extension of prime degree (>2) over 0, then K is
not a normal extension of 0.
5 Let M be a Galois extension of F and let c e Gal(M/F). Suppose that
Show that
cx(K) = L if ct
-1= Gal(M/F)).
6 Let K be a field. Define automorphisms s, t e Aut(K(x)) by: s(x)= l/x
and t(x) = I — x e K(x). Show that the subgroup S = <s, t) of Aut(K(x)) S3.
Show that k = (x2 — x + l)3/x2(x — 1)2 is in Fix S. Deduce that
[K(x) : Fix 5] [K(x): K(k)]. But kx2(x — l)2_ (x2 — x + = 0. Hence x satisfies
a polynomial of degree 6 over K(k). Deduce that Fix S= K(k), precisely.
7 Show that if f(x) is irreducible in F[x], then Gal(S1/F) is transitive on the
zeros off(x) in [A group G of permutations on a set X is transitive on X if,
for each pair y, z of elements in X, there is a permutation in G under which y is
sent to z.]
308 A brief excursion into Galois Theory

8 Letf(x) e F[x] withf(x) irreducible and characteristic F = 0. Show that if an

extension field E of F containing one zero off(x) is a subfield of some radical
tower R, say, then Sf is a subfield of some (other) radical tower. Loosely: If one
zero of f(x) is radically expressible, then they all are.

9 Let f(x) be irreducible of prime degree p over F (of characteristic 0) with

splitting field S1 over F. It can be shown, [15], that G = Gal(Sf/F) is soluble if G
is isomorphic to some subgroup of the group of exercise 7.8.7 which includes all
transformations with i= 1. Use this result to prove the 'only if' part of what
Galois saw ([83]) as his most important result, namely: Letf(x) be as above and
let a, b be zeros of f(x) in S1. Then f(x) is soluble by radicals Vf S1 = F(a, b).

10 In replace 6 by 61=c0+dc1+(e2)'c2+ ...

7.10.4
(0 I). Prove that so that bfe K. Adding the p equalities
obtain 60+ +öp_1 =pc0. Deduce that not all can be in K—and so
L=K(6,), for some i.

11 Noting that C is a splitting field over for each polynomial in k[x], and
that Gal(CRR)= C2, how can you reconcile 7.10.5 with exercise 7.9.5?

12 (a) Give two examples of fields KcL with [L:K]= 10 and such that: in (i)
all intermediate fields are normal over K; in (ii) not all intermediate fields are
normal over K.
(b) Find_the Galois groups over 0 of: (i) x3—3; (ii) x4+4: (iii) x60— 1;
(iv) + In (i), (ii), (iii) identify the normal subfields of the correspond-
ing splitting field.

13 Why are the following soluble by radicals over 0? (i) x6 + x4 — 2x2 —2;
(ii) x8—4x5—2x3+8; (iii) x6—2x5+4x4—8x3+ 16x2—32x+64?

7.11 Non-soluble polynomials: grow your own!

To show the existence of particular polynomials g(x) not all* of whose zeros are
expressible in radical form only requires us to find a polynomial whose Galois
group is not soluble. A good way to do this is to return to Galois' idea (see
exercise 7.2.11) of regarding Gal(S1/F) as a group of permutations on the zeros
off(x) in Sf. This suggests, in the first instance, that we look for a polynomial
with Galois group (since A5 is not soluble—see 6.6.1). However, it seems
easier (see exercise 6) to construct, instead, a polynomial whose Galois group is
(isomorphic to) S5.
To be certain what is required, let us suppose that f(x) is irreducible in F[x]
with 5, = F(tx1,. . , nj being, as usual, the splitting field of f(x) over F, the ;
.

being the distinct zeros off(x) in S1. Since each element U of Gal(S1/F) effects a

* Hence 'none of. . .'. See exercise 7.10.8.

 Non-soluble polynomials: grow your own! 309

permutation U,, say, on the set . , x,} (exercise 7.2.11), the obvious map
.

F: Gal(Si/F)—+Sn generates the 'obvious' theorem, namely

Theorem 7.11.1 The map F is a 1—1 homomorphism of Gal(S1/F) into That

is, Gal(S1/F) is isomorphic to a subgroup of

Proof Exercise 7.2.11.

Since is generated (see exercise 5.6.25) by the two cycles (12. . . n) and (12),
the next two lemmas almost suggest themselves.

Lemma 7.11.2 Let F, f(x), be as usual with f(x) irreducible over F and of
degree p, p being a prime. Then Gal(51/F) contains a p-cycle.

Proof Let be a zero of f(x) in S1. Then [Si: F] = [Si: F(cz)][F(oc) : F]. Since
[F(cx) : F] =p (see 4.3.4) we see that[Si: F] = I Gal(51/F)I. Hence Gal(51/F) has
an element of order p. But Gal(51/F) is a subgroup of S, (why?) and the only
elements of order p in 5,, are p-cycles. (Exercises I (below) and 5.6.8(b).)

Perhaps the easiest way to ensure that Gal(51/F) also contains a transposition
is to specialise F[x] to Q[x] and prove

Lemma 7.11.3 If f(x) e Q[x] is irreducible and has exactly 2 non-real zeros,
then Gal(51/F) contains a 2-cycle.

Proof Here we may, of course, assume that S1 C. Then complex conjugation

fixes all the real zeros in and permutes the complex ones (by exercise 2). This
suffices.

Combining 7.11.1, 7.11.2 and 7.11.3 we have:

Theorem 7.11.4 If f(x) e O[x] is irreducible of degree p (p being a prime 5)

andf(x) has exactly 2 non-real zeros, then Gal(Sf/Q) = 5,,. In particular 5,, is not
a soluble group and f(x) is not soluble by radicals.

Using 7.11.4 you can construct your own non-soluble equations ad nauseam!
To do so first note that a real polynomialf(x) of degree 5 has three real zeros
iff'(x) has just two (distinct) zeros a, h, (say) such that (say)f(a)> 0 andf(b) <0
as in Figure. 7.1
310 A brief excursion into Galois Theory

Fig. 7.1 Part of the graph of y=x5—4x+2, with zeros H. S. Tnearto —1.52, 0.51.
arid 1.24

To keep the arithmetic easy let's assume that f'(x) = x4 — c where c is some
positive real number to be determined. Then f(x)—x5/5—cx+d for some
constant d. Set y—c114, the real positive 4th root of c. Nowf'(x)=O iffx= ±7.
Alsof(—y)= _y5/5 +747+d=I?5 +d whilstf(7)= +d. Thus if dO then
f(—y)>O whilstf(y).cO if d.c4gy5. Since we wantf(x) to be irreducible in 0(x) we
choose c, d so that Eisenstein's criterion will check this requirement quickly.
Now Sf(x) = — 5cx + Sd, so to use Eisenstein let's choose d= I and then c = 16,
say. Consequently x5—80x+5 is a polynomial of the required kind! We leave
other choices, for example c=*, to you.

Exercises
1 Show that the only elements of order p in S,, are p-cycles.

2 Let M be a subfield of C which contains, to each its complex con-

jugate i Show that complex conjugation on C induces an automorphism of
order 2 on M.
3 Find two polynomialsf(x), g(x) in 0[x] such that Gal(S1/0) and Gal(Sg/0)
are isomorphic (as abstract groups) but are not isomorphic as permutation
groups. [{I, (ab), (cd), (ah)(cd)} and {I, (ah)(cd), (ac)(bd), (ad)(hc)} are two such
groups. Can you see why they can't be regarded as being isomorphic as
permutation groups?]
4 Can you prove the following? (Cf. 7.11.4.) Letf(x) e 0{x] be irreducible of
degree S and have five real zeros including the pair and
where a, b e 0. Then f(x) isn't soluble by radicals over 0.
5 (Cf. 7.11.4.) Find an irreducible polynomial of degree 6 in 0[x] which has
just two non-real zeros and yet is soluble by radicals over 0.
 Galois theory—old and new 311

6 Let F be a field (characteristic 2). Let f(x) E F[x] be separable and let
S=F(tx1 cr3 be a splitting field. Put 5= fi Show that (i) 52 e F
1

and (ii) Gal(S/F) is isomorphic to a subgroup of if S e F.

7 Exhibit a field F, containing 0, and a polynomial f(x) e F[x] such that

8 Findf(x), in Q[x], of degree 5 and with all six coefficients non-zero such that
f(x) is not soluble by radicals over 0. [Hint:f(x+ 1).]
9 x5 — 4x + 2 and x5 — 6x +3 are frequently chosen as examples of polynomials
of degree 5 which are not soluble by radicals over 0. Confirm they are indeed
both insoluble over 0.
10 True or false? A soluble irreducible equation of prime degree over F(
has exactly one real zero—or all zeros real. [Hint: Use the result of exercise
7.10.9.]
11 Test the following for solubility over 0.
(i) x5—8x+3; (ii) x5—7x+3; (iii)
x6—x5+x4—x3+x2—x+1;
(iv) x5—3x3—3x+3; (v) x5—4x2+2; (vi) x5—4x3-3-2; (vii) x5—4x4+2;
(viii) 2 (p 5 prime); (ix) +x+ [Hint: for (ix)
Eisenstein shows irreducibility. Use the rational root test on the derivative to
help locate local max and mm.].
12 (a) Find [Hint: L=GF(f) is a Galois extension of Zr,.
Hence I Aut(L)I = [L: 74] = n. Show that, for L, the Frobenius automorphism 4)
(exercise 7.5.14) has order n.]
(b) Draw the subgroups/subfields diagram for the finite field of order p12.

7.12 Galois theory—old and new

Having spent some time seeing where our noses led us, let us quickly review
where Galois' nose led him.
Galois was interested in equations of the formf(x)=a0 +a1x+
where the a1 are given numbers (as distinct from the 'unknowns' used in 7.9.5)
whose n zeros are pairwise distinct.* [For ease of communication we shall
suppose—as in the presentations [12] and [II] of 1926 and 1930 respectively—
that we are working inside C and we shall let F denote the least subfield of C
(namely 0(a0, a1,. . ,aj) which contains 0 and all the cii. As already men-
.

tioned in Section 7.5, Galois seems not to have worried too much about where
or what the zeros off(x) were.]
Let us designate the zeros off(x) by cr1, cr2,. . Whereas, for Lagrange, the
.

cr1 were formally distinct symbols, so that cr1 +cr2cr3 and cx2+1x3cr1, for example,
were likewise formally distinct, Galois had to accept that + and cr2 +

* There is no loss in generality in this latter assumption; see exercise 7.5.1.

312 A brief excursion into Galois Theory

could be numerically equal. Nevertheless, under the (obviously necessary)

assumption that the ; were numerical/v distinct, Galois observed (without
proof) that integers z1, z2 z,, could be found so that the quantity

(7.12.1)
would take n! distinct numerical values on formally permuting the ; (equiva-
lently the z1) in all n! possible ways.
With t so chosen, Galois showed that all the ; (I i n) could be expressed as
rational functions of (i.e. a ratio of polynomials in) t. [In our notation, this
means ;)= 0(t). (Why not Galois now forms 'the
equation for C—that is, the equation
(7.12.2)
where the t the n!
of the
of R(x): that is, they are symmetric functions of the;. [Hence, in
our language, they belong to F. Why?]
Now R(x) may factorise (into a product of irreducibles) in F[x], R(x)
=R1(x)R2(x). . Rp(X), say. Suppose (without loss of generality) that R1(x)
.

=(x—t1)(x—t2). . is the factor with t as a zero. Then

Definition The Galois group of f(x) over F[xj is the set of permutations on the
cc, which arise when 1 is mapped into '2 4 in turn.
One can show that this finite set has the properties required for it to be a
group and that the group is independent of the particular t chosen at the outset.
Galois now says that if he 'adjoins to the given equation' the zero of some
auxiliary equation (that is, if we replace the 'base' field F by a larger field F1,
say, and think of f(x) as being in F1[x]), then either the Galois group of f(x)
(over F1[x]) will be the same as that off(x) (over F[x]) or will be a subgroup of
the latter—cf. the proof of 7.10.2. [The reason is easily seen: In F1[x] the
previously irreducible R1(x) may, itself, now factorise into a product of factors
irreducible in F1[x]. If R1(x)=R11(x)R12(x).. .Rim(x) with R11(x)=(x—t1)
(x — t2). . (x — t1), say, where I k, the corresponding Galois group comprises
.

only those permutations of the ; which arise on mapping t to t1 t,.

If, by a succession of such adjunctions——i.e. by repeatedly enlarging the
(current) base field F—we arrive at a field F2, say, in which we can factor R(x)
into a product of linear factors, then the Galois group off(x) (over F2) will be
trivial and, since 1 belongs to F2, so too will all the zeros .,cc, of
f(x).
So the question is: What must the above sequence of subgroups (from that of
the Galois group of f(x) over F down to <e)) look like if we insist that each
'adjunction' is of a p1th root some prime) of a quantity previously
constructed (or given)?
Galois' answer, is that the (Galois) group must have a series of subgroups as
in (*) in 6.5.7: that is G must be soluble.
 Galois theory—o/d and new 313

Although, as we noted earlier, Galois' ideas were still being followed as late as
1930 [11], the 'modern' viewpoint sees the subject of Galois Theory as that of
investigating the connection between the subfields of certain types of extensions
(in the first instance Galois extensions) K of a given field F and those subgroups
of the automorphism group of K which 'fix' F. The conditions relating to the
solubility of equations by radicals then result from an application of the general
theory. [Indeed Artin, [3], will have nothing to do with equation solving, leaving
applications to an appendix written by a coauthor!] Of course Artin's approach
didn't suddenly emerge from nowhere; one can trace various ideas via, for
example, Dedekind (who obtained the result of Artin's lemma working
throughout in C), Weber (who discussed groups and fields of 'things') and
Steinitz (who noticed that a wholesale generalisation of Galois' ideas to these
general fields wouldn't necessarily go through and introduced the concepts of
normality and separability to clarify when it would). The account of this thread
of history given in [108] is very interesting but do follow this up by reading the
account [126] of later development by one who was involved.
Naturally enough, attempts have been made to extend Galois Theory to other
settings, for example to commutative rings, to division rings and to differential
equations (see, for instance, [4]). Even in its most 'primitive' setting there are
still huge gaps in our knowledge. Perhaps the most difficult problem is the
determination of precisely which (finite) groups can be Galois groups of the
form Gal(S1/Q). It is believed that every finite group is of this kind. Shafarevich
showed (1954) that every finite soluble group is a Galois group over Q so,
naturally, attention turned to the finite simple groups. In 1984 Thompson
showed that the 'monster', the largest of the 26 sporadic simple groups, is a
Galois group over 0. Since then the majority of the sporadic groups have been
proved to be Galois over 0 by one or more of Hoyden-Siedersieben, Hunt,
Matzat and Pahlings.

Exercises
I (The Primitive Element Theorem) Let F E be fields of characteristic 0.
Suppose E=F(a,b), where a,b are algebraic over F (see p. 153). Then there
exists d e E such that E = F(d). (That is, E can be generated over F by a single
element. We say E is a simple extension of F.) Proceed as follows. Let f(x),
g(x) be the minimum polynomials over F of a, b respectively and let
K=F(a1,. where a=a1,b=b1, bea splitting field forf(x)g(x)
over F, the a1, being its distinct zeros. Choose c e F such that c #0 and
for all i#k and j#l. [Why can we do this?] Claim: a+bc
(= d, say) works as required. For, clearly, F(d) F(a, b) (= E). Conversely, let
h(x) =f(d— cx) e F(d)[x]. Prove that h(h) =g(b) = 0 in E. Setting
t(x) = gcd(h(x),g(x)) e F(d)[x] we have deg t(x) 1. [Why? What if t = I?] In
K[x], .(x—bj. Hence any other zero of t(x) in K is a
h/f> I). But then so i.e. d—cbJ=ak for some k. This
is impossible (why?). Hence t(x) = x — b, whence h e F(d).
314 A brief excursion into Galois Theory

(This result was for years one of the cornerstones in the proofs of the main
results of Galois Theory. In particular the vital concept of normal extension was
made to depend on it. Artin showed how to reshape the definition of normal
extension without having to introduce the somewhat artificially chosen primi-
tive element.)
2 Deduce from exercise 1 that, if K L where [L : K] < cc and characteristic
K = 0, then L = K(ct) for suitable x e K.
3 (a) Find a suitable primitive element for (i) (ii)
(iii) the splitting field over 0 of x3 —2.
(b) In all honesty (!) which 'presentation' of the splitting field of (x2 — 2)(x2 + 1)
is the more natural (i) 2, i) or (ii) + i)?
4 Which result in Section 7.10 corresponds to Galois' idea of 'reducing the
group of the equation' by extending the 'base' field?
5 Use exercise 2 to show, under those assumptions, that Gal(L/K)I c [L: K].
Partial solutions to the
exercises

The following 'answers' are only rarely complete. They are in general meant to
give a good clue without doing all the work for the reader—and they should
certainly not be taken as the epitome of good mathematical style. Where a
counterexample is given, you should aim to produce another.

Exercises in Chapter 0
1 (a) False; (c) true; (e) false Math. Gazette, p. 220, Oct. 1987).
2 (i) True; (iii) false; (v) false; (viii) true; (x) true.
3 BnC={3};AuB={x:xer
and v=2\/(—2+\J5).
5 (b) Let xe(AuB)nC. Then (xeA or xeB) and xeC. Hence (xeA and
xeC) or (xeB and xeC). Consequently xe(AnC)u(BnC).

6 flT,={O};
reP n1
7
(i) and (ii). Do (iv) similarly.
8 a,&c. Then
Suppose {{a}, {a,b}}={{c}, {c,d}} and that Therefore
{
= {c, d} and so a = c( = d), contradiction! Consequently a = c and so
a}
{a}={c}. If now Therefore {a,b}={c}. Hence....
9 (ii) No! Take, for example, A={l,2}, B={a,b} and S={(1,a),(2,b)}.

10 (i) 1024(=2b0).

11 Assume m is odd. Then m=2t+ I for some integer t. Hence m2=4t2+4t

+ 1, a contradiction! Consequently m must be even.

15 Note that each integer is of the form 4k or 4k+ I or 4k+2 or 4k+3. Then
use (4k)2 = 8(2k2), (4k+ 1)2 = 8(2k2 +k)+ 1, etc.

16 Informally: For each integer x and each integer y, the product xy is the
square of an integer.
17 (i) is true; (ii) is false.
316 Partial solutions to the exercises

Exercises 1.2
1 Writing x for b+c we deduce (a+(b+c))+d=(a+x)+d=a+(x+d)
= a + ((h + c) + d). The others are similar.
3 means So
(The reasons for each step are left to you.)
5 From we obtain (using Axiom

a + ( — b) = a b (using Axioms A2 (where?) A4

and A3).
7 Let z and —f be the functions defined for all xeC[O,1] by: z(x)=O;
(—f)(x) = — (f(x)). Then z, —fe CEO, I]. z is the function satisfying (the ana-
logue of) the A3 axiom; givenf, then —f satisfies A4. To show axiom Z isn't
satisfied:
1 0
0

Then F z and 6 z. (Draw yourself a picture of F(x) and 6(x)!)

9 or a>O (i.e. —aeN or aeN: see 1.24). Then
=a'aeN. Hence Oa2.
10 If, for example, —aeN and heN then Hence
—(a and so neither is h. (Now tidy up this argument!)
ii (i) Use (a1—h1)(a2—b2)>O.
13 0. (Here take N as the positive rationals.)
14 If there is no '0' then A3, A4 will not be satisfied. Note also that Axiom I
can't fail if the set under consideration has no 'I'. So you might look for a subset
of 1, say, containing neither 0 nor I.
15 All ten are satisfied.
16 (ii) Let U be the set of all positive integers n such that 1 +3+
+ (2n — 1) = n2. For n = I the left-hand side (lhs) is the sum of all odd integers
from I to 2n— I (i.e. from I to I in this case). Hence the sum is 1—and this is the
value of the rhs when n= I. Hence I eU. Now suppose ke UcN; that is,
suppose lhs=rhs if n=k. For n=k+llhs is
+ (2(k + I) — I). By the induction hypothcsis this sum is k2 + (2(k + 1)— 1), that
is, (k-f-I)2. Thus k+ eU. Hence U=N by property I.
I

17 Fact: I + + + can be madc as big as you wish by taking sufficiently

many terms. (See, for example, G. H. Hardy's famous Pure Mathematics)
18
 Exercises 1.4 317

19 Let U be the set of positive integers n for which I n. Then 1 e U. Now let
keU_cN. Then (k+l)—l=keN so that l<k+l. By property I, U—N.
Hence if c is an integer with 0< c then I c. Hence c < 1 is impossible. Finally,
prove 0<t—kc I using exercise 8.
20
I —(ac)e N, a contradiction. Hence O<b. By exercise 19,
l a, l b. If l.ch then, by 1.2.6, (l )al ab= I—a contradiction
22 Look to see if the argument purporting to show that S(k) = S(k + 1) always
holds is valid in the case k= I.
23 (i) Show that Usk=SUSk_4+3Usk_5.

Exercises 1.3
1 Ifb=ua and a=vb then b=u(vb)=(uv)b. Hence, if* b$0 then 1 =uv [why?].
Thus u, v are units. [*What if b = 0?]
3 Use induction on the number r of factors in the product b1b2. . .

=b1(h2. . . b,jx, say, the case ofr=2 being given. Begin by saying: 'Ifalx then
oralh2...br'.
5 Suppose x=&x+u, y=5f3+v, z=5y+w, where u,v,we{—2, —1, l,2}.
(Why these remainders?) Show that 5,j'u2 + v2 + w2.
6 The first six irreducibles are 5, 9, 13, 17, 21, 29. Note that 9 is irreducible
(essentially because 3 H) but is not prime in H (since 693 = = 21 . 33 and
yet 9t21 and 9,f'33 in H). On the other hand each H-prime is an H-irreducible.
In fact, if h e H, then h is prime in H if h is prime in 7.
7 Prove that if . . q1, where each is of the form 4k+ I, then so is
Na—a contradiction. Thus at least one of the q, must be of the form 4k + 3 and
hence must be one of the Finish the proof as in 1.3.10.
8 [Cf. exercise 7] For the 6k+5 case consider .. =q1 . .
1 .

Show that each q1 is of the form 6k + I or 6k +5. A similar proof for primes of
the form 8k +7 fails because numbers of the form 8k +7 need not have any
prime factors of that form. (What is the smallest example of this?) [In fact there
do exist infinitely many primes of the form 8k + 7—see [42], p. 53.]

Exercises 1.4
2 Since c is a common divisor of a and b and since d is a greatest common
divisor, we have cid. Likewise dic. Now use exercise 1.3.1.
4 There exist such that cxa+yc=l, flb+öc=1. Then 1=
(cza + yc)(flh + öc) = ccflah + c(ctha + yfib + yôc). Hence, if diab and dic, then ...?
318 Partial solutions to the exercises

5 Show that if dI(a,c) then dia and dla+h. Hence di...? But (a,h)= 1.

6 Let u=(a,(b, c)). Then u divides a and band c. Hence u

divides (a, b) and c. Consequently ui((a, b), c). Conversely ...?
8 Assume without loss of generality that Or1 From b(m1—m2)
=r2—r1 wededuceblr2—r1. Butthisforcesr2—r1=O----sinceOr2—r1<Ibi.
10 IfS doesn't comprise 0 alone then there exists in S a non-zero integer a, say.
Then 0( = a — a) e S and so 0— a e S. Hence S contains at least one positive
integer. Let c be the smallest positive integer in S. [How do we know such a c
exists?] Now c —( — c) = 2c e S and likewise (or, preferably, by induction!) mc e S
for all m e /. Conversely, suppose w e S. Write w = mc + r where 0 r c c. Then
r e S (since w and mc are both in 5). Hence, by choice of c, r 0, whence w = mc.
Thus each multiple of c is in S and, conversely, each member of S is a multiple of
c, as claimed.
11 Ifd=(s,t)then dcisa and dcith so that dclsa+th=c. Hence d(>0)=l.
12
1—for example.
13 Put c=(a,b) and d=(b,r). Then cia, cib, so that cia—mh=r. Hence clb
and cir and so cid. Likewise dib and dir imply dia. Consequently dic. Thus
d=c.
14 Havingfound(901,527)=17=12527—7901,supposel7=527u-1-901v.
Then Hence Therefore

15 The common divisors of 441 and 693 in H are 1, 9, 21. So 21 is numerically

the greatest but is not a multiple (in H) of 9.
16 (i) Ifk=rs then 2'— 1 1. (Cf. x— I 1 for polynomials. Now put
= 2r.) (ii) List the divisors of 2k - 'p assuming p is a prime.
17 Put
(p\ p(p—l)...(p+l—i)
'\i) 1

Then u i! =p(p— 1). (p+ 1 —i). Now clearly a is an integer. (One can prove
. .

this by induction or observe that it is the number of ways of choosing i things

from a set of p objects.) Also piu i! But p,f'i! [Why not?] Hence piu as required.

Exercises 1.5
I No—since each of the factors is prime. See 1.5.1.
4 If a=md1=nd2 and iNtl1 +fJd2=1 then a=cxd1(nd2)+f1d2(md,).
5 sa+ th= I thc—c a(sc+tm)=c—assuming hc=ma. Therefore
a c.
 Exercises 1.9 319

6 Let a . and h =
.
. .
where the are pairwise distinct
primes. Suppose, further, that c = . . where the Uk are pairwise distinct
primes, and that all the ;, h are positive integers. Then c2
. . . .up'. By 1.5.1 the Uk must be and in some
order and each ;, 13j is some 2Yk and hence even. Thus both a and b are squares.

Exercises 1.6
2 (b)uov=(0,2,6,—l,—13,7,14,—2,-—15,6,0,0,...)
5 l+3x+x2—x3=(l+x)(l+2x—x2).Sotakeu=(l,l,0,0,...),v=...?
6 Assuming u=(u0,u1,...), v=(v0,v1,...) have zero entries from Urn and
onwards respectively, then U$V has only zero entries from the tth place
onwards, where t Is max{m, n}, the greater of m and n. (Note that it's easiest to
regard U0, for example, as being in the 0th place.) For the product, the infinite
succession of zeros starts at the ((m — l)(n — 1) + 1)th place. (Correct?)

Exercises 1.7
2 UO(v$w) has as its kth term
+1
4 Take N 1(N2) as the set of all polynomials of the form arf + +

+ +a/ (where such that ar>0(as>0).

S If(a0,a1,...)Q(h0,b1,...)=(l,0,0,...), then a0b0=l. Hence Con-
versely, given use this to define b0,b1, b2,... inductively by I =a0b0;
(for 1>0). Thus, at each stage, b1 is
defined in terms of the given and the already evaluated hk (0 k t — I).

Exercises 1.8
2 Formally x is ((0,l,0,...),(0,0,.. .),(0,0 ) whilst y is represented
formallyby((0,0,...),(l,0,0,...),(0,0

Exercises 1.9
1 (x2 — 2)(x2 + 2), (x —\/2)(x + ]2)(x2 + 2) in 7L[xJ and R[x] respectively.
3 (i) reducible: 2 and (I + x + 3x2) are non-units in Z[x]; (ii) irreducible—note
that 2 is a unit in 0; (iii) irreducible; (iv) reducible (why?).
S (i) Yes. First show that p is neither the zero element nor a unit in Z[x].
Finally note that ifpf(x)g(x) in Z[x], then p dividesf13g0,f0g1 +f1g0, etc. in 1. If
320 Partial solutions to the exercises

and in Z[x] then there exist a least i and j such that f1 and are
not divisible byp. Then (cf. the proof of 1.9.10) the coefficient of inf(x)g(x)
is not divisible by p—contradiction.
8 (b) M3x2+2x+42).
9 Assuming that and (so that degree is defined) we have
deg (fg) = deg f+ deg g; deg(f+ g) max{degf, deg g}—if, also, — g.

10 (a) Irreducible (Eisenstein with p = 2); (c) Eisenstein is no use with f as

given. Butf_ succumbs to Eistenstein and provesf_1 and hencef irreducible.
1

11 f(x) = x4 + 1 + I) = x4 + 4x3 + 6x2 + 4x + 2—irreducible by Eisenstein

(p = 2 again!) Hence f(x)=g(x)h(x) is impossible—by the usual argument.
12 Let g(x)=l+x"+ Then Suppose
g(x+ l)h(x+ l)=k(x+1), say, where k(x+ ...+p2x
has coefficients all of which (except ) are multiples of p. Likewise all
coefficients of h(x + 1) = (x + I )" — I are divisible by p—except for that of
exercise 1.4.17. Comparing coefficients in g(x + 1)h(x + 1) and k(x + I)
show that all the coefficients in g(x + l)—except that of - multiples
of p. Now use Eisenstein.
14 Take F = + x. Clearly F is irreducible in Z[x]—but no prime divides 1 and
1

1. (Note that (A B) doesn't imply A B) — as some students occasio-

nally believe!)
15 6 + 6x + 3x2 is reducible in Z[x] — but irreducible in O[x] (Eisenstein, p = 2
yet again!!)—essentially because 3 is a unit in 0 but not in /. Proof fails within
first three lines. No G, H with deg G, deg H <n may be available.
17 (Cf. the proof of 1.4.4.) Since f, g e S we see that S contains a non-zero
polynomial. Let h be any polynomial in S from amongst those of smallest degree
in S. (Such an h obviously exists once we know S contains some non-zero
polynomial.) Let h0=cxh(cce0) be monic. Now let weS. Write w=mh0-+-r
where m,reO[x] and either (i) r=0 or (ii) n#0 and degrcdegh0. Then
r=w—mh0eS (since
w=sf+t1g and h0=s2f+12g=t.r=(s1—ms2)f+(t1--mt2)g
clearly lies in S). But, by choice of h0, we are forced to deduce r=0 whence
w=mh0.

Exercises 1.10
1

2 f(x) = x, g(x) = 2x. Second part: There exist m, r e 0[x] such that f= mg + r.
Writing m = M/z, r = R/z for suitable z e 7 we get the required result — with
degR=degrcdegg (if r0).
 Exercises 2.2 321

3 (a) x2 + 2x + 1; (b) Repeated division algorithm seems to yield 392/1849 as a

gcd of the given polynomials. Hence the unique monk gcd is...?

Exercises 1.11
2 Note thatf' is just the derivative of in the usual calculus sense. Thus, if
f= (x — a)2g, then f' = 2(x — a)g +(x — a)2g'. Hence (x—a)If'.
3 Letf=x4—4x3+4x2+17.Thenf'=4x(x—1)(x—2).ButnoneofO,l,2isa
root off.
4 Withf=x4+4x2—4x—3, we have (f,f')= 1 in Q[x]. Hence I =sf+tf' with
s, e O[x]. If were a repeated complex root off we would =f'(cz) = 0
in C, whence I = s(cx)f(cx) + = 0. Contradiction.

5 f—g is of degree at most n (if it's not zero), but has n+ 1 roots. Now use
1.11.4.
7 (a) By the rational root test the only possibilities are ± I, ± ±
(b) is a root here is a root in part (a). (d) No roots here. (This polynomial
plays a prominent role in Section 4.6.)
9 x—(2—i) is a factor and so is x—(2+i)----by exercise 8. Hence (x—2+i)
(x—2—i)=(x—2)2—i2=x2—4x+5 is a factor of The
other quadratic factor is therefore x2 + x + I with roots x = ( — ± — 3)/2.
1

10 The only potential rational roots are ± 1—but neither is. So try x4—x2
+1=(x2+ax±1)(x2+bx±l). This shows a+b=0, ab±2=—l. Hence no
rational solution. (Or: Solve the quadratic in x2!!)

Exercises 2.2
1 A binary relation on A isjust a subset ofA xA. IfIAI=n then IA x A1tn2.
Now use exercise 10 of Chapter 0.
2 (ii) Note that nI(a—c)b+c(b—d); (iii) From nlma—mc we deduce nla—c
provided (n, m) = I (exercise 1.5.5). As an example 4 2(mod 2) but
2* l(mod2).
4 (i) Hence x=5,13 will do. (ii) 4x=7
+ 8m is impossible—since 4,]'7.
5 (i) t, a. (a? Yes—for, if not, there must exist x, y e 1 such that x <y and y <x
and x But there is no such pair satisfying even the first two of the three
conditions; (iv) r, s.
8 EisallofAxA except(1,4), (2,4), (3,4), (4,1),(4,2),(4,3).
322 Partial solutions to the exercises

9 The singleton S = {( 1, 2)} won't do—since condition t is satisfied, as we say,

vacuously. (For every two distinct or identical pairs to be found in S (!!) the t
condition holds.)
11 I claim the relation c comprises 81 ordered pairs. Am I right?
12 LetS={a,h}.Define _cby:hca,b_ch,a_ca.

not an order relation.

Exercises 2.3
1 Take {l}, {2}, {3}, {4} and the rest of / as equivalence classes.
2 Try {l}, {2,3}, {4,5,6}, ... etc.

4 (i)(x1,y1)R(x2,y2)iff(x1—1)2+(y1—l)2=(x2—l)2+(y2--1)2.
5 52 (Am I correct?)

6 iffz1/z2=1z1/z21, i.e. iffz1/z2 is real and positive. Thus the

equivalence classes are all half lines sprouting out of the origin.

Exercises 2.4
2 l7Ix— or
1 1 1. (Why?) Hence ± l(mod 17). That is,
there are essentially two solutions in 7L—exactly two in has 4 1

solutions, namely 1, 3, 7 in Put another way: all odd integers are solutions
of the congruence l(mod 8).
4 (a) The argument goes through since /,, is a field (as are 0, C). In fact
1.11.5 is valid too. (Do 1.11.4 and 1.11.5 hold for integral domains—e.g. /'?)
8 By inspection: and 4 will each do. (Any more?)
10 (ii) 5 8 1 (mod 13). So 8 '= 3 in (iv) 8x = I + 34k is impossible in /;
(v) By the Euclidean Algorithm I =

12 Choosing elements almost at random note that and But, by

(proposed) definition 0 = whereas 0 = = k.

Exercises 2.5
1 (i) Hence Consequently Thus
x=23.
 Exercises 2.6 323

3 Hence It follows that

1(mod 341).

6 Suppose is composite with 1<n1,n2<n. Show n11(n—1)!+l. But

n1 I(n— I)! Hence n1 1,contradiction.
8 In the range 1,2,... ,p,. .,2p,..
those numbers divisible by p are
p,2p,. Thusthere are of them. Hence there are pr_pr-i which
are not divisible by (that is, coprime to) p.
9 Assuming ar"
1)p"
1(mod/) look at (at" ')"=(l and then
reduce mod after expanding by the binomial theorem.

Exercises 2.6
1 (ii) Yes; (iii) no.
2 (a) (ii) No; (iii) yes: (b) (from A to B).

3 fisnotonto;therangeofgislV;hisl—l.
4 g is not onto; h is both 1—I and onto.

6 Iff(x)=g(x) for all x in 0 then for i=l,2,...,n+l, where

n=max{degf,degg}. Hencef=g by exercise 1.11.5 (which uses 1.11.4).
7 {O, 7}f -' —15}. (Note that 7 is not a value of 5—x2 for xe It)

8 (ii) x ef(Sn T) x =f(u) for u e Sn T Therefore x ef(S)nf(T). For an

example to show that equality does not hold in general, try to make ISnTI
'small' whilst making f(S)I=If(T)1 'large'.
(iv) xef - '(UnV) =tfrf(x)e Un V x ef - 1(U)nf - 1(V). Conversely,
xef1(U)nf1(V)=i.f(x)e U and f(x)e UnV
9 (ii) '1 then (af )g=(bf )g. Hence af=bf (sincegis
1 ness': If a(fog)=b(fog)
1-1). Consequently a=b (since ?). Thusfog is 1—1.
10 fcgflof—since l(gof), for example.
11 Let heB. Then b=af for some aeA—sincef is onto. If(b,a1)=(b,a2)
ef1 then b=a1f=a2fby definition. Butf is 1—1. Therefore a1 =a2. Hencef'
is a function from B to A. Since each a yields an afe B,f onto.
12 If a1f=a2f then (a1f)g=(a2f)g, that is allA=a2IA, In other words
a1=a2. Thereforef is 1—1. Now suppose acA. Then Hence g
sends af to a and so g is onto.
14 The answer to the second part is—No! (For iffog = lA(A finite) thenf is 1—1
(exercise 12) and hence onto (exercise 13). Likewise g is onto and, consequently,
1—1. Now if b(gof)#b and if af=b (since f is onto) then
a(fogof) =((af )g)f= whereas af= b.
324 Partial solutions to the exercises

Exercises 2.7
1 The negative integers provide an example. (Can you find others?)
2 In (iv) and (v) the identity functions are, respectively; (iv) I given by 1(x) =0
for all XE (v) 1 given by 1(x)=x for all XEX.
4 choose a,beX(afl). Now definef,geE by af=bf=a; ag=bg
=h. Then show that
7 (i) (iii) (v)
A? x I
C? x I x
I? x x x

9 In an n x n multiplication table there are n(n + l)/2 entries on or above the

'main diagonal'. Once these entries are inserted the rest are automatically
determined—if the operation is given to be commutative. The number of
commutative operations is therefore
11 We need a non-associative (but commutative) example. Try (vi) of exercise
7.

12 The operation is associative and has an identity element but is not

commutative. (Try a =2, b = c = d= I: now find a 'nicer' counterexample.)

Exercises 3.2
1 (b) is a cheat (sorry!) since + isn't a binary operation on the given set; (c) all
axioms hold—it's a field; (e) M4 fails.
2 Note that Al, A2, A3, A4 must hold in each case. For the second example
look at axiom D.
5 All elements of Y(S), except S itself, are zero divisors (look at XnX'). For
<Y(S), (p HF, 0> we need (now) only check the axioms involving H. Clearly
would have to be the element satisfying A3. Now try A4.
6 The first part is essentially exercise 1.2.7. For <F, ®, a> there is no need to
check the ® axioms again so just check M2 and D. For D try functionsf, g, h
wheref(t)= 1 for each t e It (In fact <F, $, a> would be a ring except it just fails
one 'half' of D.)
8 Write d and n in place of 0 and 1. We then have (with slight rearrangement)
d 0 n d
n
ddn n d
dnd
n n n

(Recall that 0 and 1 are only symbols; the tables define their relationship.)
 Exercises 3.3 325

This now looks like the set {E,O} of exercise 1.2.15 (In fact ii and dwere
chosen being the last letters of 'even' and 'odd'.) Thus, by exercise 1.2.15, all
axioms are satisfied and, as defined, <{0, l}, ®, Q> is a field.
9 This construction is important but proofs are quite straightforward. For
example (ZR, Z5) is the additive identity element; (—r, is the additive inverse
of(r,s), etc.
10 R=4, c=I. Now find another example.
11 l802n=36n=34nEB2n=2n. It is the only unity—see exercise 3.3.1(iii).

13 and in the first case—and with 'hats' on in the second

Try ( (?
case.
14 Both and Qk/d] l(mod4) or not) clearly satisfy
Al, A2, A3, A4, M2, D, MI, M3, Z since they are sets of complex numbers [Note
that each is closed with respect to + and ] So to check is a field we only
need to check that the multiplicative inverse of a + is again in
15 Here we are assuming that f/g =f h/gh if h #0 (the zero polynomial), etc.
Note thatf and g are to be manipulated asformal symbols—no discussion of the
roots of g is required. In particularf/g is undefined when and only when g is the
zero polynomial.

Exercises 3.3
1 (a) (i) a+c=b+c. By A4 there exists such that c+c*=ZR. Then
(a+c)+c*=(b+c)+c*. ByA2,a+(c+c*)=b+ (c+c*). ByA4,a+ZRb+ZR.
Hence a=b, by A3. (a) (iii) If e(f) satisfies M3 then e Therefore
e =f. Suppose a', a" satisfy M4. Now a' a") = (a' a) a" (by A2). Hence
. .

and so a'=a". (b) If we have

(a+b*)=c . a+(c . the first b)* [by 3.3.2(iii)]=Zj.
Since C#ZJ we deduce a+b* =Zj (axiom Z). Therefore b=ZJ+b=(a+b*)+b
=a+(b*+b)=a+ZJ=a.
3 For A3 use for A4 use (—r1, —Z1). M2 and D2 are messy but
straightforward. They make heavy use of the associative and distributive laws in
R and /. (Look out for their uses. Are any other properties of R or /
used?)
5 Suppose b, c e R are such that b = Z (and so we're
finished) or b $ Z. In the latter case there exists, since M4 holds, an element
b'eR such that b' b=e. But then c=e c=(b' . c=b' c)=b' Z=Z.
7 a+a=(a+a)2=a2-I-2aa+a2=a+2a+a.Therefore2a=OR.
8 For x,y choose (almost!) any pair of matrices from M2(7L).
326 Partial solutions to the exercises

9 (a) isn't equal to (even though the notation may tempt you to believe
it is). (b) The moral is: Use suggestive symbolism—but with care!
19 (cf. the proof of 3.3.4) If the elements of R are denoted by 14R f,,
then f0f1,f1f1,. ,f,f1 are t+ 1 distinct elements of R. (f1f1
. .

=°R and R has no zero divisors.) Thus

f1 =f1f1 for some i. Likewise, R, we have, usingf1 on the left,fJ=flfk.
Thenflfk=flflfk, and a left identity. In a
similar way there exists a right identity,f,, say, for R. Thenf1f1=f,=f1. Hencef1
is a multiplicative identity for R. Finally, R, on multiplying all
elements of R by fm' there exists an such that fmfn so that f, is a right
inverse forfm.

Exercises 3.4
I (a) No; (e) Yes—each is!; (e) no—sum may fail; (g) yes.
2 (ii) {a/b; a,he7L and for some
u{O}}.

4 (Use 3.4.2 rather than 3.4.1) : From a, h e S we find — he S. Consequently

a+(—b)eS, that is a—beS (and
given). c=: From a,heS we find
OR=a—aeS, hence —b=OR—heSandsoa—(—h)ES.
7 Even if R,S both have identity elements 1R' so that
we cannot infer that 'n Is by cancellation. (Additive cancellation is
S
always allowed, multiplicative cancellation only rarely.)
9 For all t e T contains at least one non-zero
element—see 3.4.1(F) and 3.2.3(iv)) given there exists eF. But then
Therefore 1F1T 1F1f from which 1T
12 (i) Since 15s+21t=3(5s+7t) we have [15,21]c[3]. Since 3s= 15(—4s)
+21(3s) we have [3]c[l5,2l].
13 Only (b), (c), (f), (g) and (h)(ii) are subrings (5, say). (b) is not an ideal since
I e 5, e Q, 4 S. Only the first set in (c) is an ideal. (g) is the principal ideal [15].
(h)(ii) is an ideal.
14 (2nd part) 'only if' done in first part! 'If': Let XE R(x#O). Then xR(=[x])
is an ideal containing Hence xR=R (and not {O}). In particular there
exists re R such that xr= Consequently M4 holds and R is a field.
16 Any ideal containing the a1 must contain all the z1a1, s1a1, a1t1 and u1a1v1 and
all sums of such where z1 e 7, t1, u1, V1 e R. Checking that the given set is an
ideal—and hence the unique smallest ideal containing a1,.. , am—is left to you.
.
 Exercises 3.6 327

r= + + e R). Since R is commutative, sa and uav can be written as as

and a(uv).
17 When R is commutative with a unity, 11={a1r1+ +amrm:rjeR},
12={b1s1 + Henceeachelementoflil2isasum
+amrjbisj + ...
of elements of the form (a1r1 + ... and hence a sum of
elements of the form R is commutative. Thus 'i'2
a2b1,. .. , The reverse inclusion is clear.
19 Given a, he U, suppose (without loss of generality) that ae be with
ij. Then Therefore a+b,
(O 0'\ (a
20 Let I be an ideal of If! then there exists with
Oj \c dj
not all of a, b, c, d equal to zero. By the given identity (or similar ones) we find
ci (u=a, b, c or d). We may, therefore, assume that Then, for

check that and

0)ei. Likewise ( and Hence I

22 (b)If[x,2]={pr:rel[x]} then x=pri, 2=pr2 forsuitable r1,r2elL[x]. The

latter implies that p = I or — I or 2 or —2 whilst the former rules out 2 and —2.
Hence p = or — I. This implies [x, 2] = Z[x] which is impossible since I [x, 2].
1

(Why not?)

Exercises 3.5
I (Cf. exercise 1.5.6.) Care is needed if n is even. For example, 36=(—4)(—9)
with coprime factors, neither of which is a square!
3 A solution for n = uv yields a solution for v. Thus if FC is true for 4 and each
odd prime p then it is also true for n = 4v and n =pv. But every integer n 3
either has an odd prime factor or is a power of 2.

Exercises 3.6
I (b) From h=ma and c=na we obtain hx+cy=a(mx+ny). Hence albx+cy.
(ci 1.3.4.)
2 Ifc is a unit in R then cd= where c/c RcR[x]. Hence c is a unit in
11[x]. Conversely if cp(x) = 'RIx] = then p(x) e R (by a degree argument) and so
c is a unit in R.
328 Partial solutions to the exercises

Suppose c is irreducible in R. Then unit in R and, hence, not in R[x].

If c =p(x)q(x) e R[x], then p(x), q(x) R and one is a unit in R—hence in R[x].

3 Ifv=2+V3 then uv=1 in But ukvk=(uv)k=1 so uk is a unit for each

ke V. Since 0< u<1 we have if i=j.
4 In M2(R) the units are those matrices with non-zero determinant. In
(7L[x])[y] the units are the units of /[x] (by exercise 2) and these are just 1 and

5 If 3= (a + 5)(c + then (taking norms) 9 = (a2 + 5b2)(c2 + 5d2).

Now a2+5b2$3 (ifa,b€ Hence a2+5b2=1 or 9. Consequently
a unit. However 3 is not prime in
since 31(1 —2\/--5)=21 but 3tl in

6 If then
Therefore (a2 + 3h2)(c2 + 3P) = 16. Consequently a2 + 3b2 = 4 (why not 1, 2, 8
nor 16?). It follows that a= ±2, b=0, or a= ± I, h= ± 1. These six possibilities
do all yield units.

7 4i and in 2/12 (the even integers mod 12) but 4, are not associates
since 2/12 has no multiplicative identity! (Cf. exercise 3.2.11.)
Even better is the following, due to Al Hales. 7 define and 0
by: (a, m)®(b, n) = (a + b, m + n), (a, m)0(b, n) = (an + hm, mn). (Ci exercise
3.3.3.) You can check that R is a commutative ring with unity (... what?) and
that (1,0), (2,0) divide each other in R but are not associates.

8 (i) in the units are 1, 3, 1, 11. Irreducibles are 2, 10. Primes are 2, *, Iii.
(iv) This example is fascinating! First note that f is a unit if f(x) 0 for all x.
There are no irreducibles: for any suchf must havef(a) =0 for some a. Defining
I by: I(a)=0; I(x)= 1 if we findf(x)I(x)=f(x) for all x, i.e.f=f I. Now
show thatf is prime ifff has exactly one zero. (Show that iff has more than one
zero, then you can find functions g and h such that f=gh but f,j'g and fth.)

10 Yes! (2i) Let it be a prime and u and unit in /[i]. Then unit.
Finally—if untz=41 then or intl so uinluvcx or uxluvfl, where uv= I.

12 (a) Follow 1.3.7. End more formally by: 'It follows that asc = bc. But hc = a.
Hence asc=a=ae. Therefore a(sc—e)=z and so sc—e=z (why?). Conse-
quently sc(=cs, why?)=e; that is, c is a unit and a is irreducible.' (b) 2 is
irreducible (see text) but 2 isn't prime since 21(1 + — 3)( I — —3) yet
2,f'l in /[\/_3]. (c) Since is irreducible the only divisors are u and
ccu where u (= ± 1) is a unit. Likewise for fi. Hence the only common divisors are
I, — I. (Note that /7 are not squares since they are irreducibles.)

13
 Exercises 3.7 329

Exercises 37
1 (a), (b), (c). They are all domains. That 3.7.1(I) holds for ZbJ—i], ZW—21
has already been noted (Theorem 3.7.4). The proof for is similar via 3.6.3.
To check (II) look at (d=—l,—2,2). Let
a
Then = —==x+y\/d, where x,yeQ. Now choose X,
h u+v\/du—vVd
Yel such that 0 We obtain a=(X+ where

r={(x—X)+(y—
But N(r)=I(x—X)2—d(y— Y)2IN(b)=MN(b), say. For
where 0 h2 Thus, 0 M (Can you see why?) (A similar proof
copes with d= — 1, —2 and 3 but not with d= —3. Again, can you see why?).
(d) (II) For write a=mb+r where —IbI/2r<IbI/2. Then
so that b(r) < ... (f) For (II) note that if a = x1(a1 + 1x +
what? =
where are units then
(ifj i). (What if icj?)
2 Given define i5(a) = Then checking (I) and (II) of 3.7.1 is straight-
forward. Does ö (in 3.7.1) imply the existence of q as defined here? Try R=7L.
Now prove that and t(9)tj(3).
3 (ii) Fn=3+\/—2, r= The gcd is
5 Write Rf+ Sg = and then write R = mg + r(m,
1 r e Q[x] and deg r c degg.
r=0 is impossible. Why?) Then I =rf+(S+mf)g.
6 (fljlsing 2 is irreducible but not prime in
(ii) If then N(r)=1a2—10b21. Putting
± 4, 5(mod 10) show that a2 — lOb2 = ± 2 has no solutions. Thus N(r) #2. Show
similarly that N(r) #5. Hence show that 2,5, are irreducible in ZR/b].
7 There is no contradiction. (Perhaps the factors with equal norms are
associates?) The equality a2 — 2b2 = (a + — shows that elements
with norm I are units. (Cf. exercise 9(u).)
8 (i) Since (77,91,143)= 1, z= I will do. Which other z will do?
9 (i) From 3.7.1(I), 45(l)f5(la) if a#0. (ii) If also ab=1 then
=i(l).

11 If d12 and dl I in then N(d')14. Show that N(d)= 1 SO that

d= ± I .Any 1cm, k say, would have to divide the common multiples 4 and
2(1 + —3) and, hence, N(k) = 4, 8 or 16. There is no contradiction; the pair
doesn't have a gcd in Z[%J—3] (exercise 3.6.12(d)). Now note the
word 'each' on line I of the question!
12 If each irreducible is prime in D copy the proof of 3.7.1 1 to find D is a UFD.
If D is a UFD and if irc=ab, with it irreducible, write a, b,c as products of
330 Partial solutions to the exercises

irreducibles and use uniqueness of factorisation to show that it is an associate of

an irreducible factor of a or b.
13 This is rather long. You can find the proof in [9], Vol. I.

Exercises 3.8
I One of u,v must be odd, the other even: u=2r+ l,v=2s, say.
2
2k)(2k+l
3 If p = 4k +3 is not a prime in /[i] then it is not irreducible either. Hence
p = (a + ib)(c + id) with neither factor being a unit. Then from
p2 = (a2 + b2)(c2 + cE) would follow p = a2 + b2 = 4k +3, contradicting exercise I.
Ifp=4k+1 is a prime in 1, then p=a2+b2=(a+ib)(a—ib) where
4 (a2 + h2)(c2 + d2) = NOxjI) = hd)2 + (ad± bc)2.
11 009=(102+ 12)(l02+32).
7 Note that at step (ix) we would need an equality of the form r4 + s4 = t4.
9 Note that (x2+4, 16)= 1 in 7 since x is odd. Deduce that x+2i, x—2i are
coprime in /[i]. Note also that if y3 = c41 with /3 coprime in /[i], then = us3,
1i=vt3, whereu,ve {l, —l,i, —i}. Butu,vareeachcubesofunits, hencecz,/Jare
cubes.

Exercises 3.9
1 'Only if': Supposing that n is prime in 7 and that abe[n], then ab=tn for
some tel and so nia or nib. Therefore a (or b)e[n].
2 [x] is prime since iff1f2e[x] then f1f2—xg where we think of f1,f2,g as
polynomials in x with coefficients in l[y, z]. Nowf1,f2 can't both have non-zero
'constant' terms in l[y,z]. (Why not?) Thus or xif2.
3 (Last part) If[a] is prime, and 1 =b cm R, then be[a] orce[a];
that is, b=ar or c=as. Therefore 1 =arc or a unit,
so that a is irreducible.
4 There is no contradiction since R is not a ... what? (Note that (0, 1)
=(0,0).)
5 Each ideal is principal (3.7.16). If I=[f] thenf is irreducible, by
exercise 3. Hence [f] is maximal—for if [f ] [gJ O[x] then f=gh, where h is
not a unit in O[x] since [g] [f]. (Couldn't g be a unit?)
6 (ii) If {3(a+h\/—5)+(—1+\/—5)(c+d\/—5)}=[2] then I

(3a—5d—c=l and 3b—d+c=0)=t.3i1 in 7. Hence N(cx)$l. But (3=cxy and

 Exercises 3.10 331

— I + = cth)(N(cx) 19 and N(x)L61 whence N(cx) = 3. Impossible! (iii)

AB [3 . 3,3(1—5 1), 3R'—5 1), (V 5 +
+ 5—1)]. Thus 9,6 e AB;
hence [3]cAB. But ABc[3]....
8 (i) abe [x] then xr=ah=a1 . a,,,b1 . . . say, in terms of irreducibles. By
. .

uniqueness of factorisation x is an associate of an a This implies that xla

or xih; that is a e [x] or be [x]. (ii) Cf. exercises 3 and 5. (iii) Each PID is an
UFD. Now use (i) and (ii).
10 Let aeI1\12,beI2\11. Then aheI1nI2 but
ii If and select aeA\P,beB\P and proceed as in exercise 10.
12 No contradiction—the hypotheses of 3.9.6 don't hold!
14 Prove first that each non-zero ideal I of 74 is of the form I {201: le 4,},
where dim and dis the least positive integer such that e I. Next show that if I is
a prime ideal then dis prime in /. (Then I must be maximal—for if? e 7Lm\I then
djr in 1. Therefore I = for suitable &, JI e 74. (Cf. exercise 3.6.8(i).)
15 Let f= xk(ak + a,, + 1x + .) = say, where 0. If the non-zero ideal I
of has the elementf as one of its elements for which k is minimal then,
certainly, I =

Exercises 3.10
1 (i) Ycs!; (iii) no (check multiplication); (v) not I I. and have the same
image.); (vi) yes. (Perhaps you should try to find proofs!)
2 Checking M2: (xoy)Qz—(xy+x+y)Oz——(xyz+xz+yz)+(xy+x+y)+z
whcreas xQ(yoz)=xQ(yz+y+z)=xyz+xy+xz+x+yz+y+z. <Z,$,Q>
is indeed a ring with 'z' = —1 and 'e' =0. Thus for an isomorphism 0 we would
need zO = 0, eO = I. (Can you 'extend' this 0 to the whole of 7'!)
3 Yes. This is part of the (well-known) 'block multiplication' of matrices.
4 Intuition: has an element whose square is 1 + I (i.e. 2); surely
hasn't? Proof: For anx isomorphism 0 we'd have 0(1) =1 hence 0(2) = 2.
Suppose 0R/2)= a + bV3. Then 2 =(a +

X
6 Since (\-y xJ 'corresponds' to x+iy in 3.10.2 we should choose
x = cos(2rr/17), y = sin(2ir/17)—using De Moivre's theorem.
7 (i) °R4 (OR + = °RØ$°R4. Now add e(ORØ) to each side and use the
associative law. (ii) Let s e S. Since 0 is onto S, there exists r e R such that rqS =s.
332 Partial solutions to the exercises

Then (lR4)Os= lR4iOrØ=(lR r)4=rØ=s. (iii)

$(—a)4. Hence (— a)4 is an (hence the unique) additive inverse for a4. (vi)
çb:R—+S is 1—i and onto. Hence 4r':S—÷R exists, is 1-I and onto (2.6.5).

then rç5=s10s2=(r1 . Therefore (s10s2)01 =r=r1 r2 (why?)

=s141'
8 For Ø:Q—÷Q show that lçb= I and hence that and (n/m)ç5=n/m for
all For note that if Ocr—a then Oc(r—a)q5; i.e.
açA<rq5.

10 (i) (ii) {xt(b0+b1x+ S"): tel,

12 For each prime p take (Why are no two such
isomorphic?)
13 No. For example l[x] has only two units whereas O[x] has infinitely many.
14 (Last part) {2r, 2s} determines the element r/s of 0.
17 Let T be any subfield of F. By considering Sn T show that S T. Now
show that S T is impossible if Sc T.
18 If na=OR for all a, then n I Conversely, if I + 1 + + 1 =°R then
a(l + 1 + + l)=aOR=OR. If mn is the characteristic of R then (ml)(nl)
=(mn)l But if R is a domain then ml =°R or... . For the next part use the
binomial expansion and exercise 1.4.17. For the last part try

Exercises 3.11
2 (i) Certainly U must be an integral domain. Let u be a non-zero non-unit in
U (if there are any such). Then u=u1u2. . . u, is a product of irreducibles in U[x]
(why?), hence in U (see exercise 3.6.2). If u=u1 . . . . as products of
.

irreducibles of U (hence of U[x]; why?) then the u1, pair off as associates in
U[x], hence in U (why? why?). (ii) If u e U(u is an irreducible element with
no multiplicative inverse in U form the ideal I={xf+ug:f,ge U[x]}. Continue
as in exercise 3.4.22(b).
3 For the first part follow the text for l[x]. For the first part of the second
paragraph regard 0[x,y] as Q[x}[y], i.e. as polynomials in y with coefficients in
the UFD Q[x]. Note that xis prime and that 3x2, xI2x but x2,f'2x. The
final polynomial factorises.
4 Tryl+L
5 Any 'rational' root is of the form r/s where (r, s) = I and and si I in l[y].
The second polynomial has content y2 — 1.
 Exercises 4.2 333

6 1fF1 = u1P1, F2 = u2P2 with P1, P2 primitive and iff is a common divisor of
degree 1 of F1 and F2, in then there existsge such that gjP1 and
gjP2 in Now (ci 1.9.14), if P1=gh1 and P2=gh2 and ifg=(a/b)G then
GIP1 and Gil'2 in U[x].

8 Is Q[{x]] a UFD? See 3.7.14 and exercise 3.9.15. In fact the two rings aren't
isomorphic: P(x, y) = + xy + x2y2 + x3y3 +
1 lies in one ring but not the
other. (So what? They might be isomorphic in some unnatural way. Note that
P(x, y) has a multiplicative inverse in is a UFD. See [22].)

Exercises 3.12
1

2 Let be any real number which is not a root of any polynomial in Z[x]. (iv
and e are such numbers.) Set f(x) e R + ifff(cx) >0 in It (Cli exercise 1.7.4.) In
particular look at, say, x—3 for rx=e and ri—ic. Now have fun!

6 1eR4 implies that 1+1, 1+1+1, etc. are all in Rt Hence 0eV if
0 =I+1+ + I (n summands, for some n). Contradiction.
8 Clearly (?) Fj is not empty. If 0< FaIb and 0< FeId then 0< Dab and 0< Dcd.
Therefore 0< Dab& + cdb2 = (ad+ bc)(bd). Hencethe sum of... The answer to
.

the question is 'no'. (Find a counterexample in 0 with a=b= — 1.)

10 Using P2, 7L[x] is not Archimedean since there is no n e 7 such that
Note that, with P1, Z[x] is Archimedean.
11 Try 71[x] under P2. (How do you know that there is no 'funny' isomorphism
which shows that

Exercises 4.2
1 (a) Each is. Intuitively, for the first one: 'replace x by x+2 and add' is the
same as 'add first and then replace x by x + 2'. Same for multiplication.
Formally, ...? (b) Look at 10 and i4 iq5. .

2 For (i), (ii), (iii) the image 10 of I determines hO for each 11. Also
1Ü=â i0=1U=à. (In 4 this implies à=Ô or 1, for example.) For
the final part think in terms of nim.
3 Question asks: Is det(AB)=detA is det(Ad-B)=detA+detB?
4 (i) RU isn't empty (why not?). Given s2 RU let r1, r2 e R be such that
Then s1cais2=r1UEBr2U=(r1 +r2)0e RU. Likewise for 0. That
es1 RU follows from 4.2.6(u). Now use 3.4.2. For an example try R = 7, S 0
or 7[x]. (ii) Use r1Uer2U=(r1—r2)U.
334 Partial solutions to the exercises

5 For a, b eR, (a+b)(O a =(aO$hO)Vi =(aO)qi FH =

a(th/i)FJRb(Oi/i). (Why is each equality valid?) Multiplication is similar.
6 Let SinceOisonto,s=rOforsomereR. Nowcheck that
eos=sOe=s. Thus e is a (hence the) multiplicative identity for S. 'Ontoness' is
essential: try zO e
=
7 Given just ( you can say almost nothing. (That's why arbitrary maps are of
little use in comparison to homomorphisms.) If V is a homomorphism use
exercise 6.
9 would be in kerO if were in 0. Clearly k(x)=
kerO. Then {k(x)h(x):h(x)e 0[x]}
c ker 0. If f(x) e ker 0 write f(x) = m(x)k(x) + r(x) with deg r(x) I. Prove
r(x)0=0. In the Z[x] case k(x) would not be in kerO but 441k(x) would be!

10 (One half) If ker0={OR} and if r10=r20 then r1—r2ekerO by exercise

4(u). Hence r1 and so 0 is 1—1.
11 Since F has only two ideals {0} and F, the images are: (i) isomorphic to F
(see exercise 10); (ii) the ring with one element (exercise 3.2.4.)
12 —l+67x+I. 4+67x+x2+I is another.
13 rI=rl+ORerl+I=r2+I. Therefore r1—r2+i for some iel. For the
converse consult the proof of 4.2.14.
14 OR+I satisfies A3 and, given a+I, —a+l satisfies A4.

15 For properties (r), (s), (t) use r1 —r1 El; r1 —r2 El r2 —r1 el;
respectively.
16 x+S=1 +x+S (since 1 ES) but their squares x2-l-S and ??? aren't equal.

17 (i) 0—since x has been 'reduced to zero' in 0[x]. [Define 0:Q[x]—÷Q by

(1+ [x])0 =f(0)]; (iii) 4—since both x and 4 been 'reduced to zero' in Z[x].
[Define 0: /[x]/Z[x, 4] —+4 by (f+ [x, 4])0 =f(0)]. (i) gives a field, (v) doesn't.
18 'Only if': Tfg=hk where degh and degk I, then h+[g], k+[g] would be
zero divisors. For the converse see the proof of 4.2.1. (What if degg I or
g=0?) 4{x]/[g] is the field {a+bx+[g]; a,beZ5}, so it has 25 elements.
19 Iff1,f2 E andf1 thenf1 + [g] sincef1 f2 [g}.

20 Use the map given by (a+b12)0=a+bx+I. Check 0 is

well defined, onto and 1—I. Finally

((a + + = ((ac + 2bd) + (bc +
=(ac+2bd)+(hc+ad)x+I=(a+bx+l)(c+dx+ I)
—since bdx2 +1 =bd(x2 —2)+ 2bd+I = 2bd+l.
 Exercises 4.4 335

21 After finding r, s using the Euclidean Algorithm you should find the
required inverse to be
22 will do—but what is the smallest such extension?

Exercises 4.3
1 U maps R onto the subring RU_c S. Now use 4.3.1.
2 'Deduce':rekerUiffr+A1=0+A1andr+A2=0+A2,i.e.iffreA1 and
re A2. Note that which has mn elements since (m,n)= 1, is isomorphic

7/
,
[m]n{n]

to a subring of—s----.
[m] En]

3 Define R/K —* S by (r + = rU. Don't forget to check well-definedness,

i.e. that
4 0 is 1—1: if A1U=A20 and if a1eA1, then a10=a20 for some a2eA2.
Consequently a1—a2eI_cA2, so a1eA2. Therefore A1_cA2. By symmetry
Note that since IU={05}cB.
Let where A is an ideal of R. Let aeA and seS. Since U is onto,
there exists r e R such that rU = s. Let t = aU e AU. Then ts = aUrU = (ar)U e AU.
(Why?). Likewise st e AU.
Deduction: If R/M had any ideals strictly between R/M and {0R/M} it would
correspond to an ideal of R strictly between R and M (why?).
5 Since R is commutative and has a I so does R/P. Then RIP is a domain if
{x(=r1 +P),y(=r2 +P)e RIP and x y0R/p} {x=OR/p or y=OR/p}
i.e. if
rlr2+P=OR/P
7 (i) Since P is a prime ideal, R/P is a domain (exercise 15) and hence a field
(3.3.4). Now use the last part of exercise 4.
F[x]
9 Use

Exercises 4.4
1 IfyeB put D=y2—2>0. Ify4put v=V. Ify<4 put v=y—D/3(sO
0<vcy). Then v2—2=D—4Dy+D2/9>0 since 9—6y+D>0.
3 (i) Let s1=(a1,a2,...), s2=(b1,b2,...), s3=(c1,c2,...) be Cauchy conver-
gent sequences. For transitivity note that, given k >0, there exist M1, M2 such
336 Partial solutions to the exercises

that Ia,—bJ <k/2 for all and <k/2 for all j>M2. Then Ia,—c,I
Ia,—b11+lb,—c11<k for all t>max{M1,M2}. (iii) To shows is Cauchy first
prove, by induction, that 1 ; 2 for each x,. Then show that, for n > rn,
Hence

4 This is quite involved. See [50].

5 (i), (ii), (iii) of 4.4.1 are easy to check. The zero element is the upper section
of positive rationals; ev is the set of all rationals t such that —te C\V except
that if Q\V contains a rational a such that xa for all xeQ\V, then —a is
omitted from ev.
7 . ... —1a01, we have

(a(0,0), (1,—b\
8

+ +
..is meaningless only when 22
0) satisfy A3 and M3: For M4 (a, h) has multiplicative inverse
which .
a+b= 0, i.e. when (a,b) is
the element (0,0).
9 The map 4:C—dR[x}/{x2+l] given by (a,b)4=a+bx+[x2+l] shows this
(cf. exercise 4.2.20). 0 is (fairly clearly) well defined, onto, 11 (why?) and is
easily checked to be a homomorphism.
11 No! Use (1,0) to show that axiom Z fails.
12 Check the multiplication axioms.
13 (ii) Field! (Find, by experimenting, specific zero divisors in the other two
cases.)
14

17 What we're doing is constructing / formally from Vu{0},just as we built

0 from 7, l1 from 0 and C from l1. <rn,n> represents rn—n (even if rncn).

Exercises 4.5
1 al + b\/2 + + d\/5 = 0 + + cAJS)2. Deduce that
u2 and are rational.
2 (f1—g1)w1+
3 Since x3 —2 is M9 (the minimum polynomial for 0 = the result is
immediate from 4.3.5 (i) and the uniqueness part of 4.3.4.
4 Let 1 and letf1,f2 be the non-zero elements of the field (F,
say). Let b be any one of theft. Thenf1b,. . is the same set of elements (ef.
.
 Exercises 4.6 337

3.3.4). Thusf1 12. . 1bf2b. . Cancellingf1f2.. we get 1 =b'. Thus

x has u + I (distinct) roots in F and, clearly, no subfield of F has this
property.
S Writing q—1=mk+n(0n.ck) we find x"=l, so that n==O and, hence,
Clearly (ct/3)kt=1. If(txfl)51 then sIkL So s=uv, where ux=k and
pkv
vy = But 1. = (xfl)'"'
1 = I. But JJVY = I. Hence jJV =1 (why?). Thus
v= Likewise
1. u = k. The rest is reasonably straightforward.
6 Each cubic is irreducible. (Show neither has a root in 73.) Thus the factor
rings are fields by exercise 4.2.18. They are isomorphic by 4.5.8(iii).)
7 Clearly x + [P] isn't of order 3, nor of order 5 since
(x+ [P])5 =x(x4 +x+ 1) +x2 +x+[P} #0+ [P].
You can check the order is 15.
8 Thinking of the 9 elements of H as the 9 polynomials a + bx with a, b e 74

looks hopeful!
9 (i) Since say, we have x"—l=(xrn—l)g(x) where g(x) is of
degree u — in. Now x" —1 has u distinct roots in whilst xm — I and g(x)
have at most in and u — m roots respectively.
11 Recall: the non-zero elements of GF(8) satisfy the equation x7 = I.
12 Note that, if g(x) = r(x)f(x) + s(x) e then
0+s(u) g(x)# 1.

13 Cf. exercise 7.4.22.

Exercises 4.6
I (a) In Fig. 4.8 take a= 1. Then c= 1/b. (b) Use Fig. 4.8 to get -.Ju from u
(u=3, u=2+13, etc.). (c) (a,b) is constructible from the pair (0,0),(l,0) if
each of a, b lies in some field in some sequence 0 = 0 c R, where
each 0,2+1 = for some r e O,. The problem is to show (see (3.4.2(F)) that
each of a + b, ab, etc. lies in some such sequence. (Cf. exercise 3.4.19.)
2 Suppose
+cx15vju1+(cz21v1+ ...

Then each bracket is equal to °E since the form a basis for K over E. But then
each 0Qj = °F (why?)
3 First part: If (u, v) is the point of intersection of ax + by + c = 0 and
kx + ly + in = 0 where a, b, c, k, 1, in e then (in general) u = (bin — k)/(al— bk)
338 Partial solutions to the exercises

e Likewise for v. Conversely: If Or—C 0r+ 0 are subfields of 11 with

:0,]2 for each i and if a,, h, e 0, then the point P1 = (a,, b,) is construct-
ible from P(={pi'...'pr}). [Proof: If[O,+1:0,]=2 then there exists a (real)
r e 0, + such that r2 = a e 01. Clearly (r, 0) is constructible if (a, 0) is.)

4 Note that o(-

2) isn't a subfield of 0(cos 20°) (why not?). In any case
01=2. (So what?)

6 Put u=2cos(2ir/5)=z+ l/z where Then

since z4+z3+z2+z+ 1=0.

7 Provef(x) is irreducible over 0 directly—or note that, iff(x) = g(x)h(x), then

y'g(y) . y'h(y) where i= deg g(x) and j= deg h(x), yields a factorisation of
+1.

8 a) Use the fact that am — hn = 1 for suitable a, h e 1. (b) For (ii): notice how
p—i . .

the arises in exercises 6 and 7. Apply the same technique to (i).

2

9 (a) For (i) use angle or arc bisection. For (ii) use exercise 8. (b) 2° is not
constructible; 3° is! (Reason?)
10 (a) Use Pythagoras' Theorem.
11Given the angle 0, split 0 into equal parts for any n. Then look at
0—0/2+0/4— ...

Exercises 4.7
I (a) —353; (d) 5s5—3s4s1 +5352. (The method of exercise 4 is quite quick
here.)
2 (i) Coefficient of x2 in the new equation is _((_3)2_2.7); (iii) Use
I I I 1 1 1

a
+-+ =—'
b c 53 ab
+—+
bc ca 53

3 (i) follows since m is chosen as small as possible. Then . . , Sk) = 0 yields

on putting xk=O.

5 Put Ur(Xi xr+,÷i,O,...,O)=tui(xi,...,xr+,)xt+,+i. Then

Ur(Xi, . . . . . ,0)0 u0(x1,. .

. ,xr+t)O.
 Exercises 5.3 339

Hence

Ur(xi,...,xr+,+i,O,...,O)=xr÷t+1V(xi,...,xr+,+i)[forsome V]=

• ,Xr+t_i,Xr+t+i)#+r.

Therefore

O=Ur(Xi,...,Xr+t_i,O,xr+t+i,O,...,O)(why?)
=xr+,+ixr+tvo(xi, . .

Consequently Ur(Xi,...,Xr+t+i,O,...,O)=XY+t+1X,+1W(Xi,...,X,+I+i) for

some W. Continuing in this way yields
Ur(xi,. . . ,Xr+t+i,O, . . . ,O) Xr+t+ixr+t.. .x1Z(x1,. . .

for some Z. But this is impossible! (Why? Look at degrees.)

Exercises 5.2
I The coefficient of x in the new equation is s1s3 —4s4.
2 The constant term in the equation for the first function is —s3s2s1.

Exercises 5.3

I (a) No: is a problem; (c) Yes: identity is 37; (d) No. 0 is the identity
but a has no multiplicative inverse ifa=... what?; (e) Yes: g is the identity; (f)
Yes; (h) No; (j) Yes; (I) Yes: f1(x) is the identity.
2 (b) Not abelian; finite order (in fact how many elements?); (e) abelian—
order 4; (g) not abelian—order 48 (why?); (k) abelian—infinite order; (1) not
abelian (look atf2f4,f4f2)—order 6.
3 Clearly (C) is satisfied. Also
(g1, h1) {(g2, h2) (g3, h3)} =(g1, h1) . (g2og3, h2sth3)
=(g1o(g2og3), h1s4h2*h3))
=((g10g2)0g3, (h1*h2)*h3)=
={(g1, h1) (g2, h2)} • (g3, h3)
—hence associative. (eG, eH) is the identity and (g, h)1 1,
h
- 1). Hence (N)
and (I) hold.
5 Yes. Prove it! (e6 is still the identity; likewise for inverses.)
6 Order is surely 12 (or is it 24???).
340 Partial solutions to the exercises

7 Let Aut(F) denote the set of all automorphisms of F. Then <Aut(F), o) is a

group. (C and I follow from exercise 3.10.7 (vii) and (vi).)
8 Closure holds by exercise 3.8.4. Axioms A, N, I are fairly trivially satisfied.
10 For all xe 7,b(7.d(x))=a(cx+d)+b= Tac.ad+b(X). T10 is the identity.
12 There are 3, two being the identity permutation and (123).
13 If IS> 1 then <S,o) is not a group. (Look at e°x, e being the supposed
identity.)
14 No! Think along the lines of exercise 13.

Exercises 5.4
I
3 Let a,beG. Then a=a', b=b', and so ab=(ab)'=b'a1=ba.
5 Hence the g1g (1 comprise n
distinct elements of G. But I G = n....I

9 (i) Look at a specific aeG. Now,

1)—i
a 'a=(a'a)e=(a 'a)(a1(a') ')=a1a)a 1)(a
=a'(a1)' =e.
Now show ea = a. Therefore the right neutral and inverse elements are double
sided.

Exercises 5.5
1 Try lf=2, 2f=3, 3f=? (Is there a 'smaller' example?)

2 (12463)(58)
3 (a)(1 547 3)(2 6);(b)(l 547 3)(2 6);(c)is?

4
1234567
3 4 7 5 2 6)

S X and Y 'XY have the same 'cycle shape'. Indeed see exercise 6.

6 (i)f'(x1 xr)f sends each xjsuccessively to to to i.e.

xj to 1f overall. (ii) To get the correct 'general' f, try experimenting with
specific examples.
7 (i)Ifftistheidentityandiffl=ctfl(0r<t)thencx=flft_r.
 Exercises 5.6 341

8 If y is common to the orbit of cx and $ then cx, $ are in the same orbit—_namely
y's—by exercise 7(i).
9 For the first part see 5.5.8. The even permutations include (123) and
(12)(34). By 5.5.11 you should find 12 altogether.
11 Note that the second (fourth) equality follows from the first (third).
12 To see if (i) can be changed into (ii) first change (i) (or (ii)), legally, by
moving the blanks to the same 'place'. Then (ii) will be obtainable from (i)—
legally—if the corresponding permutation
(14 11 7 1 12 2 B 9 6
12 7 13 4 11 B 14 6
is even. (See Thomas Fournelle, The permutation game. Pi Mu Epsilon Journal,
5, 1973, 425—9).

Exercises 5.6
1 If S is a subgroup and if a, be S then a' e 5, hence a tb €S. (5.6.5 (ii) and
(i).) For the converse, first show e eS, then b' e 5, finally ab eS.
2 Assume that a, b e S ab e S. Deduce that aMeS for each de t and, since
b'=e for some that alft eS.
3 If and look at ab with aeA\B and beB\A.
4 If am=e and b"=e look at (ab 1)mn (Is the set in question non-empty?)
5 Use exercise 2. TI=l.1.3!.lO! (correct?)

7 Clearly T G and coincides with the restriction of o to T x T. Also

<T, = <T, is a group—it is given as a subgroup of <5,

8 (a) Lagrange's Theorem will tell us we need only look for subgroups of
orders 1, 2,4, 8, 16. <a4> and <bat> (i =0, 1,. . , 7) are those of order 2 (since
.

<a2>, (i=0,1,. .,7)are those of order 4. (b) In S4:

.

The different cyclic ones are: <(1234)>, <(1243)>, <(1324)>. (Why no more?
What about non-cyclic ones?) In 55 each (labca) generates a 5-cycle where
{a, b, c, d} = {2, 3,4, 5}—surely not all different?

9 For 56: Note that each element is a product of disjoint cycles of lengths
x, y,... involving 6 letters. Now show that one cannot write 6= x +y +
where lcm{x,y, .
.
.} = 1.2.

11 The order is inn if (in, n) = I. Otherwise be careful. For example, it's

possible for a, b each to have order 6 and for ab to have order 1 or 2 or 3 or 6.
0)
12
342 Partial solutions to the exercises

14 (ii) ab—b '(ba)b. Now use part (i).

16 0/! is infinite since, for each n e V. 1/n has order n exactly.
17 If x=(abc. . .)(. . .)( is a product of disjoint cycles then x(ab)

18

19 (i) (12), (13), (123){ =(12)(13)}, (132){ =(123)2}, (23){ =(123)(12)} all lie in
the subgroup. Any more? (ii) The order is 12.
20 {a,ba}, {a3,h} are two such.
21 For n=3 choose 6, 10, 15. (Why? Factorise these integers into primes!)
22 If {a1/b1: I i t} generated and if p,{'h,, 1 i t, p being prime, how
could you show that l/pe
24 flSFBut<U)isoneofthes2....
ACA

25 (12.. .n)=(23) and (23)(l2)(23)=(13). Obtain (34) and

.

(14); (45) and (15) etc. similarly. Then use products of transpositions.

Exercises 5.7
1 S—SI, S(12)(34)=S(243)=S(143), S(13)(24), S(14)(23) are the distinct right
cosets.
3 Therefore gh*HcgH__and conversely since
e
4 cecH. So, if cH=dH then c=dk for suitable keH. Conversely t'c
5.7.4).
S
(i) eH? Try a=(12), b=(123), H=??. For (iii) think of
53—again!
6 If texAnyB then xA=tA, yB=IB (why?). Hence xAnyB=tAntB=
t(AnB) (Why?) Then follows, by counting! If
then each xHnyK must be a coset of HnK, hence
non-empty. Thus if heH, keK we have Hence h'k=k1h1,
where h1 e H and k1 e K. From this HK KH, which is therefore (proof?) a
subgroup of G. Now prove, by similar means that HK is actually equal to G.
(The converse, namely that if G=HK then IG:HnKI=IG:HIIG:KI follows
using exercise 7 and the equality IH:HnKI=IHK:KI.)
 Exercises 5.8 343

7 Given G=g1Kug2Ku ug,,,K and K=k1Huk2Hu uk,,H as Unions

im.j=n
of distinct cosets, prove that G = U is a Union of distinct cosets (cf.
i= 1,j= 1
4.6.1).

8 The a1H (i=0, 1,... ,n) can't all be pairwise distinct. Hence arH=aW for
some r, s. Note that t4'n tic possible.
9 If G={g1h:l then

10 What is it that '... divides the order of a group.'?

11 The order is the 1cm of {1, 2, 3, ..., 12} (=27 720?)

Exercises 5.8
1 Ifa,beG=<x), cyclic, then a=f, h=xs for some r,s. So
—9

3 If (x3)" = e then 1013k. The smallest such k is 10.

4 y=xtforanytsuchthat(t,15)=l.
5 0(n)—since = iffy = x' with (t, n) = I and 1 t n. For (ii), note that
0(n) =3 is impossible (use exercise 2.5.8).
6 (88, 28) = 4. Hence n = 4 will do. (So will n = 2. Why?)

7 Let a (,Ee)eG. Then <a)G. Hence G is cyclic. But, also, <a2>=e or

<a2> = G. Hence G isn't infinite cyclic.

9 Try our old friend (the smallest non-abelian group) again!

10 If <0, +> were cyclic with generator then x/2 0. Now let a=x/z,
b=y/z where x,y,ze land z>0. Let (x,y)=u=rx+sy for suitable r,se7L and
set c= u/z. Then ce <a, b>. Conversely, writing x= cxu, y—flu(e 7) we have
a=czc, b=fic. Hence a,be <c>.
12 Np! If a + b12 were the proposed generator what power of it would equal,
say, or 1+12?
14 Clearly is ihfinite and a group. If P='(t> is cyclic, then I for
some k — impossible! Let S be a subgroup of P. Then either S contains fth roots
of unity for arbitrarily large n or there is a k such that S contains no psth roots of
unity for any s>k.
15 The given number is prime!
344 Partial solutions to the exercises

Exercises 5.9
1 Let <G, > = be the (multiplicatively written) infinite cyclic group
generated by its element x. Define 0: <1, + > —* <G,) by 0(n)=x". Then 0 is an
isomorphism. If and <H, 0> are each infinitc cyclic, use 5.9.4.
2 See 3.10.2 (iii) and exercises 3.10.7 (vi), (vii).
3

4 The subgroup <(1 2 3 4 5 6)> is certainly abelian. Is either <(123), (45)> or

<(123), (34)> non-abelian of order 6?
5 One is—the other isn't. (I leave you to find out which is which!)
6 One is isomorphic to C4—the other is not.
7 <M(14), 0> is abelian and has order 4$14). $> is cyclic of order 6.
8 4i:<Z,c>—*<Z, +> given by Ø(z)=z+7 (or is it z—7 ??) looks a likely
candidate.
9 Look at the map given in 3.10.2(i).
10 Cf. exercise 3.10.6.
11 (a) Yes. Try the obvious(?) map. (b) Suppose . 7), the p1
being distinct (positive) primes. Show that there is an isomorphism between
and <Z[x], +> via +oçf. Now try to use the fact that
each of is, like 7, a UFD.
12 No! (Use the fact that Q[x] is countable whilst 1k is not—or determine, if 0
were an isomorphism with q5(r) = x, what would be.)
13 The fields are isomorphic since each is
isomorphic to Q[x]/[x3 —2]. Hence their multiplicative subgroups of non-zero
elements are isomorphic.
14 Let IA I = 6. If A has an element of order 6 then A C6. Otherwise A can
only have elements of orders 1, 2 and 3. If it has only elements of orders 1 and 2
then A is abelian (exercise 5.4.3) and has a subgroup of order 4 (t6). Neither can
all elements e) have order 3 (since i,j 2)} produces 9 (distinct?)
elements). So suppose a,heA with IaI=2, hI=3. Prove that
0j 2} is a set of 6 distinct elements of A. Hence show that a 'ba—b or
b2(=b 1) [why?] In the former case A is abelian. In the latter A
15 (i) No. (Look at the order of each element and then at or and ca). (ii) No.
(Assume so. Find the orders of a and T. Deduce the 'group' would
be C6.) (iii) If so then it must be cyclic of order 5 with 0 as the identity. Is it?
16 Define
 Exercises 5.10 345

17 Taking the origin at the centre of the cube, note that the larger group
contains the rotations of exercise 5.3.1 together with 24 isometries r1R,
where R denotes reflection in the origin. (What about the rR if R is a reflection
in the y—z plane? Doesn't that yield more isometries of the cube?)
18 For the first part recall exercise 5.3.7. For x,yeG, (xy)i/ig==g'xyg
Now prove that tug is 1-1 and onto. For all xeG,
Now show Hence
Jnn(G) is a subgroup.

19 See, for example, [8].

20 For the deduction use the first part and 5.9.6.

21 There are only finitely many essentially different n x n multiplication tables.
Alternatively ask yourself: How many subgroups (roughly) has

Exercises 5.10
I (ii) Use det(AB)=detA detB.
2 Omcanbeontoforexampleif<A, +>=<Q, +)orif<A,
(m,n)=.., guess what?

3 Let u =gç&, v = hi/i e Gi/i — a homomorphic image of G (abelian). Then

u. v=g hi/i =(gh)çli= (hg)çli=h =v u. Hence G is abelian. If =e then
Hence the order of gçli divides n. One specific instance is
trivial (and that's a hint!)
4 Let S3q5 be a homomorphic image of S3. By 5.10.6, kercb=<e) or A3 or 53.

Thus S3q5 = S3 or some group (!) of order 2 or the trivial group.

5 (ii) For any such i/i, li/i must have order dividing both S and 12. Hence
(iii) <M(14), 0> under 4, say. (See exercise 5.9.7.) Hence
i/,=Ao4:l—.M(14) will do—where
7 Note that i/i can't be onto (by 5.10.10 (iv)). Try t/i:C2—*53.
8 Map odd permutations to — I, etc.
9 If then is determined
since the are known.
10 Ifx,yekerçfr then(x_1)i/i=(xçfr)_'=e=ewhilst(xy)çfr=xyi/iee
= e. (Question: Is ker i/i non-empty?)

II (i) .c=: Let N=g1Ng. Then, given gnegN, we find gn=g(f'n1g)—n1g

for suitable n1eN. Thus gNcNg. (ii) Let meM. Then g1mg=neN.
Therefore m=gng' egNg'. Hence McgNg'.
346 Partial solutions to the exercises

12 If SA (abelian) and if SES, aeA, then a'sa=sa1aeS. 5.9.2(iv)

provides the smallest example of the desired sort.
13 (a) Each element of 5.3.4(1) is of the form where i=0, I;
Hence (since htab=a1).
14 ForgeG and
15 For geG, g'Sg is a subgroup of G of order 20—hence g'Sg=S. The
number 20 has no significance.
16 If ae<HuK>, then a=h1k1...h)c, Hence g'ag
17 V<A4 (example 5.3.4(m)). Now usc 5.10.9.
18 Let 06 Aut(G) and !/i9 e Inn(G). Inn(G) Aut(G) (exercise 5.9.18). For
all xeG, Therefore
U Inn(G).
19 Clearly each Hg1g is one of the Prove that Hg1g = if Hg1 =
(use exercise 5.7.4). Thus is a permutation on X. Clearly Hence
(uv)Vi = = = uiliorn/i. Now k e ker Vi if = if
(h)g1kg1'(h')eH—for all g1 and all h, that is, if keg1Hg for all geG.
Finally, if NH with NCG, then N= flg'Ngc flg'Hg.
gEG gcG

20 The subgroup of index 4 leads to a homomorphism i/i of G into the group

S, of order 4! Consequently ker Vi is not G—nor <e> (why not?).
21 For n x n matrices A, B it is easy to check that tr(AB)=tr(BA). It follows
that tr(C'

Exercises 5.11

2 Put x=(T) eS4. Then x - 1(12)(34)x=(ab)(cd) [cf. exercise 5.5.6] cV.

or S3. But
3

4 If (aT)" = T then a" e T. Therefore a" (and hence a) has finite order.
Consequently a e T, i.e. aT = T.
5 H<G then
SG/N=t'S=K/N: Let K={k:kNeS}. Trivially Now use
=k1N(k2N) Hence K G (and K/N=S, by definition).
6 Let gA = Then A is a homomorphism, since = 'P (exercise 5.9.18).
gA is the identity map if (i.e. if g 1xg=x) for all XE G.
 Exercises 6.2 347

7 (a) Take N1 <N2cG, G being infinite cyclic: (b) Try G=D4 with
=1N21=4.
8 Let leT, so that l=a* for some aeG. Then gl/i=t iffg=ak where keK.
Thus I KI elements map to each element of T.
9 Given gN = hN show that gK = hK. This shows that 0 is well-defined.
ker 0= {IN: tK =eG/K} = {tN: 1€ K} = K/N. By 5.11.6,

Exercises 6.2
gP&
I Let g e G. Then = eH = eH = eG for some fi.
2 No prizes for guessing which group!
3 (a) (Transitivity) For subsets U, 1", WofG: Vand Vh= W)=t. U9h= 14'
4 Let S G. The map A. given by u2=g 1ug (u eS) establishes the isomor-
phism between S and g 'Sg.
5 Ifx,ye NG(H) then W=H, IP=H and H
so that xy -' e NG(H). (Have we yet used that H is a subgroup?); (b) Take H = G
(any G!)—or, for example, {I,(l2)} in 53; (d) Look for a suitable subset of 53 to
act as K.
6 For h e H and v e NG(H), v - 'hv e = H. Since NG(H) is the set of allg such
that = H, 'unique largest'-ness follows.
7 For x,ye T, W=IP i.e. iffxyt eN6(H)nT.
8 U is a union of subsets, each of size HI and each
geG
containing eG. So the union contains at most IHI+(IG:HI—l)(IHI-—1)
elements.

9 Look at g '(1234)g=(1234) where There are 4 such g.

10 g e fl CG(x) means that g centralises every element x of G.

xeG

12 (6.2.6). Hence there is such that fyI=p. But <y)<G

(exercise 5.10.14). Set
13 For 54, if not exercise 5.5.6 might help. D4 has classes {I}, {a2}—and
three others, each with two elements.
14 Let xeHnC1 (x#e). Then H<G and is a
complete conjugacy class. It follows that H C,. But I C,1 =f for some integer
c 0 and H contains at least one class with just one element.
348 Partial solutions to the exercises

15 Let x e ((G)\S. Then KS, x> G (since <S, x) 1> SI) and so each element of
G is of the form x1s, where seS (why?). Since eS we
deduce that S<<S,x). Note that exercise 12 says 'there exists'; exercise 15 asks
you to show 'for all'.
16 Take n=4 in 5.3.4(1). (There is another example in Section 5.9.)
17 Let teNG(S), and seS. Prove that
=r'x(tsr')t {why?}=(r1xt)s. It follows that r1xieC6(S). Let teN6(S).
Then defined by gives an automorphism of S (cf. exercise 5.9.18).
The map 0: Aut(S) given by tO = A, is a homomorphism with ker 0
={t:r'st—s for all seS}.
18 Note that for eachgeG.
19 No!
20 If P is a Sylow p-subgroup in G then surely also in each T for which
P T G. If P, Q are Sylow p-subgroups of NG(P) then Q = P' for some
teN6(P) (by 6.2.12). But P'=P.
21 Let K = {P = P0, P1,. . .
be the set of all Sylow p-subgroups. Put
if P7 = for some s e S. As in 6.2.12 the conjugacy classes have IS:
(as power of p) elements in them. Thus at least one P1 is in a class of its own.
Consequently S N6(P,) for some P,. Now look at P1S to deduce S P,. Hence
for somegeG, by 6.2.12.
22 Suppose x'NG(P)x=NG(P). Then x1Pxx1NG(P)x=NG(P). But P
= x - 'Px (by exercise 20). Therefore xc N6(P).
23 Look at the number of Sylow 3-subgroups. 5. The
number of Sylow 5-subgroups is I or 6. If 6, count the number of Sylow 3-
subgroups. In the last case follow the proof of 6.2.l6(iii).
24 One of the subgroups is example 5.3.4(m). (Recall that conjugate sub-
groups are isomorphic.) There is no contradiction (since 8124). [Do you see why
I'm saying this?]
25 whereh1eHnK and k1 belongs to asetofcoset
representatives of K modulo HnK. Thus {hk:h e H, k e K} contains at most
I HI K: HnK distinct elements. (Are they pairwise distinct?)

Exercises 6.3
1 Generalize the solution to exercise 5.3.3.
3 (a) Yes. (The proof is easy.); (c) strictly, 'no' (n-tuples of matrices aren't
matrices!), but if we allow the words 'isomorphic to' then the answer is 'yes'—
for example, with A e GLm(R), Be GLJIR) associate
fA 0\
in GLn+m(11).
B)
 Exercises 6.3 349

4 (a) No—the largest order is 30; (b) A5 has no element of order 15.
5 (a) IfS4=A x B with Al 1BI, then IAI=2, 3 or4. But then S4 would have a
non-trivial centre—contradicting exercise 5.6.17. (b) Does C2 x D3 help? (c) Try
to generalise (b). (e) No! Count the subgroups of order p.
6 Show that if Cr, are generated by x,y respectively, then xy generates Crs.
8 (a) Map (a, (c, d)) to (a, c, d).

9 If g = . . . . . = k1 . ks..
. . . as given then, using the commutativity of
we get
n<H1,. . ,.H,,.
. . . , Hj = <e>. [Here denotes that the element below it is
'missing'.]
10 is one way. Any more?
Il I claim 16. Am I right?

12 No! Find an example inside C2 x C2.

13 (a) Prove is the internal direct sum A®B where A=<(1,0)>,

B=<(l, 1)> are infinite cycles. Now N B. Hence
(b) Z$7L—AeB where B=<(2,3)) and A=??
14 Map a + ib( = re'°) to (r, 0). [This won't quite work—can you see how to
amend the 0 bit?]

15

16 N=N1 x N2 is very easy. For the rest, take the map G1 x G2 to

G1/N1 x G2/N2 given by (g1,g2)—+(g1N1,g2N2). The kernel is
{(n1,n2):n1 eN1,n2eN2}—N. In particular N<G1 xG2.
17 The kernel is {geG: gN1=N1 and gN2=N2}={geG: geN1nN2}. If
N1nN2—'(e) the map is a 1—1 homomorphism into G/N1 x GIN2.

18 ((A x B) = ((A) x ((B). The proof is straightforward.

19 G has 1 +kp Sylow p-subgroups, where I (so that 1 +kp= 1, q or

q2). Now q, q2 are impossible since — I. Likewise for q. Thus G has unique
and hence normal) Sylow r-subgroups for r =p, q and so their direct product G
is then abelian since the subgroups are (see 6.2.7).
20 Left to you. Note that, even if all the G can still have elements
of infinite order whereds ?? can't have.
21and22 A=C2xC4xC4x...;B=C4xC4xC4x....ThesearenotisO-
morphic. (Look for an element in A which has a property not possessed by any
element of B. Try to phrase this formally.) Now let 0: A —÷ B map C2 to <e) and
the other C4s in the obvious way; let cli: B —. A map the first C4 to C2 etc.
350 Partial solutions to the exercises

Exercises 6.4
1 then

2 Try X=<(12)), Y=<(34)>, Z=<(12)(34)>.

3 b) 218 =2 x 109. Hence C2®C 109 is the only possibility—and it's cyclic
(why?). (c) The seven 'partitions' 1 + I + I + I + 1, 1 + 1 + 1 +2, 1 +2+2 etc. of 5
lead to seven different isomorphism classes.

5 (a) To have just 3 elements of order 2 the 2-subgroup must be C2® C8 or

C4ff3C4.

6 Since, for each odd integer a, a4rr 1 (mod 16), <M(16), 0> is certainly not
cyclic. Hence M(16) = C2 x C4 or C2 x C2 x C2. Which?

7 Example (which much better describes what happens than a formal explana-
tion can!): If A = C9$C8ff3C5 with generators x,y, z respectively, then one such
series, in additive notation, is (with corresponding factors underneath):
A=<x,y, z>><3x,y, z>>(y, z)><2y, z>><4y,z>><z>><e>.
C3 C3 C2 C2 C2 C5

8 Clearly we can concentrate on the Sylow p-subgroups of A and B. Suppose

.. . Theelementswith orders ")are
precisely those in the subgroups D1®D2EJ
where each D1(E1, . .) C1 is cyclic of order p(p2,p3, .). Thus if
. . .

= C'2 $ C'2 $... ® etc. one sees that m = n and, more generally, that has
neither more nor less cyclic factors of order p2 than does

9 Write A as a direct sum of its Sylow subgroups, S,,. Then each S,, is
cyclic- —or else S,, has more than p elements of order p.

Ii A can be generated by a and a—b. Try A=<a>G3<a—b> with a, a—h of

orders 4 and 2 respectively. Does this work?
12 Again, an example to show the way. If A is the direct sum of cycles of
orders 2,2,2,22,26; 32,33; 74,75,77 choose the n1 'backwards' as n5
=26.33.5277, n4=223275, n3=274, etc. For the last part see, for
example, [13].)

13 Suppose (F*,.> is S, x x S,—the direct product of its Sylow

subgroups—and that S,,, say, is not cyclic. Then S,, has more than p elements of
order p, all satisfying = e. This is impossible by 1.11.4.
14 Let G=C1® ®D5 where the are prime power
cycles. The pairing off of the direct summands of G®G with those of H®H
given by the uniqueness of the factors is easily seen to lead to a similar pairing of
the factors of G and H.
 Exercises 6.5 351

Exercises 6.5
2 [a,bc]=a 'c1b1abc:
4 [(142),(235)]=?
5 Useg1[a,b]g=[g'ag,g'bg].Notethata1(b'ab)eA;(a1V'a)beB.
6 Use (a'h'ab)O—(aO)'(bO)'aOhO and G'=<[a,b]: aeA, beB>.
7 implies "=[H°", Show that
= (GO)°° implies + "0 = [G°°0, by induction.
8 (a) Assuming that = x show that "= <[(a, h), (u, v)]) where
a, u E A and b, v e B. But [(a, b), (u, v)] = ([a, u], [b, v]). (b) G/(HnK) is isomorphic
to a subgroup of G/H x G/K (by exercise 6.3.17).
Now use part (a) and exercise 7.
9 'If': Assume that each 1/G1 is abelian. Then 1
(by 6.5.4). Now
prove, by induction, that for each i and deduce that = <e>.
10

11 Take the (restricted) direct product of groups of increasing soluble

length—see 6.5.6(iv).
12 Any such series Z = G0> G1> would have to have each finite
cyclic (since simple abelian—see exercise 5.8.7).
13 Take H,K of order 2 inside S3. If K<G then
'(h2k1k ')E HK.
14 C180> C90> C45 > > C5> <e>: C180> C60> > C4> C2 > <e> are
two such.
15 Let G/H>G1/H> ... >Gk/H>H/H and H>H1> >H,=<e> be com-
position series for G/H and H respectively. Consider G> G1> ... > Gk > H
>H1>...>H,=<e>.
16 Such a sequence would imply the existence of a normal subgroup of order 2
or 3 in S4. Contradiction.
17 Let 0:G GIN. Then Since the length
of GiN is m, we have G(m)N = N, i.e. G(m) N. But then = <e> implies that
<e>.

19 If GI .
.
the being primes then each composition series of G
has cx1+22+ +; terms—including G and <e). But (Or: let
> Gr = <e> be a composition series for G. Show that the series
G = G0> G1>
H=H0?HnG1 ... becomes a composition series for H when
the repeated terms are removed. Use the isomorphisms HnG,/HnG,+1
(HnG,)G141/G1÷1.)
352 Partial solutions to the exercises

20 C3 and C5 are Sylow {3, 5} subgroups of A5 since A5 has no elements of

orders 32, 52 nor 15. Note that 511A5: and C5 that neither index
is a if-number.

Exercises 6.6
1 Use exercise 5.8.7.

3 (Cf. exercise 6.3.19.) The number of Sylow q- (respectively p-) subgroups is

of the form 1 + kq (resp. 1 + kp) and divides p (resp. q). Hence there is one such
(the latter sincep>q (!!)). Hence G is their direct product (why?). Ifqlp— I, then
we still have P<G and G/P is abelian. Now use exercise 6.5.17.

4 [WLOG is forbidden since p2q is not symmetrical in p and q.} Now G has
N=l+kq(M=l+kp)Sylow q- (p-) subgroups where N=l orp orp2,(M=l
or q). Thus if M= I then there exists P'CG such that so that G is soluble.
If M=q then Thus N= I (in which case argue as in the case P<G) or
N =p2. But then p2 distinct q cycles leave just p2 — elements to form, with e, a
1

single subgroup of order p2.

5 All groups with orders less than 60 have orders of the form p", pq, p2q, p2q2
except for 24, 30, 40, 42, 48, 54, 56, Orders 42,48 and 56 are dealt with in 6.2.16.
Now, if I G 1 24, then G has a subgroup of index 3 and so there exists (exercise
5.10.19) a non-trivial homomorphism into S3. If = 40 then G has 1 +5k
Sylow 5-subgroups.
6 If A5 and N are distinct normal subgroups of index 2 in S5, then A5nN is a
normal subgroup of index 2 in A5. D4 has two such subgroups.
7 Each (finite) even permutation on 1 ÷ may be regarded as being in for
some n so that with Ar<As if r<s. If N<P, then is normal in

each in particular for n 5. If N > <e>, then NnA1> <e> for all t> some to.
Then, fors>max{t0,5} we have that NA5. Hence NP;
i.e. P is simple.
8 Show that for some n. (See exercise 5.9.20.)
10 Let H be a group with composition series of length 35. Take G as a suitable
(see exercise 8).
11 Each group of order n <60 is soluble (exercise 5) and so has a composition
series with prime cyclic factors. For n =60 compare the composition series of
C60 and A5.

In solutions to Chapter 7 we may sometimes assume a given polynomial is

monic without expressly saying so.
 Exercises 72 353

Exercises 72
1 The method reduces x11 — I to a quintic which may (or may not) be soluble
by radicals.
2 sof(x)=x4—2x2—l is the minimum poly-
nomial of This has two non-real zeros. But E2 c It
4 (ii) x4 — I Ox2 + 1 has zeros ± ± 124). 0 c OR/24) c 0(1(5 + =
gives the required splitting field and radical tower. (Why is — in

0(1(5 + 124))?). (iii) sin 72° =] 8 =

say. Then O(i2) contains

and hence = cos 72°. Hence S1 = O(i2) D 0(15) D 0. (Why is S1 equal to

(iv) Use exercise 4.6.7 to reduce the problem to a cubic and the methods
of Section 5.2 to solve the cubic. Then continue as in (vii)!!. (vi) the zeros are
cosO+isinfl where O=tir/6, t= 1,3, 5, 7,9, II. Thus i is a zero ofx6+ I and so
= (vii) Determine 2= — as

in Section 5.2. Then (Note

that fl= — 1/32.)
S Calculate (x + + \/3)(x + 12— \/3)(x — + \/3)(x — - 13) (for a sur-
prise!). Show that 13 e 0(12 + 13) (why?) so that 0(12, = 0(12 +
Note also that ±12±13c 0(12+13) (why?). [In particular 0(12+13)
=0(1(5+124)) cf. exercise 4]. For (ii) try replacing x by x+ 1.
-

7 Use R c C and 4.8.1. For the next part see exercises 7.9.
9 (a) Use exercise 4.5.5; (b) is irreducible over 77. Let 2 be a zero in
GF(73) (see pp. 166, 168). Then, in fact, in
GF(73)[x].

11 Each element of S is a ratio of polynomials in the ; with coefficients in F. If

the ; are the zeros of f(x) =f0 +f1x+ ... then, clearly, so too are the
a(crj. Hence, if S = each a(cx,) is one of the Thus a permutes the cxi.
12 (a) In 7.2.1 (cf. 4.3.4). Hence is finite.
GaI(R/F) is finite since, with R = F(r1, r2,. .. , rj, each must map to a zero of
its minimum polynomial in F[x]. (b) By the proof of 4.2.1, f(x) factorises as a
product of a linear factor and a polynomial f1(x), say, of degree n — I in F[x],
where [E:F] n. Now use induction on n.
14 Let a e Gal(S1/F) and 2 be a zero Then a(cx) is zero off1(x). If otx)
is also a zero of then (why not a(cx)?) in S1[x]. But
1= + v(x)fjx) for suitable u(x), v(x) e F[x]. (Why?) Substituting 2 gives
1
354 Partial solutions to the exercises

16 The modification is needed since F(y1), F(y2) need not be equal—only

isomorphic.
17 (a) Naïvely, just 'change each I to a 2'. A more highbrow way to establish A
is to map F1 [x] F2[x] -4 and show that the kernel of 2op is the
idcal
(b) (Loosely) Any factorisation of g2(x) in F2[x] can be 'dragged back' via
to an equivalent factorisation of g1(x) in F1[x].
18 so sends to
19 (a) The other zeros (in C) of the minimum polynomial of are
—12), —1(1+12), —1(1—12). Thus a non-identity automorphism of
0(1(1 +12)) must take J(l +12) to —1(1 +12). This then implies that 12 is
sent to ]2. So answers are: No; None!
20 Follow the hint—or show that a basis for L over K is mapped, by a, to a set
of elements of L which are independent over K. Thus the dimension of a(L)
over K is at least as great as that of L over K. Hence a(L) L is impossible.

Exercises 7.3
1 Try to get the answer by examining and
2 0(r) (cC) contains all the zeros r, r2 r" of the polynomial f(x)
1

+x+l, which is irreducible over 0. Use 7.2.3' with F1=F2=0

andf1(x) =f2(x) =g1(x) =g2(x) =f(x).
3 Since the answer is...?
4 The answer to one part is 'yes' and to the other 'no'. Which is which?
5 Informally!: (i) Identity and i\/3—÷ —i13; (iii) 8 of them. Briefly: i—+±i;
-

or (v) The minimum polynomial of over 0(13) is

x2 + 13. Hence there are two automorphisms (one being complex conjugation).
(vii) Note that 0(i13)=0(w) Hence there are three auto-
morphisms given by: l,w,w9.
6 (aJ There are two ways of extending 2:0 —÷0 to an automorphism of
0(12) to 0(12) (viz. —÷ ± then two ways of extending each such
to an automorphism v of 0(12,13). E.g. if = —12, then v sends
(12,13) to (—12,13) or (—12, —13). - -
(b) There are 8 automorphisms of 0(12,13,15) [but note, not of
0(12,13,16) so take care!].
7 (i) Looks as though the answer must be 3 x 3 x 2.
 Exercises 7.4 355

8 Calling the zeros r, wr, w2r one finds that, for example, the map given by
I r wr w2r
r—+wr yields the permutation 1
\cor r cur
2

9 (ii) 6 (ci x3—2); (iii) 8 (cz—*2, —oc, —ioc: —i where

10 (i) 4. (The zeros of x4+ I are (, where is a primitive eighth root

of 1.) Since x4+ I is irreducible over 0 (exercise 1.9.11), can map to any in
Sf. The image of determines the rest. (iii) x3 + 2x — is irreducible over 0 and
1

has only one real zero a, say. (Why?). So we have 0(a, /3, 71) = By 7.2.3' there
exists an automorphism a, say, mapping a to /3. There is also an automorphism
r, say, of order 2 due to complex conjugation. See which permutations of the
three zeros arise, recalling that Gal(51/0)S3 sincef(x) has degree three.
12 where 4 ($ 1)is a primitive pth root of unity and Q2 is
the (unique) real pth root of 2. 1
=p — and x" —2 is irreducible
1

over 0w). Hence, to each automorphism p of 0(() there are p automorphisms v

of 0((, extending p. Thus jGal(Sf/0)I =p(p— 1).
13 Take So [R:0]=65elGal(R/0)j. For (i),
note that = 0(]2, a)). Hence I Gal(Sf/0(\/2)) =6.
14 The minimum polynomial is x4—2x2— I (see exercise 7.2.2.). It is
irreducible over 0 and splits over I), i'jR/2— 1))
[= 0k1W2 + 1), i)—why?J. Hence I Gal(S1/0) 1=8.

15 Cos72 . So
=

Exercises 7.4
1 x2—2ex+(e2+ir2). lGal(S1/11)I=2—since Sf =...?
2
and belongs to Aut(L) and fixes F elementwise. So then, rather trivially,
does x. Thus Fix
4 (i) 0(12); (iii) The answer is not (why not?)
5 The given field is the splitting field over 0 of (x3 — 2)(x4 — 3). Use 7.4.8.
6 0, 0(13), Any more? What about 0('j2 + 13)?
7 (i) 0(x2); (ii) 0; (iii) 0.
8 IfcxeFix(<uG1>) then, certainly cxeFixG1 for each i. Hence cxen(FixGJ.
The converse is just as easy. For the other part try G1 = <a>, G2 = <t> where a
and r are as in exercise 7.
356 Partial solutions to the exercises

9 = Hence is the identity map on s1• Clearly a fixes and hence

also
10 See exercises 7.9.
11 (iii) Let cx=a0+a1r+ be a fixed element. If a1 is the
automorphism given by a1(r) = rt, then a1(cx) has coefficient of r1equal to a1
(since l,r, ..., form a ...? for 0(r) over 0). Hence

(To use 7.48 you must show f— I separable over 0.)

12 (vi) R = is a real field and so only two of the zeros of x8 —2 are in R.
Clearly, then, Fix(Gal(R/F)) =
13 (v) For Gal(R/F) I = 6 see exercise 7.3.6: for [R : F] = 6 use the basis 1, w,
to2, r, cur, w2r.

14 (i) No; (ii) Gal(R/0) = 4.

I I

15 By definition, each element of Gal(L/K) is an automorphism of L which

extends the identity map on K. Then 7.4.3, with K1 = K2 = K and L1 = L2 = L,
suffices.
16 This is straightforward (once each f(x) is provcd separable) since, for
example, in (iv) I, \/3, is a basis for over 0(=F).
18 Let g(x) be an irreducible factor off(x)=x"+ ... +x+ 1. Sincef(x) is of
+ 1)),
degree k and has k distinct zeros in its splitting field SO,- = we deduce
that g(x) can have no repeated zeros in its splitting field S9 S,. (This is much
improved in 7.5.1!)
19 Note that F c E, S1(F) Sf(E). Let be an irreducible factor off(x) in
E[x] and let .(x—cç) with the ; in S1(F). Let gp(x) be that
irreducible factor of f(x), in F[x], which has as a zero in 51(F). Then
in E[x]. (Otherwise I = for suitable r(x), s(x)
eE[x]—and then 1=0 in 51(E). [Cf. exercise 7.2.14.]) Thus in
51(E)[x]. But by hypothesis, = (x—/31). (x—flj with distinct from
. .

51(E). Thus the; are some of the and so are pairwise distinct in S1(E).
22 If is a zero of x"—t in then &'=t. Thus, in Sf[x], x"—t=x"—f
= (x —

Exercises 7.5
1 If f(x) = (g1(x))2' . in F[x] with the g1(x) being distinct and
. .

irreducible, then f(x)/d(x)=g1(x).. .

3 (The first part becomes trivial if one uses exercise 9.) Let g(x) e L[x] be
irreducible and let be a zero of g(x) in M. Let M2(x) be the minimum
 Exercises 75 357

polynomial over K. Show g(x)jMjx) in L[x} (cf. exercise 7.4.19). But Mjx)
factors into linear factors in M[x]. (Why?) Hence so does g(x). For the last part
let h(x) be irreducible in K[x] and have a zero, say, in L. Since h(x) splits into
linear factors in M{x] we only have to show cach zero j3, say, lies in L. But there
exists a GaI(M/K) such that a(cz) = /3 (why?). Hence /36 L (why?)
S so [L:FJccc. If [L:F}>l let g(x) be the minimum polynomial of
L\F. Then consider, as in 7.5.2, h(x)=fl(x—a(cz)) as a ranges over
Gal(L/K). Continue—as in 7.5.2!—deducing that etc. (see 7.6.3).
6 Clearly L is separable over K. To show M is separable over L let x e M and
let g(x) (rcsp. h(x)) be its minimum polynomial over L (resp. K). Show
in L[x] (cf. exercise 7.4.19). Let SL be a splitting field for h(x) regarded as an
element of L[x]. Then SL contains Si—a splitting field for h(x) regarded as an
element of K[x]. In SK[x], . (x—;) with distinct cc. How does
.

g(x) factorise in SL[x]?

7 That Artin's definition implies ours follows from exercise 5. Conversely take
S to be the splitting field of x2 — it over 72(t). The field S is normal in our sense
over 72(t) {look at {S:72(t)]}, but not separable over 72(t) (exercise 7.4.22).
Indeed Gal(S/72(i9) = <e) and so Fix...?
8 (ii)—*(iii) is just 7.4.7; (iii)—>(iv): If then [L:K}
(why?)[L:F] (by 7.4.4). But [L:KI<[L:F] is
impossible; (iv)—*(ii) is esscntially 7.5.2. For: Gal(L/K)j <cc (as in exercise
7.2.12). Now let g(x) be any polynomial in K[x] with a zero in L. Forming
h(x) = fl(x — as in 7.5.2, show that g(x) is a product of distinct linear factors
in L[x].
10 Clearly {L1 nL2 : K} < cc. Let g(x) e K[xJ be irreducible and have a zero in
L1nL2. Then for cceL1,
Thus the ; belong to For L1uL2: by exercise 9, L1 and L2 are splitting
fields off1(x),f2(x), say, over K. Hence <L1uL2) is a splitting field over K of
fl(x)f2(x).
11 The hypotheses imply that L is the splitting field over K of
M2(x). . . Now use exercise 9.
12 A non-separable splitting field? That of x2 — it over 72(t). For the other
part, there are three examples in 7.3.1. (See 7.4.2 if stuck.)
13 (a) Even in K[x], iff(x) (of degree 1) has a repeated zero in its splitting
field, then (f(x),f'(x)) 1. This is impossible (why?) unless f'(x) = 0, that is,
unless f(x) has the given form. Conversely: Suppose f(x) is as given. If
f(4=(x—2)g(x) then 0=f'(x)=g(x)-i-(x—cz)g'(x). Hence (b) If
(IC(—p' then, in (a) each But
358 Partial solutions to the exercises

has repeated factors.

14 etc. Hence 4) is a homomorphism. Trivially 4) is
1 1 (why?) and hence onto if K is finite.

Exercises 7.6
1 G(F(H)) = Gal(L/Fix H) H is immediate. Hence F(G(F(H))) F(H). But
is also immediate. Choosing M=F(H) we get
9

2 Write H for Gal(L/K)—which is finite. Then, by 7.6.3, L is a Galois

extension of Fix H = F. 7.6.4 then gives the required map.
4 There must be such a k by exercise 3. In the sccond part k can be chosen to
be one of I, \/2, \/3, \/6 (why?)
5 The may be chosen as non-trivial solutions to the system of m equations in
n unknowns (with m<n).
6 Trivially (H = )Gal(K/<E1 uE2>) Gal(K/E1)nGal(K/E2) = H1 nH2. If
then, by 7.6.4, there would be a field F such that
which is impossible.
7 Since K=FixG=Fix<G,cli> we have, by 7.6.3, IGI=1L:FixGI
Hence i/ieG.
9 The 'lower' (i.e. smaller) field is G(ir2); the 'upper' (i.e. smaller) group is
{ a3}.

10 I'll leave the pictures to you! That for (iii) is pretty beastly! The subgroups
joined to <e> are <a2), <a1t) (i=O, 1, 2, 3, 4, 5). Those joined to
Gal(51/Q) are <a), <a2, t), <a3, at>, <a3, at>, <a3, a2t), where, briefly,
a —÷ as), a being a primitive sixth root of
I—so that
11 (i) is very like 7.6.5; (ii) the zeros are ±r, ±ir where Working asin
7.6.5 shows that Gal(O(i, = C2 x C2. The lattices look like

0(i. r)\ where 4a3 cr4

0(i) 0(r) r r r —r
-i -i

12 S is also the splitting field of (x2 + 2x — 1 )(x3 — 2). Hence =

where SQ is the splitting field of g(x) = x2 + 2x — I. Thus Gal(S1/0) is C2 x S3.
 Exercises 7.7 359

13 x4 + 2 = (x + I)(x + 2)(x2 +1) in 4. So GF(9) is a splitting field for 1(x).

Since GF(9) is a Galois extension of degree 2 over GF(3) we find that Gal(51/7z3)
has order 2.
15 As many as there are subgroups of C2 x C2 x C2! (Why?)
16 S1 = 0(C) where C = has minimum polynomial x4 + 1 with zeros
(7• Define c, r by c(Ø=C3,
C, t(C)=C5. Show that Gal(51/Q)=
Since a(a+bC+cC2+dC3)=a+bC3—cC2+dC we see that Fix{c}=
{a+b(C+C3):a,heO} Find Fix{r}, Fix{0t} similarly.

17 The statement on fields follows from the fundamental I—I correspondence

and the analogous results concerning cyclic groups (see 5.8.2(i)).

Exercises 7.7
1 Surely [0(54:0(F)] = [Si: F] [0(Sf) : F].
2 n=9.
3 Gal(L/K) is trivial, since x5 —5 has only one zero in L. Thus Gal(K/0)
= S3= Gal(L/0) and Fix(Gal(L/0)) = DQ = Fix(Gal(K/0)).
4 The minimum polynomial of cx = + over 0 is x6 — 6x3 +7 (see
exercise 7.3.16). It has one other real zero, /1= and four non-real
ones. But /3 Q(cz). Hence Gal(0(cx)/0) = <e>. Note that = C2.
5 T is the least field containing V and 51=F(cx1,. . . , cxj. Since
T = V0x1,. . . , ;). If cx e T\ V then there exists 0 e Gal(R/F) = Gal(R/ V) such
that OOx) cx. But 0 acts as an automorphism U (say) on T (since 0(V) V and
0(S4c51) and so

6 0(12,13), 0(12,13,15), etc.

7 (This generalises 7.7.3 slightly.) M is a splitting extension of K (exercise

7.5.9) and hence of L and c(L). Now apply 7.2.4.
8 Let cx=J(1 +12). Then M2(x)=x4—2x2-- 1 which has zeros ±1(1 +12),
±1(1 —12). But](l —12)e0(,J(l (look at (1(1 On the
contrary 1(1 +s,/3),i).
9 L is a splitting extension of K (exercise 7.5.9) and so, as in 7.7.2, each
Oe Gal(M/K) restricts tp an element O of Gal(L/K). 0—. O provides a homomor-
phism of GaI(M/K) into Gal(L/K) with kernel the set of those automorphisms
of M which fix L.
10 By definition, m(x) splits in Sm=R(Ui, ..., ..., r5,u1, ..., uj.
Since r1 u1 are all zeros of m(x) in 5m' we can see that 5m is a
splitting field for m(x) over F.
360 Partial solutions to the exercises

Exercises 7.8
I The minimum polynomial Nrj(X), say, divides x" — and Mr1(X) and has
degree > (why?). Hence two of its zeros are (dcx,
1 where (is a primitive p1th
root of unity, cz is a pith root of and This shows that (e SM.
2 x12—1=(x6—l)(x6+l)=(x6—1)(x2+1)(x4—x2-F1). Hence b12(x)=x4
—x2+ 1 =fl(x—(1), where the (i= 1, 5, 7, II) are the primitive 12th roots
of 1.
3 where (isa primitive 12th root of 1. The other zeros of 412(x) are
('' as in exercise 2. The elements of G = Gal(51/O) are Id, p, a, t, where
and One shows easily that p2=a2=t2=Id and
that r=po to deduce that x C2. The fields are 0, O('J— I),
0(s) —3) and 0(s)3). [Hint: For Fix a use 1 + (4 = (2, for example.]
4 G?,(x) = fl(x — (1) where (i runs over all primitive nth roots of 1. The maps ai
given by = (i are the elements of the Galois group G, say, and the map
A: G —÷ 74, given by =1 (mod n) establishes the asserted isomorphism. (A is
clearly i—i and is onto since, if (i,n)= 1, then (1isa primitive nth root of 1 and
so, for that value of i, a1 exists.)
5 (ii) Sf=O(() where (is a 10th root of I whose minimum polynomial is
1)/(x+ 1)=x4—x3+x2—x+ 1. (Why is
this irreducible over 0?) Hence Gal(51/0) = 53,57, a9}, where a,(() =
Since = 59 and = 57 we have Gal(Sf/Q) = C4.
6 For 7.8.3 the map n) is a 1—1 homomorphism of Gal(L/K) into
<1n' +). The other assertion follows as in exercise 4 with the use of exercise
4.5.5.
7 Sf = 0(w, cx) where w is a primitive pth root of 1 and is the (or a) real pth
root of a. The elements of Gal(Sf/0) are the maps cr,j defined by a11(w) = ci,
(1 1,0 1). Now ak,ajj(04=wth whilst
= = S"co'cx. Therefore 5kl5ij= 51k Now associate with the trans-
formation on 77,, given by

8 Cf. exercise 7.6.10.

9 Exercise 7.7.9 says that if and L is normal over K, then
Gal(M/L) <Gal(M/K). Here = 0((, r) D 0(() D 0, where r is a real nth root
of a and (is a primitive nth root of I. Now use 7.8.3 and 7.8.2 on Gal(Sf/O(O)
and
10 Let (be a primitive mnth root of I. Then u = (" and v = (" are primitive mth
and nth roots of I. Define Pt by = (1 where (i, mn) = I. Then induces t1
given by a1(u) = u1, 'r1(v) vt—clements of Gal(Sf/0) and Gal(59/0) respectively.
Show that the map r1) of Gal(Sh/0) to Gal(Sf/0) x Gal(Sq/0) is 11
and a homomorphism. It is onto since (Why?)
 Exercises 710 361

Exercises 7.9
2 Since the a1 are independent over F, exercise 4.7.3 shows that we may
identify F(a0,a1,...,a4) with F(s1,s2,...,s5) of exercise 1. Thus as in
exercise 1. Hence Gal(S1/F(a0,a1
3 It doesn't mention over which field the polynomial isn't soluble by radicals!
Indeed all polynomials in Q[x] can be solved by radicals over IL but, of course,
it is not true that all polynomials in Q[x] can be solved by radicals over Q.
4 Each group G is isomorphic to a subgroup of some (Theorem 5.9.6). Now
use 7.6.4.

5 We need five real numbers which are completely algebraically independent

over 0. Take a0 =ir, say (since it satisfies no polynomial equation over 0 (see
(C) on p.174). Now take a1 to be any real number which is a zero of no
polynomial in Q(ic)[x] = 0(a0)[x]. Such an a1 exists since 0, 0(m), 0(ir)[x] and,
also, the set of all (real) zeros of polynomials in 0(ic)[x] are all countable sets,
whereas is not. Continue in this way to find a2, a3, a4. Then apply exercise 1.

Exercises 7.10
1 'Only if' is the normality part of 7.7.3 (or see exercise 7.7.9). 'If' is really
7.10.3 with F,K,E replacing [Or see exercise 5 below.]
2 The only such N is 0(w) since S3 has only one normal subgroup other than
53 and <e>.
3 Gal(51/0) = C6 (exercise 7.8.6). C6 has a normal subgroup H = {Jd, a} of
order 2, a being complex conjugation. Then [Fix H: 01=3. Suppose
Fix H = 0(fi), where /3 e C and fjfl 0 (n 3). [Why just one term in this radical
tower?] If M(x) is the minimum polynomial of /3 over 0 then M(x) has degree 3
and divides .? — pn Hence M(x) has zeros /3, where is an nth root of 1.
Then e Fix H. This is impossible since Fix H P (why?)

4 Assume that K = 0(fi), where /3" e 0, is normal over 0. Then G =

Gal(K/0) = C,,. Look at the a(fi)(c e G) which are distinct pth roots of /3.
5 'If': Suppose cx(K)= T. Then
c L. (Why equality7)
6 For Aut(F) 53 see exercise 5.3.1(1). That k e Fix S is trivial—and then the
rest is easy, too.
7 Let a, b be any two zeros of f(x) in Se.. According to 7.2.3' and 7.2.4 the
identity automorphism on F extends to one between F(a) and F(b), which itself
extends to an automorohism of S
362 Partial solutions to the exercises

8 Extend R to a normal radical tower N, say, over F. Sincef(x) has one zero in
we deduce thatf(x) splits in N[x].
9 The result stated implies that if the zeros off(x) are labelled 1, 2, . , p then
. .

the elements of Gal(S1/F) form a subgroup S, containing all v' in the solution
of exercise 7.8.7. Now show that ifcru,v(a)=a and Cu,v( b) —=b then u= I and v=0.
This proves that the only element of Gal(S1/F) fixing (F and) a and b is ea. But
Fix(<e6>) =
10 On adding the use the equality I ... +(S)" 1=0.
11 There is no contradiction:f(x) is soluble if all elements of P are given to you
'for free'—but not necessarily so if the only 'given' elements are those in

12 (a) Choose LDK such that Gal(L/K) is: in (i) C10; in (ii) D5. (D5 is a
subgroup of the group of exercise 7.8.7 when p=5.)
(b) (ii) Sf=0(i); (iii) try exercise 7.8.10; (iv)
exercise 7.3.13. For (i), the only normal subfield other than 0 and to) is
0(w).
13 (i) and (ii) factorise (how?) In (iii) replace x by — 2y.

Exercises 7.11
1 Use: Each permutation can be written as a product of disjoint cycles.
2 The real (sorry!!) point is: complex conjugation maps M to itself.
3 If one has zeros ± ± what do you think the other is?
4 No! Thisf(x) factorises in 0[x].
5 What about using a cubic, irreducible over 0 and with three real zeros, one
being negative: now replace x by x2?
6 Regard C 6 Gal(S/F) as a permutation of the cxi. Clearly c(8) = 8 or —8
according as C is an even/odd permutation. For (i) use C(82) = C(8)C(8). For (ii)
note that Gal(S/F) if each C fixes 8, i.e. if 8€ Fix(Gal(S/F)) =?
7 Combine exercises 7.9.1 and 7.11.6.
9 The values suggested in the text for c, d does the first! The second is similar.
10 True. By exercise 7.10.9, if more than one zero is real then
11 (i), (ii): Look for quadratic factors—there being no linear ones. (iv) to (ix)
are not soluble. (The derivative of (ix) has two rational zeros which can be
found by the rational root test. Investigate (ix) at these turning points.)
 Exercises 7.12 363

12 (a) If 4) hasorder d then 4A(cx) = for each 2 e L. But

cx
= 2pd
Consequently cx" =e. Now K* is cyclic of order 1 (exercise 4.5.5). Hence
d= n. It follows that Aut(L) is a cyclic group of order ii generated by 4).

Exercises 7.12
2 Note that if [L:K]< cc then L= K(cx1,. ..,;) for (finitely many) suitable
; e L (with each algebraic over K). Now apply exercise 1 n — I times.
3 (a) (iii) Recall that the splitting field is w).

4 7.10.2.
5 Suppose that [L: K] = n and that, by exercise 2, L = K(cx). Then cx has
minimum polynomial of degree n over K (see 4.3.4). By 7.2.3' there are exactly
= [L: K]) ways of extending the identity map on K to an element of
Gal(L/K)....
Bibliography

Despite its length this bibliography has no pretentions to completeness, as a

visit to your university/polytechnic/college library will show. All the items
listed are either mentioned in the text or are items you might enjoy reading
just for fun. The point of producing such a list is neither to dazzle nor depress
you with its length but to give you a wide choice of where to start reading.
Because of its length the list has been split into subsections. In some cases
items listed under one heading could easily have gone under several others.
There exist a number of biographies/obituaries in book and paper form of
several of the mathematicians referred to in the text. We have not included
mention of these as most can be tracked down through [134].

Algebra
1 Adamson, lain T. Introduction to Field Theory. Oliver and Boyd. Edin-
burgh, 1964.
2 Albert, A A (ed). Studies in Modern Algebra. Mathematical Association
of America, 1963.
3 Artin, Emil. Galois Theory. University of Notre Dame Press, Indiana,
1964.
4 Bastida, Julio R. Field Extensions and Galois Theory. Addison-Wesley,
California, 1984.
5 Birkhoff, Garrett and MacLane, Saunders. A Survey of Modern Algebra
(3rd edition). Macmillan, New York, 1965.
6 Burton, David M. A First course in Rings and Ideals. Addison-Wesley,
Reading, Massachusetts, 1970.
7 Childs, Lindsay. A Concrete Introduction to Higher Algebra. Springer-
Verlag, New York, 1979.
8 Clifford, A H and Preston, G B. The Algebraic Theory of Semigroups, Vols
1, 2. American Mathematical Society, Rhode Island, 1961.
9 Cohn, P M. Algebra Vols 1, 2. Wiley, London, 1974 and 1977.
10 Coxeter, H S M and Moser, W 0 J. Generators and Relations for Discrete
Groups. Springer-Verlag, Berlin, 1965.
11 Dehn, Edgar. Algebraic Equations. Columbia University Press, New York,
1930.
12 Dickson, L E. Algebraic Theories. Dover Publications, New York, 1959.
13 Fraleigh, John B. A First Course in Abstract Algebra. Addison-Wesley,
Reading, Massachusetts, 1967.
 Bibliography 365

14 Gaal, Lisi. Classical Galois Theory: with Examples. Markham, Chicago,

1971.
15 Garling, D J H. A Course in Galois Theory. Cambridge University Press,
Cambridge, 1986.
16 Gorenstein, Daniel. Finite Simple Groups: An Introduction to their Class jti-
cation. Plenum Press, New York, 1982.
17 Hadlock, Charles Robert. Field Theory and its Classical Problems. Carus
Monograph 19. Mathematical Association of America, 1978.
18 Herstein, I N. Topics in Algebra (2nd edition). Wiley, New York, 1975.
19 Herstein, I N. Noncommutative Rings. Carus Monograph 15. Mathemati-
cal Association of America, 1968.
20 Jacobson, Nathan. Lectures in Abstract Algebra Vols 1, 2, 3. Van Nos-
trand, Princeton, NJ, 195 1—64.
21 Jacobson, Nathan. Basic Algebra I, IL Freeman, San Francisco, 1974—80.
22 Kaplansky, I. Commutative Rings. University of Chicago Press, Chicago,
1974.
23 Ledermann, W. Introduction to Group Theory. Oliver and Boyd, Edin-
burgh, 1973.
24 McCoy, Neal H. Fundamentals of Abstract Algebra. Allyn and Bacon,
Boston, 1973.
25 McCoy, Neal H. Rings and Ideals. Carus Monograph 8. Mathematical
Association of America, 1948.
26 McCoy, Neal H and Berger, Thomas R. Algebra: Groups, Rings and
other Topks. Allyn and Bacon, Boston, 1977.
27 Macdonald, Ian D. The Theory of Groups. Oxford University Press,
Oxford, 1968.
28 MacLane, Saunders and Birkhoff, Garrett. Algebra. Macmillan, New
York, 1967.
29 Magnus, Wilhelm, Karrass, Abraham and Solitar, Donald. Combinatorial
Group Theory: Presentations of Groups in terms of Generators and Rela -
tions. Wiley—lnterscience, New York, 1966.
30 Meldrum, J D P. Near-rings and Their Links with Groups. Pitman, London,
1985.
31 Miller, G A, Blichfeldt, H F and Dickson, L E. Theory and Application
of Finite Groups. Dover, New York, 1961.
32 Mostow, George D, Sampson, Joseph H, and Meyer, Jean-Pierre. Funda -
mental Structures of Algebra. Holt, Rinehart and Winston, New York,
1966.
33 Pilz, Günter. Near-Rings: the Theory and its Applkations. North-
Holland, Amsterdam, 1977.
34 Rotman, Joseph J. The Theory of Groups. An Introduction (2nd edition).
Allyn and Bacon, Boston, 1973.
35 Sawyer, W W. A Concrete Approach to Abstract Algebra. Freeman, San
Francisco, 1959.
36 Scott, W R. Group Theory. Prentice-Hall, Englewood Cliffs, 1964.
37 Shapiro, Louis. Introduction to Abstract Algebra. McGraw-Hill, London,
1975.
38 Stewart I. Galois Theory (2nd edition). Chapman and Hall, London, 1989.
366 Bibliography

39 van der Waerden, B L. Algebra Vols 1, 2 (7th edition). Ungar, New

York, 1970.
40 Weber, Heinrich. Lehrbuch der Algebra. Friedrich Vieweg und Sohn,
Braunschweig, 1895—6.
41 Zariski, Oscar and Samuel, Pierre. Commutative Algebra Vols 1, 2. Van
Nostrand, New York, 1958.

Number theory (including Number Systems)

42 Allenby, R B J T and Redfern, E J. Introduction to Number Theory with
Computing. Edward Arnold, London, 1989.
43 Burton, David M. Elementary Number Theory, Revised Printing, Allyn
and Bacon, Boston, 1980.
44 Carmichael, Robert D. Theory of Numbers and Diophantine Analysis.
Dover Publications Inc., 1914.
45 Davenport, H. Higher Arithmetic: An Introduction to the Theory of
Numbers. Hutchinson, London, 1962.
46 Edwards, Harold M. Fermat's Last Theorem. A Genetic Introduction to
Algebraic Number Theory. Springer-Verlag, New York, 1977.
47 Hardy, G H and Wright, E M. An Introduction to the Theory of Numbers
(4th edition). Oxford University Press, Oxford, 1960.
48 Isaacs, G L. Real Numbers. A Development of Real Numbers in an
Axiomatic Set Theory. McGraw-Hill, New York, 1968.
49 Le Veque, William J. Topics in Number Theory Vols 1, 2. Addison-
Wesley, Reading, Massachusetts, 1956.
50 Mendelson, Elliot. Number Systems and the Foundations of Analysis.
Academic Press, New York, 1973.
51 Pollard, Harry. The Theory of Algebraic Numbers. Carus Monograph 9.
Mathematical Association of America, 1961.
52 Ribenboim, Paulo. Algebraic Numbers. Wiley—Interscience, New York,
1972.
53 Ribenboim, Paulo. 13 Lectures on Fermat's Last Theorem. Springer-
Verlag, New York, 1980.
54 Shanks, Daniel. Solved and Unsolved Problems in Number Theory. Spar-
tan Books, Washington DC, 1962.
55 Stewart, Ian and Tall, David. Algebraic Number Theory. Chapman and
Hall, London, 1979.
56 Uspensky, J V and Heaslett, M A. Elementary Number Theory. McGraw-
Hill, New York, 1939.

Logic, set theory

57 Exner, Robert M and Rosskopf, Myron F. Logic in Elementary
Mathematics. McGraw-Hill, New York, 1959.
58 Halmos, Paul R. Naive Set Theory. Van Nostrand, New York, 1960.
 Bibliography 367

59 Rosser, J Barkley. Logic for Mathematicians. McGraw-Hill, New York,

1953.
60 Stoll, Robert R. Sets, Logic and Axiomatic Theories. Freeman, San
Francisco, 1961.

Applications (in a broad sense)

61 Birkhoff, Garrett and Bartee, Thomas C. Modern Applied Algebra.
McGraw-Hill, New York, 1970.
62 Boardman, A D, O'Connor, D E and Young, P A. Symmetry and its
Applications in Science. McGraw-Hill, London, 1973.
63 Buerger, M J. Elementary Crystallography. Wiley, New York, 1956.
64 Buerger, Martin J. Crystal-structure Analysis. Wiley, New York, 1960.
65 Ellis, Andrew and Treeby, Terence. Algebraic Structure. John Murray,
London, 1974.
66 Gillman, Land Jerison, M. Rings of Continuous Functions. Van Nostrand,
Princeton, NJ, 1960.
67 Hadley, George. Linear Algebra. Addison-Wesley, Reading,
Massachusetts, 1961.
68 Liebeck, Hans. Algebra for Scientists and Engineers. Wiley, London,
1969.
69 Lockwood, E H and Macmillan, R H. Geometric Symmetry. Cambridge
University Press, Cambridge, 1978.
70 Miller Jr, Willard. Symmetry Groups and their Applications. Academic
Press, New York, 1972.
71 Murnaghan, Francis D. The Theory of Group Representations. The Johns
Hopkins Press, Baltimore, 1938.
72 Schonland, David S. Molecular Symmetry. Van Nostrand, London, 1965.
73 Schwarzenberger, R L E. N-Dimensional Crystallography. Research
Notes in Mathematics, 41. Pitman, London, 1980.
74 South, G F. Boolean Algebra and Its Uses. Van Nostrand Reinhold,
London, 1974.
75 Whitesitt, J Eldon. Boolean Algebra and its Applications. Addison-
Wesley, Reading, Massachusetts, 1961.

History
76 Bell, Eric Temple. Development of Mathematics (2nd edition). McGraw-
Hill, New York, 1945.
77 Bell, Eric Temple. Men of Mathematics. Simon and Schuster, New York,
1962.
78 Bell, Eric Temple. Mathematics, Queen and Servant of Science. G Bell,
London, 1952.
79 Bourbaki, Nicolas. Elements d'Histoire des Mathématiques. Hermann,
Paris, 1960.
80 Boyer, Carl B. A History of Mathematics. Wiley, New York, 1968.
368 Bibliography

81 Crowe, Michael J. A History of Vector Analysis. University of Notre

Dame Press, Indiana, 1967.
82 Dickson, L F. History of the Theory of Numbers Vols 1, 2, 3. Carnegie
Institute of Washington, Washington, 1919.
83 Edwards, Harold M. Galois Theory. Springer-Verlag, New York, 1984.
84 Eves, Howard. An Introduction to the History of Mathematics (3rd edi-
lion). Holt, Rinehart and Winston, New York, 1969.
85 Kline, Morris. Mathematical Thought from Ancient to Modern Times.
Oxford University Press, New York, 1972.
86 Ore, Oystein. Number Theory and its History. McGraw-Hill, New York,
1948.
87 Novy, Lubos. Origins of Modern Algebra. Noordhoff International,
Netherlands, 1973.
88 Struik, Dirk J. A Concise History of Mathematics. 0 Bell, London, 1954.
89 van der Waerden, B L. A History of Algebra. Springer-Verlag, Berlin,
Heidelberg, 1985.

Papers
The following is just a sample. I could have quadrupled the list with no
difficulty. I leave you the excitement of discovering some of these other papers
(in the journals mentioned below and elsewhere) for yourself.
The abbreviations AMM and AHES used below stand for American
Mathematical Monthly and Archive for History of Exact Sciences.
90 Aschbacher, Michael. The Classification of the Finite Simple Groups.
The Mathematical Intelligencer 3, 2, 1981, 59—65.
91 Ayoub, Raymond 0. Paolo Ruffini's Contributions to the Quintic.
AHES 23, 1980, 253—77.
92 Birkhoff, Garrett. Current Trends in Algebra. AMM 80, 1973, 760—82.
93 Cohn, P M. Unique Factorisation Domains. AMM 80, 1973, 1—18.
94 Cohn, P M. Rings of Fractions. AMM 78, 1971, 596—615.
95 Conway, J H. Monsters and Moonshine. The Mathematical Intelligencer
2, 4, 1980, 165—71.
96 Cooke, Roger. Letter to the editor. AMM 91, 1984, 382.
97 Dickson, L F. Fermat's Last Theorem and the Origin and Nature of
the Theory of Algebraic Numbers. Annals of Mathematics, Series 2,
18, 1916—17, 161—87.
98 Dieudonné, J. The Historical Development of Algebraic Geometry.
AMM 79, 1972, 827—66.
99 E, H. A Short History of the Fields Medal. The Mathematical Intelligen -
cer 1, 3, 1978, 127—9.
100 Edwards, Harold M. The Genesis of Ideal Theory. AHES 23, 1980,
321—78.
101 Fefferman, C. An Easy Proof of the Fundamental Theory of Algebra.
AMM 74, 1967, 854—5.
102 Fisher, Charles S. The Death of a Mathematical Theory: a Study in the
Sociology of Knowledge. AHES 3, 1966—7, 137—59.
 Bibliography 369

103 Hamburg, Robin Rider. The Theory of Equations in the 18th Century:
The Work of Joseph Lagrange. AHES 16, 1976, 17—36.
104 Hawkins, Thomas. The Origins of The Theory of Group Characters.
AHES 7, 1971, 142—70.
105 Hawkins, Thomas. The Theory of Matrices in the 19th Century. Proc.
Intern at. Congress of Mathematicians, Vancouver, 1974. Vol. 2
Canadian Math Congress, 1975, 561—70.
106 Henkin, Leon. On Mathematical Induction. AMM 67, 1960, 323—38.
107 Hungerford, Thomas W. A counterexample in Galois Theory. AMM 97,
1990, 54—7.
108 Kiernan, B Melvin. The Development of Galois Theory from Lagrange
to Artin. AHES 8, 197 1—2, 40—154.
109 Koppelman, Elaine. The Calculus of Operations and the Rise of
Abstract Algebra. AHES 8, 197 1—2, 155—242.
110 Mead D G. The Missing Fields. AMM 94, 1987, 87 1—2.
Ill Miller, G A. History of the Theory of Groups to 1900. The Collected
Works of George Abram Miller. University of Illinois, Urbana, 1935—59.
112 Moran, W and Pym, J S. On the Construction of the Real Number
System. Mathematics Magazine, 43, 1970, 257—9.
113 Motzkin, Th. The Euclidean Algorithm. Bulletin Amer. Math. Soc. 55,
1949, 1142—6.
114 Pierpont, James. Lagrange's Place in the Theory of Substitutions. Bul-
letin Amer. Math. Soc. 1, 1895, 196—204.
115 Pierpont, James. Early History of Galois' Theory of Equations. Bulletin
Amer. Math. Soc. 4, 1898, 332—40.
116 Rothman, Tony. Genius and Biographers: The fictionalization of Evariste
Galois. AMM 89, 1982, 84—10.
117 Ruchte, M F and Ryden, R W. A Proof of the Uniqueness of Factorisa-
in the Gaussian Integers. AMM 80, 1973, 58—9.
tion
118 Samuel, Pierre. Unique Factorization. AMM 75, 1968, 945—52.
119 Samuel, Pierre. About Euclidean Rings. Journal of Algebra 19, 1971,
282—301.
120 Schwarzenberger R L E. The 17 plane symmetry groups. Mathematical
1974, 123—31.
Gazette, 58,
121 Seidenberg, A. Did Euclid's Elements, Book I, develop Geometry
Axiomatically? AHES 14, 1974—5, 263—95.
122 Simmons, G J. The Number of Irreducible Polynomials of Degree n
over GF(p). AMM 77, 1970, 743—5.
123 Stark, H M. On the Problem of Unique Factorization in Complex
Quadratic Fields. Proc. of Symposia in Pure Mathematics, XII. American
Math. Soc., 1969, 41—56.
124 Stewart, Ian. The Truth about Venn Diagrams. Mathematical Gazette
60, 1976, 47—54.
125 Szabó, Arpád. Greek Dialectic and Euclid's Axiomatics. Problems in
the Philosophy of Mathematics (Imre Lakatos, ed.). North-Holland,
Amsterdam, 1967.
126 van der Waerden, B L. Die Galois-Theorie von Heinrich Weber bis Emil
Artin. I41-!ES9, 1972, 240—8.
370

127 Weyl, Hermann. A Half-Century of Mathematics. AMM 58, 1951,

523—53.
128 Wigner, Eugene P. Symmetry Principles in Old and New Physics.
Bulletin Amer. Math. Soc. 74, 1968, 793—8 15.
129 Wilder, Raymond L. The Role of the Axiomatic Method. AMM 74,
1967, 115—27.
130 Wilson, J C. A Principal Ideal Ring that is not a Euclidean Ring. Math.
Magazine, 46, 1973, 74—8.
131 Youschkevitch, A P. The Concept of Function up to the Middle of the
19th Century. AHES 16, 1976, 37—85.

Other references
132 Mal'cev, A I. Groups and Other Algebraic Systems in Mathematics: its
Contents, Methods and Meaning Vol 3. MIT Press, Cambridge,
Massachusetts, 1963.
133 Newman, James R (ed). The World of Mathematics Vols 1, 2, 3, 4.
Allen and Unwin, London, 1960.
134 Dictionary of Scientific Biography. Charles Scribner's Sons, New York,
1970—80.
135 Encyclopedic Dictionary of Mathematics. MIT Press, Cambridge,
Massachusetts, 1977.
Notation
Because of the limited number of (sensible) symbols available to us, it is not uncommon
for one symbol to be used to denote totally distinct concepts. For example, note that
here the symbol (yi, . , yj will denote (i) a gcd of the elements y1, 3½, . .. , or
.

(ii) an ordered n -tuple or (iii) a cyclic permutation, depending on the context. This is
at worst irritating since there will rarely be any cause for confusion.

Z,Q,R,C 1
36,176
k=1
E,t' 1
Q[{x]] 39
(likewise 2, 8 O[x, y] 42
{x:P(x)} 2 (z0,z1 zn);
{a,b,c,. ..} 2 (x1,x2 xr);
{a,b} 2,128 (gi,g2 g,,) 44,134;197;250
Ac,B2A 2 degf 45
AcB,BDA 2 f(c) 52
2 ss
r
2 [xJ;[ai],[ai,a2,...,an] 56;99
AB 2 57,117
0 2 aRb,aRi, 59
AnB,AuB 3 e.r. 61
a 61
flSn,flTc,
oEA
4,8,98
Zn 63
A\B 4 q5(m) 70
RX(likewiseox C° F°) 4,188 f
74
AxB 5,252
(a,b) 5,26 af,f(a) 74
(a,b,c) 5,32 a+b 75
Sf,f(S) 75,77
—A 6 75,77
Az>B 6 75
7
fIs 75
3,V 8 fog 76
apb 78
UjRn 8
a", na, a_n, etc. 81, 95, 196
O[x] (likewise Z[x], R[x], (R, +, -) (and similar) 85
10,36,43,67 (a11 a12\
87
C 15, 137 \a21
b—a 15,94 88
aj 16,63,109 2Z 89
20,42,104 R®S 90
aA'b,ftg 20,42 91
32 0R,1R,(a)R 95,97
101
(a0,a1,a2,...) 36,39,157 N(a) 105
®,O 36,etc. ER 108
372 Notation

Z[p] 108 192

UFD,PID 112 200
126,132,217 200
FD 128 203
Z[\/2] 130 C(G) 206
F(a,S,y,...) — 130 (U),(a,b,c,...) 206
0(x), Q(x, y) 130 gH,Hg 210
137 G:Hl 211
kere 143,225 InnG,AutG 223
r+I 145 225,243
R/!,G/N 145,229 NcnG 225
'a b Xg 243
det( N0(H)
d) 243
Ma 154 C'0(x),C'0(S) 243,249
[E:F] 165
.

GF(p") 167 . 250

(G,o)(and similar) 186 [a,b] 259
M(n) 188 259
188,189;254 [A,B] 259
'a b 260
189 HK 263
a
etc.
101 189 Gal (Sf/F), Gal (L/K), etc.
Ca 190, 215 Fix E, Fix (E), etc.
190
<xuy><Ux.>
Other notation used includes:
(i) to denote the number of elements in a (finite) set or group. (Cf. Definition
5.3.7.)
(ii) Ma(X), where X is a ring, to denote the set (ring) of all n x n matrices with
entries from X. [For ii = 2, X = C see Definition 3.2.5j
(iii)(x, y; x2 = 1, xy2 = y2x) (etc.), to denote the group generated by elements x, y
subject only to the relationships x2 = 1 and xy2 = y2x (etc.). Seep. 219 and pp. 238—41.
Index

Abel N H, 187, 303 Euclidean, 29, 51, 110

Abelian groups, 187 Kronecker's, for factorisation in Q[x], 56
finite, 254ff. Al-Khowarizmi Mohammed, xiii
fundamental theorem of, 255 Alternating group, 200
structure theorems for finitely is simple, 268
generated, 256 Analytic deficiency of 0, 156
Abel—Rufilni theorem, 303 Angle,
Absolute value (see modulus) bisection of, 170
Abstraction, method of, xiv trisection of, 170, 173, 174
Addition, Antisymmetric binary relation, 59
in a ring, 84 Archimedean integral domain, 139
in field of fractions, 128 Arithmetic, fundamental theorem of, 33
of complex numbers, 160 Artin E, 216
of cosets of an ideal, 146 Artin's Lemma, 292
of integers mod n, 64 Associate elements in 1, 0[x], etc., 21, 42,
of matrices, 87 104
of polynomials, 36 Associative binary operation, 79
tables for 4, 64, 66 Associative law(s), 11, 37, 65, 85, 187
Additive group of ring, of field, 188 generalised, 80, 196
Additive, Associativity of composition of
identity in a ring, 84, 85 functions, 79
inverse in a ring, 84 Automorphism(s),
notation for groups, 187 fixed field of, 283
Affine geometry, example of, 193 group of a group, 223
Algebra, identity, of a ring, 127
abstract, xiii inner, 223
classical, xiii of a field, 194
fundamental theorem of, 54, 178ff. of a group, 223
Lic, example of, 90 of an algebraic system, 127
linear, xxv, 163 of0,ofR,ofC, 132
modern, xiii ofl, 138
Peacock's arithmetic and symbolic, xviii Axiomatisation, method of, xiv
possible modern definition of, xiii, 57 Axioms,
use of, xxv as 'rules of the game', xvi
Algebraic, for groups, 186
deficiency of 0, R, 156 for rings and variants, 84
element over a field, 153 for the integers, 11
geometry, xii, 99, 102
integer, number, 91
Algebraically closed field, 180 Babylonians, xiv, xvi
complex number field is an, 180 Basis,
field of all algebraic numbers is an, 180 for field over subfield, 164
Algebraist, definition of, 128 theorem of Hubert, 99
Algebraist's, Bernoulli D, 73
dream, 140, 168, 254 Bijection, 75
task, 128 Binary operation, 57, 78
Algorithm, associative, 79
division, 27, 49, 107 commutative, 79
374 Index

Binary operation (cont) Commuting elements in a group, 206

identity neutral) element for, 79 Compass and straightedge (ruler), 170ff.
subset closed under, 78 Complete ordered field, 158, 161
Binary relation(s), 58 Complex number(s), I
antisymmetric, 59 algebraically closed, 180
of congruence mod n, 57, 59 basis for over 164
product of, 61 defined, 160
reflexive, 59 modulus of, 63
symmetric, 59 Complex roots,
transitive, 59 of a real equation, 55
Bisection of an angle, 170 ofl, 190,219
Black hole, ring theoretic, 98 Composite integer, 66
Bolzano B, 162 Composition,
Boolean ring, example of, 90 factor, 262
Burnside W, 266 of functions, 76
is associative, 79
CAIN, 187 series, 262
Calculus of operations, xviii Conditions, necessary and sufficient, 7
Cancellation law(s), Congruence,
in a group, 195 mod n, 57, 62
in an integral domain, 95 notation in general ring, 117
in a ring, failure of, 90 Conjugacy class(es), 243
in!, 12 Conjugate,
Cantor G, I complex roots of a real equation, 55
Cartesian product, 5, 250 elements/subsets/subgroups in a group,
Cases, proof by, 7 243
Cauchy A L, 159 Constant,
Cauchy convergent sequence, 157 function (map), 75
Cayley A, 220 polynomial, 42
Cayley's theorem, 221 Constructible point(s), 171
generalisation of, 224 Constructible polygon, 174
Centraliser, Gauss's theorem, 295
of an element, 243 Construction,
of a subgroup, 249 ofCfromfl, 159
Centre, ofiqfromo, 157
of a group, 206 of a regular n-gon, 174
of a p-group, 244, 245 with compass and straightedge, I 70ff.

Characteristic, Content of a polynomial, 45, 135

of an integral domain, 133 Contradiction, proof by, 6
of a ring, 133 Contraposition, proof by, 6
Circle, squaring the, 174 Contrapositive of an assertion, 6
Class, Converse of an assertion, 6
conjugacy, 243 Coprime relatively prime), 31, 104
equation of a group, 244 Correspondence, one to one (1—I), 75
equivalence, 62 Correspondence theorem for rings, 155
Closed, Cosets,
algebraically, 180 left (right), of a subgroup in a group, 210
subset under a binary operation, 78 of an ideal in a ring, 145
Codomain of a function (map), 77 sum/product of,
Collection (see set) in group, 229
Collineation of an affine geometry, 193 in ring, 146
Combination, linear, 164 Countable set, 78
Common divisor in!, 20 Counterexample,
Commutative, proof by minimum, 23
binary operation, 79 use of, 7
group (see abelian group) Crystallographic,
law(s), 11,65, 187 point group, 237
ring, 85 restriction, 235
Commutator, 259 space group, 233
subgroup, 259 Cube, duplication of, 170, 173
 Index 375

Cubic, Division ring, 85, 94

equation (polynomial), solution by of Hamilton's quaternions, 88, 132
radicals, 183ff, 306 Divisor, 20, 42
polynomial, irreducible in Q[x], 54 common, 20
Cut, Dedekind, 157 greatest common, 26, 51, 104
Cycle (see cyclic permutation) Divisor of zero zero divisor), 12, 85
Cyclic group, 213 Domain,
finite, characterisation of, 219 integral (ID), 85, 86, 94
generator of, 203, 207 of a function (map), 75
infinite, characterisation of, 219 principalideal (ND), 112
of roots of 1, 190, 219 unique factorisation (UFD), 112
subgroup of, 215 Duplication of the cube, 170, 173
Cyclic permutation(s), 197
disjoint, 198 Eisenstein F G M, 46
length of, 197 Eisenstein's test, 46, 49, 136, 144
Cyclic subgroup, 203 Element(s),
Cyclotomic polynomial, 47, 301 algebraic over a field, 153
associate, 21, 42, 104
d'Alembert J le R, 73 centraliser of, 243
Dedekind cut, 157 commuting, 206
Dedekind J W R, 121 conjugate of, 243
Definition by induction, 16 identity neutral), 11, 38, 79, 85, 187
Degree of a polynomial, 45 inverse, 11, 84, 187
de Moivre A, 190 irreducible, 21, 42, 43, 104
de Moivre's theorem, 190 norm (or valuation) of, 105, 108
de Morgan A, 8 ofset, 1

de Morgan's laws, 8 order of, in a group, 204

Denial negation) of an assertion, 6 positive, 15, 137
Derivative of a polynomial, 55 prime, 21,42,43, 104
Derived, relatively prime coprime), 31, 104
series, 260 separable, 289
commutator) subgroup, 259 theorem of primitive, 313
Descartes R, 183 Elementary symmetric polynomial, 176
Descent, Fermat's method of, 118, 120 Embedding of,
Determinant (map), 150, 224 an ID in a field, 128
Dickson L E, 271 aringinaringwith 1, 127
Difference of sets, 4 one ring in another, 127
Dihedral group, 190 Empty set, 2
Dimension of one field over another, 165 Epimorphism, 155
product formula for, 172 Equality of,
Direct product of groups, 250 complex numbers, 160
external, 251 functions (maps), 75
internal, 251 ordered pairs, 5
proper, 252 polynomials, 36
restricted, 253 sets, 2
unrestricted, 253 Equation(s),
Direct proof, 6 class, 244
Direct sum, cubic, 183ff.
of (abelian) groups, 251, 255ff. quadratic, xiii, xiv, xvi
of rings, 90 quartic, 185, 306
Discrete group of isometries, 234 quintic, 274, 304, 308
Disjoint cycles, 198 Equinumerous, 78
Disquisitiones Arithmeticae, xvii, 68, 69, Equivalence,
174 class, 62
Distributive law(s), 11, 39, 84 relation (er.), 59, 61ff.
Divides, 20, 42, 104 Euclid, 23, 29, 32
Division algorithm, Euclid's
for 1, for Q[x], etc., 27, 49, 107, 109, Elements, xiv
110 parallel postulate, xv
fails in Z[x], 43, 51 fails, xxvi
376 Index

Euclidean, 15-puzzle, 202

algorithms, 29, 51, 110 Finite abelian groups, fundamental theorem
norm valuation), 108 of, 255
ring(ER), 108, 115 Finite,
ringisaUFD, 112 dimensional extension, 165
Euler L, 69 field,162ff.
Euler's group, 189
phi-function 0-function, totient), 70 integral domain, 94
theorem, 70, 73 set, examples of, 2
Even permutation, 200 Finitely generated,
Exponents, laws of, 81, 95, 197 abelian group, 256
Exponential isomorphism, 220 group, 206
Extension, First isomorphism theorems, 152, 231
finite dimensional, 165 Fixed field, of an automorphism, 283
Galois, 289 Four square theorem of Lagrange, 7
normal, 288 Fractions,
ofa field, 141 field of, of an ID, 128, 129
of a homomorphism, examplc of, 144 partial, 110
radical, 274, 276 Fraenkel A A, 95
separable, 289 Frieze pattern, 241
simple, 313 Frobenius, monomorphism, (133), 291
split(ting), 276 Function(s) (see mapping(s))
External direct product, 251 polynomial, 77
ring of, on R, 90
Factor, Fundamental theorem
group group), 229 of algebra, 54, 178ff.
highest common (see gcd) of arithmetic, 33
in composition series, 262 fails in 105, 106
ring (_= quotient ring), 146 of finite abelian groups, 225
Fermat P de, 102 of Galois Theory, 307
three problems of, 116ff. on symmetric polynomials, 176
Fermat's
Conjecture (FC), 103, 121, 122 Galois E, 224, 268, 275
proofofforn=3,4, 117, 120 Galois,
little theorem, 32, 69 extension, 289
method of descent, 118, 120 field, 167
Fibonacci L, 16 group, 194, 261, 276, 312
Fibonacci sequences, 16 as a group of permutations, 276, 279, 308
Field, 67, 85 of a polynomial, 276, 312
algebraically closed, 180 Galois' theorem on solution of equations by
automorphism of a, 194 radicals, 261, 306
characteristic of (see integral domain) Galois Theory, Fundamental Theorem of, 00
complete ordered, 158, 161 Gap, 161
element algebraic over, 153 Gauss C F, 68, 174
extension, 141 Gaussian integers, 108
finite, 162ff. Generalised associative law, 80, 196
Galois, 167 General linear group, 192
number Zahlkorper), xx Generating set for a (sub)group, 206
of algebraic numbers, 91 Generator(s) of,
of complex numbers, 160 a cyclic group, 203, 207
of fractions of an ID, 128, 129 an ideal, 99
of rational numbers, 130 a principal ideal, 99
of realnumbers, 158 Geometric constructions, 'rules of the
ordered, 139 game', 171
prime (sub)field, 98, 131 Geometry,
quotient (see field of fractions) affine, example of, 193
skew (see division ring) algebraic, xii, 99, 102
splitting, of a polynomial, 149, 276 non-Euclidean, xv, xxiii (and xxvi)
uniqueness of, 276 Glide reflection, 233
Fields medal, 266 Gordan P, 99
 Index 377

Greatest common divisor (gcd), 26, 51, 104 symmetric, 189, 197ff.
as a linear combination, 27, 44, 109 transitive, 307
Greeks, xiv, xvi trivial subgroup of, 203
Group(s), with no trivial) centre, 206
abelian, 187, 245ff.
fundamental theorems, 255, 256
Hall P, 265
abstract theory of, 221
Hall it-subgroup, 265
additive, of a field, ring, 188
Hall's theorem, 265
additive, of integers, v, 187
Hamilton W R, 87
alternating, 200, 268
Highest common factor hcf), (see gcd)
automorphism of, 223
Hubert D, 25
automorphism group of, 223
Hilbert's
axioms, 186
basis theorem, 99
centre of, 206
Fl-numbers, 24
class equation of, 244
Holder 0, 263
commutative (see abelian groups)
Homomorphic image, 142, 224
conjugate elements/subsets/subgroups
Homomorphism (E homomorphic
of, 243
map), 141, 224
crystallographic point, 237
extension of, example, 144
crystallographic space, 233
kernel of, 143, 225
cyclic, 213
trivial, 141
dihedral, 190
direct product of, 250
direct sum of, 251 Ideal(s), 98, 123
discrete, 234 generated by a set, 99
factor quotient), 229 intersection of, 101
finite, 189 in Z, in Q[xJ, in an ER are principal, 98,
finitely generated, 206 99, 113
Galois, 194, 261, 276, 312 maximal, 124
general linear, 192 numbers, xx
generated by, 206 prime, 123
homomorphic image of, 224 principal, 99
infinite, 189 product of, 101
inner automorphism of, 223 sum of, 101
isomorphic, 217 union of increasing sequence of, 101
Lie, example of, 192 union, not a subring, 101
multiplication table of, 218 zero, 99
non-abelian, 189 Identity neutral) element(s)
of complex nth roots of 1, 190, 219 in 7, in 0[x], in ring, in group, etc., 11,
of inner automorphisms, 223 38, 79, 85, 187
of isometries, 190 Identity function (map), 75
oforderp, 214 If, 7
of orderp2, 244 1ff (if and only if), 7
of orderpq, 269 Ill-definedness (see well-definedness),
of orderp2q2, 270 examples of, 68, 151
of orders 4,6,8,218 Image,
of permutations, 188, 197ff. homomorphic, 142, 224
of symmetries of a regular n-gon, 190 inverse, of subset, 75
188 of element/subset under map, 75
order of, 189 Inclusion,
p-, 242 function (map), 75
it-, 265 of sets, 2
plane symmetry, 234 Independent, linearly, 164
presentation of, 219 Indexing set, 4
quotient (see factor) Index of a subgroup in a group, 211
set of generators for, 206 product formula for, 212
simple, 247, 268ff. Indirect proof, 6
soluble, 258ff., 260, 261, 267 Induction,
space, 233 definition by, 16
special linear, 192, 271 principle of mathematical, Il, 16, 17
378 Index

Infinite, groups, 217

group, 189 rings, 126
set, 2 lsomorphism,
Infinitude of primes, 23, 48 exponential, 220
Injection map, 75 is an equivalence relation, 132, 222
Inner automorphism, 223 of complete ordered fields, 158
Insolubility of general quintic, 274, 304 theorems, 152, 153, 231, 232, 263
Integer(s), 'up to', 130, 152, 217
algebraic, 91
axioms for, 11
characterisation of, 137 Jordan C, 261
composite, 66 Jordan-Holder theorem, 263
division algorithm for, 27
Gaussian, 108
irreducible, 21 Kernel,
mod n, 63ff. of a group homomorphism, 225
negative, 15 of a group homomorphism is a normal
positive, 15 subgroup, 225
prime, 21 of a ring homomorphism, 142, 143
relatively prime coprime),. 31, 104 of a ring homomorphism is an ideal, 143
square-free, 91 'Killing off', 147
unique factorisation in, 33 Kronecker L, 44
Integral domain (ID), 85, 86, 94 Kronecker's
Archimedean, 139 method of polynomial factorisation, xxi, 56
characteristic of, 133 theorem, 141, 147
finite, 94 Kummer E E, 122
ordered, 137 Kummer's ideal factor, 122
well ordered, 137
well ordered, uniqueness of, 137
Internal direct product, 251 Lagrange J L, 182
Interpolation, Lagrange's formula, 55 Lagrange's
intersection of, four square theorem, 7
ideals, 101 interpolation formula, 55
(sub)sets, 3 theorem, 211
subfields, 98 converse for abelian groups, 230
subgroups, 205 converse false in general, 211, 231
subrings, 98 LaméG, 103
Into function (map), 75 Lattice of points in plane, 235
Inverse image of a subset, 75 Laws (see associative, cancellation,
Inverse clement(s), commutative, distributive, de Morgan's,
additive, in a ring, 84 mysterious, exponents, zero-divisor)
multiplicative, in a group, 187 Least common multiple (1cm), 115
Inverse of a function (map), 75 Left coset, 210
Invertible non-singular) matrix, 192 LeibnizGW, 69
Irrationality of 6 Lemma,
Irrational number, I Artin's, 292
Irreducible, definition of, 12
element in a ring, 104 Length of,
element is not always prime, 106 cycle, 197
element is prime in an ER, 110 series, 262
integer, 21 Less than, II
polynomial(s), 42, 43, 54 LieMS, 90
polynomial(s) in R[x], in C[x], 54 Lie,
Irreducibility tests for polynomials, algebra, example of, 90
Eisenstein's, 46, 49, 136, 144 group, example of, 192
rational root test, 53 Lindemann C L F, 140
Isometry, Linear,
improper,234 algebra, xxv, 163
proper, 234 combination, 164
Isomornhic. Linearly independent set, 164
 Index 379

Mal'cevAl, 130 Noether E, 99

Map(s), mapping(s) (see pp. 75, 76 for Noetherian ring, 99
definitions of all concepts concerning Non-abelian group, 189
mappings) Non-Archimedean ID/field, 139
Mappings, compositon is associative, 76 Non-associative ring, 84, 90
Mathematical induction, Non-Euclidean geometry, xv, xxiii (and xxvi)
principle of, 11 Non-singular invertible) matrix, 192
equivalent versions of, 16, 17 Non-soluble quintics, 274, 304, 308ff.
Mathieu E L, 271 Non-unique factorisation, example of, 25, 105
Matrix, matrices, Norm,
addition of, 87 of an element in 105
additive inverse of, 88 valuation) on an ER, 108
determinant of, 150 Normal,
invertible non-singular), 192 extension, 288
multiplication of, 88 and normal
zero, 88 subgroup, 307
Maximal, subgroup, 225
ideal, 124, 156 Normaliser, 243
normal subgroup, 262 Number(s),
Member element) of a set, algebraic, 91
Method of descent, Fermat's, 118, 120 complex, definition of, 160
Minimum, field Zahlkorper), xx
polynomial, 154 irrational, I
counterexample, proof by, 23 perfect, 32
Mod(ulo) n, it-, it'-, 265
addition/multiplication, 64 rational, definiton of, 130
congruence, 58, 62 real, definition of, 158
Modulus absolute value) of, ring Zahlring), 83
complex number, 63
integer, 16 Odd permutation, 200
Monic polynomial, 48 One to one (1—1),
Monomorphism, 280, 284 correspondence, 75
Frobenius, (133), 29! function (map), 75
Moore E H, 167 Only if, 7
Multiple, Onto function (map), 75
least common (1cm), 115 Operation, binary, 57, 78
root, 53 associative, 79
Multiplication, commutative, 79
in field of fractions, 128 identity element for, 79
of complex numbers, 160 subset closed under, 78
of cosets of an ideal, 146 Orbit of permutation, 198
of cosets of a normal subgroup, 229 Order
ofintegersmodn, 64 of a group, 189
of matrices, 88 of a group element, 204
of permutations, 188, 198ff. of Galois group, theorem on, 285
of polynomials, 36 of a (sub)group, 211
table of a group, 218 relation, 59
tables for 74, 64, 66 Ordered,
Multiplicative inverses, 11, 66, 84, 187, Archimedean ID, 139
194, 195 field, 139
Multiplicity of a root, 53 field, complete, 158, 161
Mysterious law, 12 integral domain, 137
pair, 5
n-gon, symmetry group of regular, 190 triple, 5
nth roots of I, (cyclic) group of, 190, 219 well-, 17, 137
Near-ring, 86 Ordering on elements of 7, Q[x], etc., 15,
Necessary condition(s), 7 39, 136ff.
Negation denial) of assertion, 6
Negative integer(s), 15 p-group, 242
Neutral element (see identity element) centre of, 244, 245
380 Index

p-primary component, 254 value of, 52

p-subgroup, 247 zero, 37
Sylow, 246 zero root) of, 52, 53, 67, 274
Pair, ordered, 5 Positive,
Partial fractions, 110 elements in ordered ID, 137
Partition, 61 integers, 15
Pascal B, 16 are well ordered, 17
Peacock's principle of permanence of Power,
equivalent forms, xviii series (ring), 39, 125
Pentagon, regular, 175 set, 5
Perfect number, 32 Presentation of a group, 219
Permutation(s), 75 Primary component of abelian group, 254
Cauchy's notation for, 189 Prime(s),
cyclic, 197 element in /, 0[x], etc., 21, 42, 104, 110
length of, 197 ideal, 123, 156
disjoint (cyclic), 198 infinitude in /, O[x], 23, 48
even, 200 integer, 21
group(s), 188, 197ff. polynomial, 42
multiplication of, 188, 198ff. relatively, 31, 104
odd, 200 (sub)field, 98, 131
orbit of, 198 which are sums of two squares, 116
transposition, 199 Primitive,
Phi-function, Euler's, 70 element theorem, 313
it-number, iC-number, 265 polynomial, 44, 134
it-subgroup, 265, 268 rootmodn, 214,216
Plane symmetry group, 234 root of unity, 281
Point(s), constructible, 171 Principal,
Polygon, constructible, 174 ideal, 99
Gauss' theorem, 000 ideal domain (PID), 112
Polynomial(s), isaUFD, 112
addition of, 36 Principle,
constant, 42 of mathematical induction, I I
content of, 45, 135 equivalent versions of, 16, 17
cubic, 54 of permanence of equivalent forms, xviii
cyclotomic, 47, 301 well ordering, 17
degree of, 45 Product,
derivative of, 55 cartesian, 5, 250
division algorithm for, 49 direct, 250ff.
equality of, 36 of cosets, 146, 229
elementary symmetric, 176 of ideals, 101
function, 77 of integers mod ii, 64
in 2 letters, 42 of matrices, 88
in n letters, 133 of subsets in group, 263
irreducible, 42, 43, 54 Proofs, methods of, 5ff.
irreducibility tests for, 46, 49, 53, 136, Proper,
144 subfield, 129
minimum, 154 subgroup, 203
monic, 48 subring, 129
multiplication of, 36 subset, 2
over 0, etc., 10, 35
prime, 42, 43
primitive, 44. 134 Quadratic equation, solution of, xiii, xiv, xvi
Gauss' theorem on, 44, 135 Quartic equation (polynomial), solution by
reducible, 42 radicals, 185, 306
root zero) of, 52, 53, 67 Quaternions, division ring of, 88, 132
separable, 285, 287 Quintie equation (polynomial), insolubility by
splitting field of, 149 radicals, 274, 304, 308ff.
substitution in, 52 Quotient group, ring (see factor group,
symmetric, 176ff. ring)
unique factorisation in, 48, 133ff. Quotient field (see field of fractions)
 Index 381

Radical(s), multiplicity of, 53

extension of a field, 274, 276 of unity, 190,219
soluble by, 185, 261, 276, 306 primitive, 281
solution of cubic, quartic by, 181ff, 306 primitive mod n, 214, 216
tower, 276 rational root test, 53
Range of function (map), 75 (a zero) of a polynomial, 52, 53, 67, 274
Rational numbers, I Rotation, 233
defined and characterised, 130 Ruffini P, 274
polynomials over, 10, 35, etc. Ruffini—Abel theorem, 303
Rational root test, 53 Ruler (a straightedge) and compass
ReaJ numbers, 1 construction, 170ff.
defined and characterised, 158 Russell paradox, 2
Reducible polynomial, 42
reductio ad absurdu,n, 6
Reflection, 233 Second isomorphism theorems, 152, 263
Reflexive binary relation, 59 Semigroup, 194
Regular 5-, 7-, l7-gon, 174, 175 Separable element, 289
Regular polygon, extension, 289
Gauss' theorem, 295 polynomial, 285, 287
symmetry group of, 190 Sequence,
Relation, Cauchy convergent, 157
binary (see binary relation) Fibonacci, 16
equivalence, 59, 61ff. Series,
of congruence mod n, 57, 62 composition, 262
order, 59 derived, 260
Relatively prime (a coprime), 31, 104 power, 39, 125
Remainder theorem, 52 Set(s), I

Restriction of function (map), 75 cartesian product of, 5, 250

Restricted direct product, 253 countable, 78
Right coset, 210 difference of, 4
Ring(s), element (a member) of,
axioms for, 84 empty, 2
Boolean, example of, 90 equality of, 2
characteristicof, 133 equinumerous, 78
commutative, 85 finite, example of, 2
direct sum of, 90 inclusion of, 2
division, 85, 86, 94 indexing, 4
Euclidean (ER), 108, 115 infinite, 2
factor (a quotient), 146 intersection of, 3
homomorphism of, 141 of generators for a group, 206
isomorphism of, 126 power, 5
near-, 86 (proper) subset of, 2
Noetherian, 99 singleton, 2
non-associative, 84, 90 spanning—(of vectors), 164
number- (a Zahiring), 83 union of, 3
of algebraic integers, 91 Simple,
of functions on R, 90 extension, 313
of matriccs, 88, 91 group, 247, 268ff.
power series, 39, 125 group of order 60 is A5, 270
quotient (see factor ring) ring, 102
relations between different kinds of, 86, Simplicity of alternating groups, 268
89 Singleton set, 2
simple, 102 Skew field (see division ring)
units in a general, 104 Soluble,
units in the number- +, >, 105 by radicals, 185, 261, 274, 276, 302ff.
with unity, 85 group, 258ff., 260, 261, 267
Ring-theoretic black hole, 98 group of length m, 260
Root(s), Space group, 233
adjoining, to a field, 141, 147, 149 Span(ning set), 164
conjugate complex, of a real equation, 55 Special linear group, 192, 271
382 Index

Splits, 149 Surjection, 75

Splitting field extension), 149, 276 Sylow P L M, 242
uniqueness of, 276 Sylow
Square-free integer, 91 p-subgroup, 246
Squares,sumof, 7,116 it-subgroup, 268
Squaring the circle, 174 Sylow's theorems, 245, 246
Steinitz E, 131 Symmetric,
Straightedge ruler) and compass binary relation, 59
construction, I 70ff. group, 189, t97ff.
Subfield(s), 98 polynomial(s), 176
intersection of, 98 elementary, 176
prime, 98, 131 fundamental theorem on, 176
proper, 129 Symmetry,
test for, 98 group of n-gon, 190
Subgroup(s), 203 plane, group, 234
centraliser of, 249
conjugate, 243 Tables
coset of, 210 for 4, 64, 66
cyclic, 203 group multiplication, 218
commutator), 259 Test,
direct product of, 251 Eisenstein's. 46,49, 136, 144
generated by, 206 for subfield, 98
generator of cyclic, 203, 207 for subgroup, 204, 207
Hall it-, 265 for subring, 96, 100
index of, 211 rational root, 53
intersection of, 205 Third isomorphism theorems, 153, 232
maximal normal, 262 Totient (see phi-function)
non-trivial, 203 Tower, radical, 276
normal, 225 Transformation (see mapping)
normaliser of, 243 Transitive,
of cyclic groups, 215 binary relation, 59
order of, 189,211 group, 307
p-, 247 Translation(s), 233
proper, 203 Transposition, 199
set of generators for, 206 Triple, ordered, 5
Sylow Triseetion of angle, 170, 173, 174
p-, 246 Trivial,
ii-, 268 homomorphism, 141
test for, 204, 207 subgroup, 203
trivial, 203 Twin primes problem, 58
union of, not a subgroup, 207
Subring(s), 96, 202 Union
intersection of, 98 of ideals not a subring, 101
proper, 129 of increasing sequence of ideals, 101
test for, 96, 100 of subgroups not a subgroup, 207
union of ideals not a, 101 of (sub)sets, 3
Subset(s), 2 Unique faetorisation, failure in
closed under binary operation, 78 105, 106
conjugate, in a group, 243 Unique faetorisation theorem for,
proper, 2 abelian groups, 255, 256, 258
restriction of map to, 75 Euclidean rings, ill
Substitution in polynomial, 52 integers, 33
Sufficient condition(s), 7 principal ideal domains, 112
Sum, polynomials over any field, 112, 133, 135
direct, of abelian groups, 251, 255ff. polynomials over 0, 47
of cosets in a ring, 146 Unique factorisation domain (UFD), 112
of ideals, 101 Unit(s),
of integers mod n, 64 in a power series ring, 106
of matrices, 87 in a rin,g, 104
of squares, 7, 116 in 105
 Index 383

ini,Q[x], 21,42 Well ordered, 17

Unity! integral domain, 137
complex roots of, 190, 219 Well ordering principle, 17
element in a ring, 85 Wilson, Sir J, 71
Unrestricted direct product, 253 Wilson's theorem, 71
Upper section, 158 converse of, 72
'Up to isomorphism', 130, 152, 217 WLOG, 22
Valuation norm) on an ER, 108
Value, x2+2=y3, 119
absolute (see modulus) x2+4=y3, 120
of a polynomial, 52 x3+y3+z3=0, 117
Venn diagram, 3 x4+y4=z2, 120
Venn J, 3

Wallpaper pattern, 234 Zahlkörper number field), xx

Wantzcl P, 174 Zahiring number ring), 83
Waring E, 70 Zero,
Webcr H, 274 divisor law, 11, 38
Wedderburn J H M, 94 divisors in general ring, 85
Well-definedness (see also ill-dcfinedness) divisors in i0, 66
in 65 element in ring, field, group, 85, 187
of coset sums and products, 146, 152, polynomial, 37
229 root) of a polynomial, 52, 53, 67, 274