Abstract Algebra - 3rd Edition by I.N. Herstein

Third Edition
l o N Hen-stein
Late Professor of Mathematics
University of Chicago
JOHN WILEY & SONS, INC.
NEW Y ORK CHI CHESTER WEI NHEI M BRI SBANE SI NGAPORE TORONTO
Cover Photograph: Charnl ey Residence, entryway.
Photo by Nick Merrick/Hedrich-Bl essing
This book was typeset in 10/12 Times Ten Roman by University Graphics, Inc.
The paper in this book was manufactured by a mil l whose forest management programs incl ude
sustained yiel d harvesting of it timberl ands. Sustained yiel d harvesting principl es ensure that the
number of trees cut each year does not exceed the amount of new growth.
Copyright 1999 by J ohn Wil ey & Sons, Inc. Al l rights reserved.
No part of this publ ication may be reproduced, stored in a retrieval system or transmitted in any
form or by means, el ectronic, mechanical , photocopying recording, scanning or otherwise,
except as permitted under Sections 107 or 108 of the 1976 United States Copyright Act, without
either the prior written permission of the Publ isher or authorization through payment of the
appropriate per-copy fee to the Copyright Cl earance Center, 222 Rosewood Drive, Danvers, MA
01923, (508) 750-8400, fax (508) 750-4470. Requests to the Publ isher for permission shoul d be
addressed to the Permissions Department, J ohn Wil ey & Sons, Inc. 605 Third Avenue, New
York, NY 10158-0012, (212) 850-6008, E-mail : PERMREQ@WI LEY .COM. To order books
or for customer service cal l l -800-CALL-WILEY(225-5945).
Library of Congress Cataloging in Publication Data
Herstein, I. N.
Abstract al gebra / I.N. Herstein. 3rd ed.
p. cm.
Incl udes index.
ISBN 0-471-36879-2
1. Al gebra, Abstract. I. Titl e
QA162.H47 1995 95-21470
512'.02^dc20 CIP
Printed in the United States of America.
10 9 8 7 6 5. 4 3 2
To Biska
Preface ix
U Thi ngs Fami l i ar and Less Fami l i ar 1
1 A Few Prel i mi nary Remarks 1
2 Set Theo ry 3
3 Mappi ngs 8
4 A(S) (The Set of 1-1 Mappi ngs of S o nto I tsel f) 16
5 The I ntegers 21
6 Mathemati cal I nducti o n 29
7 Co mpl ex Numbers 32
JM Gr o u p s 4 0
1 Definitions and Exampl es of Gro ups 40
2 So me Simpl e Remarks 48
3 Subgro ups 51
4 Lagrange's Theo rem 56
5 Ho mo mo rphi sms and No rmal Subgro ups
6 Facto r Gro ups 77
7 The Ho mo mo rphi sm Theo rems 84
8 Cauchy's Theo rem 88
9 Di rect Pro ducts 92
10 Finite Abel i an Gro ups (Opti o nal ) 96
11 Coniugacy and Svl ow's Theo rem (Opti o nal )
66
101
Co n t en t s
Th e Sy mmet r i c Gr o u p 10 8
1 Prel i mi nari es 108
2 Cycl e Deco mpo si ti o n 111
3 Odd and Even Permutati o ns 119
Ri n g Th e o r y 1 2 5
1 Definitions and Exampl es 125
2 So me Simpl e Resul ts 137
3 I deal s, Ho mo mo rphi sms, and Quo ti ent Ri ngs 139
4 Maxi mal I deal s 148
5 Po l yno mi al Rings 151
6 Pol ynomi al s over the Rati o nal s 166
7 Fiel d of Quo ti ents of an I ntegral Do mai n 172
Fi el ds 1 7 6
1 Exampl es of Fiel ds 176
2 A Brief Excursi on i nto Vecto r Spaces 180
3 Fiel d Extensi o ns 191
4 Fi ni te Extensi o ns 198
5 Constructibil ity 201
6 Ro o ts of Pol ynomi al s 207
Sp ec i al Topi c s ( Op t i o n al ) 2 1 5
1 The Simpl icity of A 215
2 Finite Fiel ds I 221
3 Finite Fiel ds I I : Existence 224
4 Finite Fiel ds I I I : Uni queness 227
5 Cycl otomic Po l yno mi al s 229
6 Liouvil l e's Cri teri o n 236
7 The I rrati onal i ty of IT 239
I ndex 243
PREFACE TO THE THIRD EDITION
When we were asked to prepare the thi rd edi ti on of this bo o k, it was our con-
sensus that it shoul d no t be al tered in any significant way, and that Herstei n's
informal styl e shoul d be preserved. We feel that o ne of the bo o k's virtues is
the fact that it covers a big chunk of abstract al gebra in a co ndensed and in-
teresting way. At the same ti me, wi tho ut trivial izing the subject, it remai ns ac-
cessibl e to mo st undergraduates.
We have, ho wever, co rrected mi no r erro rs, strai ghtened o ut inconsis-
tencies, cl arified and expanded so me proofs, and added a few exampl es.
To resol ve the many typo graphi cal pro bl ems of the second edition,
Prenti ce Hal l has had the bo o k co mpl etel y retypesetmaki ng it easier and
mo re pl easurabl e to read.
It has been po i nted o ut to us that so me instructors woul d find it useful
to have the Symmetri c Gro up S and the cycl e no tati o n avail abl e in Chapter
2, in o rder to pro vi de mo re exampl es of groups. Rather than al ter the
arrangement of the contents, thereby disturbing the original bal ance, we sug-
gest an al ternate ro ute thro ugh the materi al , which addresses this concern.
After Section 2.5, one coul d spend an ho ur discussing permutati o ns and thei r
cycl e deco mpo si ti o n (Sections 3.1 and 3.2), l eaving the proofs until l ater. The
students might then go over several past exampl es of finite groups and expl ic-
itl y set up i so mo rphi sms with subgro ups of S. This exercise wo ul d be mo ti -
vated by Cayl ey's theo rem, quo ted in Section 2.5. At the same ti me, it wo ul d
have the beneficial resul t of maki ng the students mo re comfortabl e with the
co ncept of an i so mo rphi sm. The instructor coul d then weave in the vari ous
subgro ups of the Symmetri c Gro ups S as exampl es thro ugho ut the remai n-
x Pr ef ac e t o Th i r d Ed i t i o n Ch . 6
der of Chapter 2. If desi red, o ne coul d even i ntro duce Sections 3.1 and 3.2
after Section 2.3 or 2.4.
Two changes in the fo rmat have been made since the first edition. First,
a Symbol List has been incl uded to facil itate keepi ng track of termi nol ogy.
Second, a few pro bl ems have been marked with an asterisk (*). These serve
as a vehicl e to i ntro duce concepts and simpl e arguments that rel ate in so me
i mpo rtant way to the discussion. As such, they shoul d be read careful l y.
Final l y, we take this o ppo rtuni ty to thank the many individual s who se
col l ective efforts have hel ped to i mpro ve this edition. We thank the review-
ers: Kwangil Ko h from No rth Carol i na State University, Do nal d Passman
from the University of Wisconsin, and Ro bert Zi nc from Purdue University.
And, of course, we thank Geo rge Lobel l and El ai ne Wetterau, and o thers at
Prenti ce Hal l who have been most hel pful .
Barbara Co rtzen
Davi d J. Wi nter
I n the l ast hal f-century or so abstract al gebra has beco me increasingl y i mpor-
tant no t onl y in mathemati cs itsel f, but al so in a vari ety of o ther discipl ines.
Fo r i nstance, the i mpo rtance of the resul ts and concepts of abstract al gebra
pl ay an ever mo re i mpo rtant rol e in physics, chemistry, and co mputer science,
to cite a few such outside fiel ds.
I n mathemati cs itsel f abstract al gebra pl ays a dual rol e: that of a unify-
ing l ink between di sparate parts of mathemati cs and that of a research subject
with a highl y active l ife of its o wn. I t has been a fertil e and rewardi ng research
area bo th in the l ast 100 years and at the present mo ment. So me of the great
acco mpl i shments of o ur twenti eth-century mathemati cs have been precisel y
in this area. Exciting resul ts have been pro ved in gro up theory, co mmutati ve
and no nco mmutati ve ring theo ry, Li e al gebras, J o rdan al gebras, combi na-
torics, and a host of o ther parts of what is kno wn as abstract al gebra. A sub-
ject that was once regarded as esoteri c has beco me consi dered as fairl y down-
to -earth for a l arge cross section of schol ars.
The purpo se of this bo o k is twofol d. Fo r tho se readers who either want
to go o n to do research in mathemati cs o r in so me al l ied fiel ds that use al ge-
braic noti ons and metho ds, this bo o k shoul d serve as an i ntro ducti o nand,
we stress, onl y as an i ntro ducti o nto this fascinating subject. Fo r interested
readers who want to l earn what is going on in an engaging part of mo dern
mathemati cs, this bo o k coul d serve that purpo se, as wel l as pro vi de them with
so me highl y usabl e tool s to appl y in the areas in which they are i nterested.
The choice of subject matter has been made with the objective of intro-
ducing readers to so me of the fundamental al gebraic systems that are bo th in-
xi i Pr ef ac e t o Fi r st Ed i t i o n
teresti ng and of wide use. Mo reo ver, in each of these systems the aim has
been to arrive at so me significant resul ts. There is l ittl e purpo se served in
studying so me abstract object wi tho ut seeing so me nontrivial co nsequences of
the study. We ho pe that we have achi eved the goal of presenti ng i nteresti ng,
appl icabl e, and significant resul ts in each of the systems we have cho sen to
discuss.
As the reader wil l soon see, there are many exercises in the bo o k. They
are often divided into three categories: easier, middl e-l evel , and harder (with
an occasional very hard). The purpo se of these pro bl ems is to al l ow students
to test thei r assimil ation of the materi al , to chal l enge thei r mathemati cal inge-
nuity, to prepare the gro und for materi al that is yet to co me, and to be a
means of devel opi ng mathemati cal insight, intuition, and techni ques. Readers
shoul d no t beco me di scouraged if they do no t manage to sol ve al l the pro b-
l ems. The i ntent of many of the pro bl ems is that they be tri edeven if no t
sol vedfor the pl easure (and frustration) of the reader. So me of the pro b-
l ems appear several ti mes in the bo o k. Trying to do the pro bl ems is undo ubt-
edl y the best way of going abo ut l earning the subject.
We have strived to present the materi al in the l anguage and to ne of a
cl assroom l ecture. Thus the presentati o n is so mewhat chatty; we ho pe that
this wil l put the readers at thei r ease. An attempt is made to give many and
reveal ing exampl es of the vari ous concepts discussed. So me of these exam-
pl es are carried fo rward to be exampl es of o ther pheno mena that co me up.
They are often referred to as the discussion progresses.
We feel that the bo o k is sel f-contained, except in o ne secti o nthe sec-
o nd l ast o ne of the bo o kwhere we make impl icit use of the fact that a pol y-
nomi al over the co mpl ex fiel d has co mpl ex roots (that is the cel ebrated Fun-
damental Theorem of Algebra due to Gauss), and in the l ast section where we
make use of a l ittl e of the cal cul us.
We are grateful to many peo pl e for thei r co mments and suggestions o n
earl ier drafts of the bo o k. Many of the changes they suggested have been in-
co rpo rated and shoul d i mpro ve the readabil ity of the bo o k. We shoul d l ike to
express o ur special thanks to Professor Marti n I saacs for his highl y useful
co mments.
We are al so grateful to Fred Fl o wers for his usual superb j o b of typing
the manuscri pt, and to Mr. Gary W. Ostedt of the Macmi l l an Co mpany for
his enthusi asm for the project and for bringing it to publ i cati on.
Wi th this we wish al l the readers a happy voyage o n the mathemati cal
j o urney they are abo ut to undertake into this del ightful and beautiful real m
of abstract al gebra.
I .N.H.
aES a is an el ement of the set S, 3
aS a is no t an el ement of the set S, 3
SCT,TDS S is a subset of the set T, 3
S=T The sets 5 and Tare equal (have the same el ements), 4
0
The empty set, 4
AUB The uni o n of the sets A and B, 4
AC)B The i ntersecti on of the sets A and B, 4
{s E 5 | s satisfies P] The subset of el ements of S satisfying P, 4
A - B The difference of the sets A and fi, 4
A' The co mpl ement of A, 5
(a, b) Ordered pair consisting of a, b (see al so bel ow), 5
AX B The Cartesi an pro duct of A and B, 5
U The set of real numbers, 8
f : S- > T Functi o n from the set S to the set T, 8
m
I mage of the el ement s under the functi o n/, 8
i: S S, i
s
The identity function o n S, 9
I nverse i mage of t under f, 10
r
l
(A)
I nverse i mage of a subset A of T under / : S T, 10
fg, fg
Co mpo si ti o n or pro duct of functions / andg, 11,18
A(S) Set of 1-1 mappi ngs from a set S to S, 16
s
Symmetri c gro up of degree n, 16,109
n\ n factorial , 17
Z Set of i ntegers, 21
0(s) Orbi t of s rel ative to mappi ng /, 21
xi v Sy mb o l Li st
N Set of positive integers, 21
m | n m divides n, 22
m | n m does no t divide n, 22
(a, b) Greatest co mmo n divisor of a, b (see al so abo ve), 23
C Set of co mpl ex numbers, 32
i,i Square ro o ts of 1, 32
z = a + bi Co mpl ex number z with real part a and i magi nary
part b, 32
z = a - bi Conj ugate of co mpl ex number z = a + bi, 32
IIz I nverse of the compl ex number z, 33
|z| Abso l ute val ue of compl ex number z, 34
?(cos 9 + / sin 6) Po l ar form of a co mpl ex number, 35
9 Primitive nth ro o t of unity, 36, 42
Q Set of rati onal numbers, 42
E
n
Gro up of nth ro o ts of unity, 42
|G | Order of the gro up G, 42
C(a) Central i zer of a in G, 53,102
(a) Cycl ic gro up generated by a, 53
Z(G) Center of gro up G, 53
a ~ b a is equi val ent to b in a specified sense, 57
a = b mo d n a is co ngruent to b mo dul o n (l ong fo rm), 57
a = b(n) a is co ngruent to b mo dul o n (short fo rm), 57
[a] Cl ass of al l b equi val ent to a, 58
cl (a) Conjugacy cl ass of a, 58,101
o(a) Order of el ement a of a gro up, 60
i
G
{H) I ndex ofHin G, 59
Z Set of i ntegers mo d n, 61
U Gro up of invertibl e el ements of Z, 62
<p(n) Eul er cp function (phi function), 62
Hb Ri ght coset of subgro up H, 58
aH Left coset of subgro up H, 64
G ~ G' Gro up G is i so mo rphi c to gro up G', 68
cp(G) I mage of ho mo mo rphi sm, 70
Ker cp Kernel of the ho mo mo rphi sms 9, 70,140
N <i G TV is a no rmal subgro up of G, 72
GIN Quo ti ent of a gro up G by a subgro up N, 78
AB Pro duct of subsets A, B of a gro up, 79
G
1
X G
2
X . . . X G Di rect pro duct of G
h
G
2
, . . . , G, 93
fab . . . c
N
\ii v ... \v
/
(a, b,... , c) Cycl e sending a to b,. .. , c to a, 111
A Al ternati ng gro up of degree n, 121,215
a
0
+ ct^' + a
2
j + a
3
k Quaterni o n, 131
det x Determi nate of the 2X2 matri x x, 136
Permutati o n sendi ng a to 11, b to v,..., c to w, 110
Sy mb o l Li st xv
H{F) Ri ng of quaterni o ns over F, 136
R$
Di rect sum of rings R, S, 146
(a) I deal generated by a in a co mmutati ve ring, 145
F[x] Po l yno mi al ring over the fiel d F, 152
deg p(x) Degree of po l yno mi al p(x), 153
I deal generated by g(x) in a po l yno mi al ring, 157
g(x)\f(x) Po l yno mi al g(x) divides f(x), 157
R[x] Po l yno mi al ring over ring R, 163
F(x) Fiel d of rati onal functions in x over F, 177
v G V Vecto r v in a vecto r space V, 180
av Scal ar a ti mes vecto r v, 180
a
1
v
1
+ . . . + av
n
Li near co mbi nati o n of vectors v
x
,.. . , v, 181
(v
u
v
2
, ...,v) Subspace spanned by v
t
, v
2
, , v, 181
VW Di rect sum of vector spaces V, W, 181
di m
F
(V) Di mensi o n of V over F, 186
U+W Sum of subspaces U, W of V, 190
[K:F] Degree of K over F, 191
F[a] Ri ng generated by a over F, 196
F{a) Fiel d extensi o n o btai ned by adjoining to a to F, 196
E(K) Fiel d of al gebraic el ements of K over F, 198
Fo rmal derivative of po l yno mi al /(x), 227
4>M
nth cycl otomic pol ynomi al , 230
ABSTRACT ALGEBRA
TH1M
'M/llMlt
f
AMI UAi
1. A FEW PREL I MI NA RY REMA RK S
Fo r many readers this bo o k wil l be thei r first contact with abstract mathe-
matics. The subject to be discussed is usual l y cal l ed "abstract al gebra," but
the difficul ties that the reader may enco unter are no t so much due to the "al -
gebra" part as they are to the "abstract" part.
On seeing some area of abstract mathematics for the first time, be it in
anal ysis, topol ogy, or what-not, there seems to be a co mmo n reaction for the
novice. This can best be described by a feel ing of being adrift, of not having
something sol id to hang on to. This is no t too surprising, for whil e many of the
ideas are fundamental l y quite simpl e, they are subtl e and seem to el ude one's
grasp the first time around. One way to mitigate this feel ing of l imbo, or asking
onesel f "What is the point of al l this?," is to take the concept at hand and see
what it says in particul ar cases. I n other words, the best road to good under-
standing of the notions introduced is to l ook at exampl es. This is true in al l of
mathematics, but it is particul arl y true for the subject matter of abstract al gebra.
Can o ne, with a few strokes, quickl y describe the essence, purpo se, and
backgro und for the materi al we shal l study? Let's give it a try.
We start with so me col l ection of objects S and endo w this col l ection
with an al gebraic structure by assumi ng that we can co mbi ne, in one or sev-
eral ways (usual l y two ), el ements of this set S to obtai n, once mo re, el ements
of this set S. These ways of co mbi ni ng el ements of S we cal l operations o n S.
2 Th i n g s Fami l i ar an d Less Fami l i ar Ch . 1
Then we try to condition or regul ate the nature of S by i mposi ng certai n
rul es on ho w these o perati o ns behave on S. These rul es are usual l y cal l ed the
axioms defining the parti cul ar structure on 5. These axioms are for us to de-
fine, but the choice made co mes, historical l y in mathemati cs, from noticing
that there are many co ncrete mathemati cal systems that satisfy these rul es or
axioms. We shal l study so me of the basic axiomatic al gebraic systems in this
bo o k, namel y groups, rings, and fields.
Of course, o ne coul d try many sets of axioms to define new structures.
What woul d we requi re of such a structure? Certainl y we woul d want that
the axioms be consistent, that is, that we shoul d no t be l ed to so me nonsensi -
cal contradi cti on co mputi ng within the framewo rk of the al l owabl e things the
axioms permi t us to do . But that is no t eno ugh. We can easil y set up such al -
gebraic structures by i mposi ng a set of rul es on a set S that l ead to a patho -
l ogical or weird system. Furthermo re, there may be very few exampl es of
so methi ng obeying the rul es we have l aid down.
Ti me has shown that certain structures defined by "axi o ms" pl ay an im-
po rtant rol e in mathemati cs (and o ther areas as wel l ) and that certai n o thers
are of no interest. The o nes we menti o ned earl ier, namel y groups, rings, and
fiel ds, have stood the test of ti me.
A wo rd abo ut the use of "axi o ms." I n everyday l anguage "axi o m"
means a sel f-evident truth. But we are no t using everyday l anguage; we are
deal ing with mathemati cs. An axi om is no t a universal truthbut o ne of sev-
eral rul es spel l ing o ut a given mathemati cal structure. The axi om is true in
the system we are studying because we have forced it to be true by hypo the-
sis. It is a l icense, in the parti cul ar structure, to do certain things.
We return to so methi ng we said earl ier abo ut the reacti o n that many
students have on thei r first enco unter with this ki nd of al gebra, namel y a l ack
of feel ing that the materi al is so methi ng they can get thei r teeth i nto. Do no t
be di scouraged if the initial expo sure l eaves you in a bit of a fog. Stick with
it, try to understand what a given co ncept says, and most i mpo rtantl y, l o o k at
particul ar, concrete exampl es of the concept under discussion.
P R O B L E M S
1. Let S be a set having an o perati o n * which assigns an el ement a * b of S
for any a, b G S. Let us assume that the fol l owing two rul es hol d:
1. If a, b are any objects in S, then a * b = a.
2. If a, b are any objects in S, then a * b = b * a.
Show that S can have at most o ne object.
Sec . 2
Set Th eo r y 3
2. Let S be the set of al l integers 0, 1, 2, . . . , n,.... Fo r a, b in S define
* by a*b = a b. Verify the fol l owing:
(a) a * b j= b * a unl ess a = b.
(b) (a * b) * c + a * (b * c) in general . Under what condi ti ons on a, b, c is
(a * b) * c = a * (6 * c)?
(c) The i nteger 0 has the pro perty that a * 0 = A for every a in 5.
(d) Fo r a in S, a * a = 0.
3. Let 5 consist of the two objects and A. We define the o perati o n * o n S
by subjecting and A to the fol l owing conditions:
1. n * A = A = A * Q
2. * = .
3. A * A = .
Verify by expl icit cal cul ation that if a, b, c are any el ements of S (i.e., a, b
and c can be any of or A ), then:
(a) a * b is in S.
(b) (a * b) * c = a * (b * c).
(c) a * b = b * a.
(d) There is a parti cul ar a in S such that a * b = b * a = b for all b in S.
. (e) Given b in 5, then b * b = a, where a is the parti cul ar el ement in Part
(d).
2. SET THEORY
Wi th the changes in the mathemati cs curri cul um in the school s in the Uni ted
States, many col l ege students have had so me expo sure to set theo ry. This in-
tro ducti o n to set theo ry in the school s usual l y incl udes the el ementary no -
tions and o perati o ns with sets. Go i ng o n the assumpti o n that many readers
wil l have so me acquai ntance with set theo ry, we shal l give a rapi d survey of
those parts of set theo ry that we shal l need in what fol l ows.
First, however, we need so me no tati o n. To avoid the endl ess repeti ti o n
of certai n phrases, we i ntro duce a sho rthand for these phrases. Let S be a
col l ection of objects; the objects of S we cal l the elements of S. To deno te
that a given el ement, a, is an el ement of S, we wri te a G Sthis is read "a is
an el ement of 5." To deno te the co ntrary, namel y that an object a is not an
el ement of S, we write a S. So, for i nstance, if S deno tes the set of al l posi-
tive i ntegers 1, 2, 3, . . . , , . . . , then 165 E S, whereas - 13 S.
We often want to kno w or pro ve that given two sets S and T, o ne of
these is a part of the other. We say that S is a subset of T, which we wri te
S C T (read "5 is co ntai ned in 7"") if every el ement of S is an el ement of T.
4 Th i n g s Fami l i ar an d Les s Fami l i ar Ch . 1
I n terms of the no tati o n we no w have: S C T if s G S impl ies that s G T. We
can al so deno te this by writing T D S, read "T contains S." (This does no t ex-
cl ude the possibil ity that S = T, that is, that S and T have exactl y the same
el ements.) Thus, if T is the set of al l positive integers and S is the set of al l
positive even integers, then S C T, and 5 is a subset of T. I n the definition
given abo ve, S D S for any set S; that is, S is al ways a subset of itsel f.
We shal l frequentl y need to show that two sets S and T, defined per-
haps in distinct ways, are equal , that is, they consist of the same set of el e-
ments. The usual strategy for pro vi ng this is to show that bo th S C T and
T C S. Fo r instance, if S is the set of al l positive integers having 6 as a factor
and T is the set of al l positive i ntegers havi ng bo th 2 and 3 as factors, then
S = T. (Prove!)
The need al so arises for a very pecul i ar set, namel y o ne having no el e-
ments. This set is cal l ed the null or empty set and is deno ted by 0 . I t has the
pro perty that it is a subset of any set S.
Let A, B be subsets of a given set S. We no w i ntro duce metho ds of con-
structing o ther subsets of S from A and B. The first of these is the union of A
and B, wri tten A U B, which is defined: A U B is that subset of S consisting
of tho se el ements of S that are el ements of A or are el ements of B. The "o r"
we have just used is so mewhat different in meani ng from the o rdi nary usage
of the word. Here we mean that an el ement c is in A U B if it is in A, o r is in
B, or is in both. The "o r" is no t meant to excl ude the possibil ity that bo th
things are true. Co nsequentl y, for i nstance, A U A = A.
If A = {1, 2, 3}and B ={2, 4, 6,10}, then A U B ={1, 2, 3, 4, 6,10}.
We no w pro ceed to o ur second way of constructing new sets from ol d.
Agai n l et A and B be subsets of a set S; by the intersection of A and B, writ-
ten A D B, we shal l mean the subset of S consisting of tho se el ements that
are bo th in A and in B. Thus, in the exampl e abo ve, A n B = {2}. It shoul d
be cl ear from the definitions invol ved that A n B C A and A n B C B.
Parti cul ar exampl es of i ntersecti ons that hol d universal l y are: A D A = A,
A n S = A, A D 0 = 0 .
This is an o ppo rtune mo ment to i ntro duce a no tati o nal devi ce that wil l
be used ti me after ti me. Gi ven a set S, we shal l often be cal l ed on to de-
scribe the subset A of S, who se el ements satisfy a certai n pro perty P. We
shal l write this as A = {s G 5 | s satisfies P). Fo r i nstance, if A, B are subsets
of S, then Al)B = [s(ES\s&A or s<=B] whil e AnB = {sGS\sE.A
and^G B).
Al tho ugh the no ti o ns of uni o n and intersection of subsets of S have
been defined for two subsets, it is cl ear ho w o ne can define the uni o n and in-
tersecti on of any number of subsets.
We no w i ntro duce a third o perati o n we can perfo rm on sets, the differ-
ence of two sets. If A, B are subsets of 5, we define A - B = {a G A\a $L B\.
Sec . 2 Set Th eo r y 5
So if A is the set of al l positive i ntegers and fl is the set of al l even integers,
then A B is the set of al l positive o dd integers. I n the parti cul ar case when
A is a subset of 5, the difference S A is cal l ed the complement of A in S
and is wri tten A'.
We represent these three o perati o ns pictorial l y. If A is @ and B is ,
then
1. A U B =mWM is the shaded area.
2. ^ H B = ( A M a l i s the shaded area.
3. A B = Kg,4it ) b ) is the shaded area.
4. B - A ={ A [ KBVA is the shaded area.
No te the rel ati on amo ng the three o perati o ns, namel y the equal ity
A U B = (A n B) U (A - B) U (B - A). As an il l ustration of ho w one goes
abo ut pro vi ng the equal ity of sets co nstructed by such set-theoreti c construc-
tions, we pro ve this l atter al l eged equal ity. We first show that (A fl B) U
(A - B) U (fl - A) C A U fl ; this part is easy for, by definition, A D B C A,
A - B C A, and B - A C B, hence
{A n fl ) U {A - B) U (5 - A) C A U A U 5 = A U B.
No w for the o ther direction, namel y that A U B C (A D B) U (A - 5) U
(5 - A). Gi ven u E A U 5, if 4 and G 5, then u G i f l B , so it is cer-
tainl y in (A n B) U (A - fl ) U (fl - A). On the o ther hand, if u G A but
w ^ fl , then, by the very definition of A fl , u G A 5, so again it is cer-
tainl y in (A n fl ) U {A - fl ) U (fl - A). Final l y, if u G fl but u A, then
m G fl - A, so again it is in (A Pi fl ) U (A - fl ) U (fl - A). We have thus
co vered al l the possibil ities and have sho wn that A U fl C (A Pi fl ) U
(A - fl ) U (fl A). Havi ng the two o ppo si te contai ni ng rel ations of A U fl
and (A n fl ) U (A - fl ) U (fl - A), we o btai n the desired equal ity of these
two sets.
We cl ose this brief review of set theo ry wi th yet ano ther constructi on
we can carry o ut on sets. This is the Cartesian product defined for the two
sets A, B by A X fl = [(a, b) \ a G A, b G fl }, where we declare the o rdered
pai r (a, b) to be equal to the o rdered pai r (a
l 5
b{) if and onl y if a = a
x
and
b = b\. Here, to o , we need not restrict oursel ves to two sets; for instance, we
can define, for sets A, B, C, thei r Cartesi an pro duct as the set of o rdered
tripl es (a, >, c ) , where a G A, b G B, c G C and where equal ity of two or-
dered tripl es is defined co mpo nent-wi se.
P R O B L E M S
Easier Problems
1. Descri be the fol l owing sets verbal l y.
(a) S = (Mercury, Venus, Earth, . . . , Pl uto).
(b) S = {Al abama, Al aska, . .. , Wyomi ng}.
2. Descri be the fol l owing sets verbal l y.
(a) S = {2, 4, 6, 8, . . . }.
(b) S = {2, 4, 8,16, 32, . ..}.
(e) S= {1,4,9,16,25,36,...}.
3. If A is the set of al l resi dents of the Uni ted States, B the set of al l Cana-
di an citizens, and C the set of al l wo men in the worl d, descri be the sets
Ar\BnC,A-B,A-C,C-A verbal l y.
4. If A = {1, 4, 7, a} and B = {3, 4, 9, 11} and you have been tol d that
A n B = {4, 9}, what must a be?
S.lfACB and B C C, pro ve that A C C.
6. If AC B, pro ve that i U C C S U C f o r any set C.
7. Show that AUB = BUA and AnB = BC\ A.
8. Pro ve that (A - B) U (B - A) = (A U B) - (A n 5). What do es this
l ook l ike pictorial l y?
9. Pro ve that A (1 (B U C) = (A D B) U (A D C).
10. Pro ve that A U (B f) C) = {A U 5) n (A U C).
11. Wri te down al l the subsets of S = {1, 2, 3, 4}.
Middle-Level Problems
*12. If C is a subset of S, l et C deno te the co mpl ement of C in S. Pro ve the
De Morgan Rules for subsets A, B of S, namel y:
(a) (A n B)' = A' U 5' .
(b) (A U B)' =A'n B'.
* 13. Let S be a set. Fo r any two subsets of S we define
A + B = (A - B) U (B - A) and A B = A H B.
Sec . 2 Set Th eo r y 7
Pro ve that:
(a) A + B = B + A.
(b) A + 0 = A.
( c ) A A = A.
(d) A + A = 0.
(e) A + (B + C) = {A + B) + C.
(f) If A + B = A + C, then B = C.
( g ) A (fl + C) =A B + A C.
*14. If C is a finite set, l et m(C) deno te the number of el ements in C. If A, B
are finite sets, pro ve that
m(A U fl) = m(A) + m(B) - m(A n fl).
15. Fo r three finite sets A, fl , C find a fo rmul a for m(A U fl U C). (Hi nt:
First consider D = fl U C and use the resul t of Pro bl em 14.)
16. Take a shot at finding m{A
x
U A
2
U U 4 ) for n finite sets A
U
A
2
, ... ,
A.
17. Use the resul t of Pro bl em 14 to show that if 80% of al l Ameri cans have
go ne to high school and 70% of al l Ameri cans read a dail y newspaper,
then at least 50% of Ameri cans have bo th go ne to high school and read a
dail y newspaper.
18. A publ i c o pi ni o n pol l shows that 93% of the po pul ati o n agreed with the
go vernment o n the first decision, 84% on the second, and 74% on the
third, for three decisions made by the go vernment. At l east what per-
centage of the po pul ati o n agreed with the go vernment on al l three deci-
sions? (Hi nt: Use the resul ts of Pro bl em 15.)
19. I n his bo o k A Tangled Tale, Lewis Carrol l pro po sed the fol l owing riddl e
abo ut a gro up of disabl ed veterans: "Say that 70% have l ost an eye, 75%
an ear, 80% an arm, 85% a l eg. What percentage, at least, must have l ost
al l fo ur?" Sol ve Lewis Carrol l 's pro bl em.
*20. Show, for finite sets A, fl , that m(A X fl ) = m(A)m(B).
21. If S is a set having five el ements:
(a) Ho w many subsets does S have?
(b) Ho w many subsets havi ng four el ements does S have?
(c) Ho w many subsets having two el ements does S have?
Harder Pro bl ems
22. (a) Sho w that a set having n el ements has 2" subsets.
(b) If 0 < m < n, ho w many subsets are there that have exactl y m el e-
ments?
3. MA PPI NGS
One of the trul y universal concepts that runs thro ugh al most every phase of
mathemati cs is that of a function or mapping from o ne set to ano ther. One
coul d safel y say that there is no part of mathemati cs where the no ti o n does
no t arise or pl ay a central rol e. The definition of a function from o ne set to
ano ther can be given in a formal way in terms of a subset of the Cartesi an
pro duct of these sets. I nstead, here, we shal l give an informal and admi ttedl y
no nri go ro us definition of a mappi ng (function) from o ne set to ano ther.
Let S, T be sets; a function or mapping f from S to T is a rule that as-
signs to each el ement s G S a unique el ement t G T. Let's expl ain a l ittl e
mo re tho ro ughl y what this means. If s is a given el ement of S, then there is
only one el ement t in T that is associated to s by the mappi ng. As s varies
over S, t varies over T (in a manner dependi ng on s). No te that by the defini-
tion given, the fol l owing is not a mappi ng. Let S be the set of al l peo pl e in
the worl d and T the set of al l countri es in the worl d. Let / be the rul e that as-
signs to every perso n his or her co untry of citizenship. Then / is no t a map-
pi ng from S to T. Why no t? Because there are peo pl e in the wo rl d that enjoy
a dual citizenship; for such peo pl e there woul d no t be a unique co untry of cit-
izenship. Thus, if Mary J o nes is bo th an Engl ish and French citizen, / wo ul d
no t make sense, as a mappi ng, when appl i ed to Mary J o nes. On the o ther
hand, the rul e / : (R R, where R is the set of real numbers, defined by
f(a) = a
1
for a G R, is a perfectl y go o d function from R. to R. I t shoul d be
no ted that/ ( - 2) = ( - 2)
2
= 4 =/(2), and/ ( - a) =f{a) for al l a G R.
We deno te that / is a mappi ng from S to T by / : S T and for the
f G T menti o ned abo ve we wri te t = f(s); we cal l t the image of s under /.
The co ncept is hardl y a new o ne for any of us. Since grade school we
have constantl y enco untered mappi ngs and functions, often in the form of
formul as. But mappi ngs need no t be restricted to sets of numbers. As we see
bel ow, they can occur in any area.
Exampl es
1. Let 5 = {al l men who have ever l ived} and T = {al l wo men who have ever
l ived}. Define f:S^>T by f(s) = mo ther of s. Therefo re, /(J o hn F. Ken-
nedy) = Ro se Kennedy, and according to o ur definition, Ro se Kennedy is
the i mage under / of J o hn F. Kennedy.
2. Let S = {al l l egal l y empl o yed citizens of the Uni ted States} and T = {posi-
tive integers}. Define, for s S , f(s) by f(s) = Social Security Number of s.
(Fo r the purpo se of this text, l et us assume that al l l egal l y empl o yed citizens
of the Uni ted States have a Social Security Number.) Then / defines a map-
pi ng from . to T.
Sec. 3 Map p i n g s 9
3. Let S be the set of al l objects for sal e in a grocery store and l et T = {al l
real numbers}. Define / : 5 >T by f(s) = pri ce of s. This defines a mappi ng
from S to T.
4. Let S be the set of al l i ntegers and l et T = S. Define f:S> Thy f(m) =
2m for any i nteger m. Thus the i mage of 6 under this mappi ng, /(6), is given
by/ ( 6) = 2 6 = 12, whil e that of - 3, / ( - 3) , is given by / ( - 3) = 2(- 3) =
- 6. If s
u
s
2
G S are in S and f(s
x
) = f(s
2
), what can you say abo ut s
1
and s
2
7
5. Let S = T be the set of al l real numbers; define / : S - T by /(s) = ^
2
.
Do es every el ement of T co me up as an i mage of so me s G SI If not, ho w
woul d yo u describe the set of al l i mages \f(s) \s G 5}? When is f(s{) =
6. Let 5 = T be the set of al l real numbers; def i ne/: S >Tb y f(s) = s
3
. Thi s
is a function from 5 to T. What can yo u say abo ut \f(s) \ s E S}? When is
7. Let T be any no nempty set and l et S = T X T, the Cartesi an pro duct of T
with itsel f. Def i ne/ : T X T T b y f(t
l
, t
2
) = t
L
. This mappi ng from T X T
to T is cal l ed theprojection of I X T o nto its first co mpo nent.
8. Let S be the set of al l positive i ntegers and l et T be the set of al l positive
rati o nal numbers. Define / : S X 5 -> T by / (m, ) = m/n. Thi s defines a
mappi ng from S X 5 to T. No te that / ( l , 2) = | whil e / ( 3, 6) = f = =
/ ( l , 2), al tho ugh (1, 2) ^ (3, 6). Descri be the subset of 5 X S consisting of
tho se (a, b) such that/(a, b) = |.
The mappi ngs to be defined in Exampl es 9 and 10 are mappi ngs that
occur for any no nempty sets and pl ay a special rol e.
9. Let 5, T be no nempty sets, and l et r
0
be a fixed el ement of T. Defi ne
f:S> T by f(s) = 1
0
for every s G S; / is cal l ed a constant function from
S to T.
10. Let S be any no nempty set and define i:S*Sby i(s) = s for every
s G S. We cal l this function of S to itsel f the identity function (or identity map-
ping) o n S. We may, at times, deno te it by i
s
(and l ater in the bo o k, by e).
No w that we have the no ti o n of a mappi ng we need so me way of i denti -
fying when two mappi ngs from o ne set to ano ther are equal . This is no t
Go d given; it is for us to deci de ho w to decl are f = g where f:S->T and
g : S - T. What is mo re natural than to define this equal ity via the actions of
/ and g o n the el ements of S? Mo re precisel y, we decl are that / = g if and
onl y if f(s) = g(s) for every s G S. If S is the set of al l real numbers and / is
defined o n S by f(s) = s
z
+ 2s + 1, whil e g is defined on S by g(s) =
(s + l )
2
, o ur definition of the equal ity o f / and g is merel y a statement of the
famil iar identity (s + l )
2
= s
z
+ 2s + 1.
10 Th i n g s Fami l i ar an d Less Fami l i ar Ch , 1
Havi ng made the definition of equal i ty of two mappi ngs, we no w want
to singl e out certain types of mappi ngs by the way they behave.
Definition. The mappi ng / : S T is onto or surjective if every t G T
is the image under / of so me s G S; that is, if and onl y if, given t G T, there
exists an s G 5 such that t = f(s).
I n the exampl es we gave earl ier, in Exampl e 1 the mappi ng is no t o nto ,
since no t every wo man that ever l ived was the mo ther of a mal e chil d. Simi-
l arl y, in Exampl e 2 the mappi ng is not o nto , for not every positive integer is
the Social Security Number of so me U.S. citizen. The mappi ng in Exampl e 4
fail s to be o nto because no t every integer is even; and in Exampl e 5, again,
the mappi ng is no t o nto , for the number 1, for instance, is no t the square of
any real number. Ho wever, the mappi ng in Exampl e 6 is o nto because every
real number has a uni que real cube root. The reader can deci de whether or
no t the given mappi ngs are o nto in the o ther exampl es.
If we define f(S) = {f(s) G T \ s G S), ano ther way of saying that the
mappi ng / : 5 >T is o nto is by saying that f(S) = T.
Ano ther specific type of mappi ng pl ays an i mpo rtant and parti cul ar
rol e in what fol l ows.
Definition. A mappi ng/: .V 7' is said to be one-to-one (wri tten 1-1)
or injective if for s, + s
2
in S, /(i 'j ) + f(s
2
) in T. Equi val entl y, / is 1-1 if
/ (
5
i )
=
f (
s
2 ) impl ies that s
t
= s
2
.
I n other words, a mappi ng is 1-1 if it takes distinct objects into distinct
images. I n the exampl es of mappings we gave earl ier, the mappi ng of Exampl e
1 is no t 1-1, since two bro thers woul d have the same mo ther. Ho wever in Ex-
ampl e 2 the mappi ng is 1-1 because distinct U.S. citizens have distinct Social
Security numbers (provided that there is no goof-up in Washi ngton, which is
unl ikel y). The reader shoul d check if the various other exampl es of mappi ngs
are 1-1.
Given a mappi ng / : S >T and a subset A C T, we may want to l o o k at
B = {s G 5 \f(s) G A}; we use the no tati o n f~\A) for this set B, and cal l
F~
X
(A) the inverse image of A under f. Of parti cul ar i nterest is f~
l
{t), the in-
verse image of the subset {f} of T consisting of the el ement t G T al o ne. If
the inverse i mage of \t\ consists of onl y o ne el ement, say s G 5, we coul d try
to define f~
l
(t) by defining /
_ 1
( 0 = s. As we no te bel ow, this need not be a
mappi ng from Tto S, but is so if / i s 1-1 and o nto . We shal l use the same no -
tation f~
l
in cases of bo th subsets and el ements. This f~
l
does not in general
define a mappi ng from T to S for several reaso ns. First, if / i s no t o nto , then
there is so me t in T which is no t the i mage of any el ement s, so /
- 1
( f ) = 0 .
Second, if / is no t 1-1, then for so me t E T there are at l east two distinct
s
1
= s
2
in S such that f(s^) = t = f(s
2
). So f~
l
(t) is not a uni que el ement of
Ssomething we requi re in o ur definition of mappi ng. Ho wever, if / i s bo th
1-1 and o nto T, then/
- 1
i ndeed defines a mappi ng of To nto S. (Verify!) Thi s
brings us to a very i mpo rtant cl ass of mappi ngs.
Definition. The mappi ng / : S * 7' is said to be a 1-1 correspondence
or bijection if / i s bo th 1-1 and o nto .
No w that we have the no ti o n of a mappi ng and have singl ed out vari ous
types of mappi ngs, we might very wel l ask: "Go o d and wel l , but what can we
do with them?" As we shal l see in a mo ment, we can i ntro duce an o perati o n
of combi ni ng mappi ngs in certai n ci rcumstances.
Co nsi der the situation g : S T and / : T ^> U. Gi ven an el ement s E S,
then g sends it into the el ement g(s) in T; so g(s) is ripe for bei ng acted on
by / Thus we get an el ement f(g(s)) E U. We cl aim that this pro cedure pro -
vides us with a mappi ng from 5 to U. (Verify!) We define this mo re formal l y
in the
Definition, if g : S --> 7' and f:T^>U, then the composition (or prod-
uct), deno ted by fg, is the mappi ng fg:S^> U defined by (fg)(s) =
f(g(s)) for every sGS.
No te that to co mpo se the two mappi ngs / and gthat is, for fg to
have any sensethe terminal set, T, for the mappi ng g must be the initial set
for the mappi ng /. One special time when we can always compose any two
mappings is when S = T = U, that is, when we map S into itself. Al tho ugh
special , this case is of the utmo st i mpo rtance.
We verify a few pro perti es of this co mpo si ti o n of mappi ngs.
L emma 1.3.1. If /;: .S' T, g : T U. and / : (/ ^ V, then =
(f
a
g) It-
Proof. Ho w shal l we go abo ut provi ng this l emma? To verify that two
mappi ngs are equal , we merel y must check that they do the same thing to
every el ement. No te first of al l that bo th f (g h) and ( / g) h define map-
pings from S to V, so it makes sense to speak abo ut thei r possibl e equal ity.
Our task, then, is to show that for every s E S, (f(gh))(s) =
((/ g)
n
)(
s
)- We appl y the definition of co mpo si ti o n to see that
( / o (go h))(
S
)= / ( ( g o / , ) ( , ) ) = f(g(h(s))).
Unravel i ng
{(f
0
g)h)(s) = (fg)(Ks))=f(g(h(s))),
we do i ndeed see that
( / ( r #) = ((fg)hm
for every s E S. Co nsequentl y, by definition, f (g h) = (f g) /?.
(The symbol wil l al ways indicate that the proof has been co mpl eted.)
This equal ity is descri bed by saying that mappi ngs, under co mpo si ti o n,
satisfy the associative law. Because of the equal ity invol ved there is real l y no
need for parentheses, so we write f (g
0
h) as f g h.
Lemma 1.3.2. Tandf : T-> U are bo th 1-1, then/ g : 5 - U
is al so 1-1.
Proof. Let us suppo se that (fg)(si) = (/g)(s
2
); thus, by definition,
f(g(*i)) = f(g(
s
2))- Si nce/i s 1-1, we get from this that g(
Sl
) = g(s
2
); ho w-
ever, g is al so 1-1, thus s
x
= s
2
fol l ows. Since (fg)(si) = (fg)(s
2
) forces
S\ = s
2
, the mappi ng/
0
g is 1-1.
We l eave the proof of the next Remark to the reader.
Remark. If g : S T and / : T-> U are bo th o nto , then fg : S -> U is
al so o nto .
An i mmedi ate co nsequence of combi ni ng the Remark and L emma
1.3.2 is to o btai n
L emma 1.3.3. If g : S T and f:T^ U are bo th bijections, then
Uis al so a bijection.
If / i s a 1-1 co rrespo ndence of S o nto T, then the "o bj ect" f~
x
:T-^S
defined earl ier can easil y be shown to be a 1-1 mappi ng of T o nto S. I n this
case it is cal l ed the inverse of /. I n this situation we have
Lemma 1.3.4. I f/: S > T is a bijection, then/ / ~' = /
r
and / ^ / =
i
s
, where i
s
and i
r
are the identity mappi ngs of S and T, respectivel y.
Proof. We verify o ne of these. If / G T, then {fr
l
){t) = f{f~\t)).
But what is /
_ 1
( 0? By definition, /
_ 1
( f ) is that el ement s
Q
G S such that
t =/ (s
0
). So / ( / - ' ( f ) ) =/Oo ) = f. I n o ther wo rds, (/ /
- 1
)(f ) = r for every
t 7; hence/ /
_ 1
=/
r
, the i denti ty mappi ng on T.
We l eave the l ast resul t of this section for the reader to pro ve.
L emma 1.3.5. I f / : S > T and i
T
is the identity mappi ng of T o nto it-
sel f and i
s
is that of S o nto itsel f, then i
T
f =/ and/
0
i
s
= f.
P R O B L E M S
Easier Problems
1. Fo r the given sets S, T determi ne if a mappi ng / : S T is cl earl y and un-
ambi guousl y defined; if not, say why not.
(a) S = set of al l wo men, T =set of al l men, f(s) =husband of s.
(b) S =set of positive i ntegers, T = S, f(s) = s - 1.
(c) S = set of positive integers, T = set of nonnegative integers, f(s) =
s - 1.
(d) S = set of no nnegati ve i ntegers, T = S, f(s) = s - 1.
(e) 5 =set of al l integers, T = 5, /(s) =5 1.
( f ) 5 =set of al l real numbers, T = S, f(s) = Vs.
(g) 5 =set of al l positive real numbers, T = S, f(s) = 'Vs.
2. I n tho se parts of Pro bl em 1 where / does define a function, determi ne if
it is 1-1, o nto , or bo th.
*3. If / is a 1-1 mappi ng of S o nto T, pro ve that /
_ 1
is a 1-1 mappi ng of T
o nto S.
*4. I f /i s a 1-1 mappi ng of S o nto T, pro ve that/
- 1
/ =/y.
5. Gi ve a pro o f of the Remark after L emma 1.3.2.
*6. If / : S -> T is o nto and g:TÛ and h:T>U are such that gf=hf,
pro ve that g =h.
*7. If g: S ->r, :>S T, and i f/: T U is 1-1, show that if / g = fh,
then g =
8. Let 5 be the set of al l i ntegers and T = {1, 1); / : S ->T is defined by
/(s) =1 if 5 is even, /(j ) = - 1 if 5 is o dd.
(a) Do es this define a function from S to 77
(b) Show that /($! +j
2
) =f(si)f(s
2
). What does this say abo ut the i nte-
gers?
(c) I s f(s
1
s
2
) = f{si)f(s
2
) al so true?
9. Let S be the set of al l real numbers. Define f:S-^> S by f(s) = s
2
, and
g:S->Sbyg(s)=s + l.
(a) Fi nd/ g.
(b) Fi nd
(c) I s/ g = g o / ?
10. Let S be the set of al l real numbers and for a,b G S, where a = 0; define
(a) Show that f
b
f
c
_
d
= / for so me real u, v. Gi ve expl icit val ues for
u, v in terms of a, b, c, and d.
(b) I s/
f l ; 6
/,,
r f
=f
c
,
d
/,
6
al ways?
(c) Fi nd al l f
a>b
such that /
a >i
/
u
=/
u
f
c
,,
b

(d) Show that /7J, exists and find its form.
11. Let S be the set of al l positive integers. Define / : S -> 5 by / ( l ) = 2,
/(2) = 3, /(3) = 1, and f(s) = s for any o ther s G 5. Sho w that fff =
i
s
. What i s/ "
1
in this case?
12. Let S be the set of no nnegati ve rati o nal numbers, that is, S = {m/n | m, n
no nnegati ve integers, n + 0}, and l et Tbe the set of al l integers.
(a) Do es f:S> T defined by f(m/n) = 2"'3" define a l egi ti mate function
from S to r?
(b) If not, ho w coul d you modify the definition of / so as to get a l egiti-
mate function?
13. Let S be the set of al l positive i ntegers of the form 2"'3", where m > 0,
n > 0, and l et T be the set of al l rati o nal numbers. Define / : 5 T by
/(2'"3") = m/n. Pro ve that / defines a function from S to T. (On what
pro perti es of the i ntegers does this depend?)
14. Let f:S-> S, where S is the set of al l integers, be defined by f(s) =
as + b, where a, b are i ntegers. Fi nd the necessary and sufficient condi-
tions on a, b in o rder that/
0
/ = i
s
.
15. Fi nd al l /o f the form given in Pro bl em 14 such t hat / / / = i
s
.
16. I f /i s a 1-1 mappi ng of S o nto itsel f, show that ( /
_ 1
)
_ 1
= /.
17. If S is a finite set having m > 0 el ements, how many mappi ngs are there
of S i nto itsel f?
18. I n Pro bl em 17, ho w many 1-1 mappi ngs are there of S i nto itsel f?
19. Let S be the set of al l real numbers, and define / : S S by f(s) =
s
2
+ as + b, where a, b are fixed real numbers. Pro ve that for no val ues
at a, b can / be o nto or 1-1.
20. Let S be the set of al l positive real numbers. Fo r positive real s a, c and
no nnegati ve real s b, d, is it ever possibl e that the mappi ng / : S S de-
fined by f(s) = (as + b)/(cs + d) satisfies ff= z
s
? Find al l such a, b, c, d
that do the trick.
21. Let S be the set of al l rati onal numbers and l et f
a b
: S > S be defined by
f
ab
(s) = as + b, where a #0, b are rati onal numbers. Find al l f
C(i
of this
form satisfying f
c> d
a
f
ib
= /,
b
/
c
,
d
for every /
fl b
.
22. Let S be the set of al l integers and a, b, c rati o nal numbers. Define
f:S>S by f(s) = as
2
+ bs + c. Fi nd necessary and sufficient conditions
o n a, b, c, so that/def i nes a mappi ng o n S [Note: a, b, c need not be i nte-
gers; for exampl e, f(s) = %s(s + 1) = ^s
2
+ s does al ways give us an in-
teger for i ntegral s.}
Harder Pro bl ems
23. Let S be the set of al l integers of the form 2"'3", m > 0, n > 0, and l et T
be the set of al l positive integers. Show that there is a 1-1 co rrespo ndence
of S o nto T.
24. Pro ve that there is a 1-1 co rrespo ndence of the set of al l positive i ntegers
o nto the set of al l positive rati o nal numbers.
25. Let S be the set of al l real numbers and T the set of al l positive real s.
Fi nd a 1-1 mappi ng / of S o nto T such that/I s, + j '
2
) =/ (s
]
)/ (s
2
) for al l
s
u
s
2
G S.
26. Fo r the / i n Pro bl em 25, find/
-1
expl icitl y.
27. If / g are mappi ngs of S into S and f g is a constant function, then
(a) What can you say abo ut / if g is o nto ?
(b) What can you say abo ut g if / i s 1-1?
28. If S is a finite set and / is a mappi ng of S onto itsel f, show that / must be
1-1.
29. If S is a finite set and / is a 1-1 mappi ng of S into itsel f, sho w that / must
be surjective.
30. If S is a finite set and/ i s a 1-1 mappi ng of 5, show that for some i nteger
n > 0,
. / / / / = ^.
n times
31. If 5 has 7 el ements in Pro bl em 30, find an n > 0 (in terms of m) that
wo rks si mul taneousl y for al l 1-1 mappi ngs of S into itsel f.
4. A(S) (THE SET OF 1-1 MA PPI NGS OF S ONTO ITSELF)
We focus our attenti on in this section on particul arl y nice mappings of a no n-
empty set, S, into itsel f. Namel y, we shal l consider the set, A(S), of al l 1-1 map-
pings of S onto itsel f. Al tho ugh most of the concern in the bo o k wil l be in the
case in which S is a finite set, we do not restrict oursel ves to that situation here.
When S has a finite number of el ements, say n, then A (S) has a special
name. I t is cal l ed the symmetric group of degree n and is often deno ted by S.
I ts el ements are cal l ed permutations of S. If we are i nterested in the structure
of S, it real l y does no t matter much what o ur underl yi ng set S is. So, you
can think of S as bei ng the set {1, . . . , n). Chapter 3 wil l be devo ted to a
study, in so me depth, of S. I n the investigation of finite groups, S
n
pl ays a
central rol e.
There are many pro perti es of the set A (S) on which we coul d concen-
trate. We have chosen to devel o p tho se aspects here which wil l mo ti vate the
notion of a gro up and which wil l give the reader so me experi ence, and feel -
ing for, working in a gro up-theo reti c framework. Gro ups wil l be discussed in
Chapter 2.
We begin with a resul t that is real l y a co mpendi um of so me of the re-
sul ts o btai ned in Section 3.
L emma 1.4.1. A (5) satisfies the fol l owing:
(a) fgGA (S) impl ies that fgEA (S).
(b) /, g, h G A(S) impl ies that (fg) h = f (g h).
(c) There exists an el ementthe identity mappi ng isuch that f i =
z o / =/ for every / G A(S).
(d) Given / G A(S), there exists a g G A(S) (g = such that fg =
8f= i-
Proof Al l these things were do ne in Section 3, ei ther in the text mate-
rial or in the pro bl ems. We l eave it to the reader to find the rel evant part of
Section 3 that wil l verify each of the statements (a) thro ugh (d).
We shoul d no w l ike to kno w ho w many el ements there are in A(S)
when S is a finite set having n el ements. To do so, we first make a sl ight di-
gression.
Suppo se that you can do a certai n thing in r different ways and a sec-
o nd i ndependent thi ng in s different ways. I n ho w many distinct ways can
you do bo th things to gether? The best way of finding o ut is to pi cture this in
Sec . 4 A{S) ( Th e Set of 1-1 Ma p p i n g s of S On t o It sel f ) 17
a co ncrete context. Suppose that there are ;highways running from Chicago
to Detro i t and s highways runni ng from Detro i t to Ann Arbo r. I n how many
ways can yo u go first to Detro i t, then to Ann Arbo r? Cl earl y, for every ro ad
you take from Chicago to Detro i t yo u have s ways of conti nui ng on to Ann
Arbo r. Y o u can start yo ur trip from Chicago in r distinct ways, hence you
can co mpl ete it in
s + s + s + + s = rs
r ti mes
different ways.
I t is fairl y cl ear that we can extend this from doi ng two i ndependent
things to doi ng in i ndependent o nes, for an i nteger m > 2. If we can do the
first things in \\ distinct ways, the second in r
2
ways, . . . , the mth in r, dis-
tinct ways, then we can do al l these to gether in i\r
2
.. . r, different ways.
Let's recal l so methi ng many of us have al ready seen:
Definition. If n is a positive i nteger, then nl (read "n factorial") is de-
fined by nl = 1 2 3 n.
L emma 1.4.2. If S has n el ements, then A (S) has nl el ements.
Proof. Let / G A(S), where S = \x
1
, x
2
, . . . , x}. Ho w many choices
do es/ have as a pl ace to send x
x
l Cl earl y n, for we can send x
l
under/ to any
el ement of S. But no w / is not free to send x
2
anywhere, for since / is 1-1, we
must have f(xA = f(x
2
). So we can send x
2
anywhere except onto f(xi).
Hence / can send x
2
into n 1 different images. Co nti nui ng this way, we see
that / can send x, into n (i 1) different images. Hence the number of
such/'s is n(n ~ l)(n - 2) 1 =nl
Exampl e
The number n\ gets very l arge quickl y. To be abl e to see the pi cture in its en-
tirety, we l ook at the special case n = 3, where nl is stil l quite smal l .
Consider A(S) = S
3
, where S consists of the three el ements x
h
x
2
,x
3
. We
l ist al l the el ements of 5
3
, writing out each mappi ng expl icitl y by what it does
to each of x
u
x
2
, x
3
.
1. i: Xi X\, x
2
> x
2
, x
3
> x
3
.
2. f:x\>x
2
,x
2
-^>x-
i
,x
2l
^X\.
3. g:x^ x
2
,x
2
~+ x
u
x
3
^> x
3
.
4. g / : * i ->x
u
x
2
-*x
3
,x
3
->x
2
. (Verify!)
5. f g : A " ! - > a - 3 , a :
2
x
2
, A ' 3 - > a c j . (Verify!)
6. / " / : a c 1 ^ a c 3 , a c 2 ^ a c 1 , a - 3 - > a :
2
. (Verify!)
Since we have l isted here six different el ements of S
3
, and S
3
has onl y
six el ements, we have a co mpl ete l ist of al l the el ements of S
3
. What do es
this l ist tel l us? To begi n with, we no te that fg^gfso o ne famil iar rul e
of the ki nd of ari thmeti c we have been used to is viol ated. Since g G S
3
and
g G S
3
, we must have gg al so in S
3
. What is it? If we cal cul ate gg, we
easil y get g g = i. Simil arl y, we get
(fg)(fg) = ' = (gf)(g.f)-
No te al so that f(ff) = i, hence f
1
= ff. Final l y, we l eave it to the
reader to show that gf= f^
1
g.
I t is a l ittl e cumberso me to write this pro duct in A (S) using the . From
now on we shall drop it and write f g merely as fg. Al so , we shal l start using
the sho rthand of expo nents, to avoid expressions l ike fff---f. We de-
fine, for / G A(S), f = i, f
2
= ff = ff, and so on. Fo r negati ve expo nents
n we define by f~" = ( /
_ 1
) ", where n is a positive integer. The usual
rul es of expo nents prevail , namel y /'/' = f
r+s
and (f')
s
= We l eave
these as exerci sesso mewhat tedi o us ones at thatfo r the reader.
Example
Do no t j ump to concl usions that al l famil iar pro perti es of expo nents go over.
Fo r instance, in the exampl e of the /, g G S
3
defined abo ve, we cl aim that
(fg)
2
^ f
2
g
2
. To see this, we no te that
fg '
X
l ~^
X
3'
X
2
X
2>
X
3 ~*
X
L)
so that (fg)
2
: x
t
> x
u
x
2
~> x
2
, x
3
> x
3
, that is, (fg)
2
= i. On the o ther
hand, f
2
/ and g
2
= i, hence f
2
g
2
= f
2
/, whence (fg)
2
^ f
2
g
2
in this case.
Ho wever, so me o ther famil iar pro perti es do go over. Fo r i nstance, if
/, g, h are in A (S) and fg = fh, then g = h. Why? Because, from fg = fh we
have f~\fg) = f-\fh); therefo re, g = ig = (f^ftg = / ^( / g) = r\fh) =
(f
1
f)h
=
ih = h. Simil arl y, gf = hf impl ies that g = h. So we can cancel an
el ement in such an equati o n pro vi ded that we do not change sides. I n S3 o ur
/, g satisfy gf = f~
1
g, but since / f~
l
we cannot cancel the g here.
P R O B L E M S
Recal l that/g stands for fg and, al so, what/'" means. S, wi tho ut subscripts,
wil l be a no nempty set.
Sec . 4 A(S) ( Th e Set o f 1-1 Ma p p i n g s of S On t o It sel f ) 19
Easi er Pro bl ems
1. If s
{
= s
2
are in S, show that there is an/ G A(S) such that f(s
{
) = s
2
.
2. Ifsi G S, l et / / = {/ G A (5) = s,}. Show that:
(a) / G //.
(b) I f / . g G W.then/i ,'G //.
(c) I f / G #, t hen/
_ 1
G # .
3. Suppo se that s\ + s
2
are in S and f(s{) = s
2
, where/ G A (5). Then if H is
as in Pro bl em 2 and K = {g G A( 5) | g( s
2
) = s
2
}, show that:
(a) If g&K, then f-tgfGH.
(b) If h G //, then there is so me g G K such that /z = f~
1
gf.
4. I f/, g, h G. A(S), show that ( /
_ 1
g/ ) ( / ~ V ) = f~
x
(gh)f. What can yo u say
abo ut ( r ' g / y ?
5. If /, g G A (5) and/g = g/ show that:
(a) {fgf-fg
2
.
(b> ( / g r w - v
1
-
6. Push the resul t of Pro bl em 5, for the same / and g, to show that (fg)'" =
f"'g"' for al l integers m.
*7. Verify the rul es of expo nents, namel y ff
s
= / '
+
' ' and (f'Y = / " for
/ G A (S) and positive integers r, s.
8. If / g G A ( 5) and (fg)
2
=/
2
g
2
, pro ve that/g = gf.
9. If 5 = {x^x
2
, x
3
, x
4
}, l et/, g G S
4
be defined by
and
g : X j - x
2
, x
2
->x
1 ;
x
3
x
3
, x
4
-> x
4
.
Cal cul ate:
( a) /
2
, /
3
, /
4
.
(b) g
2
, g
3
.
( c ) / g .
(d) g/
(e) (fg)\ (gf)
3
.
( f ) r
1
. *
- 1
10. I f / G S
3
, show that/
6
= z.
11. Can yo u find a positive i nteger m such that /"' = i for all f G S
4
?
* 12. If / G S, show that there is so me positive integer k, dependi ng o n/ , such
that/''" = i. (Hi nt: Co nsi der the positive po wers o f/.)
*13. Show that there is a positive integer / such that / ' = i for all f G S.
14. If m < n, show that there is a 1-1 mappi ng F: S, >S such that F(fg) =
F(f)F(g) for al l / gG 5
m
.
15. If 5 has three or mo re el ements, show that we can find / g G /I (5) such
that/g #g/ .
16. Let S be an infinite set and l et M C A(S) be the set of al l el ements
/ G A (S) such that f(s) # s for at most a finite number of s G S. Pro ve
that:
( a) / g G M impl ies that/g G M.
(b) / G Mimpl ies that/
-
' G M .
17. Fo r the situation in Pro bl em 16, show, if / G A (5), that /
_ 1
M / =
G
^}
m u s t
equal Af.
18. Let S T and consider the subset [7(7) = { / G A (5) | /(f) G Tf o r every
t G T}. Show that:
( a) / G U(T).
(b) / g G U(T) impl ies that/g G U(T).
19. If the 5 in Pro bl em 18 has n el ements and T has m el ements, ho w
many el ements are there in U(T)7 Show that there is a mappi ng
F: U(T) S, such that F(fg) = F(f)F(g) f o r/ g G U(T) and Fi s o nto
c
20. If / 7 ? < , can i
7
in Pro bl em 19 ever be 1-1? If so, when?
21. I n S show that the mappi ng / defined by
[i.e., /(*;) = A.-
/ +]
if i < n, f(x) = A^] can be written as / = gj g
2
g_!
where each g
f
G 5 interchanges exactly two elements of S = {x
x
, . . . , A'),
l eaving the o ther el ements fixed in S.
Harder Problems
22. I f / G S, show that/ = h\h
2
'' ' for so me E S such that /r
2
= /.
*23. Cal l an el ement in S
n
a transposition if it i nterchanges two el ements, l eav-
ing the o thers fixed. Show that any el ement in S
n
is a pro duct of transpo -
sitions. (This sharpens the resul t of Pro bl em 22.)
24. If n is at l east 3, show that for so me / in S, / canno t be expressed in the
f o rm/ = g
3
for any g in S.
Sec . 5 Th e I n t eg er s 21
25. If / G S is such that / it i but /
3
= j , show that we can number the
el ements of S in such a way that / ( . t j ) = .v
2
, f(x
2
) = -v
3
, f(x
3
) = X ) ,
/(x
4
) =x
5
,f(x
5
) = x
6
, f(x
6
) = x
4
, ... ,f(x
3k+1
) = x
3k+2
,f(x
3k+2
) = x
3 k + 3
,
f(x
3k+3
) = x
3 k + 1
for some k, and, for al l the o ther x, G 5, f(x) = x
t
.
26. Vi ew a fixed shuffl e of a deck of 52 cards as a 1-1 mappi ng of the deck
o nto itsel f. Show that repeati ng this fixed shuffl e a finite (positive) num-
ber of ti mes wil l bring the deck back to its original order.
*27. If / G A(S), cal l , for s G S, the orbit of s (rel ative to / ) the set 0(s) =
{ f
j
(s) | al l integers / }. Show that if s, t G S, then either 0(s) n 0(t) = 0 or
0(s) = 0(t).
28. If 5 = [xi, x
2
,..., x
12
] and/ G 5
1 2
is defined by/(x,) =x,
+1
if i = 1, 2, . . . ,
11 and/ ( x
n
) =x
1
, find the orbits of al l the el ements of 5 (rel ative to / ) .
29. If / G A (S) satisfies /
3
= ;', show that the orbit of any el ement of S has
o ne or three el ements.
*30. Recal l that aprime number is an integer p>l such that p canno t be fac-
to red as a product of smal l er positive integers. If / G A(S) satisfies f
p
~ i,
what can you say about the size of the orbits of the el ements of S rel ative to
/ ? What property of the pri me numbers are you using to get your answer?
31. Pro ve that if S has mo re than two el ements, then the onl y el ements / in
A(S) such that/
0
/ = ff
0
for al l / G A(S) must satisfy f
0
= i.
*32. We say that g G A(S) commutes with / G A(S) if fg = gf. Fi nd al l
the el ements in A(S) that co mmute with f:S -> S defined by
/(*,) = x
2
,f(x
2
) = x
x
, and/ (s) =s if s # x
u
x
2
.
33. I n 5 show that the onl y el ements co mmuti ng with / defined by f(x) =
x
j+
i if / < n, f{x,) =Xi , are the po wers off, namel y i = f, f f
2
,.. . ,/""'.
34. F o r / G A(S), l et C( / ) = [g G A(S) \fg =gf). Pro ve that:
(a) g, h G C( / ) impl ies that gh G C(.f).
(b) g G C(f) impl ies that g
- 1
G C( / j .
(c) C( / ) is no t empty.
5. THE I NTEGERS
The mathemati cal set mo st famil iar to everybo dy is that of the positive inte-
gers 1, 2, . . . , which we shal l often cal l Equal l y famil iar is the set, Z, of al l
i ntegersposi ti ve, negati ve, and zero . Because of this acquai ntance with Z,
we shal l give here a rather sketchy survey of the pro perti es of Z that we shal l
use often in the ensuing materi al . Mo st of these pro perti es are wel l kno wn to
al l of us; a few are l ess wel l kno wn.
The basic assumpti on we make abo ut the set of integers is the
Well-Ordering Principle. Any no nempty set of no nnegati ve i ntegers
has a smal l est member.
Mo re formal l y, what this principl e states is that given a no nempty set V
of no nnegati ve integers, there is an el ement v
Q
G V such that v
0
< v for every
v G V. This principl e wil l serve as the fo undati o n for o ur ensui ng discussion
of the integers.
The first appl ication we make of it is to show so methi ng we al l kno w
and have taken for granted, namel y that we can divide o ne i nteger by an-
o ther to get a remai nder that is smal l er. This is kno wn as Euclid's Algorithm.
We give it a mo re formal statement and a proof based o n wel l -orderi ng.
T heorem 1.5.1 (Euclid's Al go ri thm). If m and n are i ntegers with
n > 0, then there exist i ntegers q and r, with 0 < r < such that m = qn + r.
Proof. Let W be the set of m - tn, where t runs thro ugh al l the
integers, i.e., W = [m tn 11 G Z}. No te that W contains so me no nnegati ve
integers, for if t is l arge eno ugh and negati ve, then m tn > 0. L et
V = {v G W | v ^ 0}; by the wel l -orderi ng principl e V has a smal l est el ement,
r. Since r G V, r 0, and r = m - qn for so me q (for that is the form of al l
el ements in W D V). We cl aim that r < n. If no t, r = m - qn > n, hence
m - (q + > 0. But this puts m - (q + l)n in V, yet m - (q + l )n < r,
contradi cti ng the mi ni mal nature of r in V. Wi th this, Eucl i d's Al go ri thm is
pro ved.
Eucl i d's Al go ri thm wil l have a host of co nsequences for us, especial l y
abo ut the no ti o n of divisibil ity. Since we are speaki ng abo ut the integers,
be it understood that all letters used in this section will be integers. This wil l
save a l ot of repeti ti on of certai n phrases.
Definition. Gi ven i ntegers m 0 and n we say that m divides n, writ-
ten as m | n, if n = cm for so me i nteger c.
Thus, for i nstance, 2 [ 14, (- 7) | 14, 4 | (- 16). If m | n, we cal l m a divi-
sor o r factor of n, and n amultiple of m. To indicate that m is no t a divisor of
n, we write 7 7 7 1n; so, for i nstance, 3 f 5.
The basic el ementary pro perti es of divisibil ity are l aid o ut in
Lemma 1.5.2. The fol l owing are true:
(a) 1 I n for al l n.
(b) If 7 7 7 * 0, then m \ 0.
Sec. 5 Th e I n t eg er s 23
(c) If m | n and n | q, then m \ q.
(d) If m | and m | q, then m | (un + vq) for al l it, v.
(e) If m 11, then m = 1 or m =1.
(f) If m | n and | m, then 7 7 7 = n.
Proof. The proofs of al l these parts are easy, fol l owing i mmedi atel y
from the definition of m \ n. We l eave al l but Part (d) as exercises but pro ve
Part (d) here to give the fl avor of ho w such proofs go.
So suppo se that m | n and m \ q. Then n = cm and q = dm for so me c
and d. Therefo re, un + vq u(cm) + v(dm) = (uc + vd)m. Thus, from the
definition, m \ (un + vq).
Havi ng the concept of a divisor of an i nteger, we no w want to i ntroduce
that of the greatest common divisor of two (or mo re) integers. Simpl y
eno ugh, this shoul d be the l argest possibl e i nteger that is a divisor of bo th in-
tegers in questi o n. Ho wever, we want to avoid using the size of an i nteger
for reaso ns that may beco me cl ear much l ater when we tal k abo ut rings. So
we make the definition in what may seem as a strange way.
Definition. Gi ven a, b (not bo th 0), then thei r greatest common divi-
sor c is defined by:
(a) c >0.
(b) c I a and c \ b.
(c) lid\a and d \ b, then d \ c.
We wri te this c as c (a, b).
I n other words, the greatest co mmo n divisor of a and b is the positive
number c which divides a and b and is divisibl e by every d which divides a
and b.
Defining so methi ng does no t guarantee its existence. So it is i ncumbent
o n us to pro ve that (a, b) exists, and is, in fact, uni que. The proof actual l y
shows mo re, namel y that (a, b) is a nice co mbi nati o n of a and b. This co mbi -
nati o n is no t uni que; for i nstance,
(24,9) = 3 = 3- 9 + ( - 1) 24 = ( - 5) 9 + 2- 24.
Theo rem 1.5.3. If a, b are no t bo th 0, then thei r greatest common divi-
sor c = (a, b) exists, is uni que, and, mo reo ver, c = rnâ + n
0
b for so me suit-
abl e mo and n
n
.
Proof. Since no t bo th a and fa are 0, the set A = {ma + nb \ m, n G Z)
has no nzero el ements. If .v G A and x < 0, then x is al so in A and x > 0,
for if x = m
x
a + n
{
b, then x = (m
{
)a + (n^fa, so is in A Thus has
positive el ements; hence, by the wel l -orderi ng principl e there is a smal l est
positive el ement, c, in A. Since c G / i , by the form of the el ements of A we
kno w that c = m
0
a + n
Q
b for so me / 7 J f ) ,
We cl aim that c is o ur requi red greatest co mmo n divisor. First no te that
if rf | and d \ fa, then d \ (m
Q
a + n
0
b) by Part (d) of L emma 1.5.2, that is,
d | c. So, to verify that c is o ur desi red el ement, we need onl y show that c \ a
and c | b.
By Eucl i d's Al go ri thm, a = qc + ;, where 0 < / < c, that is, =
q(m
Q
a + /?
0
/3) + r. Therefo re, ;= gn
a
b + (1 qm
0
)a. So r is in A. But
r < c and is in A, so by the choice of c, r cannot be positive. Hence r = 0; in
o ther words, a = qc and so c | a. Simil arl y, c | b.
Fo r the uni queness of c, if t > 0 al so satisfied r ] a, 11 /j>and d | f for al l d
such that d | a and d | fa, we wo ul d have t \ c and c 11. By Part (f) of L emma
1.5.2 we get that t = c (since bo th are positive).
Let's l ook at an expl icit exampl e, namel y a = 24, fa = 9. By direct exami-
nati on we know that (24, 9) = 3; no te that 3 = 3- 9 + (-1)24. What is
(- 24, 9)?
Ho w is this do ne for positive numbers a and fa which may be quite l arge?
If fa > A, interchange a and fa so that a > fa > 0. Then we can find (a, fa) by
1. observing that (a, fa) = (fa, r) where a ~ qb + r with 0 < r < fa (Why?);
2. finding (fa, r), which no w is easier since o ne of the numbers is smal l er
than before.
So, for exampl e, we have
(100, 28)
( 28, 16)
( 16,12)
(28,16)
(16, 12)
(12, 4) since 16
since 100
since 28
3 (28) + 16
1 (16) + 12
1 (12) + 4
This gives us
(100, 28) (12, 4) = 4.
I t is possibl e to find the actual val ues of m
a
and n
0
such that
4 = m
0
100 + n
0
28
Sec . 5 Th e I n t eg er s 25
by going backwards thro ugh the cal cul ations made to find 4:
Since 16 = 1 (12) + 4,
Since 28 = 1 (16) + 12,
Since 100 = 3 (28) + 16, 16 = 100 +
12= 28 +
4 = 16 +
(
(
(
3) 28
1) 12
1) 16
But then
4 = 16 + (- 1) 12 = 16 + (- 1)(28 + (- 1) 16)
= (- 1) 28 + (2) 16 = (1) 28 + (2)(100 + (- 3) 28)
= (2) 100 + (- 7) 28
so that m
0
= 2 and
n
= 7.
This shows ho w Eucl i d's Al go ri thm can be used to co mpute (a, b) for
any positive integers a and b.
We shal l incl ude so me exercises at the end of this section on o ther
pro perti es of (a, b).
We co me to the very i mpo rtant
Definition. We say that a and b are relatively prime if (a, b) = 1.
So the integers a and b are rel ativel y pri me if they have no nontrivial
co mmo n factor. An i mmedi ate corol l ary to Theo rem 1.5.3 is
Theo rem 1.5.4. The i ntegers a and b are rel ativel y pri me if and onl y if
1 = ma + nb for suitabl e integers in and n.
Theo rem 1.5.4 has an i mmedi ate co nsequence
Theo rem 1.5.5. If a and b are rel ativel y pri me and a | be, then a \ c.
Proof. By Theo rem 1.5.4, ma + nb = 1 for so me in and n, hence
(ma + nb)c = c, that is, mac + nbc = c. By assumpti o n, a \ be and by obser-
vati o n a | mac, hence a | (mac + nbc) and so a \ c.
Corol l ary. If b and c are bo th rel ativel y pri me to a, then be is al so rel -
ativel y pri me to a.
Proof. We pick up the proof of Theo rem 1.5.5 at mac + nbc = c. If d =
(a, be), then d | a and d\ be, hence d | (mac + nbc) = c. Since d \ a and d \ c
and (a, c) = 1, we get that d = 1. Since 1 = d = (a, be), we have that be is
rel ativel y pri me to a.
We now singl e o ut an ul tra-i mpo rtant cl ass of positive integers, which
we met before in Pro bl em 30, Section 4.
Definition. A prime number, o r a prime, is an integer p > 1, such that
for any integer a ei ther p \ a o r p is rel ativel y pri me to a.
This definition coincides with the usual one, namely that we cannot fac-
tor p nontrivially. Fo r if p is a pri me as defined abo ve and p = ab where
1 < a < p, then (a,p) = a (Why?) and p does not divide a since p > a. I t fol -
l ows that a = 1, so p = b. On the o ther hand, if p is a pri me in the sense that
it canno t be factored nontrivial l y, and if a is an integer no t rel ativel y pri me to
p, then (a, b) is not 1 and it divides a and p. But then (a, b) equal s p, by o ur
hypothesi s, so p divides a.
Ano ther resul t co mi ng o ut of Theo rem 1.5.5 is
Theo rem 1.5.6. If p is a pri me and p\(a
l
a
2
a,), then p \ a
f
for
so me i with 1 <; <; ? .
Proof. If p | a
u
there is no thi ng to pro ve. Suppo se that p\a
{
; then p
and ffj are rel ativel y pri me. But p \ a
l
(a
2
a), hence by Theo rem 1.5.5,
p | a
2
a
n
. Repeat the argument just given on a
2
, and co nti nue.
The pri mes pl ay a very special ro l e in the set of i ntegers l arger than 1 in
that every integer n > 1 is ei ther a pri me o r is the pro duct of pri mes. We
shal l show this in the next theo rem. I n the theo rem after the next we shal l
show that there is a uni queness abo ut the way n > 1 factors into pri me fac-
tors. The proofs of bo th these resul ts l ean heavil y o n the wel l -ordering prin-
cipl e.
Theo rem 1.5.7. If n > 1, then ei ther n is a pri me or n is the pro duct of
pri mes.
Proof. Suppo se that the theo rem is fal se. Then there must be an
intger m > 1 for which the theo rem fail s. Therefo re, the set M for which the
theo rem fail s is no nempty, so, by the wel l -ordering principl e, M has a l east
el ement m. Cl earl y, since m G M, m canno t be a pri me, thus m = ab, where
1 < a < m and 1 < b < m. Because a < m and b < m and m is the l east
el ement in M, we canno t have a G M or b G M. Since a ^ M, b M, by the
definition of M the theo rem must be true for bo th a and b. Thus a and b are
Sec. 5 Th e I n t eg er s 27
pri mes or the pro duct of pri mes; from m = ab we get that m is a pro duct
of pri mes. Thi s puts m outside of M, contradi cti ng that m G M. This pro ves
the theo rem.
We asserted abo ve that there is a certai n uni queness abo ut the deco m-
position of an integer into pri mes. We make this precise now. To avoid trivi-
al ities of the ki nd 6 = 2 3 = 3 2 (so, in a sense, 6 has two factorizations into
the pri mes 2 and 3), we shal l state the theo rem in a particul ar way.
Theo rem 1.5.8. Gi ven n > 1, then there is o ne and onl y one way to
write n in the form n = p
a
^p
c
2
p
a
k
k
, where p
x

, a
k
are al l positive.
Proof. We start as we did abo ve by assuming that the theo rem is
fal se, so there is a l east integer m > 1 for which it is fal se. This m must have
two distinct factorizations as m = pVpT ' ' ' Pt
k =
<7i
1(
72
2
' ' ' where
P \
 q \
< a
2
<
' ' '
< a
e are pri mes and where the expo nents
a
x
, ... , a
k
and b
x
, .. . , b
(
are al l positive. Since p
x
\ pi
1
p
a
k
k
= q\
l
q
b
e
',
by Theo rem 1.5.6 p
x
\ qf> for so me /; hence, again by Theo rem 1.5.6, p
x
| q
h
hence p
x
= cp. By the same to ken q
x
= p- for so me /; thus p
x
^ /? =
q
x
< q, = p
x
. This gives us that p
x
= q
v
. No w since m/p
x
< m, ni/p
x
has the uni que factorization pro perty. But m/p
x
= p"
1
^
1
p
2
- p"
k
k
=
Pi
1 1
<?2
2
' ' ' q
h
e
c a n
d since m/p
1
can be facto red in o ne and onl y one way in
this form, we easil y get k = , p
2
= q
2
, .. . , p
k
= q
k
, a
x
1 = b
x
1,
a
2
= b
2
, ... , a
k
= b
k
. So we see that the pri mes and thei r expo nents arising
in the factorization of m are uni que. This contradicts the l ack of such uni que-
ness for m, and so pro ves the theo rem.
What these l ast two theo rems tel l us is that we can buil d up the i ntegers
from the pri mes in a very precise and wel l -defined manner. One woul d ex-
pect from this that there shoul d be manythat is, an infinityof pri mes.
This ol d resul t goes back to Eucl id; in fact, the argument we shal l give is due
to Eucl id.
Theo rem 1.5.9. There is an infinite number of primes.
Proof. If the resul t were fal se, we coul d enumerate all the pri mes in
p
x
, p
2
, ... ,p
k
. Consi der the i nteger q = 1 + p
x
p
2
p
k
. Since q > p, for
every i = 1, 2, . . . , k, q canno t be a pri me. Since p
s
\ q, for we get a remai n-
der of 1 o n dividing q by p
t
, q is no t divisibl e by any of p
x
,.. . , p
k
. So q is
no t a pri me no r is it divisibl e by any pri me. This viol ates Theo rem 1.5.7,
thereby provi ng the theo rem.
Resul ts much sharper than Theo rem 1.5.9 exist abo ut ho w many pri mes
there are up to a given poi nt. The famous pri me number theo rem states that
for l arge n the number of pri mes l ess than or equal to n is "mo re o r l ess"
nl\og
e
n, where this "mo re or l ess" is precisel y described. There are many
o pen questi ons abo ut the pri me numbers.
P R O B L E M S
Easi er Problems
1. Fi nd (A, b) and express (a, b) as ma + /ib for:
(a) (116, - 84).
(b) (85,65).
(c) (72, 26).
(d) (72, 25).
2. Pro ve al l the parts of L emma 1.5.2, except part (d).
3. Show that (ma, mb) = m(a, b) if m > 0.
4. Show that if a \ m and b | m and (a, b) = 1, then (ab) | m.
5. Facto r the fol l owing into pri mes.
(a) 36.
(b) 120.
(c) 720.
(d) 5040.
6. If 7 7 7 = p'l
1
p"
c
k
and n = p\
x
p
b
k
k
, where p
u
. , . , p
k
are distinct
pri mes and a, .. . , a
k
are no nnegati ve and b
1
, . . . , b
k
are no nnegati ve,
express (m, n) as p^
1
p'jj
1
by describing the c's in terms of the a's and
/j's.
* 7. Define the least common multiple of positive i ntegers m and n to be the
smal l est positive i nteger v such that bo th 7 7 7 1 v and 7 7 1 v.
(a) Show that v = mn/(m, 71).
(b) I n terms of the factorization of m and n given in Pro bl em 6, what is
u?
8. Fi nd the l east co mmo n mul ti pl e of the pairs given in Pro bl em 1.
9. If 7 7 7 , 7 7 > 0 are two integers, show that we can find i ntegers u, v with
7 7 / 2 < 7 ; < 7 7 / 2 such that 7 7 7 = 7 7 7 7 . + v.
10. To check that a given i nteger 7 7 > 1 is a pri me, pro ve that it is eno ugh to
show that 7 7 is no t divisibl e by any pri me p with p < V77.
Sec. 6 Mat h emat i c al I n d u c t i o n 29
11. Check if the fol l owing are pri me.
(a) 301.
(b) 1001.
(c) 473.
12. Starting with 2, 3, 5, 7, . . . , construct the positive integers 1 + 2- 3,
1 + 2- 3- 5, 1 + 2 3 5 7, . . . . Do you al ways get a pri me number this
way?
Mi ddl e-Level Pro bl ems
13. If p is an o dd pri me, show that p is of the form:
(a) An + 1 o r An + 3 for so me n.
(b) 6n + 1 or 6n + 5 for so me n.
14. Adapt the proof of Theo rem 1.5.9 to pro ve:
(a) There is an infinite number of pri mes of the fo rm An + 3.
(b) There is an infinite number of pri mes of the form 6n + 5.
15. Show that no i nteger u = An + 3 can be wri tten as u = a
2
+ b
2
, where
a, b are integers.
16. If T is an infinite subset of N, the set of al l positive integers, show that
there is a 1-1 mappi ng of T onto N.
17. If p is a pri me, pro ve that o ne canno t find no nzero integers a and b such
that a
2
= pb
2
. (This shows that V p is irrational.)
6. MA THEMA TI CA L I NDUCTI ON
If we l o o k back at Section 5, we see that at several pl acesfor instance, in
the proof of Theo rem 1.5.6we say "argue as abo ve and co nti nue." Thi s is
not very satisfactory as a means of nail ing do wn an argument. What is cl eat-
is that we need so me techni que of avoiding such phrases when we want to
pro ve a pro po si ti o n abo ut all the positive i ntegers. This is pro vi ded for us by
the Principle of Mathematical Induction; in fact, this wil l be the usual metho d
that we shal l use for provi ng theo rems abo ut al l the positive integers.
Theo rem 1.6.1. Let P(n) be a statement abo ut the positive integers
such that:
(a) P(l) is true.
(b) If P(k) happens to be true for so me i nteger k S: 1, then P(k + 1) is al so
true.
Then P(n) is true for al l n > 1.
Proof. Actual l y, the arguments given in provi ng Theo rems 1.5.7 and
1.5.8 are a pro to type of the argument we give here.
Suppo se that the theo rem is fal se; then, by wel l -ordering, there is a
l east i nteger m S: 1 for which P(m) is not true. Since P(l) is true, m # 1,
hence m > 1. No w 1 < m - 1 < m, so by the choice of m, P(m - 1) must be
val id. But then by the inductive hypothesis [Part (b)] we must have that P(m)
is true. This contradicts that P(m) is no t true. Thus there can be no i nteger
for which P is no t true, and so the theo rem is proved.
We il l ustrate ho w to use i nducti on with so me rather diverse exampl es.
Exampl es
1. Suppo se that n tenni s bal l s are put in a straight l ine, to uchi ng each o ther.
Then we cl aim that these bal l s make n. 1 contacts.
Proof. If n = 2, the matter is cl ear. If for k bal l s we have k 1 con-
tacts, then adding o ne bal l (o n a l ine) adds o ne contact. So k + 1 bal l s wo ul d
have k contacts. So if P(n) is what is stated abo ve abo ut the tenni s bal l s, we
see that if P(k) happens to be true, then so is P(k + 1). Thus, by the theo -
rem, P(ri) is true for al l n > 1.
2. If p is a pri me andp \ a
x
a
2
a, thenp \ a-, for so me 1 < i < n.
Proof. Let P(ri) be the statement in Exampl e 2. Then P( l ) is true, for
if p | a
x
, it certainl y divides a
f
for so me 1 < < 1.
Suppo se we kno w that P(k) is true, and that p | a
x
a
2
a
k
a
k + x
. Thus,
by Theo rem 1.5.6, since p \ (a
x
a
2
a
k
)a
k+1
ei ther p \
Uk+i (
a
desi red con-
cl usion) o r p | a i a
k
. I n this second possibil ity, since P(k) is true we have
that p\a
f
for so me 1 1.
3. Fo r n > 1, 1 +2 + + =\n(n + 1).
Proof. If P(n) is the proposi ti on that 1 +2 + +n = \ n(n + 1), then
P(l) is certainl y true, for 1 =\ (1 + 1). If P(k) shoul d be true, this means that
1 + 2 + + k = \k(k + 1).
The questi on is: I s P(k + 1) then al so true, that is, is 1 + 2 + + k +
(k + 1) = |( / c + l )((/c + 1) + 1)? No w 1 + 2 + + k + (k + 1) =
(1 + 2 + k) + (k + 1) =\k(k + 1) + (k + 1), since P(k) is val id. But
|J fc(fc + 1) + (/c + 1) = |(/c(/c + 1) + 2(k + 1)) = + \){k + 2), which
assures us that P(k + 1) is true. Thus the pro po si ti o n 1 + 2 + + n =
\n(n +'T) is true for al l n > 1.
Sec . 6 Mat h emat i c al I n d u c t i o n 31
We must emphasi ze o ne po i nt here: Mathemati cal induction is not a
metho d for finding resul ts abo ut integers; it is a means of verifying a resul t.
We coul d, by o ther means, find the fo rmul a given abo ve for 1 + 2 + + n.
Part (b) of Theo rem 1.6.1 is usual l y cal l ed the induction step.
I n the pro bl ems we shal l give so me o ther versions of the principl e of in-
ducti on.
PROBL EMS
Easi er Pro bl ems
1. Pro ve that l
2
+2
2
+ 3
2
+ + n
2
= \n(n + l)(2n + 1) by induction.
2. Pro ve that I
3
+2
3
+ +
3
=\n
2
(n + l )
2
by induction.
3. Pro ve that a set having n s 2 el ements has \n(n 1) subsets having ex-
actl y two el ements.
4. Pro ve that a set having /? > 3 el ements has n(n l)(n 2)/3! subsets
havi ng exactl y three el ements.
5. If n > 4 and S is a set having n el ements, guess (from Pro bl ems 3 and 4)
ho w many subsets havi ng exactl y 4 el ements there are in S. Then verify
yo ur guess using mathemati cal induction.
*6. Co mpl ete the proof of Theo rem 1.5.6, repl acing the l ast sentence by an
i nducti o n argument.
7. If a 1, pro ve that 1 +a + a
2
+ + a" = (a"
+1
- 1)1 (a - 1) by i nduc-
tion.
8. By i nducti on, show that
J - + J - + . . . +
1
=
- 2 L _
1 - 2 2 - 3 n(n + 1) n. + 1 "
*9. Suppo se that P(n) is a pro po si ti o n abo ut positive integers /( such that
P(n
0
) is val id, and if P(k) is true, so must P(k + 1) be. What can yo u say
abo ut P(n)l Pro ve yo ur statement.
*10. Let P(n) be a pro po si ti o n abo ut i ntegers n such that P(l) is true and
such that if P(j) is true for al l positive integers / < k, then P(k) is true.
Pro ve that P(n) is true for al l positive integers n.
11. Gi ve an exampl e of a pro po si ti o n that is not true for any positive integer,
yet for which the i nducti o n step [Part (b) of Theo rem 1.6.1] hol ds.
12. Pro ve by i nducti on that a set having n el ements has exactl y 2" subsets.
13. Pro ve by i nducti on o n n that n
3
n is al ways divisibl e by 3.
14. Using induction on n, general ize the resul t in Pro bl em 13 to: If p is a pri me
number, then n
p
n is al ways divisibl e by p. (Hint: The binomial theorem.)
15. Pro ve by i nducti o n that for a set having n el ements the number of 1-1
mappi ngs of this set o nto itsel f is nl.
7. COMPL EX NUMB ERS
We al l kno w so methi ng abo ut the integers, rati onal numbers, and real num-
bersi ndeed, this assumpti o n has been made for so me of the text materi al
and many of the pro bl ems have referred to these numbers. Unfo rtunatel y,
the compl ex numbers and thei r pro perti es are much l ess kno wn to present-
day col l ege students. At o ne time the compl ex numbers were a part of the
high school curri cul um and the earl y col l ege one. This is no l o nger the case.
So we shal l do a rapi d devel o pment of this very i mpo rtant mathemati cal set.
The set of complex numbers, C, is the set of al l a + bi, where a, b are
real and where we declare:
1. a + bi = c + cli, for a, b, c, d real , if and onl y if a = c and b = d.
2. (a + bi) (c + di) = (a c) + (b d)i.
3. (a + bi)(c + di) = (ac - bd) + (ad + bc)i.
This l ast pro pertymul ti pl i cati o ncan best be remembered by using
i
2
= - 1 and mul tipl ying out formal l y with this rel ati on in mi nd.
Fo r the co mpl ex number z = a + bi, a is cal l ed the real part of z and b
the imaginary part of z. If a is 0, we cal l z purely imaginary.
We shal l write 0 + Oi as 0 and a + Oi as a. No te that z + 0 = z, zl = z
for any compl ex number z.
Gi ven z = a + bi, there is a co mpl ex number rel ated to z, whi ch we
wri te as z, defi ned by z = a - bi. Thi s co mpl ex number, z, is cal l ed the
complex conjugate of z. Taki ng the co mpl ex co nj ugate gives us a mappi ng
of C o nto itsel f. We cl aim
L emma 1.7.1. If z, w G C, then:
(a) \J)=z.
(b) (z + w) = z + w.
(c) (zw) = zw
(d) zz is real and no nnegati ve and is, in fact, positive if z i= 0.
Sec . 7 Co mp l ex Nu mb e r s 33
(e) z + z is twice the real part of z.
(f) z ~ z is twice the imaginary part of z ti mes i.
Proof. Mo st of the parts of this l emma are straightforward and merel y in-
vol ve using the definition of compl ex conjugate. We do verify Parts (c) and (d).
Suppo se that z = a + bi,w = c + di, where a, b, c, d are real . So zw =
(ac - bd) + (ad + bc)i, hence
(zw) = (ac bd) + (ad + bc)i = (ac bd) (ad + bc)i.
On the other hand, z = a - bi and w = c - di, hence, by the definition of the
pro duct in C, z w = (ac - bd) (ad + be) i. Compari ng this with the resul t
that we obtained for (zw), we see that i ndeed (zw) = z. w. This verifies Part (c).
We go next to the proof of Part (d). Suppo se that z = a + bi #0; then
z = a bi and zz = a
2
+ b
2
. Since a, b are real and no t bo th 0, a
2
+ b
2
is
real and positive, as asserted in Part (d).
The pro o f of Part (d) of L emma 1.7.1 shows that if z = a + bi' + 0, then
zz = a
2
+ b
2
#0andz ( l / ( V + b
2
)) = 1, so
acts l ike the inverse 1/z of z. This al l ows us to carry o ut division in C, staying
in C whil e doi ng so.
We no w l ist a few pro perti es of C.
L emma 1.7.2. C behaves under its sum and pro duct according to the
fol l owing: If u, v, w G C, then
(a) u + v = v + it.
(b) (u + v) + w = u + (v + w).
(c) uv = vu.
(d) (uv)w = u(vw).
(e) u # 0 impl ies that u~
l
= Vu exists in C such that uu~
l
= 1.
Proof. We l eave the proofs of these vari ous parts to the reader.
These pro perti es of C make of C what we shal l cal l a field, which we
shal l study in much greater depth l ater in the bo o k. What the l emma says is
that we are al l owed to cal cul ate in C mo re or l ess as we did with real num-
bers. Ho wever, C has a much ri cher structure than the set of real numbers.
34 Th i n g s Fami l i ar an d Less Famil iar Ch . 1
\uv\ = V(uv)(uv) - V(MU) (i<v) (by Part (c) of L emma 1.7.1)
= \/(uu)(vv) (by L emma 1.7.2)
= \/uu\/vxJ = \u\ \v\.
Ano ther way of verifying this l emma is to write u = a + bi, v = c + di,
u v = (ac bd) + (ad + bc)i and to no te the identity
(ac - bd)
2
+ (ad + be)
2
= (a
2
+ b
2
)(c
2
+ d
2
).
No te several smal l poi nts abo ut conjugates. If z G C, then z is real if
and onl y if z = z, and z is purel y i magi nary if and onl y if z = z. If
z, w G C, then
(zw + zw) = zw + zw = zw + zw,
so zw + zw is real . We want to get an upper bo und for \zw + zw\; this wil l
co me up in the proof of Theo rem 1.7.5 bel ow.
But first we must digress for a mo ment to o btai n a statement abo ut
quadrati c expressions.
Lemma 1.7.4. Let a, b, c be real , with a > 0. If act
2
+ ba + c > 0 for
every real a, then b
2
4ac 0.
Proof. Consider the quadrati c expression for a = -b/2a. We get
a(-b/2a)
2
+ b(-bl2a) + c > 0. Simpl ifying this, we obtai n that (4ac - b
2
)/4a
> 0, and since a > 0, we end up with 4ac - b
2
>0, and so b
2
- 4ac < 0.
We use this resul t i mmedi atel y to pro ve the i mpo rtant
T heorem 1.7.5 (T riangle Inequality). Fo r z,w E.C, \z + w \ <\z\ + \w\.
Proof. If z = 0, there is no thi ng to pro ve, so we may assume that z j= 0;
thus zz > 0. No w, for a real ,
We no w i ntro duce a "si ze" function on C.
Definition. If z = a + bi G C, then theabsolute value of z, wri tten as
\z\, is defined by \z\ =Vzi " = V o
2
+ b
2
.
We shal l see, in a few mo ments, what this l ast definition means geo met-
rical l y. I n the meanti me we pro ve
Lemma 1.7.3. If u, v G C, then \uv| = \u \ \v\.
Proof. By definition, \u \ = \Zial and |u | = Vuu. No w
0 < \az + w\
2
= (az + w) (az + w) = (az + w)(az + w)
= crzz + ctfzw + zw) + ww.
If a = zz > 0, b = zw + zw, c = wW, then L emma 1.7.4 tel l s us that
b
2
- 4ac = (zw + zw)
2
- 4(zz)(ww) < 0, hence (zw + zw)
2
< 4(zz)(ww) =
4|z|
2
|w|
2
. Therefo re, zw + zw < 2\z\ \w\.
Fo r a = 1 above,
|z + wj
2
= zz + ww + zW + zw = |zi
2
+ |w|
2
+ zW + zw
< |z|
2
+ M
2
+2\z\ \w\
from the resul t abo ve. I n o ther wo rds, |z + w \
2
s (|z | + | w |)
2
; taking square
ro o ts we get the desired resul t, |z + w \ < |z| + | w |.
Why is this resul t cal l ed the triangl e inequal ity? The reaso n wil l be cl ear
once we view the compl ex numbers geometrical l y. Represent the compl ex
number z = a + bi as the point having co o rdi nates (a, b) in the x-y pl ane.
The di stance r of this point from the origin is V a
2
+ b
2
, in o ther wo rds, |z|.
The angl e 6 is cal l ed the argument of z and, as we see, tan 6 = b/a. Al so ,
a = r cos 9, b = r sin 0; therefo re, z = a + bi = r(co s 0 + i sin 0). This rep-
resentati o n of z is cal l ed its polar form.
Gi ven z = a + bi, w = c + di, then thei r sum is z + w = (a + c) +
(b + d)i. Geo metri cal l y, we have the pi cture:
(a+c , b + d)
36 Th i n g s Fami l i ar an d Les s Fami l i ar Ch . 1
The statement \z + w\ ^ \z\ + \ w\ merel y refl ects the fact that in a triangl e
o ne side is of smal l er l ength than the sum of the l engths of the o ther two
sides; thus the term triangle inequality.
The compl ex numbers that co me up in the po l ar form cos 9 + i sin 9
are very i nteresti ng numbers i ndeed. Specifical l y,
|cos 9 + i sin Q\ =Vcos
2
9 + si n
2
9 =VI =1,
so they give us many co mpl ex numbers of absol ute val ue 1. I n truth they give
us all the co mpl ex numbers of absol ute val ue 1; to see this just go back and
l ook at the pol ar form of such a number.
Let's recal l two basic identities from tri go no metry, cos(# + i/>) =
cos 9 cos \jj sin 8 sin i/>and sin(0 + i/>) = sin 9 cos i/>+ cos 9 sin if/. There-
fore, if z = /'(cos 9 + i sin 9) and w = s(cos </>+ i sin i/>), then
zw = /-(cos 9 + i sin 9) s(cos )
= ra(cos 9 cos i/> sin 0 sin i/>) + z rs(sin 0 cos i/>+ cos 0 sin i/y)
= [co s(0 +(/>) + /si n(0 + i//)].
Thus, in mul tipl ying two co mpl ex numbers, the argument of the pro duct is
the sum of the arguments of the factors.
This has ano ther very i nteresti ng co nsequence.
Theo rem 1.7.6 (De Mo i vre's Theo rem). Fo r any i nteger n > 1,
[r(cos 9 + i sin 9)]" = r"[cos(nd) + i sin(;i0)].
Proof. We pro ceed by i nducti on o n n. If n = 1, the statement is
obviousl y true. Assume then that for so me k, [r(cos 9 + isin9)]
k
=
r
k
[cos k9 + isink9]. Thus
[/-(cos 8 + i sin 9)]
k + 1
= [r(cos 0 + i sin 0)]
/ c
r(co s 6 + i sin 0)
= r
f c
(co s kd + i sin kO) /-(cos 0 + / sin 9)
= r
k+1
[cos(k + 1)9 + isin(k + 1)9]
by the resul t of the paragraph abo ve. This compl etes the i nducti o n step;
hence the resul t is true for al l integers n s 1.
I n the pro bl ems we shal l see that De Mo i vre's Theo rem is true for al l
integers m; in fact, it is true even if m is rational .
Co nsi der the fol l owing special case:
9 = cos + i sin , where n > 1 is an integer.
Sec . 7 Co mp l e x Nu mb e r s 37
= cos 277 + / sin 277 = 1.
So 6", = 1; yo u can verify that 6'" + 1 if 0 < m < n. This pro perty of 6 makes
it o ne of the primitive nth roots of unity, which wil l be enco untered in Pro b-
l em 26.
PROBL EMS
Easi er Pro bl ems
1. Mul tipl y.
(a) (6 - 7i)(8 + i).
(b) (| + | 0 ( | - 0-
(c) (6 + 7i)(8 - 0-
2. Express z
- 1
in the form z
- 1
= a + bi fo r:
(a) z = 6 + 8/.
(b) 2 = 6- 8/ .
1
( C ) Z =
V2
+
V=2
L
_
*3 . Show that (?)
1
= (z"
1
).
4. Fi nd (cos 6 + i sin 0)
_ 1
.
5. Verify parts a, b, e, fof L emma 1.7.1.
*6. Show that z is real if and onl y if z = z, and is purel y imaginary if and
onl y if z = - z.
7. Verify the co mmutati ve l aw of mul tipl ication zw = wz in C.
8. Show that for z * 0, |z
_ 1
| = V\z\-
9. Find:
(a) 6 - 4i |.
(b) I +f i |-
( c ) j= H 7=1 .
V2 V2
10. Show that |z| = |z|.
By De Mo i vre's Theo rem,
11. Fi nd the pol ar form for
. . V2 1 .
(b) z = Ai.
i \ 6 , 6 .
(c) Z = 7= H 7= I.
,a\ 13 ^ 39 .
12. Pro ve that (co s(|0) + z sin(f 0))
2
= cos 0 + i sin 0.
13. By direct mul tipl ication show that (| + | V 3 z')
3
= 1.
Middle-Lev el Pro bl ems
14. Show that (cos 0 + i sin 9)"' = cos(m8) + i sin(m0) for all i ntegers m.
15. Show that (cos 9 + z'sin 6)'' = co s(r0) + isin(/'0) for al l rati o nal num-
bers r.
16. If z C and > 1 is any positive integer, show that there are n distinct
compl ex numbers w such that z = w".
17. Fi nd the necessary and sufficient condi ti on on k such that:
(
cos
{^r)
+1 sin
{^rj)
= 1 and
=1 if 0 < m < n.
18. Viewing the x-y pl ane as the set of al l co mpl ex numbers x + iy, show that
mul tipl ication by i induces a 90 ro tati o n of the x-y pl ane in a co unter-
cl ockwise direction.
19. I n Pro bl em 18, i nterpret geometrical l y what mul tipl ication by the co m-
pl ex number a + bi do es to the x-y pl ane.
*20. Pro ve that \z + w\
2
+ \z - w\
2
= 2(\z\
2
+ \w\
2
).
21. Co nsi der the set A = [a + bi | a, b Z}. Pro ve that there is a 1-1 co rre-
spo ndence of A o nto N. (A is cal l ed the set of Gaussian integers.)
22. If a is a (compl ex) root of the po l yno mi al
x" + axx""
1
-^ + a
n
^
1
x + a,
where the a
t
are real , show that a must al so be a ro o t, [r is a ro o t of a
pol ynomi al p(x) if p(r) = 0.]
2Trk\ , . . /27rifc\
cos + z sm
\ n \ n
Harder Pro bl ems
23. Fi nd the necessary and sufficient condi ti ons on z and w in o rder that
|z + w\ = |z| + \w\.
24. Fi nd the necessary and sufficient condi ti ons on zi , , z
k
in o rder that
|zi + + z
k
\ = |zi | + + \z
k
\.
*25. The co mpl ex number 9 is said to have order n > 1 if 0" = 1 and 0"' =h 1
for 0 < m < n. Show that if 9 has o rder and 9
k
= 1, where /c > 0, then
[ A:.
*26. Fi nd al l co mpl ex numbers 0 havi ng o rder . (These are the primitive nth
ro o ts of unity.)
1. DEFI NI TI ONS A N D EXA MPL ES OF GROUPS
We have seen in Section 4 of Chapter 1 that given any no nempty set, the set
A(S) of al l 1-1 mappi ngs of S o nto itsel f is no t just a set al o ne, but has a far
richer texture. The possibil ity of co mbi ni ng two el ements of A(S) to get
yet ano ther el ement of A(S) endo ws A(S) with an al gebraic structure. We
recal l ho w this was do ne: If /, g G A (S), then we co mbi ne them to form the
mappi ng fg defined by (fg)(s) = f(g(s)) for every s G S. We cal l ed fg the
product of / and g, and verified that/g G A(S), and that this pro duct o beyed
certai n rul es. Fro m the myri ad of possibil ities we so meho w sel ected four par-
ticul ar rul es that go vern the behavi o r of A (S) rel ative to this pro duct.
These four rul es were
1. Closure, namel y if / g G A(S), then fg G A(S). We say that A(S) is
closed under this pro duct.
2. Associativity, that is, given / g, h G A(S), then f(gh) = (fg)h.
3. Existence of a unit element, namel y, there exists a parti cul ar el ement
i G A(S) (the identity mappi ng) such that/*' = if = f'for al l / G A(S).
4. Existence of inverses, that is, given / G A(S) there exists an el ement,
deno ted by f~\ xnA{S) such that ff~
x
= f~
x
f = i.
To justify or mo ti vate why these four specific attri butes of A (S) were
singl ed out, in contradi sti ncti on to so me o ther set of pro perti es, is no t easy to
Sec . 1 Def i n i t i o n s an d Ex amp l es of Gr o u p s 41
do . I n fact, in the history of the subject it to o k qui te so me time to recognize
that these four pro perti es pl ayed the key rol e. We have the advantage of his-
torical hindsight, and with this hindsight we cho o se them not onl y to study
A(S), but al so as the chief guidel ines for abstracti ng to a much wider context.
Al tho ugh we saw that the four pro perti es abo ve enabl ed us to cal cul ate
concretel y in A(S), there were so me differences with the kind of cal cul ations
we are used to . If S has three or mo re el ements, we saw in Pro bl em 15,
Chapter 1, Section 4 that it is possibl e f o r/, g G A(S) to have fg # gf. Ho w-
ever, this did no t present us with i nsurmo untabl e difficul ties.
Wi tho ut any further pol emics we go to the
Definition. A no nempty set G is said to be a group if in G there is de-
fined an o perati o n * such that:
(a) a, b G G impl ies that a * b E G. (We descri be this by saying that G is
closed under *.)
(b) Gi ven a, b, c G G, then a * (b * c) = (a * b) * c. (This is described by
saying that the associative law ho l ds in G.)
(c) There exists a special el ement e G G such that a * e = e * a a for al l
a G G (e is cal l ed the identity or unit element of G).
(d) Fo r every a G G there exists an el ement b G G such that a * b =
b * a = e. (We write this el ement b as a~
l
and cal l it the inverse of
a in G.)
These four defining postul ates (cal l ed the group axioms) for a gro up
were, after al l , patterned after tho se that hol d in A (S). So it is no t surprising
that A (S) is a gro up rel ative to the o perati o n "co mpo si ti o n of mappi ngs."
The o perati o n * in G is usual l y cal l ed the product, but keep in mi nd
that this has no thi ng to do with pro duct as we kno w it for the integers, rati o -
nal , real s, or compl exes. In fact, as we shal l see bel ow, in many famil iar ex-
ampl es of gro ups that co me from numbers, what we cal l the pro duct in these
groups is actual l y the addition of numbers. However, a general group need
have no relation whatsoever to a set of numbers. We rei terate: A group is no
mo re, no l ess, than a no nempty set with an o perati o n * satisfying the four
group axioms.
Befo re starting to l ook into the nature of groups, we l ook at so me ex-
ampl es.
Examples of Groups
1. Let Z be the set of al l i ntegers and l et * be the ordi nary addition, +, i n Z.
That Z is cl osed and associative under * are basi c pro perti es of the i ntegers.
What serves as the unit el ement, e, of Z under *? Cl earl y, since a = a * e =
42 Groups Ch . 2
a + e, we have e = 0, and 0 is the requi red identity el ement under addi ti on.
What abo ut a"
1
? Here to o , since e = 0 = a * a
- 1
= a + a
- 1
, the a"
1
in this in-
stance is - a, and cl earl y a * ( - a) = a + (a) = 0.-
2. Let <Q be the set of al l rati o nal numbers and l et the o perati o n * o n Q be
the ordi nary addi ti o n of rati o nal numbers. As abo ve, <G> is easil y sho wn to be
a gro up under *. No te that I C Q and bo th Z and Q are gro ups under the
same o perati o n *.
3. Let Q' be the set of al l nonzero rati onal numbers and l et the o pera-
tion * on Q' be the o rdi nary mul tipl ication of rati o nal numbers. By the fa-
mil iar pro perti es of the rati o nal numbers we see that <Q' forms a gro up rel a-
tive to *.
4. Let R
+
be the set of al l positive real numbers and l et the o perati o n * on R
+
be the o rdi nary pro duct of real numbers. Agai n it is easy to check that R
+
is
a gro up under *.
5. Let E be the set of d', i = 0,1, 2 , . . . , 1, where 0 is the compl ex num-
ber 0 = cos(277-/) + /' sin(2- 7 r /n). Let 0* * B'
n
= 6*
+i
, the ordinary product of
the powers of 0 as compl ex numbers. By De Moivre's Theo rem we saw that
0" = 1. We l eave it to the reader to verify that E is a group under *. The el e-
ments of E are cal l ed the n'th roots of unity. The picture bel ow il l ustrates the
group E
h
, whose el ements are represented by the dots on the unit circl e in the
compl ex pl ane.
No te o ne striking difference between the Exampl es 1 to 4 and Exampl e
5; the first four have an infinite number of el ements, whereas E has a finite
number, n, of el ements.
Definition. A gro up G is said to be a finite group if it has a finite
number of el ements. The number of el ements in G is cal l ed the order of G and
is deno ted by \ G\.
Sec . 1
Def i n i t i o n s an d Ex amp l es of Gr o u p s 43
Thus E
n
abo ve is a finite gro up, and \E
n
\ = n.
Al l the exampl es presented abo ve satisfy the additional pro perty that
a * b = b * a for any pair of el ements. This need no t be true in a gro up. J ust
witness the case of A (S), where S has three o r mo re el ements; there we saw
that we coul d find fgE.A (S) such that fg = gf
Thi s pro mpts us to singl e o ut as special tho se gro ups of G in whi ch
a * b = b * a for al l a, b G G.
Defi ni ti o n. A gro up G is said to beabelian if a * b = b * a for al l
a, b G.
The word abelian derives from the name of the great Norwegian mathematician
Niels Henrik Abel (1802-1829), one of the greatest scientists Norway has ever
produced.
A gro up that is no t abel i an is cal l ed nonabelian, a not to o surprising
choice of name.
We no w give exampl es of so me no nabel i an groups. Of course, the
A(S) afford us an infinite famil y of such. But we present a few other exam-
pl es in which we can co mpute qui te readil y.
6. Let 1R be the set of al l real numbers, and l et G be the set of al l mappi ngs
T
b
: U > U defined by T
ab
(r) = ar + b for any real number r, where a, b
are real numbers and a 0. Thus, for i nstance, T
5
_
6
is such that T
5
_
6
(/
-
) =
5r - 6; r
5
_
6
(14) = 5 14 - 6 = 64, r
5
._
6
(7r) = 5tt '- 6. The T
a>b
are 1-1 map-
pings of U o nto itsel f, and we l et T
ab
* T
cd
be the pro duct of two of these
mappings. So
(T
a
,* r
c
,
rf
)(r) =r
a
,
6
(r
Ci rf
(r)) =r
c
,
rf
(r) + b = a(cr + d) + b
= (nc)r +(ad + b) = T,^
ad+b
(r).
So we have the formul a
T
a
,b* T
cd
= T
acad + b
. (1)
This resul t shows us that T
B
* T
c d
is in Gfor it satisfies the membershi p
requi rement for bel onging to Gso G is cl osed under *. Since we are tal king
abo ut the pro duct of mappi ngs (i.e., the composi ti on of mappi ngs), * is
associative. The el ement T
i0
= i is the identity mappi ng of R o nto itsel f.
Final l y, what is T
- 1
f c
? Can we find real numbers x # 0 and y, such that
44 Gr o u p s Ch . 2
T
a
,b * T
xy
T
x<y
* T
llb
T
l 0
?
Go back to (1) above; we thus want T
a x a y + b
= T
iQ
, that is, ax =1, ay +ft =0.
Remember no w that a + 0, so if we put x = a
- 1
and y =- a
- 1
f t, the requi red
rel ati ons are satisfied. One verifies i mmedi atel y that
a,h*
l
a-
l
-a-
,
b ~
1
a^-a^b *
1
,b ~ M , 0
So G is i ndeed a gro up.
What is T
c d
* T
a b
l Acco rdi ng to the formul a given in (1), where we re-
pl ace a by c, c by a, b by d, d by ft, we get
T
c
,a* T,
b
~ T
c i i c h + d
. (2)
Thus T
cd
* T
ih
=if r
f l i
* T
cd
and onl y if ftc +d = ad + ft. This fail s to be
true, for instance, if a 1, ft =1, c =2, d =3. So G is no nabel i an.
7. L et H C G, where G is the gro up in Exampl e 6, and H is defined by
H
=
[T
a
,b
G
G | a is rati o nal , ft any real }. We l eave it to the reader to verify
that H is a gro up under the o perati o n * defined on G. H is no nabel i an.
8. Let K C H C G, where 77, G are as abo ve and K = [T
x b
G G | ft any real }.
The reader shoul d check that K is a gro up rel ative to the o perati o n * of G,
and that K is, however, abelian.
9. Let S be the pl ane, that is, S = {(x, y) | x, y real } and co nsi der f,gEA (S)
defined by f(x, y) =(-x, y) and g(x, y) = (y, x); / is the refl ection abo ut
the y-axis and g is the ro tati o n thro ugh 90 in a countercl ockwise direction
abo ut the origin. We then define G =[fg' \ i = 0, 1 ; =0, 1, 2, 3}, and l et *
in G be the pro duct of el ements in A(S). Cl earl y, f
2
= g
4
= identity mappi ng;
(f*g)(x,y) = (fg)(x,y)=f(g(x,y))=f(-y,x) = (y,x)
and
(g*f)(x,y) = g(f(x,y)) = g(-x,y) = ( - y, -x).
So g * / =/ * g. We l eave it to the reader to verify that g * / =/ * g
- 1
and G
is a no nabel i an gro up of o rder 8. This gro up is cal l ed the dihedral group of
o rder 8. [Try to find a formul a for (fg') * (/ 'g') =fg
b
that expresses a, ft
in terms of i, j , s, and f\]
10. Let S be as in Exampl e 9 and/ the mappi ng in Exampl e 9. Let /i >2 and
l et A be the ro tati o n of the pl ane abo ut the origin thro ugh an angl e of Ittju
in the countercl ockwise direction. We then define G = [f
k
h'\k = 0, 1;
j =0, 1, 2 , 1} and define the pro duct * in G via the usual pro duct of
mappi ngs. One can verify that f
2
= h" = identity mappi ng, and fh = ft
-1
/.
These rel ati o ns al l ow us to show (with so me effort) that G is a no nabel i an
gro up of o rder 2n. G is cal l ed the dihedral group of o rder 2n.
11. Let G = {/ A(S) \f{s) # s for onl y a finite number of s S}, where we
suppo se that S is an infinite set. We cl aim that G is a gro up under the pro duct
* in A(S). The associativity hol ds automati cal l y in G, since it al ready hol ds in
A(S). Al so , i G, since i(s) = s for all s S. So we must sho w that G is
cl osed under the pro duct and if / G, then/ "
1
G.
We first dispose of the cl osure. Suppo se that / g G; then /(s) =s
except, say, for s
x
, s
2
, . . . , s
n
and g (s) = s except for s[, s
2
, , s,'
n
. Then
(fg)(
s
) = f(g(
s
)) =
s f o r a 1 1 5 o t h e r t h a n s
u s
2
, . . . , s, s{,. . .,s,' (and
possibl y even for so me of these). So fg mo ves onl y a finite number of el e-
ments of S, so fg G.
Final l y, if f(s) = s for al l s o ther than s
x
, s
2
, . . . , s, then f~
1
(f(s))
f'^s), but f-^s) = f-\f{s)) = (r\f)(s) = i(s) = s. So we obtain that
f^
1
(s) = s for al l 5 except s
u
. .., s
n
. Thus /
_ 1
G and G satisfies al l the group
axioms, hence G is a group.
12. Let G be the set of al l mappi ngs T
B
, where T
e
is the rotation of a given cir-
cl e about its center thro ugh an angl e 0 in the cl ockwise direction. I n G define
* by the composition of mappings. Since, as is readil y verified, T
#
* = T
e+ljl
,
G is cl osed under *. The o ther gro up axioms check o ut easil y. No te that
T
2
= T
Q
= the identity mapping, and Tg
1
= T_
e
= T
2lT
_
g
. G is an abel ian group.
As we did for A (S) we i ntro duce the sho rthand no tati o n a" for
a * a * a * a
n ti mes
and define a~" =(a"
1
)", for n a positive integer, and a" = e. The usual rul es
of expo nents then hol d, that is, (a'")" = a'"" and a'" * a" = a"'
+
" for any i nte-
gers m and n.
No te that with this no tati o n, if G is the gro up of integers under +, then
a" is real l y na.
Havi ng seen the 12 exampl es of gro ups abo ve, the reader mi ght get the
i mpressi on that al l , or al most al l , sets with so me o perati o n * fo rm gro ups.
This is far from true. We no w give so me exampl es of no ngro ups. I n each case
we check the four gro up axioms and see which of these fail to hol d.
Nonexampl es
1. Let G be the set of al l integers, and l et * be the ordi nary pro duct of i nte-
gers in G. Since a * b = ab, for a, b G we cl earl y have that G is cl osed and
associative rel ative to *. Furthermo re, the number 1 serves as the unit el e-
ment, since a * 1 = al = a = la = 1 * a for every a G G. So we are three-
fourths of the way to pro vi ng that G is a gro up. Al l we need is inverses for
the el ements of G, rel ative to *, to l ie in G. But this just isn't so. Cl earl y, we
canno t find an i nteger b such that 0* 5 = 0/3 = 1, since Ob = 0 for al l b. But
even o ther integers fail to have inverses in G. Fo r i nstance, we cannot find an
integer b such that 3 * b = 1 (for this wo ul d requi re that b = , and is no t
an i nteger).
2. Let G be the set of al l no nzero real numbers and define, for a, b G G, a * b
= a
2
b; thus 4 * 5 = 4
2
(5) = 80. Whi ch of the group axioms ho l d in G under
this o perati o n * and which fail to ho l d? Certainl y, G is cl osed under *. I s *
associative? If so, (a * b) * c = a * (b * c), that is, (a * b)
2
c = a
2
(b * c), and
so (a
2
b)
2
c = a
2
(b
2
c), which boil s do wn to a
2
= 1, which ho l ds onl y for
a = 1. So, in general , the associative l aw does not ho l d in G rel ative to *.
We simil arl y can verify that G do es no t have a uni t el ement. Thus even to
discuss inverses rel ative to * wo ul d no t make sense.
3. Let G be the set of al l positive integers, under * where a * b = ab, the
ordi nary pro duct of i ntegers. Then o ne can easil y verify that G fail s to be a
gro up only because it fails to have inverses for so me (in fact, mo st) of its el e-
ments rel ative to *.
We shal l find so me o ther no nexampl es of groups in the exercises.
P R O B L E M S
Easi er Pro bl ems
1. Determi ne if the fol l owing sets G with the o perati o n i ndi cated form a
gro up. If not, po i nt o ut which of the gro up axioms fail .
(a) G = set of al l i ntegers, a * b = a - b.
(b) G = set of al l i ntegers, a*b = a + b + ab.
(c) G = set of no nnegati ve i ntegers, a * b = a + b.
(d) G = set of al l rati o nal numbers 1, a * b = a + b + ab.
(e) G = set of al l rati o nal numbers with deno mi nato r divisibl e by 5 (writ-
ten so that numerato r and deno mi nato r are rel ativel y pri me), a * b =
a + b.
(f) G a set having mo re than o ne el ement, a * b = a for al l a, b G G.
2. I n the group G defined in Exampl e 6, show that the set H = {T
a h
j a = 1,
b any real } forms a gro up under the * of G.
3. Verify that Exampl e 7 is i ndeed an exampl e of a gro up.
4. Pro ve that K defined in Exampl e 8 is an abel ian gro up.
5. I n Exampl e 9, pro ve that g * f = f * g"
1
, and that G is a gro up, is no n-
abel ian, and is of o rder 8.
6. Let G and H be as in Exampl es 6 and 7, respectivel y. Sho w that if
T
llh
E G, then T_
h
*V* T~\ G H if V G 77.
7. Do Pro bl em 6 with 7/ repl aced by the gro up K of Exampl e 8.
8. If G is an abel ian gro up, pro ve that {a * fa)" =a" * b" for al l integers n.
9. If G is a gro up in which a
2
= e for al l a E G, show that G is abel ian.
10. If G is the gro up in Exampl e 6, find al l T
b
E G such that T
ab
* T
Ux
=
Ti
iX
* T
aJ
, for A// real x.
11. I n Exampl e 10, for n =3 find a formul a that expresses {fh') * {fh') as
/ " / 7
6
. Show that G is a no nabel i an gro up of o rder 6.
12. Do Pro bl em 11 for n = 4.
13. Show that any gro up of o rder 4 or l ess is abel i an.
14. If G is any gro up and a, b, c G G, show that if A * fa = a * c , then fa = c,
and if fa * a =c * a, then fa =c.
15. Express (a * fa)"
1
in terms of A "
1
and fa
-1
.
16. Usi ng the resul t of Pro bl em 15, pro ve that a group G in which a = A "
1
for every a G G must be abel ian.
17. I n any gro up G, pro ve that ( A "
1
)
- 1
=a for al l A G G.
* 18. If G is a finite gro up of even o rder, show that there must be an el ement
a + e such that a = a~
l
. (Hi nt: Try to use the resul t of Pro bl em 17.)
19. I n S
3
, show that there are four el ements x satisfying x
2
= e and three el e-
ments y satisfying y
3
=e.
20. Fi nd al l the el ements in S
4
such that x
4
= e.
21. Show that a gro up of o rder 5 must be abel i an.
22. Show that the set defined in Exampl e 10 is a gro up, is no nabel i an, and
has o rder 2n. Do this by finding the formul a for {fh') * {fh') in the
form fh
b
.
23. I n the gro up G of Exampl e 6, find al l el ements U G G such that
U * T
a
,b = T
a
,b * U for every T
a<b
E G.
24. If G is the dihedral group of order 2n as defined in Exampl e 10, prove that:
( a) If 77 is o dd and A G G is such that a * fa = fa* a for al l faG G, then a = e.
( b ) If 77 is even, show that there is an a G G, a + e, such that a * b = fa * a
for al l faG G.
(c) If 7 7 is even, find al l the el ements a G such that a * b = b * a
for al l b G G.
25. If G is any gro up, show that:
(a) e is uni que (i.e., i f / G G al so acts as a unit el ement for G, then / = e).
(b) Gi ven a G, then a
1
G G is uni que.
*26. If G is a finite gro up, pro ve that, given a G G, there is a positive i nteger
, dependi ng on a, such that a" = e.
*27. I n Pro bl em 26, sho w that there is an i nteger m > 0 such that a'" = e
for all a G G.
Harder Problems
28. Let G be a set with an o perati o n * such that:
1. G is cl osed under *.
2. * is associative.
3. There exists an el ement e G G such that e * x =x for al l x G G.
4. Gi ven x G G, there exists a y G G such that y * x = e.
Prove that G is a gro up. (Thus you must show that x * e = x and x * y = e
for e, y as above.)
29. Let G be a /imYe no nempty set with an o perati o n * such that:
1. G is cl osed under *.
2. * is associative.
3. Given a, b,c G G with a * b = a * c, then b = c.
4. Gi ven a, i>, c, G G with b * a = c * a, then b = c.
Pro ve that G must be a gro up under *.
30. Give an exampl e to show that the resul t of Pro bl em 29 can be fal se if G
is an infinite set.
31. Let G be the gro up of al l no nzero real numbers under the o perati o n *
which is the ordi nary mul tipl ication of real numbers, and l et H be the
gro up of al l real numbers under the o perati o n #, which is the addi ti on of
real numbers.
(a) Show that there is a mappi ng F: G -* H of G o nto H which satisfies
F(a * b) = F(a)#F(b) for al l a, b G G [i.e., F(a6) = F{a) + F(b)].
(b) Show that no such mappi ng /"can be 1-1.
2. SOME SI MPL E REMA RK S
I n this short section we show that certain formal pro perti es which fol l ow from
the group axioms hol d in any gro up. As a matter of fact, mo st of these resul ts
have al ready occurred as pro bl ems at the end of the precedi ng section.
Sec . 2 So m e Si mp l e Remar k s 49
I t is a l ittl e cl umsy to keep writing the * for the pro duct in G, and from
now on we shall write the product a * b simply as ab for al l a, b G G.
The first such formal resul ts we pro ve are co ntai ned in
L emma 2.2.1. If G is a gro up, then:
(a) I ts i denti ty el ement is unique.
(b) Every a G G has a unique inverse A "
1
G G.
(c) If a G G, ( a
- 1
) "
1
= a.
(d) Fo r a,bG, (A/3)"
1
= b'^'K
Proof. We start with Part (a). What is expected of us to carry out the
proof? We must show that if e, / G G and af = fa = a for al l a G G and
ae = ea = a for al l a G G, then e = f. This is very easy, for then e = ef and
/" =e/; hence e = ef = f as requi red.
I nstead of pro vi ng Part (b), we shal l pro ve a stro nger resul t (l isted
bel ow as L emma 2.2.2), which wil l have Part (b) as an i mmedi ate co nse-
quence. We cl aim that in a gro up G if ab = ac, then b = c; that is, we can
cancel a given element from the same side of an equation. To see this, we
have, for a G G, an el ement u G G such that ua = e. Thus from ab = ac we
have
u(ab) = u{ac),
so, by the associative l aw, (ua)b = (ua)c, that is, eb = ec. Hence b = eb =
ec = c, and o ur resul t is establ ished. A simil ar argument shows that if
ba = ca, then b = c. Ho wever, we cannot conclude from ab = ca that b = c;
in any abel ian gro up, yes, but in general , no .
No w to get Part (b) as an impl ication of the cancel l ation resul t. Suppo se
that b,c G G act as inverses for a; then ab = e = ac, so by cancel l ation b = c
and we see that the inverse of a is uni que. We shal l al ways write it as a
- 1
.
To see Part (c), no te that by definition A Â
- 1
) "
1
=e; but A "
1
a = e, so
by cancel l ation in ^" ' ( a"
1
) "
1
=e = a~
l
a we get that ( a
- 1
) "
1
= a.
Final l y, for Part (d) we cal cul ate
(A/>)(/3
_ 1
A
_ 1
) = ((A/3)/3"
1
)A"
1
(associative l aw)
= (A(/3/3~
1
)A"
1
(again the associative l aw)
= ( A C) A
1
= aa~
1
= e.
Simil arl y, (b~
1
a~
1
)(ab) = e. Hence, by definition, (ab)'
1
= b~
l
aT
x
.
We pro mi sed to l ist a pi ece of the argument given abo ve as a separate
l emma. We keep this pro mi se and write
L emma 2.2.2. I n any gro up G and a, b, c G G, we have:
(a) If ab = ac, then b = c.
(b) If ba = ca, then b =c.
Befo re l eaving these resul ts, no te that if G is the gro up of real
numbers under +, then Part (c) of L emma 2.2.1 transl ates i nto the famil iar
(a) = a.
There is onl y a scant bit of mathemati cs in this section; accordingl y,
we give onl y a few probl ems. No indication is given as to the difficul ty of these.
PROBL EMS
1. Suppo se that G is a set cl osed under an associative o perati o n such that
1. given a,y G G, there is an x G G such that ax = y, and
2. given a, w G G, there is a u G G such that ua = w.
Show that G is a gro up.
*2. If G is a finite set cl osed under an associative o perati o n such that ax = ay
forces ,Y = y and ua = wa forces u = w, for every a, x, y, it, w G G, pro ve
that G is a gro up. (This is a repeat of a pro bl em given earl ier. I t wil l be
used in the bo dy of the text l ater.)
3. If G is a gro up in which (ab)
1
= a'b' for three consecutive i ntegers i, pro ve
that G is abel ian.
4. Show that the resul t of Pro bl em 3 wo ul d no t al ways be true if the wo rd
"three" were repl aced by "two ." I n o ther wo rds, sho w that there is a
gro up G and consecuti ve numbers i, i + 1 such that G is not abel i an but
does have the pro perty that (ab)' = a'b' and (a/3)'
+ 1
= a'
+ l
b'
+ l
for al l
a, b in G.
5. Let G be a gro up in which (ab)
3
= a
3
b
3
and (ab)
5
= a
3
6
5
for al l a, b G G.
Show that G is abel ian.
6. Let G be a gro up in whi ch (ab)" = a"b" for so me fixed i nteger n > 1 for
al l a, b G G. Fo r al l a, b G G, pro ve that:
(a) (ab)"-
1
= b"-
[
a"-
]
.
( b ) a"b"-
1
= b"~
}
a".
(c) (afor
1
**-
1
)"*"-
1
) =e.
[Hint for Part (c): No te that (aba'
1
)
1
' = ab
r
a"
x
for al l integers;-.]
Sec . 3 Su b g r o u p s 51
3. S UBGROUP S
I n o rder for us to find out mo re abo ut the makeup of a given gro up G, it may
be too much of a task to tackl e al l of G head-o n. I t might be desirabl e to
focus o ur attenti o n on appro pri ate pieces of G, which are smal l er, over which
we have so me control , and are such that the i nformati on gathered abo ut
them can be used to get rel evant i nformati on and insight abo ut G itsel f. The
questi on then becomes: What shoul d serve as suitabl e pieces for this kind of
dissection of G? Cl earl y, whatever we cho o se as such pieces, we want them
to refl ect the fact that G is a gro up, no t merel y any ol d set.
A gro up is distinguished from an o rdi nary set by the fact that it is en-
do wed with a wel l -behaved o perati o n. I t is thus natural to demand that such
pieces abo ve behave reaso nabl y with respect to the o perati o n of G. Once
this is granted, we are l ed al most i mmedi atel y to the co ncept of a subgro up
of a gro up.
Definition. A no nempty subset, //. of a gro up G is cal l ed a subgroup
of G if, rel ative to the pro duct in G, //i tsel f forms a gro up.
We stress the phrase "rel ati ve to the pro duct in G." Take, for i nstance,
the subset A = {1, - 1} in Z, the set of integers. Under the mul tipl ication of
integers, A is a gro up. But A is not a subgro up of Z viewed as a group with
respect to +.
Every gro up G automatical l y has two obvious subgroups, namel y G it-
sel f and the subgro up consisting of the identity el ement, e, al one. These two
subgro ups we cal l trivial subgroups. Our i nterest wil l be in the remai ni ng
ones, the proper subgroups of G.
Before pro ceedi ng to a cl oser l ook at the general character of sub-
groups, we want to l ook at so me specific subgroups of some particul ar, ex-
pl icit groups. So me of the groups we consider are those we i ntroduced as ex-
ampl es in Section 1; we mai ntai n the numberi ng given there for them. I n
some of these exampl es we shal l verify that certain specified subsets are i ndeed
subgroups. We woul d strongl y reco mmend that the reader carry out such a
verification in l ots of the others and try to find other exampl es for themsel ves.
I n trying to verify whether or no t a given subset of a gro up is a sub-
gro up, we are spared checking o ne of the axioms defining a gro up, namel y
the associative l aw. Since the associative l aw ho l ds universal l y in a group G,
given any subset A of G and any three el ements of A, then the associative
l aw certainl y hol ds for them. So we must check, for a given subset A of G,
whether A is cl osed under the o perati o n of G, whether e is in A, and final l y,
given a E L A, whether a"
1
is al so in A.
No te that we can save o ne mo re cal cul ation. Suppo se that A C G is
no nempty and that given a, b G A, then ab G A. Suppo se further that given
a G A, then a
- 1
G A Then we assert that e G A. Fo r pick a G A; then a
- 1
G A
by supposition, hence aa
_ 1
G A, again by supposition. Since aa
- 1
= e, we
have that e E. A. Thus a is a subgro up of G. I n o ther wo rds,
Lemma 2.3.1. A no nempty subset A C G is a subgro up of G if and
onl y if A is cl osed with respect to the o perati o n of G and, given a G A, then
a
- 1
G A.
We no w consider so me exampl es.
Exampl es
1. Let G be the gro up Z of i ntegers under + and l et 77 be the set of even in-
tegers. We cl aim that 77 is a subgro up of Z. Why? I s 77 cl osed, that is, given
a, b G H, is a + b G 77? I n o ther wo rds, if a, b are even integers, is a + b an
even i nteger? The answer is yes, so 77 is certainl y cl osed under +. No w to the
inverse. Since the o perati o n in Z is +, the inverse of a G Z rel ative to this o p-
erati on is a. If a G 77, that is, if a is even, then -a is al so even, hence
a G 77. I n short, 77 is a subgro up of Z under +.
2. Let G once again be the gro up Z of integers under +. I n Exampl e 1, 77,
the set of even i ntegers, can be descri bed in ano ther way: namel y 77 consists
of al l mul tipl es of 2. There is no thi ng parti cul ar in Exampl e 1 that makes use
of 2 itsel f. Let m > 1 be any integer and l et 77, consist of al l mul tipl es of m
in Z. We l eave it to the reader to verify that H, is a subgro up of Z under +.
3. Let S be any no nempty set and l et G = A(S). If a G S, l et H(a) =
{ / G A(S)\f(a) = a}. We cl aim that 77(a) is a subgro up of G. Fo r if
/, g G 77(a), then (/g)(a) = /(g(a)) =/ (a) = a, since / ( a) = g(a) = a. Thus
fg G 77(a). Al so , i f / G 77(a), then/ ( a) = a, so t hat /
- 1
(/(a)) =/
- 1
( a) . But
r\f(a)) = = () =
T h u s
>
s i n c e a
= J "
- 1
(/()) =
w e n a v e
that /
- 1
G 77(a). Mo reo ver, 77 is no nempty. (Why?) Co nsequentl y, 77(a) is a
subgro up of G.
4. Let G be as in Exampl e 6 of Section 1, and 77 as in Exampl e 7. Then 77 is a
subgro up of G (see Pro bl em 3 in Section 1).
5. Let G be as in Exampl e 6, 77 as in Exampl e 7, and K as in Exampl e 8 in
Section 1. Then K C 77 C G and 7C is a subgro up of bo th 77 and of G.
6. Let C be the no nzero co mpl ex numbers as a gro up under the mul tipl ica-
tion of compl ex numbers. Let V = {a G C | \a \ is rati o nal }. Then F i s a sub-
gro up of C. Fo r if |a| and \ b\ are rati onal , then \ ab\ = |a| \b\ is rati o nal , so
ab G V; al so, \ a
1
| = 1/ \a | is rati o nal , hence a
1
G V. Therefo re, V is a sub-
gro up of C .
7. Let C and K be as abo ve and l et
U = (A G C | a = cos 0 + / sin 0, 0 any real ).
If fl = cos 6 + 7 sin 6 and = cos i/> + 2 sin I/A we saw in Chapter 1 that ab =
co s(0 + \}t) + i si n(0 + (/)), so that A6 G /, and that A "
1
= cos 0 - / sin 0 -
co s(- 0) + i si n(- e) G U. Al so , \a\ = 1, since \a\ = V co s
2
0 + si n
2
0 = 1.
Therefo re, U C V C C and (/ is a subgro up bo th of V and of C .
8. Let C , /, V be as abo ve, and l et n > 1 be an integer. Let 0 =
COS(2T77H) + i sin(277//7), and l et E = {1, 0, 0? .. . , dT'}. Since e;; = 1 (as
we saw by De Mo i vre's Theo rem), it is easil y checked that E is a subgro up
of U, V, and C , and is of o rder n.
9. Let G be any group and l et a G G. The set A = {fl' | / any i nteger) is a sub-
gro up of G. Fo r, by the rul es of expo nents, if a' G A and a' G A, then a'a' =
A '
+ ;
, SO is in A. Al so , ( A ' ) "
1
= A"', so (f l ')
- 1
G A. This makes A into a sub-
gro up of G.
A is the cycl ic subgro up of G generated by a in the fol l owing sense.
Definition. The cyclic subgroup of G generated by A is a set {a' j / any
i nteger). I t is deno ted (A).
No te that if e is the identity el ement of G, then (e) = je). I n Exampl e 8,
the gro up B is the cycl ic gro up (0) of C generated by 0.
10. Let G be any gro up; for a G G l et C(A) = [g G G | ga = ag}. We cl aim
that C(A) is a subgro up of G. First, the cl osure of C(A). If g, h G C(a), then
g A = A g and ha = a/i, thus (gh)a = g(ha) = g(ah) = (ga)h = (ag)h =
a(gh) (by the repeated use of the associative l aw), hence gh G C(a). Al so , if
g G C(a), then from ga = ag we have g~
1
(ga)g~
1
= g
1
(ag)g~\ which
simpl ifies to ag~
!
= g~
l
a\ whence g~
l
G C(a). So, C(a) is thereby a sub-
gro up of G.
These parti cul ar subgro ups C(a) wil l co me up l ater for us and they are
given a special name. We cal l C(a) the centralizer of a in G. If in a gro up
ab =fea, we say that a and b commute. Thus C(A) is the set of al l el ements in
G that co mmute with a.
11. Let G be any gro up and l et Z( G) = {z G G | z.r = xz for al l x G G). We
l eave it to the reader to verify that Z( G) is a subgro up of G. I t is cal l ed the
center of G.
12. Let G be any gro up and 77 a subgro up of G. Fo r a G G, l et a'
1
Ha =
{a~
[
ha | h G 77}. We assert that a~
]
Ha is a subgro up of G. If x = a
- 1
/ i
t
a and
y = a"
l
h
2
a where h
x
, h
2
G 77, then xy = (a~
l
h
l
a)(a~
l
h
2
a) = cr
l
(h
i
h
2
)a
(associative l aw), and since 77 is a subgro up of G, /?i/?
2
77. Therefo re,
a~\h
l
h
2
)a G a~
l
Ha, which says that xy G a~
l
Ha. Thus a~
l
Ha is cl osed.
Al so , if x = a
- 1
/ z a G a~
x
Ha, then, as is easil y verified, x
_ 1
= ( a
- 1
/ ) "
1
=
a~'
l
h~
1
a G a~
l
Ha. Therefo re, a~
l
Ha is a subgro up of G.
An even do zen seems to be abo ut the right number of exampl es, so we
go on to o ther things. L emma 2.3.1 poi nts o ut for us what we need in o rder
that a given subset of a gro up be a subgro up. I n an i mpo rtant special case we
can make a co nsi derabl e saving in checking whether a given subset 77 is a
subgro up of G. This is the case in which 77 is finite.
L emma 2.3.2. Suppo se that G is a gro up and 77 a no nempty finite sub-
set of G cl osed under the pro duct in G. Then 77 is a subgro up of G.
Proof. By L emma 2.3.1 we must sho w that a G 77 i mpl i es a'
1
G 77. If
a = e, then a~
1
= e and we are do ne. Suppo se then that a e ; co nsi der
the el ements a, a
2
, . . . , a "
+ 1
, where n = |77|, the o rder of 77. Here we
have written down n + 1 elements, all of them in H since H is closed, and 77
has only n distinct elements. Ho w can this be? Onl y if so me two of the el e-
ments l isted are equal ; put ano ther way, onl y if a' = a' for so me 1 1, a'~' G 77, hence e G 77. Ho wever, j - i - 1 > 0, so 'a'~
1
G 77
and aa'-'^
1
= a'~ ' = e, whence a
- 1
= a '
-
' "
1
G 77. Thi s pro ves the
l emma.
An immediate, but neverthel ess important, corol l ary to L emma 2.3.2 is the
Corol l ary. If G is a finite gro up and 77 a no nempty subset of G cl osed
under mul tipl ication, then 77 is a subgro up of G.
PROBL EMS
Easi er Pro bl ems
1. If A, B are subgro ups of G, show that A n B is a subgro up of G.
2. What is the cycl ic subgro up of Z generated by - 1 under +?
3. Le-t.S
3
be the symmetri c gro up of degree 3. Find al l the subgro ups of S
3
.
4. Verify that Z(G), the center of G, is a subgro up of G. (See Exampl e 11.)
5. If C(A) is the central izer of a in G (Exampl e 10), pro ve that Z(G) =
n
f l e G
C(a).
6. Show that a E Z(G) if and onl y if C(a) = G.
7. I n S
3
, find C(A ) for each A G 5
3
.
8. If G is an abel ian gro up and if H = {a G G | a
2
= e}, show that H is a
subgro up of G.
9. Give an exampl e of a no nabel i an gro up for which the H in Pro bl em 8 is
not a subgro up.
10. If G is an abel ian gro up and n > 1 an i nteger, l et A = {a" \ a G G}.
Pro ve that A,, is a subgro up of G.
* 11. If G is an abel ian gro up and H = [a G G | a"
( a)
= e for some 7 7 (a) > 1 de-
pendi ng o n a}, pro ve that His a subgro up of G.
We say that a gro up G is cyclic if there exists an a G G such that
every x G G is a po wer of A, that is, x = A
7
for so me /'. I n o ther wo rds, G
is cycl ic if G = (A) for so me a G G, in which case we say that A is a gen-
erator for G.
*12. Pro ve that a cycl ic gro up is abel i an.
13. If G is cycl ic, show that every subgro up of G is cycl ic.
14. If G has no pro per subgro ups, pro ve that G is cycl ic.
15. If G is a gro up and H a no nempty subset of G such that, given A, b G H,
then a/3"
1
G //, pro ve that His a subgro up of G.
Middle-Level Pro bl ems
* 16. If G has no pro per subgro ups, pro ve that G is cycl ic of o rder p, where p
is a pri me number. (This sharpens the resul t of Pro bl em 14.)
17. If G is a gro up and a, x G G, pro ve that C(x~
1
ax) = x~
x
C{a)x. [See Ex-
ampl es 10 and 12 for the definitions of C(b) and ofx~
1
C(a)x.]
18. If S is a no nempty set andX C S, sho w that T(X) = {fGA(S)\f(X) C
X) is a subgro up of A ( 5) if X is finite.
19. If A, B are subgroups of an abel ian group G, l et AB = {ab | a G A, b G B}.
Pro ve that is a subgro up of G.
20. Give an exampl e of a gro up G and two subgro ups A, B of G such that
is nor a subgro up of G.
21. If A, B are subgro ups of G such that b~
l
Ab C A for al l b E B, show that
A/ 3 is a subgro up of G.
* 22. If A and B are finite subgro ups, of o rders m and n, respectivel y, of the
abel ian gro up G, pro ve that AB is a subgro up of o rder inn if m and n are
relatively prime.
23. What is the o rder of AB in Pro bl em 22 if m and n are not rel ativel y pri me?
24. If 77 is a subgro up of G, l et N = C\
xSG
x~
1
Hx. Pro ve that TV is a subgro up
of G such that y~'TVy = TV for every y G G.
Harder Pro bl ems
25. Let 5, X, 7(X) be as in Pro bl em 18 (but X no l onger finite). Gi ve an ex-
ampl e of a set S and an infinite subset X such that T(X) is not a sub-
gro up of A (S).
* 26. Let G be a gro up, 77 a subgro up of G. Let Hx = [hx \ h G 77}. Sho w that,
given a,b E. G, then 77A = 77/3 or 77a D 77/3 = 0.
*27. If in Pro bl em 26 77 is a finite subgro up of G, pro ve that Ha and 77/3 have
the same number of el ements. What is this number?
28. Let M , TV be subgro ups of G such that x~
l
Mx C M and x~
l
Nx C TV for
al l x G G. Pro ve that MTV is a subgro up of G and that x~\MN)x C MTV
for al l x G G.
* 29. If M is a subgro up of G such that . Y
- 1
M X C M for al l x G G, pro ve that
actual l y MY = M.
30. If M, TV are such that .v"'Mr = M and ,Y
- 1
TVY = TV for al l x G G, and if
M n TV = (e), pro ve that inn = m for any m G M. n G TV. (Hi nt: Co n-
sider the el ement m~
l
n~
l
nm.)
4 . L AGR ANGE 'S THE ORE M
We are abo ut to derive the first real gro up-theo reti c resul t of i mpo rtance.
Al tho ugh its proof is rel ativel y easy, this theo rem is l ike the A-B-C's for fi-
nite groups and has i nteresti ng impl ications in number theo ry.
As a matter of fact, tho se of you who sol ved Pro bl ems 26 and 27 of Sec-
tion 3 have al l the necessary i ngredi ents to effect a proof of the resul t. The
theo rem simpl y states that in a finite gro up the o rder of a subgro up divides
the o rder of the gro up.
To smo o th the argument of this theo remwhi ch is due to L agrange
and for use many ti mes l ater, we make a short deto ur into the real m of set
theory.
J ust as the co ncept of "functi on" runs thro ugho ut mo st phases of math-
ematics, so al so does the co ncept of "rel ati o n." A relation is a statement aRb
abo ut the el ements a, b G S. If S is the set of i ntegers, a = b is a rel ati o n
o n S. Simil arl y, a < b is a rel ati on on S, as is a < b.
Sec . 4 L ag r an g e' s Th e o r e m 57
Definition. A rel ation on a set S is cal l ed an equivalence relation if,
for al l a, b, c G S, it satisfies:
(a) a ~ a (reflexivity).
(b) a b impl ies that b ~ a (symmetry).
(c) a ~ b, b ~ c impl ies that a ~ c (transitivity).
Of course, equal ity, =, is an equi val ence rel ati on, so the general no ti o n
of equi val ence rel ation is a general ization of that of equal ity. I n a sense, an
equi val ence rel ation measures equal ity with regard to so me attri bute. Thi s
vague remark may beco me cl earer after we see so me exampl es.
Exampl es
1. L et S be al l the items for sal e in a grocery store; we decl are a ~ b, for
a, b G S, if the price of a equal s that of b. Cl earl y, the defining rul es of an
equi val ence rel ation hol d for this ~. No te that in measuri ng this "general i zed
equal i ty" o n S we i gnore al l pro perti es of the el ements of S o ther than thei r
price. So a ~ b if they are equal as far as the attri bute of price is co ncerned.
2. Let S be the integers and n > 1 a fixed integer. We define a ~ b for a,
b G S if n | (a - b). We verify that this is an equi val ence rel ation. Since n | 0
and 0 = a a, we have a ~ a. Because n \ (a - b) impl ies that n\(b a), we
have that a ~ b impl ies that b ~ a. Final l y, if a ~ b and b ~ c, then
7 7 1 (A b) and 7 7 1 (b c); hence n \ ((a b) + (b c)), that is, n \ (a c).
Therefo re, a ~ c.
This rel ation on the integers is of great i mpo rtance in number theo ry
and is cal l ed congruence modulo n; when a ~ b, we wri te this as a = b mo d 7 7
[or, so meti mes, as a = b(n)], which is read "a co ngruent to b mo d 7 7 . " We'l l
be runni ng i nto it very often from no w on. As we shal l see, this is a special
case of a much wi der pheno meno n in groups.
3. We general ize Exampl e 2. Let G be a gro up and H a subgro up of G. Fo r
A, b G G, define a ~ b if ab~
l
G H. Since e G H and e = aa~\ we have that
a ~ fl . Al so , if A/J"
1
G H, then since His a subgro up of G, (ab~
]
)~
l
G H. But
(ab~
l
y
l
= (b~
x
y
1
a~
l
= ba'
1
, so ba~
x
G H, hence b ~ a. This tel l s us that
a ~ b impl ies that b ~ a. Final l y, if a ~ b and b ~ c, then ab~
}
G H and
/3c
_ )
G H. But (ab~
l
)(bc~
l
) = ac~
[
, whence ac"
1
G H and therefore
a ~ c. We have shown the transitivity of ~, thus ~ is an equival ence rel ati o n
o n G.
No te that if G = Z, the gro up of i ntegers under +, and H is the sub-
gro up consisting of al l mul tipl es of n, for 7 7 > 1 a fixed integer, then A/3"
1
G H
58 Gr o u p s
Ch . 2
transl ates i nto a = b(n). So co ngruence mo d n is a very special case of the
equi val ence we have defined in Exampl e 3.
I t is this equi val ence rel ati o n that we shal l use in provi ng Lagrange's
theo rem.
4. Let G be any gro up. Fo r a, b G G we decl are that a ~ b if there exists an
x G G such that b = x~
1
ax. We cl aim that this defines an equi val ence rel a-
tion o n G. First, a ~ a for a = e~
l
ae. Second, if a ~ b, then b = x~
1
ax, hence
a = (x^
1
y
1
b(x'
i
), so that b ~ a. Final l y, if a ~ b, b ~ c, then b = x~
1
ax, c =
y~
x
by for so me x, y G G. Thus c = y~
l
(x~
l
ax)y = (xyy
l
a(xy), and so a ~ c.
We have establ ished that this defines an equi val ence rel ati o n on G.
This rel ati on, to o , pl ays an i mpo rtant rol e in gro up theo ry and is given
the special name conjugacy. When a ~ b we say that "a and b are conjugate
in G." No te that if G is abel ian, then a ~ b if and onl y \fa = b.
We coul d go on and o n to give numero us interesting exampl es of equiva-
l ence rel ations, but this woul d sidetrack us from our main goal in this section.
There wil l be no l ack of exampl es in the pro bl ems at the end of this section.
We go on with o ur discussion and make the
Definition. If ~ is an equi val ence rel ation o n S, then [a], the class of
a, is defined by [a] = [b G S \ b ~ a}.
Let us see what the cl ass of a is in the two exampl es, Exampl es 3 and 4,
just given.
I n Exampl e 3, a ~ b if ab~
l
G 77, that is, if ab~
l
= h, for so me h G 77.
Thus a ~ b impl ies that a = hb. On the o ther hand, if a = kb where k G 77,
then ab'
1
= = k G 77, so a ~ b if and onl y if a G Hb = [hb | h G 77}.
Therefo re, [b] = Hb.
The set Hb is cal l ed a right coset of H m G. We ran into such in Pro b-
l em 26 of Section 3. No te that b G Hb, since b = eb and e G 77 (al so because
b G [b] = Hb). Ri ght cosets, and l eft handed co unterparts of them cal l ed left
cosets, pl ay i mpo rtant rol es in what fol l ows.
I n Exampl e 4, we defined a ~ b if b = x~
l
ax for so me x G G. Thus [a] =
[x~
l
ax | x G G}. We shal l deno te [a] in this case as cl (a) and cal l it the conju-
gacy class of a in G. If G is abel ian, then cl (fl ) consists of a al one. I n fact, if
a G Z(G), the center of G, then cl (a) consists merel y of a.
The no ti o n of conjugacy and its pro perti es wil l cro p up agai n often, es-
pecial l y in Section 11.
We shal l exami ne the cl ass of an el ement a in Exampl e 2 l ater in this
chapter.
The i mpo rtant infl uence that an equi val ence rel ati on has on a set is to
break it up and partition it into nice disjoint pieces.
T heorem 2.4.1. If -- is an equi val ence rel ati o n on S, then S = U[],
where this uni o n runs over o ne el ement from each cl ass, and where [a] #[fa]
impl ies that [a] Pi [b] = 0 . That is, ~ parti ti o ns S into equi val ence cl asses.
Proof. Since a [a], we have U
n E 5
[ a] =S. The proof of the second as-
serti on is al so qui te easy. We show that if [a] # [b], then [a] Pi [b] = 0 , or,
what is equi val ent to this, if [a] Pi [b] 0 , then [a] =[fa].
Suppo se, then, that [a] Pi [fa] 0 ; l et c G [a] Pi [fa]. By definition of
cl ass, c ~ a since c G [a] and c ~fa since c G [fa]. Therefo re, a ~ c by sym-
metry of ~, and so, since a ~ c and c ~fa, we have a ~ fa. Thus a G [fa]; if
x G [a], then x ~ A, fl ~fa gives us that x ~ fa, hence x G [fa]. Thus [a] C [fa].
The argument is obviousl y symmetri c in a and fa, so we have [fa] C [a],
whence [a] = [fa], and our assertion abo ve is pro ved.
The theo rem is no w compl etel y pro ved.
We no w can pro ve a famo us resul t of Lagrange.
T heorem 2.4.2 (Lagrange's T heorem). If G is a finite gro up and H is
a subgro up of G, then the o rder of H divides the o rder of G.
Proof. Let us l ook back at Exampl e 3, where we establ ished that the
rel ati o n a ~ fa if fl fa"' G H is an equi val ence rel ati o n and that
[a] = Ha = \ha\hEH).
Let k be the number of distinct cl assescal l them Ha
x
, ... , Ha
k
. By Theo -
rem 2.4.1, G = Ha
x
U Ha
2
U U Ha
k
and we kno w that Ha
}
Pi Ha
t
= 0 if
i * j .
We assert that any Ha, has \H\ = o rder of / / number of el ements. Map
H -> Hai by sending h fafl ,. We cl aim that this map is 1-1, for if ha
t
=
/7 'fl ,,then by cancel l ation in G we woul d have h = h'\ thus the map is 1-1. I t
is definitel y o nto by the very definition of Ha
{
. So H and Ha
t
have the same
number, | H \, of el ements.
Since G = Ha
{
U U Ha
k
and the Ha
t
are disjoint and each Ha; has
1771 el ements, we have that \G\ = k\H\. Thus |7/ | divides \G\ and La-
grange's Theo rem is pro ved.
Al though Lagrange sounds l ike a French name, J . L. Lagrange (1736-1813) was
actual l y Ital ian, having been born and brought up in Turin. He spent most of
his l ife, however, in France. Lagrange was a great mathematician who made
fundamental contributions to al l the areas of mathematics of his day.
If G is finite, the number of right cosets of 77 in G, namel y | G | /1771, is
cal l ed theindex of 77 in G and is wri tten as i
G
(77).
Recal l that a gro up G is said to be cyclic if there is an el ement a G G
such that every el ement in G is a po wer of a.
Theo rem2. 4 3 . A gro up G of pri me o rder is cycl ic.
Proof. If 77 is a subgro up of G then, by invoking Lagrange's Theo rem,
\H\ divides | G | = p, p a pri me, so | H\ = 1 or p. So if 77 + (e), then 77 = G. If
fl G G, a e, then the po wers of a form a subgro up () of G different from
(e). So this subgroup is al l of G. This says that any x G G is of the form x =
Hence, G is cycl ic by the definition of cycl ic gro up.
If G is finite and a G G, we saw earl ier in the pro o f of L emma 2.3.2 that
a
() =
e
f
o r
so me n(a) > 1, dependi ng on a. We make the
Definition. If G is finite, then the order of a, wri tten o (a), is the Zea.vf
positive integer m such that a'" = <?.
Suppo se that a G G has o rder m. Co nsi der the set A = [e, a, a
2
,..., a' "
- 1
};
we cl aim that A is a subgro up of G (since a'" = e) and that the m el ements
l isted in A are distinct. We l eave the verification of these cl aims to the
reader. Thus \ A\ = m. o(a). Since \A \ | \G\, we have
Theo rem 2.4.4. If G is finite and a G G. then o(a) | \G\.
If a G G, where G is finite, we have, by Theo rem 2.4.4, |G| = k o(a).
Thus
fl
l G|
= a
kM) = (
f l
W)* =
e
* =
e
.
We have pro ved the
Theo rem 2.4.5. If G is a finite gro up of o rder n, then a" = e for al l
a G G.
When we appl y this l ast resul t to certain special gro ups arising in num-
ber theo ry, we shal l o btai n so me cl assical number-theo reti c resul ts due to
Fermat and Eul er.
Let Z be the integers and l et n > 1 be a fixed integer. We go back to
Exampl e 2 of equi val ence rel ati ons, where we defined a = b mo d n (a
co ngruent to b mo d n) if n \ (a b). The cl ass of a, [a], consists of al l
a + nk, where k runs thro ugh al l the integers. We cal l it the congruence class
of a.
By Eucl i d's Al go ri thm, given any integer b, b = qn + r, where
0 < r < n, thus [fe] = [/]. So the n cl asses [0], [1], . . . , [ 1] give us al l
the co ngruence cl asses. We l eave it to the reader to verify that they are
distinct. \
Let Z = {[0], [1], . . . , [ - 1]}. We shal l i ntro duce two o perati o ns, +
and in Z. Under + Z wil l form an abel ian gro up; under Z wil l not fo rm
a gro up, but a certai n piece of it wil l beco me a gro up.
Ho w to define [a] + [b]l What is mo re natural than to define
[a] + [b] = [a + b].
But there is a fl y in the o i ntment. I s tjhis o perati o n + in Z wel l -defined?
What do es that mean? We can represent [a] by many a'sfor instance, if
n = 3, [1] = [4] = [2] = , yet we are using a particular a to define the
addition. What we must show is that if [a] = [a'] and [b] = [b'], then [a + b]
= [a' + b'], for then we wil l have [a] + [b] = [a + b] = [a' + b'] =
[a'] + [b'].
Suppo se that [a] = [a']; then n \ (a a'). Al so from [b] = [/>'],
n. | (b - b'), hence n \ ((a - a') + (b - b')) = ((a + b) - (a' + b')). There-
fore, a + b = (a' + b') mo d n, and so [a + b] = [a' + b'].
So we no w have a wel l -defined addi ti o n in Z. The el ement [0] acts as
the identity el ement and [-a] acts as [fl ], the inverse of [a]. We l eave it to
the reader to check o ut that Z is a gro up under +. I t is a cycl ic gro up of
o rder n generated by [1].
We summari ze this al l as
Theo rem 2.4.6. Z forms a cycl ic gro up under the addition [a] + [b] =
[ + b].
Havi ng di sposed of the addi ti o n in Z, we turn to the i ntroducti on of a
mul tipl ication. Agai n, what is mo re natural than defining
[a]-[b] = [ab]?
So, for i nstance, if n = 9, [2][7] = [14] = [5], and [3][6] = [18] = [0]. Under
this mul ti pl i cati o nwe l eave the fact that it is wel l -defined to the reader
Z does not form a gro up. Since [0][a] = [0] for al l a, and the uni t el ement
under mul tipl ication is [1], [0] canno t have a mul tipl icative inverse. Okay,
why no t try the no nzero el ements [fl ] + [0] as a candi date for a gro up under
this pro duct? Here again it is no go if n is not a prime. Fo r instance, if n = 6,
then [2] + [0], [3] * [0], yet [2] [3] = [6] = [0], so the no nzero el ements do
not, in general , give us a gro up.
So we ask: Can we find an appro pri ate pi ece of Z that wil l form a
gro up under mul tipl ication? Y es! Let U = [[a] Z | (a, n) = 1), noti ng
that (a, n) = 1 if and onl y if (b, n) = 1 for [a] = [b]. By the Co ro l l ary to
Theo rem 1.5.5, if (a, n) = 1 and (b, n) = 1, then (ab, n) = 1. So [a][b] = [ab]
yiel ds that if [a], [b] G U, then [aft] G / and U is cl osed. Associativity is
easil y checked, fol l owing from the associativity of the integers under mul ti-
pl ication. The identity el ement is easy to find, namel y [1]. Mul tipl ication is
co mmutati ve in U.
No te that if [a][b] = [a][c] where [a] G U, then we have [ab] = [ac],
and so [ab ac] = [0]. This says that n \a(b c) = ab ac; but a is rel a-
tivel y pri me to n. By Theo rem 1.5.5 o ne must have that n \ (b - c), and so
[b] = [c]. I n other wo rds, we have the cancel l ation pro perty in U. By Pro b-
l em 2 of Section 2, U is a gro up.
What is the o rder of Ul By the definition of U, \ U \ = number of in-
tegers 1 < m < n such that (m, n) = 1. This number co mes up often and we
give it a name.
Definition. The Eider cp-function, (p(n), is defined by ^(1) = 1 and,
for n > 1, <p(n) = the number of positive i ntegers m with 1 < m < n such
that (m, n) = 1.
Thus \ U \ = (p(n). If n = p, a pri me, we have <p(p) = p 1. We see
that <p(8) = 4 for onl y 1, 3, 5, 7 are l ess than 8 and positive and rel ativel y
pri me to 8. We try ano ther o ne, <p(15). The numbers 1 s m < 15 rel ativel y
pri me to 15 are 1, 2, 4, 7, 8,11,13,14, so <p(15) = 8.
Let us l ook at so me exampl es of U.
1. U
s
= {[1], [3], [5], [7]}. No te that [3][5] = [15] = [7], [5]
2
= [25] = [1]. I n
fact, U
s
is a group of o rder 4 in which a
2
= e for every a G C/
8
.
2. L/
15
= {[1], [2], [4], [7], [8], [11], [13], [14]}. No te that [11][13] = [143] =
[8],[2]
4
= [1], and so o n.
The reader shoul d verify that a
4
= e = [1] for every a G U
15
.
3. U
g
= {[1], [2], [4], [5], [7], [8]}. No te that [2]
1
= [2], [2]
2
= [4], [if = [8],
[2
4
] = [16] = [7], [2]
5
= [32] = [5]; al so [2]
6
= [2][2]
5
= [2][5] = [10] = [1].
So the po wers of [2] give us every el ement in U
9
. Thus U
g
is a cyclic gro up of
o rder 6. What o ther el ements in U
9
generate f7
9
?
I n paral l el to Theo rem 2.4.6 we have
Theo rem 2.4.7. U forms an abel ian gro up, under the pro duct
[a][b] = [ab], of o rder <p(n), where (p(ri) is the Eul er ^-function.
An i mmedi ate co nsequence of Theo rems 2.4.7 and 2.4.5 is a famous re-
sul t in number theory.
T heorem 2.4.8 (Euler). If a is an integer rel ativel y pri me to n, then
a
<p(n) = I
m o c
J
Proof. U forms a gro up of o rder <p(ri), so by Theo rem 2.4.5, a'
p(n)
=e
for al l a <EU. This transl ates i nto [a'
pin)
] = [aY
(n)
=[1], which in turn trans-
l ates i nto n | (a
v(
"
)
1) for every integer a rel ativel y pri me to p. I n o ther
wo rds, a'
Fl
"
)
= 1 mo d n.
A special case, where n = p is a pri me, is due to Fermat.
Corollary (Fermat). Up is a pri me and p \ a, then
a
p
~
l
=1 mo d p.
Fo r any i nteger b,b
p
= b mo d p.
Proof. Since <p(p) =p 1, if (fl , p) = 1, we have, by Theo rem 2.4.8,
that a
p _ 1
= l ( p) , hence a
1
a
p
~
l
= a (p), so that fl
p
= a (p). If p\b, then
bÔ(p) and ft" =0(p), so that b
p
= b (p).
Leonard Eul er (1707-1785) was probabl y the greatest scientist that Switzerl and
has produced. He was the most prol ific of al l mathematicians ever.
Pierre Fermat (1601-1665) was a great number theorist. Fermat's Last The-
oremwhich was in fact first proved in 1994 by Andrew Wil esstates that the
equation a" + b" = c" (a, b, c, n being integers) has onl y the trivial sol ution where
tf =0orfe =0o rc =0i f >2.
One final cauti o nary wo rd abo ut Lagrange's Theo rem. I ts converse in
general is not true. That is, if G is a finite gro up of o rder n. then it need no t
be true that for every divisor m of n there is a subgro up of G of o rder m. A
gro up with this pro perty is very special i ndeed, and its structure can be
spel l ed o ut qui te wel l and precisel y.
PROBL EMS
Easi er Problems
1. Verify that the rel ation ~is an equi val ence rel ati on o n the set S given.
(a) S = U real s, a ~ b if a - b is rati o nal .
(b) S =C, the co mpl ex numbers, a~b\f\a \ = \b\.
(c) S = straight l ines in the pl ane, a ~ b if a, b are paral l el .
(d) S = set of al l peo pl e, a ~b if they have the same col or eyes.
2. The rel ation ~ on the real numbers IR defined by a ~ b if bo th a > b and
b > a is not an equi val ence rel ati on. Why no t? What pro perti es of an
equi val ence rel ati o n does it satisfy?
3. Let ~ be a rel ati o n on a set S that satisfies (1) a ~ b impl ies that b ~ a
and (2) a ~ b and b ~ c impl ies that a ~ c. These seem to impl y that
a ~ a. Fo r ifa~b, then by (1), b ~ a, so a ~ b, b ~ a, so by (2), a ~ a. If
this argument is correct, then the rel ati on ~ must be an equi val ence rel a-
tion. Pro bl em 2 shows that this is no t so. What is wro ng with the argu-
ment we have given?
4. Let S be a set, {S
a
} no nempty subsets such that S = U
a
S
a
and S
a
fl Sp =
0 if a = /3. Define an equi val ence rel ation o n S in such a way that the 5
a
are precisel y al l the equi val ence cl asses.
* S. Let G be a gro up and H a subgro up of G. Define, for a, b G, a ~ b if
" '6 i /. Pro ve that this defines an equi val ence rel ati on o n G, and show
that [A] = aH = [ah\h H). The sets A / / are cal l ed left cosets of
in G.
6. If G is S
3
and 77 = [i, f}, where f:S S is defined by / ( x
:
) =
x
2 - f(
x
i)
= x
i , f(
x
i)
= x
3 , l ist al l the right cosets of H in G and l ist al l
the l eft cosets of H in G.
7. I n Pro bl em 6, is every right coset of H in G al so a l eft coset of H in G?
8. If every right coset of H in G is a l eft coset of H in G, pro ve that
aHa^
1
= Hfor al l a G.
9. I n Z
1 6
, write do wn al l the cosets of the subgro up H = {[0], [4], [8], [12]}.
(Since the o perati o n in Z is +, wri te yo ur coset as [a] + H. We do n't
need to distinguish between right cosets and l eft cosets, since Z is
abel ian under +.)
10. I n Pro bl em 9, what is the index of H in Z
1 6
? (Recal l that we defined the
index i
G
(H) as the number of right cosets in G.)
11. Fo r any finite gro up G, show that there are as many distinct l eft cosets of
H in G as there are right cosets of H in G.
12. If aH and bH are distinct l eft cosets of H in G, are Ha and Hb distinct
right cosets of H in G? Pro ve that this is true or give a co unterexampl e.
13. Fi nd the o rders of al l the el ements of U
iS
. Is U
18
cycl ic?
14. Fi nd the o rders of al l the el ements of U
20
. Is U
20
cycl ic?
* 15. If p is a pri me, show that the onl y sol utions of x
2
= 1 mo d p are x =
1 mo d p or x = 1 mo d p.
*16. If G is a finite abel ian gro up and a
,..., a are al l its el ements, sho w

that x = i f l
2
fl must satisfy x
2
= e.
17. If G is of o dd o rder, what can yo u say abo ut the x in Pro bl em 16?
Sec . 4
L ag r an g e' s Th e o r e m 65
18. Usi ng the resul ts of Pro bl ems 15 and 16, pro ve that if p is an odd pri me
number, then (p - 1)! = - 1 mo d p. (This is kno wn as Wilson's Theo-
rem.) I t is, of course, al so true if p = 2.
19. Fi nd al l the distinct conjugacy cl asses of S
3
.
20. I n the gro up G of Exampl e 6 of Section 1, find the conjugacy cl ass of the
el ement T
a b
. Descri be it in terms of a and b.
21. Let G be the di hedral gro up of o rder 8 (see Exampl e 9, Section 1). Fi nd
the conjugacy cl asses in G.
22. Verify Eul er's Theo rem for n = 14 and a = 3, and for n = 14 and a = 5.
23. I n U
4i
, show that there is an el ement a such that [a]
2
= [- 1], that is, an
i nteger a such that a
2
= - 1 mo d 41.
24. If p is a pri me number of the form An + 3, show that we cannot sol ve
x
2
= -1 mo d p
[Hint: Use Fermat's Theo rem that a
1
' '' = 1 mo d p if p \ a.]
25. Sho w that the no nzero el ements in Z fo rm a gro up under the pro duct
[a][b] = [ab] if and onl y if n is a pri me.
26. Let G be a group, H a subgroup of G, and l et S be the set of al l distinct
right cosets of H in G, T the set of al l l eft cosets of H in G. Pro ve that there
is a 1-1 mappi ng of 5 onto T. (Note: The obvious map that comes to mi nd,
which sends Ha into aH, is not the right one. See Probl ems 5 and 12.)
27. If all = bll forces Ha = Hb in G, sho w that n/Zrv"
1
= H for every A G.
28. If G is a cycl ic gro up of o rder n, sho w that there are <p(n) generato rs for
G. Gi ve their form expl icitl y.
29. If in a gro up G, aba'
1
= b', show that a'ba~
r
= b'
r
for al l positive inte-
gers /.
30. If in G a
5
= e and aba'
1
= b
2
, find o(b) if 6 # e.
*31. If o () = 7 7 7 and a
s
= e, pro ve that m | s.
32. Let G be a finite gro up, H a subgro up of G. Let f(a) be the l east positive
7 7 7 such that a'" H. Pro ve that/ (a) | o(a).
33. If z = f E. A(S) is such that /
p
= i, p a pri me, and if for so me
s S, f'(s) = s for so me 1 </ <p, show f hat/(s) =5.
34. If / A (S) has o rder p, p a pri me, show that for every s S the orbi t of
s under / has o ne o r p el ements. [Recal l : The orbit of s under / is
{/'"(*) I /any i nteger).]
35. If / G A (S) has o rder p, p a pri me, and S is a finite set having n el ements,
where (n,p) = 1, show that for so me sGS , f(s) = s.
Harder Pro bl ems
36. If a > 1 is an integer, show that n \ <p(a" - 1), where <p is the Eul er
(^-function. [Hint: Co nsi der the integers mod(fl " - 1).]
37. I n a cycl ic group of o rder n, show that for each positive integer m that di-
vides n (incl uding m = 1 and m = ri) there are <p(m) el ements of o rder m.
38. Usi ng the resul t of Pro bl em 37, show that n = 2
m
|<p( m) .
39. Let G be a finite abel i an gro up of o rder n for which the number of sol u-
tions of x'" = e is at mo st m for any m dividing n. Pro ve that G must be
cycl ic. [Hint: Let i//(m) be the number of el ements in G of order m. Sho w
that i/)(ra) ^ <p(m) and use Pro bl em 38.]
40. Usi ng the resul t of Pro bl em 39, show that U
p
, if p is a pri me, is cycl ic.
(This is a famous resul t in number theo ry; it asserts the existence of a
primitive root mo d p.)
41. Usi ng the resul t of Pro bl em 40, show that if p is a pri me of the fo rm
p = 4n + 1, then we can sol ve x
2
= 1 mo d p (with x an i nteger).
42. Using Wil son's Theo rem (see Pro bl em 18), show that if p is a pri me of
the formp = 4/7 + 1 and if
then y
2
==- 1 mo d p. (This gives ano ther proof of the resul t in Pro bl em
43. Let G be an abel ian gro up of o rder n, and a
1
,... , a
n
its el ements. L et
x = a
i
a
2
' Sho w that:
(a) If G has exactl y o ne el ement b # e such that /3
2
= e, then x =fo.
(b) If G has mo re than o ne el ement b e such that fo
2
= e, then x = e.
(c) If n is o dd, then x = e (see Pro bl em 16).
5. HOMOMORPHI SMS A ND NORMA L SUB GROUPS
I n a certain sense the subject of gro up theo ry is buil t up o ut of three basic
concepts: that of a ho mo mo rphi sm, that of a no rmal subgro up, and that of
the factor or quo ti ent gro up of a gro up by a no rmal subgro up. We discuss the
first two of these in this section, and the third in Section 6.
Wi tho ut further ado we i ntro duce the first of these.
41.)
Sec . 5
Ho mo mo r p h i s ms an d No r ma l Su b g r o u p s 67
Definition. Let G, G' be two gro ups; then the mappi ng <p : G G' is
a homomorphism if <p(ab) = cp(a)cp(b) for al l a,b G.
(Note: This cp has no thi ng to do with the Eul er <p-function.)
I n this definition the pro duct on the l eft sidein <p(ab)is that of G,
whil e the pro duct cp(a)cp(b) is that of G'. A sho rt description of a ho mo mo r-
phi sm is that it preserves the o perati o n of G. We do not insist that cp be o nto ; if
it is, we'l l say that it is. Before working o ut some facts abo ut ho mo mo rphi sms,
we present so me exampl es.
Exampl es
1. Let G be the gro up of al l positive real s under the mul tipl ication of real s,
and G' the gro up of al l real s under addition. Let cp: G G' be defined by
<p(x) = l o g
1 0
x for x G. Since l o g
1 0
(xy) = l o g
1 0
x + l o g
1 0
y, we have
<p(xy) = <p(x) + <p(y), so cp is a ho mo mo rphi sm. I t al so happens to be o nto
and 1-1.
2. Let G be an abelian gro up and l et cp: G > G be defined by cp (a) = a
2
.
Since cp(ab) = (ab)
2
= a
2
b
2
= (p(a)cp(b), cp is a ho mo mo rphi sm of G i nto it-
sel f. I t need no t be o nto ; the reader shoul d check that in U
s
(see Section 4)
a
2
= e for al l a U
8
, so cp(G) = (e).
3. The exampl e of U
s
abo ve suggests the so-cal l ed trivial homomorphism. Let
G be any gro up and G' any other; define <p(x) = e', the unit el ement of G',
for al l x G. Trivial l y, cp is a ho mo mo rphi sm of G into G'. It certainl y is no t
a very i nteresti ng one.
Ano ther ho mo mo rphi sm al ways present is the identity mapping, i, of any
group G into itsel f. Since i(x) =x for al l x G, cl earl y i(xy) = xy = i(x)i(y).
The map i is 1-1 and onto, but, again, is no t to o interesting as a ho mo mo r-
phism.
4. Let G be the gro up of integers under + and G' = {1, - 1}, the subgro up of
the real s under mul tipl ication. Define cp(m) = 1 if m is even, cp(m) = - 1 if m
is odd. The statement that cp is a ho mo mo rphi sm is merel y a restatement of:
even + even =even, even + o dd = odd, and o dd +odd = even.
5. Let G be the group of al l no nzero co mpl ex numbers under mul tipl ication
and l et G' be the group of positive real s under mul tipl ication. Let cp: G > G'
be defined by cp (a) = \a\; then cp (ab) = \ab\ = \ a\ \ b\ = cp (a) cp (b), so cp is a
ho mo mo rphi sm of G into G'. I n fact, cp is o nto .
6. Let G be the gro up in Exampl e 6 of Section 1, and G' the gro up of
no nzero real s under mul tipl ication. Define cp: G > G' by cp(T
ab
) = a. That
cp is a ho mo mo rphi sm fol l ows from the pro duct rul e in G, namel y, T
ab
T
cd
=
T
1
ac, ad + b
1. Let G = Z be the gro up of integers under + and l et G' = Z. Define
 Z by <p(m) =[/]. Since the addi ti on in Z is defined by [in] + [/*] =
[m + r], we see that <p(m + r) = <p(m) + <p(r), so cp is i ndeed a ho mo mo r-
phi sm of Z o nto Z.
8. The fol l owing general constructi on gives rise to a wel l -known theo rem.
Let G be any group, and l et A(G) be the set of al l 1-1 mappings of G onto it-
sel fhere we are viewing G merel y as a set, forgetting about its mul tipl ication.
Define T
a
: G > G by T
a
(x) = ax for every i G , W/raf K the product, T
a
T
b
, of
T
a
and T
b
as mappings on G? Wel l ,
(T
a
T
h
){x) = T
a
CT
b
x) = T
a
(bx) = a(bx) = (ab)x = T
ah
(x)
(we used the associative l aw). So we see that T
a
T
b
= T
ab
.
Define the mappi ng tp: G > A(G) by <p(a) = T
a
, for a E G. The pro d-
uct rul e for the T's transl ates into q>(ab) = T
ab
= TT
b
= cp(a)cp(b), so <p is a
ho mo mo rphi sm of G into A(G). We cl aim that cp is 1-1. Suppo se that cp(a) =
cp(/3), that is, T
a
= T
b
. Therefo re, a = T
a
(e) = T
b
(e) = b, so cp is i ndeed 1-1.
I t is no t o nto in general fo r instance, if G has o rder n > 2, then A(G) has
order n\, and since nl > n, cp doesn't have a ghost of a chance of bei ng onto.
It is easy to verify that the image of cp, cp(G) = [T
a
| a G G}, is a subgroup
o fA(G).
The fact that cp is 1-1 suggests that perhaps 1-1 ho mo mo rphi sms shoul d
pl ay a special rol e. We singl e them o ut in the fol l owing definition.
Definition. The ho mo mo rphi sm cp: G G' is cal l ed a monomor-
phism if <p is 1-1. A mo no mo rphi sm that is o nto is cal l ed an isomorphism. A n
i so mo rphi sm from G to G itsel f is cal l ed an automorphism.
One mo re definition.
Definition. Two gro ups G and G' are said to beisomorphic if there is
an i so mo rphi sm of G o nto G'.
We shal l deno te that G and G' are i so mo rphi c by writing G G'.
This definition seems to be asymmetri c, but, in po i nt of fact, it is not.
Fo r if there is an i so mo rphi sm of G o nto G', there is o ne of G' o nto G (see
Pro bl em 2).
We shal l discuss mo re tho ro ughl y l ater what it means for two gro ups to
be i somorphi c. But no w we summari ze what we did in Exampl e 8.
Sec , 5 Ho m o m o r p h i s m s an d No r ma l Su b g r o u p s 69
T heorem 2.5.1 (Cayl ey's T heorem). Every gro up G is i so mo rphi c to
so me subgro up of A (S), for an appro pri ate S.
The appro pri ate S we used was G itsel f. But there may be better
choices. We shal l see so me in the pro bl ems to fol l ow.
When G is finite, we can take the set S in Theo rem 2.5.1 to be finite, in
which case A (S) is S and its el ements are permutati o ns. I n this case, Cay-
l ey's Theo rem is usual l y stated as: A finite group can be represented as a
group of permutations.
(Arthur Cayl ey (1821-1895) was an Engl ish mathematician who worked in ma-
trix theory, invariant theory, and many other parts of al gebra.)
This is a go o d pl ace to discuss the i mpo rtance of "i so mo rphi sm." L et cp
be an i so mo rphi sm of G o nto G'. We can vi ew G' as a rel abel i ng of G,
using the l abel cp(x) for the el ement x. Is this l abel ing consistent with the
structure of G as a gro up? That is, if x is l abel ed cp(x), y l abel ed <p(y), what
is xy l abel ed as? Since cp(x)cp(y) = cp(xy), we see that xy is l abel ed as
cp(x)cp(y), so this renami ng of the el ements is consistent with the pro duct in
G. So two gro ups that are i so mo rphi cal tho ugh they need no t be equal i n
a certai n sense, as descri bed abo ve, are equal . Often, it is desirabl e to be abl e
to identify a given gro up as i so mo rphi c to so me concrete gro up that we
kno w.
We go o n with mo re exampl es.
9. Let G be any gro up, a G fixed in the discussion. Define cp: G > G by
cp(x) = a~
l
xa for al l x G. We cl aim that cp is an i so mo rphi sm of G o nto
itsel f. First,
cp(xy) = a~
x
(xy)a = a~
l
xa a~
x
ya = cp(x)cp(y),
so cp is at l east a ho mo mo rphi sm of G i nto itsel f. I t is 1-1 for if cp (x) = cp (y),
then a~
l
xa = a ~
l
ya, so by cancel l ati on in G we get x = y. Final l y, cp is o nto ,
for x = fl "
1
(fl.vfl" ')A = cp (fl xfl "
l
) for any x G.
Here cp is cal l ed the inner automorphism of G induced by a. The no ti o n
of automorphism and so me of its pro perti es wil l co me up in the pro bl ems.
One final exampl e:
10. Let G be the gro up of real s under + and l et G' be the group of al l
no nzero compl ex numbers under mul tipl ication. Define cp : G > G' by
cp(x) = co sx + i si nx.
We saw that (co sx + / si nx)(co sy + / si ny) = cos(x + y) + i sin(x + y),
hence cp (x)cp(y) = cp(x + y) and <p is a ho mo mo rphi sm of G into G'. cp is no t
f-1 because, for instance, <p(0) = <p(277) = 1, no r is cp o nto .
No w that we have a few exampl es in hand, we start a l ittl e investigation
of ho mo mo rphi sms. We begin with
L emma 2.5.2. If cp is a ho mo mo rphi sm of G into G', then:
(a) cp(e) = e', the unit el ement of G'.
(b) cp{
a
-
1
) = cp{aY
v
for al l a G G.
Pro o / Since x = xe, cp(x) = cp(xe) = cp(x)<p(e); by cancel l ati on in G'
we get cp{e) = e'. Al so , cpf/zfl "
1
) = cp(e) = e', hence e' = cp(aa~
l
) =
cp(a)cp ( a
- 1
) , which pro ves that <p(fl
_1
) = cp ( A )
- 1
.
Definition. The ('mage of cp, <p(G), is cp(G) = (cp(a) | a G G}.
We l eave to the reader the pro o f of
L emma 2.5.3. If is a ho mo mo rphi sm of G i nto G', then the i mage
of cp is a subgro up of G'.
We singl ed out certai n ho mo mo rphi sms and cal l ed them mo no mo r-
phisms. Thei r pro perty was that they were 1-1. We want to measure ho w far
a given ho mo mo rphi sm is from bei ng a mo no mo rphi sm. This pro mpts the
Definition. If cp is a ho mo mo rphi sm of G into G', then the kernel of
cp, Ker cp, is defined by Ker cp = \a G G | cp(a) = e'}.
Kercp measures the l ack of 1-1' ness at o ne point e'. We cl aim that
this l ack is rather uniform. What is W = {x G G\cp(x) = w'} for a given
w' G G'l We show that if cp(x) = w' for so me x G G, then W =
[kx\k G Kercpj = (Ker<p).\\ Cl earl y, if k G Ker cp and = w', then
cp(kx) = cp(k)cp(x) = e'cp(x) = w', so /cx G W. Al so , if <p(x) = <p(y) = w',
then <p(x) = i p(y), hence ^( y) ^^- ) "
1
= e'\ but <p (x)"
1
= cp (x ~ ') by
L emma 2.5.2, so e' = cp(y)<p(x)~
1
= cp(y)cp(x~
l
) = cp(yx
_ 1
), whence
yx~
1
G Ker cp and so y G (Ker cp)x. Thus the inverse image of any element
w ' in cp (G) G G' is the set (Ker cp)x, where x is Ay el ement in G such that
cp(x) = w'.
We state this as
L emma 2.5.4. If w' G G' is of the form cp(x) = w', then
[y G G | cp(y) = w'} = (Ker cp)x.
Sec . 5 Ho mo mo r p h i s ms an d No r ma l Su b g r o u p s 71
We no w shal l study so me basic pro perti es of the kernel s of ho mo mo r-
phi sms.
Theo rem 2.5.5. If cp is a ho mo mo rphi sm of G into G', then
(a) Ker cp is a subgro up of G.
(b) Gi ven a G G, a^f K er cp)a C Ker cp.
Proof. Al tho ugh this is so i mpo rtant, its pro o f is easy. If a, b G Ker cp,
then cp(a) = cp(b) = e', hence cp(ab) = cp(a)cp(b) = e', whence ab G Ker cp,
so Ker cp is cl osed under pro duct. Al so cp (a) = e' impl ies that cp (a~
J
) =
<p(a)~
l
= e', and so a^
1
G Ker cp. Therefore, Ker cp is a subgroup of G. If
k G Ker cp and a G G, then <p(/c) = e'. Co nsequentl y, cp(a~
1
ka) =
cp (a~
l
)cp(k)cp(a) = cp{a~
x
)e'cp{a) = ^( a
_ 1
) ^( a) = cp(cT
l
a) = <p(e) =e'. Thi s
tel l s us that a~
l
ka G Ker 9, hence a
_ 1
(K er cp)a G Ker <p. The theo rem is no w
compl etel y pro ved.
Co ro l l ary. If cp is a ho mo mo rphi sm of G i nto G', then 9 is a
mo no mo rphi sm if and onl y if Ker cp = (e).
Proof. Thi s resul t is real l y a corol l ary to L emma 2.5.4. We l eave the
few detail s to the reader.
Pro perty (b) of Ker cp in Theo rem 2.5.5 is an interesting and basic o ne
for a subgro up to enjoy. We ran into this pro perty in the text materi al and
pro bl ems earl ier o n several occasions. We use it to define the ul tra-i mpo rtant
cl ass of subgroups of a group.
Definition. The subgro up N of G is said to be a normal subgroup of
G if a~
x
Na C N for every a G G.
Of course, Ker cp, for any ho mo mo rphi sm, is a no rmal subgro up of G.
As we shal l see in the next section, every no rmal subgro up of G is the kernel
of so me appro pri ate ho mo mo rphi sm of G into an appro pri ate gro up G'. So in
a certain sense the noti ons of ho mo mo rphi sm and no rmal subgroups wil l be
shown to be equival ent.
Al tho ugh we defined a no rmal subgro up via a~
1
Na C TV, we actual l y
have a~
1
Na = N. Fo r if a~
1
Na C Nfor al l a G G, then N = a(a~
1
Na)a~
1
C
aNa'
1
= (/r' ^ T'/Vi i T
1
C TV. So N = aNa'
1
for every a G G. Transpo si ng,
we have Na = aN; that is, every l eft coset of TV in G is a ri ght coset of
N in G.
On the o ther hand, if every l eft coset of 7V in G is a right coset, then the
l eft coset aN, whi ch contains a, must be equal to the right coset contai ni ng a,
namel y Na. Thus, aN = Na and N = a~
l
Na for al l a G G, which is to say that
TV is no rmal in G.
We write "TV is a no rmal subgro up of G" by the abbrevi ated symbo l
N<G.
No te that a~
l
Na = N does not mean that a~
l
na = n for every n G TV.
No merel y that the set of al l a'
y
na is the same as the set of al l n.
We have pro ved
Theo rem 2.5.6. N < G if and onl y if every l eft coset of TV in G is a
right coset of TV in G.
Befo re going any further, we pause to l o o k at so me exampl es of kernel s
of ho mo mo rphi sms and no rmal subgro ups.
If G is abel ian, then every subgro up of G is no rmal , for a~
l
xa = x for
every a, x G G. The converse of this is not true. No nabel i an gro ups exist in
which every subgroup is no rmal . See if you can find such an exampl e of o rder
8. Such no nabel i an gro ups are cal l ed Hamiltonian, after the I rish mathemati -
cian W. R. Hami l to n (1805-1865). The desi red gro up of o rder 8 can be fo und
in thequaternions of Hami l to n, which we i ntro duce in Chapter 4, Section 1.
I n Exampl e 1, <p(x) = l o g
H )
x, and Ker cp = [x | l o g
1 0
x = 0} = {1}. I n
Exampl e 2, where G is abel ian, and cp(x) = x
2
,
Ker cp = [x G G|x
2
= e}
The kernel of the trivial ho mo mo rphi sm of Exampl e 3 is all of G. I n Exam-
pl e 4, Ker cp is the set of al l even integers. I n Exampl e 5, Ker cp =
{a G C | \a | = 1), which can be identified, from the po l ar form of a co mpl ex
number, as Ker cp = (co sx + i si nx|x real }. I n Exampl e 6, Ker cp =
{T
x
j, G G | b real }. I n Exampl e 7, Ker cp is the set of al l mul tipl es of n. I n Ex-
ampl es 8 and 9, the kernel s consists of e al one, for the maps are mo no mo r-
phi sms. I n Exampl e 10, we see that Ker cp = \2irm \ m any i nteger}.
Of course, al l the kernel s abo ve are no rmal subgro ups of thei r respec-
tive groups. We shoul d l o o k at so me no rmal subgro ups, intrinsical l y in G it-
sel f, wi tho ut reco urse to the kernel s of ho mo mo rphi sm. We go back to the
exampl es of Section 1.
1. I n Exampl e 7, H = {T
b
G G | a rati o nal }. If T
x y
G G, we l eave it to the
reader to check that T~], HT
xy
C H and so H < G.
2. I n Exampl e 9 the subgro up [i, g, g
2
, g
3
} < G. Here to o we l eave the
checking to the reader.
3. I n Exampl e 10 the subgro up H = {i, h, h
2
,..., h"~
1
} is no rmal in G. This
we al so l eave to the reader.
Sec . 5 Ho mo mo r p h i s ms an d No r ma l Su b g r o u p s 73
4. If G is any gro up, Z(G), the center of G, is a no rmal subgro up of G (see
Exampl e 11 of Section 3).
5. If G = S
3
, G has the el ements i, f, g, g
2
, fg, and gf, where f(x
1
) = x
2
,
f(x
2
) = x, , f(x
3
) = x
3
and g(x
x
) = x
2
, g(x
2
) = x
3
, g(x
3
) = X j . We cl aim
that the subgro up /V = [i, g, g
2
} <i S
3
. As we saw earl ier (or can co mpute
no w), fgr
1
= g-
1
= g
2
,fg
2
r
l
= g, (fg) gifgr
1
= fggg-'r
1
= M~
l
=
g
2
, and so on. So N <\ S
3
fol l ows.
The materi al in this section has been a rather rich diet. I t may no t seem
so, but the i deas presented, al tho ugh simpl e, are qui te subtl e. We reco m-
mend that the reader digest the co ncepts and resul ts tho ro ughl y before goi ng
on. One way of seeing ho w co mpl ete this digestion is, is to take a stab at
many of the al most infinite l ist of pro bl ems that fol l ow. The materi al of the
next section is even a richer diet, and even harder to digest. Avo i d a mathe-
matical sto machache l ater by assimil ating this section wel l .
PROBL EMS
Easi er Pro bl ems
1. Determi ne in each of the parts if the given mappi ng is a ho mo mo rphi sm.
If so, identify its kernel and whether o r no t the mappi ng is 1-1 or o nto .
(a) G = Z under +, G' = Z, cp(a) = [a] for a G Z.
(b) G gro up, cp: G > G defined by cp(a) = a
- 1
for a e G.
(c) G abel ian gro up, cp: G - G defined by cp (a) = cC
l
for a G G.
(d) G gro up of al l no nzero real numbers under mul tipl ication, G' =
{1, 1}, cp(r) = 1 if r is positive, cp(r) = - 1 if r is negative.
(e) G an abel ian gro up, n > 1 a,fixed i nteger, and cp: G - G defined by
cp(a) = a" for a G G.
2. Recal l that G G' means that G is i so mo rphi c to G'. Pro ve that for al l
gro ups Gx, G
2
, G
3
:
(a) G
x
= Gj .
(b) G
x
= G
2
impl ies that G
2
= G
t
.
(c) Gi = G
2
, G
2
= G
3
impl ies that G! G
3
.
3. L et G be any gro up and A (G) the set of al l 1-1 mappi ngs of G, as a set,
o nto itsel f. Define L
a
: G G by L
a
(x) =xa
_ 1
. Pro ve that:
(a) / G/ 1( G) .
(b) L
a
L
b
= L
f l h
.
(c) The mappi ng </>: G > A(G) defined by ip(a) = L
a
is a mo no mo r-
phi sm of G into A (G).
4. I n Pro bl em 3 pro ve that for al l a, b G G, T
a
L
b
= L
b
T
a
, where T
a
is de-
fined as in Exampl e 8.
5. I n Pro bl em 4, sho w that if V G A(G) is such that T
t
,V = VT
a
for al l
a G G, then V = L
b
for so me b G G. (Hi nt: Acti ng o n e G G, find o ut
what b shoul d be.)
6. Pro ve that if cp : G - G' is a ho mo mo rphi sm, then cp (G), the i mage of G,
is a subgro up of G'.
7. Show that <p: G >G'', where <p is a ho mo mo rphi sm, is a mo no mo rphi sm
if and onl y if Ker cp = (e).
8. Fi nd an i somorphi sm of G, the gro up of al l real numbers under +, onto
G', the gro up of al l positive real numbers under mul tipl ication.
9. Verify that if G is the gro up in Exampl e 6 of Section 1, and 77 =
{T
b
G G | a rati o nal ), then 77 O G.
10. Verify that in Exampl e 9 of Section 1, the set 77 = {;', g, g
2
, g
3
} is a no r-
mal subgro up of G, the di hedral gro up of o rder 8.
11. Verify that in Exampl e 10 of Section 1, the subgro up
H= [i,h,h
2
, ...,h"~
[
}
is no rmal in G.
12. Pro ve that if Z(G) is the center of G, then Z(G) <1 G.
13. If G is a finite abel i an gro up of o rder n and cp: G >G is defined by
(r?) = o'" for al l A G G, find the necessary and sufficient co ndi ti o n that
cp be an i so mo rphi sm of G o nto itsel f.
14. If G is abel ian and cp: G >G' is a ho mo mo rphi sm of G onto G', pro ve
that G' is abel ian.
15. If G is any gro up, TV <3 G, and cp: G -> G' a ho mo mo rphi sm of G onto
G', pro ve that the i mage, cp(N), of TV is a no rmal subgro up of G'.
16. If TV < G and M < G and MTV = [ran | m G M, n G TV), pro ve that M/V is
a subgro up of G and that MN O G.
17. If M < G, N < G, pro ve that M n TV < G.
18. If 77 is any subgro up of G and TV = D
a e G
a~
l
Ha, pro ve that N <\ G.
19. If 77 is a subgro up of G, l et /V(/7) be defined by the rel ati o n N(H) =
[fl G G | a~
l
Ha = 77). Pro ve that:
(a) 7V(77) is a subgro up of G and 7V(77) D 77.
(b) 77 < N(H).
(c) If 7C is a subgro up of G such that 77 < K, then K C 7V(77). [So 7V(77)
is the l argest subgro up of G in which 77 is normal .]
20. If M-< G, TV < G, and M n TV = (e), show that for ra G M, n G TV, ran = nra..
Sec. 5
Ho mo mo rphi sms and Normal Subgro ups 75
21. Let S be any set having mo re than two el ements and A(S) the set of
al l 1-1 mappings of S o nto itsel f. If s G S, we define H(s) =
{ / G A(S) | f(s) = s}. Prove that H(s) cannot be a no rmal subgroup of A(S).
22. Let G = S
3
, the symmetri c gro up of degree 3 and l et H = [i, / }, where
f(.X
{
) = X
2
, f(x
2
) = X j , f(x
3
) = x
3
.
(a) Wri te do wn al l the l eft cosets of If in G.
(b) Wri te do wn al l the right cosets of H in G.
(c) I s every l eft coset of H a right coset of HI
23. L et G be a group such that al l subgro ups of G are no rmal in G. If
a, b G G, pro ve that ba = a'b for so me;'.
24. If G
l 5
G
2
are two groups, l et G = Gj X G
2
, the Cartesi an pro duct of G
l 5
G
2
[i.e., G is the set of al l o rdered pairs (a, b) where a G G
l 5
b G G
2
] .
Defi ne a pro duct in G by (a
1 ;
b
t
)(a
2
, b
2
) = (a
t
a
2
, b
x
b
2
).
(a) Pro ve that G is a gro up.
(b) Show that there is a mo no mo rphi sm <pj of G
x
into G such that
cp
1
(G
1
) <1 G, given by cp
x
(aA = 62) . where e
2
is the identity el e-
ment of G
2
.
(c) Fi nd the simil ar mo no mo rphi sm cp
2
of G
2
into G.
(d) Using the mappi ngs <p
u
cp
2
of Parts (b) and (c), prove that
<P\(G
x
)(p
2
(G
2
) = G and <Pi(Gj) D <p
2
(G
2
) is the identity el ement of G.
(e) Pro ve that G
x
x G
2
G
2
X G
x
.
25. Let G be a group and l et W = G X G as defined in Pro bl em 24. Prove that:
(a) The mappi ng cp: G defined by cp (a) = (a, a) is a mo no mo rphi sm
of G into W.
(b) The i mage <p(G) in W [i.e., {(a, A) | a G G}] is no rmal in W if and onl y
if G is abel ian.
*26. If G is a gro up and a G G, define c r
f l
: G -> G by cr (<?)
= a
g
a l
We saw
in Exampl e 9 of this section that a
a
is an i so mo rphi sm of G o nto itsel f, so
cr G A(G), the group of al l 1-1 mappi ngs of G (as a set) o nto itsel f. De-
fine iff: G > A (G) by ijj(a) = a
a
for al l a G G. Pro ve that:
(a) i/>is a ho mo mo rphi sm of G i nto A (G).
(b) Ker <// = Z(G), the center of G.
27. If 9 is an auto mo rphi sm of G and TV O G, pro ve that 6(N) < G.
28. L et 9, ip be auto mo rphi sms of G, and l et 9ip be the pro duct of 9 and (// as
mappi ngs on G. Pro ve that 0i/y is an auto mo rphi sm of G, and that 0"
1
is
an auto mo rphi sm of G, so that the set of al l auto mo rphi sms of G is itsel f
a gro up.
*29. A subgro up 7 of a gro up Wis cal l ed characteristic if <p(T) C 7 for al l au-
to mo rphi sms, <p, of W. Pro ve that:
(a) M characteristic in G impl ies that M <\ G.
(b) M, TV characteristic in G impl ies that MTV is characteristic in G.
(c) A no rmal subgro up of a gro up need not be characteristic. (This is
quite hard; you must find an exampl e of a gro up G and a no ncharac-
teristic no rmal subgro up.)
30. Suppose that \G\ = pm, where p\m and p is a pri me. If 77 is a no rmal
subgro up of o rder p in G, pro ve that 77 is characteristic.
31. Suppo se that G is an abel i an gro up of o rderp " m wher ep \ m is a pri me.
If 77 is a subgro up of G of o rder p", pro ve that 77 is a characteri sti c sub-
gro up of G.
32. Do Pro bl em 31 even if G is no t abel ian if you happen to kno w that for
so me reaso n or o ther 77 O G.
33. Suppose that TV < G and M C TV is a characteristic subgro up of TV. Pro ve
that M < G. (It is not true that if M < TV and TV < G, then M must be
no rmal in G. See Pro bl em 50.)
34. Let G be a gro up, s&(G) the gro up of al l auto mo rphi sms of G. (See Pro b-
l em 28.) Let 7(G) = [a
a
\ a G G}, where cr
a
is as defined in Pro bl em 26.
Pro ve that 7(G) < si(G).
35. Show that Z(G), the center of G, is a characteristic subgro up of G.
36. If TV < G and 77 is a subgro up of G, show that 77 n TV < 77.
Harder Pro bl ems
37. If G is a no nabel i an gro up of o rder 6, pro ve that G S
3
.
38. Let G be a gro up and 77 a subgro up of G. Let S = { 77A | a G} be the
set of al l right cosets of 77 in G. Define, for b G G, T
b
: S ->5 by 7
6
(77fl )
= 77fl /3"
1
.
(a) Pro ve that T
b
T
c
= 7
6 c
for al l b, c G G [therefore the mappi ng
</>: G A (5) defined by i/>(/3) =7
A
is a ho mo mo rphi sm].
(b) Descri be Ker i/>, the kernel of i/> : G >A(S).
(c) Show that Ker i/> is the l argest no rmal subgro up of G l ying in 77
[l argest in the sense that if TV < G and TV C 77, then TV C Ker i/>] .
39. Use the resul t of Pro bl em 38 to redo Pro bl em 37.
Recal l that if 77 is a subgro up of G, then theindex of 77 in G, /
G
(77), is
the number of distinct ri ght cosets of 77 and G (if this number is finite).
40. If G is a finite gro up, 77 a subgro up of G such that n f i
G
(H)l where n -
| G j. pro ve that there is a no rmal subgro up TV + (e) of G co ntai ned in 77.
Sec , 6 Fac t or Gr o u p s 77
41. Suppo se that you kno w that a gro up G of o rder 21 contains an el ement a
of o rder 7. Pro ve that A = (a), the subgro up generated by a, is no rmal in
G. (Hi nt: Use the resul t of Pro bl em 40.)
42. Suppo se that you kno w that a gro up G of o rder 36 has a subgro up H or
o rder 9. Pro ve that either H <3 G o r there exists a subgro up N <1 G,
NC H, and |/V| = 3.
43. Pro ve that a group of o rder 9 must be abel i an.
44. Pro ve that a group of order p
2
, p a pri me, has a normal subgroup of
o rder p.
45. Usi ng the resul t of Pro bl em 44, pro ve that a gro up of o rder p
2
, p a
pri me, must be abel ian.
46. Let G be a gro up of o rder 15; show that there is an el ement a = e in G
such that a
3
= e and an el ement ft e such that ft
5
= e.
47. I n Pro bl em 46, show that bo th subgro ups A = \e, a, a
2
} and Z? =
[e, 6, ft
2
, ft
3
, ft
4
} are no rmal in G.
48. Fro m the resul t of Pro bl em 47, show that any gro up of o rder 15 is cycl ic.
Very Hard Pro bl ems
49. Let G be a gro up, H a subgro up of G such that i
G
(H) is finite. Prove that
there is a subgro up N d H,N <\ G such that i
c
(TV) is finite.
50. Construct a gro up G such that G has a no rmal subgro up TV, and N has a
no rmal subgro up M (i.e., N < G, M < TV), yet M is no t no rmal in G.
51. Let G be a finite gro up, cp an auto mo rphi sm of G such that <p
2
is the i den-
tity auto mo rphi sm of G. Suppo se that cp(x) = x impl ies that x = e. Pro ve
that G is abel ian and cp(a) = a~
x
for al l a G G.
52. Let G be a finite gro up and cp an auto mo rphi sm of G such that cp(x)
x~
l
for /7?ore than three-fourths of the el ements of G. Prove that <p(y) =
y~
l
for a// y G G, and so G is abel i an.
6. FACT OR GROUPS
Let G be a gro up and TV a no rmal subgro up of G. I n provi ng Lagrange's The-
o rem we used, for an arbi trary subgro up H, the equi val ence rel ation a ~ft if
ab~
1
G H. Let's try this o ut when TV is no rmal and see if we can say a l ittl e
mo re than o ne coul d say for just any ol d subgro up.
So, l et a ~ ft if aft"
1
G N and l et [a] = {x G G | x ~ a}. As we saw
earl ier, [a] = Na, the right coset of N in G co ntai ni ng a. Recal l that in
l ooking at Z we defined for it an o perati o n + via [a] + [ft] = [a + ft]. Why
no t try so methi ng simil ar for an arbi trary gro up G and a no rmal subgro up
TV of G?
So l et M = {[A] | a G G], where [a] = [x G G | G TV} = TVA. We de-
fine a pro duct in M via [][&] = [ab]. We shal l soon show that M is a gro up
under this pro duct. But first and foremost we must show that this product in
M is well-defined. I n o ther wo rds, we must show that if [a] = [a'] and [b] =
[b'], then [ab] = [a'b'], for this wo ul d show that [a][b] = [ab] = [a'b'] =
['][>']; equival entl y, that this product of classes does not depend on the par-
ticular representatives we use for the classes.
Therefo re l et us suppo se that [a] = [a
1
] and [b] = [>']. Fro m the defi-
nition of o ur equi val ence we have that a' = na, where n G TV. Simil arl y,
b' = mb, where m G TV. Thus a'b' = namb = n(ama~
l
)ab; since TV < G,
anuC
1
is in TV, so n(ama~
l
) is al so in TV. So if we l et n
x
= n(ama~
l
), then
rt] G TV and a'b' = n
x
ab. But this tel l s us that a'b' G Nab, so that
a'b' ~ ab, from which we have that [a'b
1
] = [ab], the exact thi ng we re-
qui red to ensure that o ur pro duct in M was wel l -defined.
Thus M is no w endo wed with a wel l -defined pro duct [fl ][6] = [ab].
We no w verify the gro up axi o ms for M. Cl o sure we have from the very def-
inition of this pro duct. If [a], [b], and [c] are in M, then [A]([/3][C]) =
[fl ][/3c] = [a(be)] = [(ab)c] (since the pro duct in G is associ ati ve) =
[A/J ][C] = ([fl ][6] ) [ c ] . Therefo re, the associative l aw has been establ i shed
for the pro duct in M. What abo ut a unit el ement? Why no t try the o bvi o us
choi ce, namel y [e]l We i mmedi atel y see that [a][e] = [ae] = [a] and
[e][a] = [ea] = [a], so [e] do es act as the unit el ement for M. Fi nal l y, what
abo ut i nverses? Here, to o , the o bvi o us choi ce is the co rrect o ne. If a G G,
then [A ][A
- 1
] = [ f l i j T
1
] = [e], hence [fl
-
'] acts as the i nverse of [a] rel ati ve
to the pro duct we have defi ned in M.
We want to give M a name, and better stil l , a symbol that indicates its
dependence on G and TV. The symbol we use for M is GIN (read "G o ver TV or
G mo d TV") and GIN is cal l ed the factor group o r quotient group of G by TV.
What we have sho wn is the very i mpo rtant
Theo rem 2.6.1. If TV < G and
GIN = {[A] | A G G) = {TVfl | A G G],
then G/TVis a gro up rel ative to the o perati o n [a][b] = [ab].
One observati on must i mmedi atel y be made, namel y
Theo rem 2.6.2. If TV O G, then there is a ho mo mo rphi sm </>of G o nto
GIN such that Ker i/>, the kernel of if, is TV.
Sec . 6 Fac t or Gr o u p s 79
Proof. The mo st natural mappi ng from G to GIN is the o ne that
do es the trick. Defi ne i/>: G - GIN by i/y (a) = [a]. Our pro duct as defi ned
in GIN makes of f a ho mo mo rphi sm, for ip(ab) = [ab] = [a][b] =
\fj(a)4>(b). Since every el ement X G GIN is of the fo rm X = [b] = i//(/3)for
so me b G G, i/>is o nto . Fi nal l y, what is the kernel , Ker ifi, of i/>? By defini-
tion, Ker i/>= [a G G | i/>(a) = 7?}, where is the uni t el ement of G/7V. But
what is El No thi ng o ther than E = [e] = TVe = TV, and a G Ker i/>if and
onl y if = TV = i/>(a) = TVA. But TV<7 = TV tel l s us that a = ea E. Na = N, so
we see that Ker which is easywe l eave to the
reader. So Ker i/>= TV.
Theo rem 2.6.2 substanti ates the remark we made in the precedi ng sec-
tion that every no rmal subgro up TV of G is the kernel of so me ho mo mo r-
phism of G o nto so me gro up. The "so me ho mo mo rphi sm" is the i// defined
above and the "so me gro up" is GIN.
This construction of the factor gro up G by TV is possibl y the singl e mo st
i mpo rtant constructi on in gro up theo ry. I n o ther al gebraic systems we shal l
have anal o go us constructions, as we shal l see l ater.
One mi ght ask: Where in this who l e affair did the no rmal i ty of TV in G
enter? Why no t do the same thing for any subgro up H of G? So l et's try and
see what happens. As before, we define
W= {[a]|f l GG} = [Ha\aEG]
where the equi val ence a ~ b is defined by ab'
1
G H. We try to i ntro duce a
pro duct in W as we did for GIN by defining [a][b] = [ab]. Is this pro duct wel l
defined? If h G H, then [hb] = [b], so for the pro duct to be wel l defined, we
woul d need that [a][b] = [a][hb], that is, [ab] = [ahb]. This gives us that Hab
Hahb, and so Ha = Hah; this impl ies that H = Haha"
1
, whence aha~
l
G H.
That is, for al l a G G and al l h G H, aha^
1
must be in H; in o ther wo rds, H
must be no rmal in G. So we see that in order for the product defined in W to
be well-defined, H must be a normal subgroup of G.
We view this matter of the quo ti ent gro up in a sl ightl y different way. If
A, B are subsets of G, l et AB = [ab \ a G A, b G B}. If H is a subgro up of G,
then HH C H is ano ther way of saying that H is cl osed under the pro duct
of G.
Let GIN = {TVa | a G G] be the set of al l right cosets of the no rmal sub-
group TV in G. Usi ng the pro duct of subsets of G as defined abo ve, what is
(Na)(Nb)7 By definition, (Na)(Nb) consists of al l el ements of the form
(na)(mb), where n, m G TV, and so
(na)(mb) = (nama~
l
){ab) = nâb,
80 Gr o u p s
Ch . 2
where n
l
= noma
1
is in TV, since /Vis no rmal . Thus (Na)(Nb) C Nab. On the
o ther hand, if n G N, then
n{ab) = (na)(eb) G (Na)(Nb),
so that Nab C (Na)(Nb). I n short, we have shown that the pro ductas sub-
sets of Gof Na and Nb is given by the formul a (Na)(Nb) = Nab. Al l the
o ther gro up axioms for GIN, as defined here, are no w readi l y verified from
this pro duct formul a.
Ano ther way of seeing that (Na)(Nb) = Nab is to no te that by the no r-
mal ity of TV, aN = Na, hence (Na)(Nb) = N(aN)b = N(Na)b = NNab =
Nab, since NN = TV (because TV is a subgro up of G).
Ho wever we view GINas equi val ence cl asses or as a set of certai n
subsets of Gwe do get a gro up who se structure is i nti matel y ti ed to that of
G, via the natural ho mo mo rphi sm \poiG o nto GIN.
We shal l see very so o n ho w we co mbi ne i nducti on and the structure of
GIN to get i nformati on abo ut G.
When G is a finite gro up and TV < G, then the number of right cosets of
TV in G, i
G
(N), is gi venas the proof of Lagrange's Theo rem sho wedby
'G(n) = | G|/|TV|. But this is the o rder of GIN, which is the set of al l the right
cosets of TV in G. Thus | G/TV| = | G | /1 TV |. We state this mo re formal l y as
Theo rem 2 .63. If G is a finite group and TV < G, then |G/TV| =
|G|/|TV|.
As an appl ication of what we have been tal king abo ut here, we shal l
pro ve a special case of a theo rem that we shal l pro ve in its ful l general i ty
l ater. The proof we givefor the abel ian casei s no t a parti cul arl y go o d
o ne, but it il l ustrates qui te cl earl y a general techni que, that of pul l ing back
i nformati on abo ut GIN to get i nformati on abo ut G itsel f.
The theorem we are about to prove is due to the great French mathematician
A. L. Cauchy (1789-1857), whose most basic contributions were in compl ex
variabl e theory.
Theo rem 2.6.4 (Cauchy). If G is a finite abel ian gro up of o rder |G|
and p is a pri me that divides \ G\, then G has an el ement of o rder p.
Proof. Befo re getting invol ved with the proof, we po i nt o ut to the
reader that the theo rem is true for any finite gro up. We shal l pro ve it in the
general case l ater, with a pro o f that wil l be much mo re beautiful than the o ne
we are abo ut to give for the special , abel ian case.
We pro ceed by induction on | G |. What does this mean precisel y? We shal l
Sec. 6 Fac t or Gr o u p s 81
assume the theo rem to be true for al l abel ian groups of order l ess than |G| and
show that this forces the theorem to be true for G. If | G | = 1, there is no such p
and the theo rem is vacuousl y true. So we have a starting point for our induction.
Suppo se that there is a subgro up (e) #N =h G. Since |7V| < \ G\, if p di-
vides | N |, by o ur i nducti on hypothesi s there woul d be an el ement of o rder p
in N, hence in G, and we woul d be do ne. So we may suppo se that p||7V|.
Since G is abel i an, every subgro up is no rmal , so we can form GIN. Because p
divides \G\ and p\\N\, and because \GIN\ = |G|/|7V|, we have that p
divides \GIN\. The gro up GIN is abel ian, since G is (Prove!) and since
N * (e), \N\ > 1, so \G/N\ = \G\I\N\ < |G|. Thus, again by induction, there
exists an el ement in GIN of o rder p. I n o ther words, there exists an a G G
such that [a]
p
= [e], but [a] + [e]. Thi s transl ates to a
p
E N, a < N. So if
m = \N\, then (a
p
)
m
= e. So (a'")
p
= e. If we coul d show that b = a'" e,
then b woul d be the requi red el ement of o rder p in G. But if a'" = e, then
[a]'" = [e], and since [a] has o rder p,p \ m (see Pro bl em 31 of Section 4). But,
by assumpti o n,p\ m = \N\. So we are done ifG has a nontrivialsubgroup.
Hut if G has no nontrivial subgro ups, it must be cycl ic of pri me o rder.
(See Pro bl em 16 of Section 3, which yo u shoul d be abl e to handl e mo re eas-
il y now.) What is this "pri me o rder"? Because p divides |G|, we must have
j G| = p. But then any el ement a = e G G satisfies a
p
= e and is of o rder p.
This co mpl etes the i nducti on, and so pro ves the theo rem.
We shal l have o ther appl ications of this ki nd of gro up-theo reti c argu-
ment in the pro bl ems.
The no ti o n of a factor gro up is a very subtl e o ne, and of the greatest
i mpo rtance in the subject. The fo rmati o n of a new set from an ol d o ne by
using as el ements of this new set subsets of the ol d o ne is strange to the neo -
phyte seeing this ki nd of constructi on for the first ti me. So it is worthwhi l e
l ooking at this whol e matter from a vari ety of poi nts of view. We consider
GIN from ano ther angl e now.
What are we doing when we fo rm G/7V? Sure, we are l o o ki ng at equi va-
l ences cl asses defined via TV. Let's l o o k at it ano ther way. What we are do i ng
is identifying two el ements in G if they satisfy the rel ati on ab^
1
G N. I n a
sense we are bl otting o ut N. So al tho ugh GIN is not a subgro up of G, we can
l ook at it as G, with TV bl o tted out, and two el ements as equal if they are
equal "up to TV."
For instance, in forming Z/N, where Z is the gro up of integers and TV is
the set of al l mul tipl es of 5 in Z, what we are doi ng is identifying 1 with 6, 11,
16, 4, 9, and so on, and we are identifying al l mul tipl es of 5 with 0. The
nice thing abo ut al l this is that this identification jibes with addi ti on i n Z
when we go over to Z/N.
Let's l o o k at a few exampl es from this po i nt of view.
1. Let G = {T
aJ)
\a * 0, b real ) (Exampl e 6 of Section 1). L et TV =
[T
lb
\b real ) C G; we saw that TV <1 G, so it makes sense to tal k abo ut G/N.
No w T
ab
and T
a0
are in the same l eft coset of TV in G, so in G/N we are get-
ting an el ement by identifying T
ab
with T
aQ
. The l atter el ement j ust depends
o n a. Mo reo ver, the T
aJ>
mul tipl y according to T
ub
T
cd
= T
a c a d + b
and if we
identify T
a
with T
ai0
, T
c d
with T
c 0
, then their product, which is T
nc a d + b
, is
identified with T
ac<0
. So in GIN mul tipl ication is l ike that of the gro up of
no nzero real numbers under mul tipl ication, and in so me sense (which wil l be
made mo re precise in the next section) GIN can be identified wi th this gro up
of real numbers.
2. Let G be the gro up of real numbers under +and l et Z be the gro up of in-
tegers under +. Since G is abel ian, Z <G, and so we can tal k abo ut G/Z.
What does G/Z real l y l o o k l ike? In forming GIT, we are identifying any two
real numbers that differ by an integer. So 0 is identified with 1, 2, 3, . ..
and 1, 2, 3, . . . ; | is identified with , , , , . . . . Every real number a
thus has a mate, a, where 0 <a < 1. So, in G/Z, the who l e real l ine has been
co mpressed into the uni t i nterval [0, 1]. But a l ittl e mo re is true, for we have
al so identified the end po i nts of this unit interval . So we are bendi ng the unit
interval aro und so that its two end points to uch and beco me o ne. What do
we get this way? A circl e, of co urse! So G/Z is l ike a circl e, in a sense that
can be made precise, and this circl e is a gro up with an appro pri ate pro duct.
3. Let G be the gro up of no nzero co mpl ex numbers and l et TV =
[a G | \a | =1} which is the uni t circl e i n the co mpl ex pl ane. Then TV is a
subgro up of G and is no rmal since G is abel ian. I n going to GIN we are de-
cl aring that any compl ex number of absol ute val ue 1 wil l be identified with the
real number 1. No w any a G, in its pol ar form, can be wri tten as a =
r(co s 8 + i sin 8), where r = \a\, and [cos 8 + i sin 0| =1. I n identifying
cos 8 + i sin 8 with 1, we are identifying a with r. So in passing to GIN every
el ement is being identified with a positive real number, and this identification
jibes with the products in G and in the gro up of positive real numbers, since
\ab\ = \a \ \b\. So GIN is in a very real sense (no pun i ntended) the gro up of
positive real numbers under mul tipl ication.
PROBL EMS
1. If G is the gro up of al l no nzero real numbers under mul ti pl i cati on and TV
is the subgro up of al l positive real numbers, wri te o ut GIN by exhibiting
the cosets of TV in G, and construct the mul tipl ication in GIN.
Sec . 6
Fac t or Gr o u p s 83
2. If G is the gro up of no nzero real numbers under mul tipl ication and
/V = {1, 1}, show how you can "identify" GIN as the group of al l
positive real numbers under mul tipl ication. What are the cosets of
/Vin Gl
3. If G is a gro up and TV <l G, show that if M is a subgro up of GIN and M =
{a G G\Na G M], then M is a subgro up of G, and MDJ V.
4. If M in Pro bl em 3 is normal in GIN, show that the A/ defined is no rmal
in G.
5. I n Pro bl em 3, show that MIN must equal M.
6. Argui ng as in the Exampl e 2, where we identified G/Z as a circl e, where
G is the gro up of real s under + and Z i ntegers, consider the fol l owing:
l et G = [(a, b)\a,b real }, where + in G is defined by (a, b) + (c, d) =
(a + c, b + d) (so G is the pl ane), and l et N = {(a, b) G G | a, b are i nte-
gers}. Show that GIN can be identified as a to rus (do nut), and so we can
define a pro duct o n the do nut so that it beco mes a gro up. Here, you may
think of a torus as the Cartesi an pro duct of two circl es.
7. If G is a cycl ic gro up and TV is a subgro up of G, sho w that GIN is a cycl ic
gro up.
8. If G is an abel ian gro up and TV is a subgro up of G, show that GIN is an
abel ian gro up.
9. Do Pro bl ems 7 and 8 by observing that GIN is a ho mo mo rphi c i mage
of G.
10. Let G be an abel ian gro up of o rder p'^P? '"' PV
!
> where p
ly
p
z
, .. . ,p
k
are distinct pri me numbers. Sho w that G has subgroups S
lt
S
2
,... , S
k
of
o rders p'(\ p
2
2
,.. ., p
a
k
", respectivel y. (Hi nt: Use Cauchy's Theo rem and
pass to a factor group.) This resul t, which actual l y hol ds for al l finite
groups, is a famous resul t in gro up theo ry kno wn as Sylow's Theorem.
We pro ve it in Section 11.
11. If G is a gro up and Z(G) the center of G, show that if G/Z(G) is cycl ic,
then G is abel ian.
12. If G is a gro up and TV <] G is such that GIN is abel ian, prove that
aba'
1
b~
L
G N for al l a,b G G.
13. If G is a gro up and N <1 G is such that
aba^b'
1
G TV
for al l a, b . G, pro ve that G/7V is abel ian.
14. If G is an abel ian group of o rder P\p
2
- p
k
, where p
ly
p
2
,... ,p
k
are
distinct pri mes, pro ve that G is cycl ic. (See Pro bl em 15.)
15. If G is an abel ian gro up and if G has an el ement of o rder m and o ne of
o rder n, where m and n are rel ativel y pri me, pro ve that G has an el ement
of o rder mn.
16. Let G be an abel ian gro up of o rder p"m, where p is a pri me and p\m.
Let P = [a G G \ a
pk
= e for so me k dependi ng o n a}. Pro ve that:
(a) P is a subgro up of G.
(b) GIP has no el ements of o rder p.
(c) |P| = p".
17. Let G be an abel ian gro up of o rder mn, where m and n are rel ativel y
pri me. Let M ={ E G a"' =e). Pro ve that:
(a) M is a subgro up of G.
(b) G/M has no el ement, x, o ther than the identity el ement, such that
x'" = unit el ement of GIM.
18. Let G be an abel i an gro up (possibl y infinite) and l et the set T =
{ G G | =e, m > 1 dependi ng on A}. Pro ve that:
(a) T is a subgro up of G.
(b) GIT has no el emento ther than its identity el emento f finite o rder.
7. THE HOMOMORPHI SM THEOREMS
Let G be a gro up and cp a ho mo mo rphi sm of G onto G'. If K is the kernel of
cp, then if is a no rmal subgro up of G, hence we can fo rm G/K. I t is fairl y nat-
ural to expect that there shoul d be a very cl ose rel ati onshi p between G' and
GIK. The First Homomorphism Theorem, which we are abo ut to pro ve,
spel l s out this rel ati o nshi p in exact detail .
But first l et's l ook back at so me of the exampl es of factor gro ups in
Section 6 to see expl icitl y what the rel ati onshi p menti o ned abo ve mi ght be.
1. Let G = [T
a b
| A = 0, b real } and l et G' be the gro up of no nzero real s
under mul tipl ication. Fro m the pro duct rul e of these T's, namel y
T
a
.hT
c d
= T
acad
b
, we determi ned that the mappi ng cp: G > G' defined
by cp(T
aJj
) = a is a ho mo mo rphi sm of G o nto G' with kernel K =
{T
t h
| b real ). On the o ther hand, in Exampl e 1 of Section 6 we saw that
GIK = [KT
a0
\a + Q real }. Since
(*T
fl
.
0
)(tfT,,
0
) = KT
ax
,
0
the mappi ng of GIK o nto G', which sends each KT
n
o nto a, is readi l y
seen to be an isomorphism of GIK o nto G'. Therefo re, GIK G'.
2. I n Exampl e 3, G was the group of no nzero compl ex numbers under mul tipl i-
cation and G' the group of al l positive real numbers under mul tipl ication.
Sec. 7
Th e Ho m o m o r p h i s m Th e o r e ms 85
Let cp:G > G' defined by cp(a) = \a\ for a G G. Then, since \ab\ =
|ff | | b|, <p is a ho mo mo rphi sm of G o nto G' (can you see why it is o nto ?).
Thus the kernel K of cp is precisel y K = {a G G | \a| = 1}. But we have al -
ready seen that if |fl | = 1, then a is of the fo rm cos 0 + z' sin 0. So the set
K = {cos 0 + z sin 0 | 0 < 0 < 2rr\. If A is any co mpl ex number, then
a = /'(cos 0 + z sin 0), where r = |fl |, is the pol ar form of a. Thus Ka =
/(/(cos 0 + z sin 0) = /((co s 0 + /' sin 0)r = i (/\ since /((co s 0 + /' sin0) = /(
because cos 0 + z sin 0 K. So GIK, who se el ements are the cosets Ka,
from this discussion, has al l its el ements of the form Kr, where r > 0. The
mappi ng of GIK o nto G' defined by sendi ng Kr o nto ;then defines an iso-
morphism of GIK o nto G'. So, here, to o , GIK G'.
Wi th this l ittl e experi ence behi nd us we are ready to make the j ump the
whol e way, namel y, to
Theo rem 2.7.1 (First Ho mo mo rphi sm Theo rem). Let cp be a ho mo -
mo rphi sm of G onto G' with kernel K. Then G' GIK, the i so mo rphi sm be-
tween these bei ng effected by the map
i / / : GIK G'
defined by ip(Ka) = <p(a).
Proof. The best way to show that G//C and G' are i somorphi c is to ex-
hibit expl icitl y an i so mo rphi sm of GIK o nto G'. The statement of the theo -
rem suggests what such an i so mo rphi sm might be.
So define i/>: GIK >G' by *jj{Ka) = cp(a) for a G G. As usual , our first
task is to show that i/> is wel l defined, that is, to show that if Ka = Kb, then
ip(Ka) = ij/(Kb). This boil s do wn to showing that if Ka = Kb, then cp(a) =
cp(b). But if Ka = Kb, then a = kb for so me /c G /(, hence cp(a) = cp(kb) =
cp(k)cp(b). Since A: G /(, the kernel of cp, then <p(/c) = e', the identity el ement
of G', so we get cp(a) = 9(6). Thi s shows that the mappi ng i/>is wel l defined.
Because maps G/Z( o nto G'.
Is </>1-1? Suppose that </>(#) = </>(^): then <p(a) = ip(Ka) =
i/z(/<:6) =cp(b). Therefo re, e' = cp{a)cp{b)~
x
= ( p ^f i r
1
) = cpiab'
1
). Be-
cause ab~
]
is thus in the kernel of cpwhich is Kwe have ab~
l
G /(. Thi s
impl ies that Ka = Kb. I n this way i/> is seen to be 1-1.
Final l y, is i/> a ho mo mo rphi sm? We check: tp((Ka)(Kb)) = 4i(Kab) =
cp(ab) = cp{a)cp{b) = ip(Ka)ij/(Kb), using that cp is a ho mo mo rphi sm and
that (Ka)(Kb) = Kab. Co nsequentl y, </ V is a ho mo mo rphi sm of GIK o nto G',
and Theo rem 2.7.1 is pro ved.
Havi ng tal ked abo ut the First Ho mo mo rphi sm Theo rem suggests that
there are o thers. The next resul t, however, is an extensi o n of the Fi rst
Ho mo mo rphi sm Theo rem, and is tradi ti onal l y cal l ed the Correspondence
Theorem. I n the context of the theo rem abo ve, it exhibits a 1-1 co rrespo n-
dence between subgro ups of G
1
and tho se subgro ups of G that co ntai n K.
T heorem 2.7.2 (Correspondence T heorem). Let the map <p: G G' be
a ho mo mo rphi sm of G o nto G' with kernel K. If H' is a subgroup of G' and if
H = [a Gj<p(a) / / ' },
then H is a subgro up of G, 77 D K, and HIK = 77. Final l y, if H' <\G\ then
H<G.
Proof. We first verify that the 77 abo ve is a subgro up of G. I t is no t
empty, since e G 77. If a, b G 77, then <p(a), <p(>) G 77', hence cp (ab) =
cp(a)cp(b) G 77', since 77' is a subgro up of G'; this puts ab in 77, so 77 is
cl osed. Further, if a G 77, then cp(a) G 77', hence cp(a~
l
) = cp(a)~
l
is in 77',
again since 77' is a subgro up of G', whence a~
l
G 77. Therefo re, 77 is a sub-
gro up of G.
Because cp(K) = {e'\ C 77', where e' is the unit el ement of G', we have
that K C 77. Since K <\ G and 7C C 77, it fol l ows that 7C < 77. The mappi ng <p
restricted to 77 defines a ho mo mo rphi sm of 77 o nto 77' with kernel K. By the
First Ho mo mo rphi sm Theo rem we get HIK 77'.
Final l y, if 77' < G' and if a G G, then cp(
a
y
1
H'cp(a) C 77', so
cp(
a
-
1
)H'cp(a) C 77'. This tel l s us that cp(
a
-
1
Ha) C 77', so a
- 1
77a C 77. Thi s
pro ves the normal i ty of 77 in G.
I t is wo rth noti ng that if K is any no rmal subgro up of G, and cp is the
natural ho mo mo rphi sm of G o nto G/X, then the theo rem gives us a 1-1 cor-
respo ndence between al l subgro ups 77' of G/K and tho se subgro ups of G
that contain K. Mo reo ver, this co rrespo ndence preserves no rmal i ty in the
sense that 77' is no rmal in G/K if and onl y if 77 is no rmal in G. (See Pro bl em
7, as wel l as the l ast concl usion of the theo rem.)
We no w state the Second Homomorphism Theorem, l eaving its pro o f
to the reader in Pro bl em 5.
T heorem 2.7.3 (Second Homomorphism T heorem). Let / / be a sub-
gro up of a gro up G and TV a no rmal subgro up of G. Then 77TV =
{hn | h G 77, n G TV} is a subgro up of G, 77 Pi TV is a no rmal subgro up of 77,
and 77/(77 fl TV) (HN)/N.
Sec . 7
Th e Ho m o m o r p h i s m Th e o r e ms 87
Final l y, we go on to the Third Homomorphism Theorem, which tel l s us
a l ittl e mo re abo ut the rel ati o nshi p between TV and N' when N' <3 G'.
Theo rem 2.7.4 (Thi rd Ho mo mo rphi sm Theo rem). If the map
cp : G G' is a ho mo mo rphi sm of G o nto G' with kernel /(then, if TV' <1 G'
and N = [a E G\ cp(a) E TV'}, we co ncl ude that G//V G7/V'. Equi val entl y,
GIN=*(G/K)/(NIK).
Proof. Define the mappi ng I/>: G - G'lN' by I/)(a) =N' cp(a) for every
a E G. Since maps G o nto G'lN'.
Furthermo re, ifi is a ho mo mo rphi sm of G o nto G'lN', for tp(ab) =
N'q>(ab) = N'q>(a)cp(b) = (N'<p(a))(N'(p(b)) = ip(a)ip(b), since TV' < G'.
What is the kernel , Af, of I/>? If a G M, then if/(a) is the uni t el ement of
G'lN', that is, tf/(a) = N'. On the o ther hand, by the definition of ip, tp(a) =
N'cp(a). Because N'cp(a) = N' we must have cp(a) E TV'; but this puts a in N,
by the very definition of N. Thus MEN. That TV C Af is easy and is l eft to
the reader. Therefo re, M = N, so ip is a ho mo mo rphi sm of G onto G'lN'
with kernel TV, whence, by the First Ho mo mo rphi sm Theo rem, GIN G'lN'.
Final l y, again by Theo rems 2.7.1 and 2.7.2, G' = GIK, N' =M/C, which
l eads us to G/7V - G'lN' = (G/K)/(N/K).
This l ast equal ity is highl y suggestive; we are sort of "cancel ing o ut" the
K in the numerato r and deno mi nato r.
PROB L EMS
1. Show that M D N in the proof of Theo rem 2.7.3.
2. Let G be the gro up of al l real -val ued functions o n the unit interval [0, 1],
where we define, for f g E G, addi ti o n by ( / + g)(x) =f(x) + g(x) for
every x E [0, 1]. If N = {f E G \ f(\) = 0}, pro ve that G/7V = real num-
bers under +.
3. Let G be the group of no nzero real numbers under mul tipl ication and l et
N = {1, - 1}. Pro ve that G/7V positive real numbers under mul tipl ication.
4. If Gi, G
2
are two gro ups and G = G
t
X G
2
= {(a, b) \ a E G
u
b E G
2
),
where we define (a, b)(c, d) = (ac, bd), sho w that:
(a) N = {(a, e
2
) | a E G
x
), where e
2
is the uni t el ement of G
2
, is a no rmal
subgro up of G.
(b) 7Y =Gi .
(c) G/N=*G
2
.
5. Let G be a gro up, H a subgro up of G, and N < G. Let the set ffl V =
[h <= H,n <E N\. Pro ve that:
(a) H H N < H.
(b) ffl V is a subgro up of G.
(c) TV C HN and TV <77/V.
(d) (HN)/N = H/(H n TV).
*6. If G is a group and N < G, show that if a G G has finite o rder o (a), then
Na in G/7V has finite o rder m, where m \ o(a). (Pro ve this by using the
ho mo mo rphi sm of G o nto GIN.)
7. If (p is a ho mo mo rphi sm of G o nto G' and N < G, show that 9 (AT) < G'.
8. CA UCHY' S THEOREM
I n Theo rem 2.6.4Cauchy's Theo remwe pro ved that if a pri me p divides
the o rder of a finite abelian gro up G, then G contains an el ement of o rder p.
We did point out there that Cauchy's Theo rem is true even if the group is no t
abel ian. We shal l give a very neat proof of this here; this proof is due to
McKay.
We return for a mo ment to set theo ry, doing so methi ng that we men-
ti o ned in the pro bl ems in Section 4.
Let S be a set, / A(S), and define a rel ation o n S as fol l ows: s ~ t if
t = f'(s) for so me i nteger /' (i can be positive, negative, or zero ). We l eave it
to the reader as a pro bl em that this do es i ndeed define an equi val ence rel a-
tion on S. The equi val ence cl ass of s, [s], is cal l ed the orbit of s under/. So S
is the disjoint uni o n of the orbits of its el ements.
When / is of o rder p,p a pri me, we can say so methi ng abo ut the size of
the orbits under /; tho se of the readers who sol ved Pro bl em 34 of Section 4
al ready kno w the resul t. We pro ve it here to put it on the reco rd official l y.
[If f
k
(s) = s, of course f
tk
(s) = s for every integer t. (Prove!)]
L emma 2.8.1. If / A(S) is of o rder p, p a pri me, then the orbit of
any el ement of S under/ has 1 or p el ements.
Proof. Let s S; iif(s) = s, then the orbit of s under /co nsi sts merel y
of s itsel f, so has o ne el ement. Suppo se then that f(s) s. Co nsi der the el e-
ments s, f(s), f
2
(s), . . . , f
p
'
1
(s); we cl aim that these p el ements are distinct
and consti tute the orbi t of s under /. If no t, then f'(s) = f'(s) for so me
0 < i <; ' </ ? - 1, which gives us that f'~'(s) = s. Let m = j - i; then
0 < m < p - 1 and/'"( s ) = s. But/"( s ) = s and si ncep \ m, ap + bm = 1 for
so me i ntegers a and b. Thus f(s) = f
ap+bm
{s) =f
ap
{f
bm
{s)) = f
ap
(s) = s,
Sec. 8
Cau c h y ' s Th e o r e m 89
since f'"(s) = f
p
(s) = s. This contradi cts that f(s) = s. Thus the orbit of s
under/co nsi sts of s, f(s), f
2
(s),... ,f
p
~
l
(s), so as p el ements.
We no w give McKay's pro o f of Cauchy's Theo rem.
Theo rem 2.8.2 (Cauchy). If/; is a pri me and p divides the o rder of G,
then G contai ns an el ement of o rder p.
Proof. If p = 2, the resul t amo unts to Pro bl em 18 in Section 1. Assume
that p + 2. Let S be the set of al l orderedp-tupl es (a
x
, a
2
, , a
p
-\, a
p
), where
a
h
a
2
, , a
p
a r e m
G and where a
x
a
2
a
p
-\a
p
= e. We cl aim that S has n?~
x
el ements where n = \G\. Why? We can choose a
x
,..., a
p
_
x
arbitraril y in G,
and by putting a
p
= (a
x
a
2
a^~\ thep-tupl e (a
x
, a
2
,..., a
p
^
x
, a
p
) then satisfies
a
L
a
2
a
p
_
x
a
p
= a
x
a
2
a
p
_
l
(a
x
a
2
cip^y
1
= e,
so is in S. Thus S has n
p
~
1
el ements.
No te that if a
x
a
2
a
p
_
x
a
p
= e, then a
p
a
x
a
2
a
p
_
x
= e (for iixy = e in
a group, then yx = e). So the mappi ng / : S ^ S defined by f(a
x
,..., a
p
) =
(a
p
, a
x
, a
2
, . .. , is in A(S). No te that/ =h /, the identity map on S, and
that f = i, so / is of o rder p.
If the orbit of s under / has o ne el ement, then f(s) = s. On the o ther
hand, if f(s) = s, we kno w that the orbit of s under / consists precisel y of p
distinct el ements; this we have by L emma 2.8.1. No w when is f(s) + si We
cl aim that f(s) + s if and onl y if when s = (a
x
, a
2
, .. . , a
p
), then for so me
i + j , a i + cij. (We l eave this to the reader.) So f(s) = s if and onl y if 5- =
(a, a,... , a) for so me a G G.
Let m be the number of s G S such that f(s) = s; since for s =
(e, e,. . ., e), f(s) = s, we kno w that m > 1. On the o ther hand, if f{s) # s,
the orbit of s consists of p el ements, and these orbits are disjoint, for they are
equi val ence cl asses. If there are k such orbits where f(s) # s, we get that
n
p~
l
=
m
+ /
C
p
;
f
o r w e
have acco unted this way for every el ement of S.
But p\nby assumpti on and p | (kp). So we must have p \ m, since m =
ii
p
~
x
- kp. Because m + 0 and p \ m, we get that m > 1. But this says that
there is an s = (a, a... ., a) + (e, e,. . . , e) in S; from the definition of S this
impl ies that a
p
= e. Since a + e, a is the requi red el ement of o rder p.
No te that the proof tel l s us that the number of sol utions in G of x
p
= e
is a positive mul tipl e of p.
We strongl y urge the reader who feel s unco mfo rtabl e with the proof
just given to carry out its detail s for p = 3. I n this case the action of / o n S
beco mes cl ear and o ur assertions abo ut this action can be checked expl icitl y.
Cauchy's Theo rem has many co nsequences. We shal l present o ne of
these, in which we determi ne co mpl etel y the nature of certai n gro ups of
o rder pq, where p and q are distinct pri mes. Other co nsequences wil l be
found in the pro bl em set to fol l ow, and in l ater materi al o n groups.
L emma 2.8.3. Let G be a gro up of o rder pq, where p, q are pri mes
and p > q. If a E G is of o rder p and A is the subgro up of G generated by a,
then A <\G.
Proof. We cl aim that A is the only subgro up of G of o rder p. Suppo se
that B is ano ther subgro up of o rder p. Co nsi der the set AB =
[xy | x G A, y G B\; we cl aim that AB has p
2
distinct el ements. Fo r suppo se
that xy = uv where x, u G A, y, v G B; then u~
l
x = vy~K But u~
1
x G A,
vy~
l
G B, and since u~
l
x = try
- 1
, we have u~
l
x G A fl B. Since B + A.and
A Pi B is a subgro up of A and A is of pri me o rder, we are forced to concl ude
that A fl B = (e) and so iC
i
x = e, that is, u = x. Simil arl y, v = y. Thus the
number of distinct el ements in AB is p
2
. But al l these el ements are in G,
which has onl y pq q). Wi th this co ntradi cti o n we
see that B = A and A is the onl y subgro up of o rder/? in G. But if x G G, B =
x~
L
Ax is a subgro up of G of o rder p, in co nsequence of which we co ncl ude
that x~
]
Ax =A; hence A<\G.O
Corol l ary. If G, a are as in L emma 2.8.3 and if x G G, then
x~
l
ax =fl ', where 0 < i < p, for so me i (dependi ng o n x).
Proof. Since e + a G A and x~'Ax =A, x~
l
ax G A. But every el ement
of A is of the form a', 0 < i </J , and x
-
'AX e. I n co nsquence, xâx =A',
where 0 < i< p.
We no w pro ve a resul t of a different fl avor.
L emma 2.8.4. If a G G is of o rder m and 6 G (7 is of o rder n, where
m and n are rel ativel y pri me and ab = /3A, then c = ab is of o rder
Proof. Suppose that A is the subgro up generated by a and B that gener-
ated by b. Because |A| =m and |B | =n and (m, ) = 1, we get A fl B =(e),
which fol l ows from Lagrange's Theo rem, for |A D 51 | n and \ A D B\ \ m.
Suppo se that c' = e, where i > 0; thus (ab)' = e. Since A/3 = ba, e =
(A&)' =a'b'; this tel l s us that a' = b~' G A n 5 = (e). So A'' =e, whence in | z',
and />' = e, whence n \ i. Because (m, n) = 1 and m and n bo th divide i, inn
divides i. So / >mn. Since (ab)'"" = a"'"b""' =e, we see that m/z is the smal l -
est positive i nteger i such that (ab)' = e. Thi s says that ab is of o rder mra, as
cl ai med in the l emma.
Sec. 8 Cau c h y ' s Th e o r e m 91
Befo re considering the mo re general case of gro ups of o rder pq, l et's
l ook at a special case, namel y, a gro up G of o rder 15. By Cauchy's Theo rem,
G has el ements b of o rder 3 and a of o rder 5. By the Corol l ary to L emma
2.8.3, b"
x
ab = a', where 0 < i < 5. Thus
b'
2
ab
2
= b'\b~
l
ab)b = b^db = (bâb)' = ( V) ' = a'
2
nnd simil arl y, b~
3
ab
3
= a'
3
. But b
3
= e, so we get a'
3
= a, whence a'
3
~
4
= e.
Since a is of o rder 5, 5 must divide i
3
- 1, that is, i
3
= 1(5). Ho wever, by Fer-
mat's Theo rem (Corol l ary to Theo rem 2.4.8), i
4
= 1(5). These two equati o ns
for /' tel l us that / = 1(5), so, since 0 < i < 5, i 1. I n short, b"
l
ab = a' = a,
which means that ab = ba. Since a is of o rder 5 and b of o rder 3, by L emma
2.8.4, c = ab is of o rder 15. This means that the 15 po wers e = c, c, c
2
,. ..,
;
1 4
are distinct, so must sweep o ut al l of G. I n a word, G must be cyclic.
The argument given for 15 coul d have been made shorter, but the fo rm
n which we did it is the exact pro to type for the proof of the mo re general
T heorem 2.8.5. Let G be a gro up of o rder pq, where p, q are pri mes
md p > q. li q \p 1, then G must be cycl ic.
Proof. By Cauchy's Theo rem, G has an el ement a of o rder p and an
el ement b of o rder q. By the Corol l ary to L emma 2.8.3, b~
l
ab = a' for so me i
with 0 0 (Pro ve!), and so b~ab<< = a'".
But b
q
= e; therefo re, a''
1
= a and so a
1
^
1
= e. Because a is of order p, we
concl ude that p\i
q
1, which is to say, i
q
= l(p). Ho wever, by Fermat's
Theorem, z
p _ 1
= l(p). Since q\p 1, we concl ude that i = l(p), and since
) < i < p, i = 1 fol l ows. Therefo re, b~
l
ab = a' = a, hence ab = ba. By
L.emma 2.8.4, c = ab has o rder pq, so the po wers of c sweep o ut al l of G.
Thus G is cycl ic, and the theo rem is pro ved.
PROB L EMS
1. I n the pro o f of Theo rem 2.8.2, show that if so me two entries in s =
(ai, a
2
, ... , a
p
) are different, then f(s) s, and the orbit of s under /
has p el ements.
2. Pro ve that a gro up of o rder 35 is cycl ic.
3. Usi ng the resul t of Pro bl em 40 of Section 5, give ano ther pro o f of
L emma 2.8.3. (Hi nt: Use for H a subgro up of o rder p.)
4. Construct a no nabel i an group of o rder 21. (Hint: Assume that a
3
= e,
b
1
= e and find so me / such that a
l
ba = a' j= a, which is consistent with
the rel ati ons a
3
= b
1
= e.)
5. Let G be a gro up of o rderp" m, wherep is pri me andp \ m . Suppo se that
G has a normal subgro up P of o rder p". Pro ve that 9(P) = P for every
auto mo rphi sm 0 of G.
6. Let G be a finite gro up with subgro ups A, B such that \A \ > V |Gj and
|5| > Vj G). Pro ve that / l n 6 # (e).
7. If G is a gro up with subgro ups A, B of o rders in, n, respectivel y, where
m and n are rel ativel y pri me, pro ve that the subset of G,
AB = [ab ) a G A, b E B), has distinct el ements.
8. Pro ve that a gro up of o rder 99 has a nontri vi al no rmal subgro up.
9. Pro ve that a gro up of o rder 42 has a nontrivial no rmal subgro up.
10. Fro m the resul t of Pro bl em 9, pro ve that a gro up of o rder 42 has a nor-
mal subgro up of o rder 21.
Harder Pro bl ems
11. If G is a gro up and A, B finite subgro ups of G, pro ve that the set AB =
{ab | a G A, b B} has (\A\ \B\)I\A f l B\ distinct el ements.
12. Pro ve that any two no nabel i an gro ups of o rder 21 are i so mo rphi c. (See
Pro bl em 4.)
13. Using the fact that any gro up of o rder 9 is abel ian, pro ve that any gro up
of o rder 99 is abel ian.
14. Let p > q be two pri mes such that q\p - 1. Pro ve that there exists a
no nabel i an gro up of o rder pq. (Hi nt: Use the resul t of Pro bl em 40 of
Section 4, namel y that U
p
is cycl ic if p is a pri me, and the idea needed to
do Pro bl em 4 above.)
15. Pro ve that if p > q are two pri mes such that q\p 1, then any two no n-
abel ian groups of o rder pq are i somorphi c.
9. DI RECT PRODUCTS
I n several of the pro bl ems and exampl es that appeared earl ier, we went
thro ugh the fol l owing construction: If G,, G
2
are two gro ups, then G =
Gi X,G
2
is the set of al l o rdered pairs (a, b), where a G G
x
and b G G
2
and
Sec . 9 Di r ec t Pr o d u c t s 93
where the pro duct was defined component-wise via (a
x
, b
l
)(a
2
, b
2
) =
(a
x
a
2
, b
x
b
2
), the pro ducts in each co mpo nent bei ng carried out in the respec-
tive gro ups G
x
and G
2
. We shoul d l ike to formal ize this pro cedure here.
Definition. If G
x
, G
2
,. . . , G are n groups, then their (external) di-
rect product G
x
X G
2
X G
3
X X G is the set of al l o rdered n-tupl es
( a , , a
2
,. . ., a) where a , G G,, for i = 1, 2,.. . , n, and where the pro duct in
G] X G
2
X X G is defined co mpo nent-wi se, that is,
(!, a
2
,. .., a
n
)(b
1
, b
2
,..., b) = (a
x
b
x
, a
2
b
2
,ab).
That G = G
x
X G
2
X X G is a gro up is i mmedi ate, with
(e
x
, e
2
,.. ., e) as its uni t el ement, where e
t
is the uni t el ement of G
h
and
where (a
x
, a
2
, >
a
)
1
= ( f \ 2
_ 1
> , fl ,r')'
G is merel y the Cartesi an pro duct of the groups G
x
, G
2
, . . ., G with a
pro duct defined in G by co mpo nent-wi se mul tipl ication. We cal l it external,
since the gro ups G
x
, G
2
,. . ., G are any groups, with no rel ation necessaril y
hol di ng amo ng them.
Co nsi der the subsets G, C G, X G
2
X X G = G, where
G,- = . . ., e
M
, a,-, e
i + 1
, . . . , e) \ a
t
G G,};
in o ther wo rds, G
;
consists of al l /i-tupl es where in the iih co mpo nent any el -
ement of G, can occur and where every o ther co mpo nent is the identity el e-
ment. Cl earl y, G,- is a gro up and is i so mo rphi c to G,- by the i so mo rphi sm
TTJ\ Gj > G
t
defined by 7r
/
(e
1
, e
2
,. . . , a
h
. . ., e
n
) = a,-. Furthermo re, not
onl y is G, a subgro up of G but G, <J G. (Prove!)
Gi ven any el ement a = (a
x
, a
2
,. . . , a) G G, then
a = (a,, e
2
,.. . , e
n
)(e
x
, a
2 )
e
3
, . . ., e) (e
1 ;
e
2
, . . . , e,,.,, a);
that is, every a G G can be wri tten as a = a j a
2
a , where each a,- G G,-.
Mo reo ver, a can be wri tten in this way in a uni que manner, that is, if a =
a
x
d
2
a = b
L
b
2
b
n
, where the a,- G G, and b,- G G,, then
a
2
= &!,... ,a
n
= b
n
. So G is buil t up from certain no rmal subgro ups, the
Gj, as G = G
G
2
G in such a way that every el ement a G G has a
unique representati o n in the form a = a j a , a with a,- G G,.
This mo ti vates the fol l owing
Definition. The gro up G is said to be the (internal) direct product of
its normal subgro ups N
u
N
2
,. . . , N if every a G G has a unique representa-
tion in the form a = a
x
a
2
a, where each a, G TV,- for = 1, 2, . . . , n.
Fro m what we have discussed abo ve we have the
Lemma 2.9.1. If G =G, X G
2
X X G is the external direct pro d-
uct of Gi, G
2
, , G, then G is the i nternal direct pro duct of the no rmal
subgro ups G
l 5
G
2
, . . ., G defined abo ve.
We want to go in the o ther direction, namel y to pro ve that if G is the in-
ternal direct pro duct of its no rmal subgro ups N
x
, N
2
,..., N, then G is iso-
morphic to N
x
X TV, X X A
7
,,. To do so, we first get some prel iminary resul ts.
The resul t we are abo ut to pro ve has al ready o ccurred as Pro bl em 20,
Section 5. Fo r the sake of co mpl eteness we pro ve it here.
Lemma 2.9.2. Let G be a gro up, M, N no rmal subgro ups of G such
that M fl N = (e). Then, given ra E M and n G N, mn = nm.
Proof. Co nsi der the el ement a =mnm~
l
n~
x
. Viewing a as bracketed
o ne way, a = (mnm~
l
)n~
l
; then, since N <I G and n G N, mnm~
l
G N, so
a ={mnm~
l
)n~
l
is al so in N. No w bracket a in the other way, a = ni(nm~
1
n~
1
).
Since M <3 G and m'
1
G M, we have nm~
l
n~
l
G M and so fl =
m(nm~
x
rC
v
) G M. Thus a G M, n N =(e), which is to say, mmn~
x
nT
x
=e.
This gives us that mn =nm, as requi red.
If G is the i nternal direct pro duct of the no rmal subgro ups N
h
N
2
,... , N, we cl aim that AT,- PI Nj = (e) for i + j . Fo r suppo se that
A E A
7
,- Pi A^; then A =e e eae e, where the a occurs in the z'th pl ace.
This gives us o ne representati on of A in G =N
1
N
2
N
n
. On the other hand,
a = e- e- -- e- a- e- --e, where the a occurs in the y'th pl ace, so a has the second
representation as an el ement of AÂ/^N. By the uni queness of the repre-
sentation, we get fl =e, and so A
7
,- Pi Nj = (e).
Perhaps things wo ul d be cl earer if we do it for n =2. So suppo se that
A
7
) <l G, N
2
<I G, and every el ement a G G has a uni que representati o n as
a Aj A
2
, where a
x
G A/j, fl
2
G Af
2
. Suppo se that a E. N
x
C\ N
2
; then A =
A e is a representati o n of a = a
x
a
2
with a
x
=a G N
x
, a
2
= e G A^. Ho w-
ever A =e a, so A =b
x
b
2
, where b
x
= e E. N
x
, b
2
= a E N
2
. By the
uni queness of the representati o n we must have a
{
= b
x
, that is, a = e.
So N
x
PI N
2
=(e).
The argument given above for A
7
,,..., N
n
is the same argument as that
given for n =2, but perhaps is l ess transparent. At any rate we have pro ved
Lemma 2.9.3. If G is the i nternal direct pro duct of its no rmal sub-
gro ups N
x
, N
2
,..., N, then, for / ;', AT,- Pi Nj =(e).
Corol l ary. If G is as in L emma 2.9.3, then if / #/ and A,- A
7
, and
cij GW-, we have A;fl y = A
;
A,-.
Sec. 9
Direct Products 95
Proof. By L emma 2.9.3, iV
;
fl TV} = (e) for i /. Since the TV's are no r-
mal in G, by L emma 2.9.2 we have that any el ement in N
t
commutes with any
el ement in TV,-, that is, a
l
a
j
= aja, for a
t
G N
h
a
;
- G TV,-.
Wi th these prel iminaries o ut of the way we can no w prove
Theo rem 2.9.4. Let G be a gro up with no rmal subgroups N
x
, N
2
,. ,
N. Then the mappi ng ip(a
x
, a
2
, ., a) = a
x
a
2
a is an i so mo rphi sm
from N
x
X N
2
X X N (external direct pro duct) o nto G if and onl y if G is
the i nternal direct pro duct of N
x
, N
2
,. . . , N.
Proof. Suppo se G is an i nternal direct pro duct of N
x
,. . ., N. Since
every el ement a in G has a representati o n a = a
x
a
2
a, with the a, N
h
we have that the mappi ng ip is o nto . We assert that it is al so 1-1. Fo r if
tjj((a
u
a
2
, . . . , a,,)) = !/>((&!, b
2
,-.., b,,)), then by the definition of
a
x
a
2
a
n
= b
x
b
2
b. By the uni queness of the representati o n of an el e-
ment in this form we deduce that a
x
= b
h
a
2
= b
2
, ... , a = b. Hence </>is 1-1.
Al l that remai ns is to show that i/>is a ho mo mo rphi sm. So, consider
(//((!, a
2
,..., )(&!, b
2
,..., b)) = ip((a
L
b
x
, a
2
b
2
,ab))
= faA)(flA) ' " (
a
b,)
= a
x
b
L
a
2
b
2
a
n
b
n
.
Since b
x
G 7V
l 5
it co mmutes with a
h
b
t
for i > 1 by the Corol l ary to L emma
2.9.3. So we can pul l the b
x
across al l the el ements to the right of it to get
a
1
b
l
a
2
b
2
ab = a
1
a
2
b
2
a
3
b
3
abb
1
. No w repeat this pro cedure with b
2
,
and so on, to get that a
y
b
1
a
2
b
2
ab = (a
1
a
2
a
n
)(b
1
b
2
b
n
). Thus
^( ( f l i , a
2
,..., a
n
)(b
t
, b
2
,..., b)) = a
x
b
x
a
2
b
2
ab
n
= {a
x
a
2
a
n
)(b
x
b
2
b)
=<K(fli,
2
. . a,))4i{{b
x
b
2
,..., b,,)).
I n o ther wo rds, i// is a ho mo mo rphi sm.
On the o ther hand, suppo se that ii/ is an i so mo rphi sm. Then the concl u-
sion that G is the i nternal direct pro duct of N
x
, N
2
,..., 7V easil y fol l ows
from the fact that />is o nto and 1-1.
Wi th this the proof of Theo rem 2.9.4 is co mpl ete.
Corol l ary. Let G be a gro up with no rmal subgro ups N
x
, N
2
. Then G
is the i nternal direct pro duct of N
x
and 7V
2
if and onl y if G = N
X
N
2
and
N
x
n N
2
= (e).
Proof. This fol l ows easil y from the fact that ifj\N
i
XN
2
> G, which is
given by i/f(fl j,
a
i)
= a
\
a
i> is
a n
i somorphi sm if and onl y if N\N
2
= G and
N
l
nN
2
= (<?)
I n view of the resul t of Theo rem 2.9.4 and its corol l ary, we dro p the ad-
jectives "i nternal " and "external " and merel y speak abo ut the "di rect pro d-
uct." When no tati o n G = N
x
X N
2
is used it shoul d be cl ear from context
whether it stands for the i nternal or external direct pro duct.
The objective is often to show that a given gro up is the direct pro duct
of certai n no rmal subgro ups. If o ne can do this, the structure of the gro up
can be co mpl etel y determi ned if we happen to kno w tho se of the no rmal
subgro ups.
P R O B L E M S
1. If Gi and G
2
are gro ups, pro ve that G
x
X G
2
G
2
X G
1
.
2. If Gi and G
2
are cycl ic groiips of o rders m and n, respectivel y, pro ve that
G
x
X G
2
is cycl ic if and onl y if m and n are rel ativel y pri me.
3. Let G be a gro up, A = G X G. I n A l et T = {(g, g)\g<= G).
(a) Pro ve that T ~ G.
(b) Pro ve that T<AH and onl y if G is abel ian.
4. Let G be an abel ian gro up of o rder p'fp'p- p'
k
k
, where p
x
, p
2
,..., p
k
are distinct pri mes and m
x
> 0, m
2
> 0, . . . , m
k
> 0. By Pro bl em 10 of
Section 6, for each i, G has a subgro up T
3
, of o rder p"'
1
. Show that
G =- Pj X P
2
X X P
k
.
5. Let G be a finite gro up, N
x
, N
2
,.. ., N
k
no rmal subgro ups of G such that
G = N
t
N
2
N
k
and |G| = \N
L
\ \N
2
\ \N
k
\. Pro ve that G is the direct
pro duct of N
x
, N
2
,. . . , N
k
.
6. Let G be a gro up, N
x
, N
2
,. .., N
k
no rmal subgro ups of G such that:
1. G = N
L
N
2
N
k
.
2. Fo r each i, Nj n (N
1
N
2
Ny- i ^+i N
k
) = (e).
Pro ve that G is the direct pro duct of N
lt
N
2
,... , N
k
.
10. FI NI TE A B EL I A N GROUPS ( OPTI ONAL )
We have just finished discussing the i dea of the di rect pro duct of gro ups. If
we were to l eave that topi c at the po i nt where we ended, it mi ght seem l ike
a nice l ittl e co nstructi o n, but so what? To give so me mo re substance to it,
Sec . 10 Fi ni t e A b el i an Gr o u p s ( Op t i o n al ) 97
we shoul d pro ve at l east o ne theo rem which says that a gro up satisfying a
certai n co ndi ti o n is the direct pro duct of so me parti cul arl y easy groups. Fo r-
tunatel y, such a cl ass of gro ups exists, the finite abel i an gro ups. What we
shal l pro ve is that any finite abel i an gro up is the direct pro duct of cycl ic
groups. Thi s reduces mo st questi o ns abo ut finite abel ian groups to questi o ns
abo ut cycl ic gro ups, a reducti o n that often al l ows us to get co mpl ete an-
swers to these questi o ns.
The resul ts o n the structure of finite abel ian gro ups are real l y special
cases of so me wider and deeper theo rems. To consider these wo ul d be goi ng
too far afiel d, especial l y since the story for finite abel ian groups is so i mpo r-
tant in its own right. The theo rem we shal l pro ve is cal l ed the Fundamental
Theorem on Finite Abelian Groups, and rightful l y so.
Befo re getting do wn to the actual detail s of the proof, we shoul d l ike to
give a quick sketch of ho w we shal l go abo ut provi ng the theo rem.
Our first step wil l be to reduce the pro bl em from any finite abel ian
gro up to o ne whose o rder is p", where p is a pri me. This step wil l be fairl y
easy to carry out, and since the gro up wil l have o rder invol ving just o ne
prime, the detail s of the pro o f wil l no t be cl uttered with el ements whose or-
ders are so mewhat compl i cated.
So we shal l focus on groups of o rder p". Let G be an abel ian gro up of
o rder p". We want to show that there exist cycl ic subgroups of G, namel y
Ai, A
2
, ,A
k
, such that every el ement x G can be wri tten as x =
b
y
b
2
---b
k
, where each b
{
A
h
in a uni que way. Otherwi se put, since
each Aj is cycl ic and generated by a
h
say, we want to show that x =
a'"
l
a
2
- a"'
k
, where the el ements a'"' are uni que.
A difficul ty appears right away, for there is not just one choice for these
el ements a
x
, ... , a
k
. Fo r instance, if G is the abel ian gro up of o rder 4 with
el ements e, a, b, ab, where a
2
= b
2
= e and ab = ba, then we can see that if
A, B, C are the cycl ic subgroups generated by a, b, and ab, respectivel y, then
G = AxB=AXC=BxC.So there is a l ack of uniqueness in the choice of
the ,. Ho w to get around this?
What we need is a mechani sm for picking a
x
and which, when appl ied
after we have picked a
x
, wil l al l ow us to pick a
2
, and so on. What shoul d this
mechanism be? Our control on the el ements of G l ies onl y in specifying thei r
orders. I t is the o rder of the el ementwhen properl y usedthat wil l give us
the means to pro ve the theo rem.
Suppo se that G = A
x
X A
2
X X A
k
, where \G\ = p" and the A's
have been numbered, so that \Aj\ = p"> and j s n
2
> > n
k
, and each A;
is cycl ic generated by a
t
. If this were so and x = fl "'
1
a'
k
k
, then
x
p
"
1
= (fl'//
1
<*)'"" = ci'{'"""a'
2
"
2P
"
1
a'
k
"
kP
"
1
because n
x
~ n
h
p"'\ p"\ so since every a"'
iP
"' = e, thus x
p
"' = e. In other
words, a
x
should be an element of G whose order is as large as it can possibly
be. Fi ne, we can no w pick a
x
. What do we do for a
2
? If G = GIA
X
, then to
get the first el ement needed to represent G as a direct pro duct of cycl ic
groups, we shoul d pick an el ement in G who se o rder is maxi mal . What does
this transl ate into in G itsel f? We want an el ement a
2
such that a
2
requires as
high a power as possible to fall into A
x
. So that wil l be the ro ad to the sel ec-
tion of the second el ement. Ho wever, if we pick an el ement a
2
with this
pro perty, it may no t do the trick; we may have to adapt it so that it wil l . The
doing of al l this is the technical part of the argument and does go thro ugh.
Then o ne repeats it appro pri atel y to find an el ement a
3
, and so on.
This is the pro cedure we shal l be going thro ugh to pro ve the theo rem.
But to smo o th out these successive choices of a
x
, a
2
,..., we shal l use an in-
duction argument and so me subsidiary prel i mi nary resul ts.
Wi th this sketch as guide we ho pe the proof of the theo rem wil l make
sense to the reader. One shoul d no t confuse the basic idea in the proof
which is qui te reaso nabl ewi th the technical detail s, which may cl oud the
issue. So we no w begin to fil l in the detail s of the sketch of the proof that we
outl i ned abo ve.
L emma 2.10.1. Let G be a finite abel ian gro up of o rder mn, where m
and n are rel ativel y pri me. If M = {x G G | x'
n
= e] and N = {x G G | x" = e],
then G = M X N. Moreover, if neither m no r n is 1, then M i= (e) and N + (e).
Proof. The sets M and N denned in the assertion abo ve are quickl y
seen to be subgro ups of G. Mo reo ver, if m + 1, then by Cauchy's Theo rem
(Theo rem 2.6.4) we readil y o btai n M (e), and simil arl y if n 1, that
N + (e). Furthermo re, since M n N is a subgro up of bo th M and N, by La-
grange's Theo rem, \ M D N \ divides \M\ = m and |jV| = n. Because m and n
are rel ativel y pri me, we o btai n \M fl N\ = 1, hence M C\ N = (e).
To finish the proof, we need to show that G = MN and G = M X N.
Since m and n are rel ativel y pri me, there exist integers r and s such that
rm + sn = 1. If a G G, then a = a
1
= a
s
"
+rm
= a
s
"a
n
"; since (a*")
1
" =
a""" = e, we have that a
s
" G M. Simil arl y, a'"
1
G TV. Thus a = a
s
"a
n
" is in
MN. I n this way G = MN. I t no w fol l ows from Corol l ary to Theo rem 2.9.4
that G = M X TV.
An i mmedi ate co nsequence is the
Corol l ary. Let G be a finite abel ian group and l et p be a pri me such
that p divides |G|. Then G = P X T for so me subgro ups P and T, where
\P\ v. p"\ m > 0, and | T\ is no t divisibl e by p.
Sec . 10 Fi ni t e A b el i an Gr o u p s ( Op t i o n al ) 99
Proof. Let P = [x G G\ x
pS
= e for so me s) and l et the subset
T = [x G G | x' = e for r rel ativel y pri me to /?}. By L emma 2.10.1, G = P X T
and /
J
(e)- Since every el ement in P has o rder a po wer of p, \P\ is not divis-
ibl e by any o ther pri me (by Cauchy's Theo rem), so \P\ = p'" for some m.
It is easy to see that p j | T\ by maki ng use of Lagrange's Theo rem. Thus
we real l y have that P is no t merel y so me subgro up of G but is what is cal l ed
a p-Syl ow subgro up of G. (See Section 11).
We no w co me to the key step in the pro o f of the theo rem we seek.
The proof is a l ittl e difficul t, but o nce we have this resul t the rest wil l be
3asy.
Theo rem 2.10.2. Let G be an abel ian gro up of o rder p", p a pri me,
and l et a G G have maxi mal o rder of al l the el ements in G. Then G =
A X Q, where A is the cycl ic subgro up generated by a.
Proof. We pro ceed by i nducti on on n. If n = 1, then \ G\ =p and G is
al ready a cycl ic gro up generated by any a j= e in G.
We suppo se the theo rem to be true for al l m < n. We first show that
the theo rem is correct if there exists an el ement b G G such that b A = (a)
and b
p
= e. Let B = (b), the subgro up of G generated by b; thus
A n B = (e) (see Pro bl em 1).
Let G = GIB\ by assumpti o n B ==(e), hence \G\ < \G\. I n G, what is
the o rder of d = 5a? We cl aim that o(a) = o(a). To begin with, we kno w
that o(a) | o(a) (see Pro bl em 6 of Section 2.7). On the o ther hand, a
o (
"
) =
e,
so fl
o(
'
7)
G 5. Since a
o(7,)
G A , we see that fl "
(
'
7)
G A f l B = (e), whence
fl
<.() _
e
tel l s us that o(a) \ o(a). Hence o(a) = o(d).
Since a is an el ement of maxi mal o rder in G, by the induction we kno w
that G = ( f l ) X T for so me subgro up T of G. By the Co rrespo ndence
Theo rem we al so kno w that T = QIB for so me subgro up Q of G. We cl aim
that G is the i nternal direct pro duct A X Q. That G = AQ is l eft to
the reader. I t remai ns to show that A (~l O = (e). Let a' A fl (J . Then
fl ' G <2A/3 = 7", and since (a) (1 T = (e), we have that d' = e. But since o(a) =
o(fl ), this impl ies a' = e. Therefo re, A f l Q = (e) and we obtain that
G = A X Q.
Suppo se, then, that there is no el ement b in G, b not in A , such that
b
p
= e. We cl aim that this forces G = A = (fl ), in whi ch case G is a cycl ic
gro up. Suppo se that G + A and l et x G G, x A have smal l est possibl e
order. Because o(x
p
) < o(x), we have, by o ur choice of x, that x
p
G A, hence
x
p
= ' for so me i.
We cl aim that p\i. Let o(a) = /r
!
, and no te that the maximal ity
100 Gr o u p s Ch . 2
of the o rder of a impl ies that x
pS
= e. But x
pS
= (x
p
)
pS
~
l
= (a
i
)
pS
~
l
= e. Since
o(a) =p'
5
, we have /? | /.
Thus x
p
= a', where p | i. Let y = fl ~'
/f
' x. Then y
p
= a~'x
p
= cT'a' = e.
Mo reo ver, y (a) = A, because x & A. But this puts us back in the si tuati on
discussed abo ve, where there exists a b 6 G, > A such that b
p
= e;m that
case we saw that the theo rem was correct. So we must have G = (), and G
is a cycl ic gro up. Thi s finishes the i nducti o n and pro ves the theo rem.
We are no w abl e to pro ve the very basic and i mpo rtant
T heorem 2.10.3 (Fundamental T heorem on Fi ni te Abel i an Gro ups).
A finite abel ian gro up is the direct pro duct of cycl ic gro ups.
Proof. Let G be a finite abel i an gro up and p a pri me that divides |G|.
By the Corol l ary to L emma 2.10.1, G = P X T, where |P| =p". By Theo rem
2.10.2, P = Ai X A
2
X X A
k
, where the A, are cycl ic subgro ups of P. Ar-
guing by induction o n |G|, we may thus assume that T = T
x
X T
2
X X T
q
,
where the T, are cycl ic subgro ups of T. Thus
G = (A
x
X A
2
X X A
k
) X (T, X T
2
X X T
q
)
= A
}
X A
2
X X A
k
x r, X T
2
X X T
q
.
This very i mpo rtant theo rem is no w pro ved.
We return to abel ian gro ups G of o rder p". We no w have at hand that
G = A
l
X A
2
X X A
k
, where the A
t
are cycl ic gro ups of o rder p"'. We
can arrange the numberi ng so that n
i
s n
2
> > n
k
. Al so , \G\ =
\A
1
X A
2
X X A
k
\ = \A
X
\ \A
2
\ \A
k
|, which gives us that
hence n = n
x
+ n
2
+ + n
k
. Thus the integers ( 1 , 2: 0 give us a partition of
n. I t can be sho wn that these integers H j , n
2
,.. . , n
k
which are cal l ed the
invariants of Gare unique. I n o ther wo rds, two abel i an gro ups of o rder p"
are i somorphi c if and onl y if they have the same i nvari ants. Granted this, it
fol l ows that the number of no ni so mo rphi c abel ian gro ups of o rder p" is equal
to the number of parti ti o ns of n.
Fo r exampl e, if /; = 3, it has the fol l owing three parti ti o ns: 3 = 3, 3 =
2 + 1,3 = 1 + 1 + 1, so there are three no ni so mo rphi c abel i an gro ups of
o rder p
3
(i ndependent of p). The gro ups co rrespo ndi ng to these parti ti o ns
are a cycl ic gro up of o rder p
3
, the direct pro duct of a cycl ic gro up of o rder p
1
by o ne of o rder p, and the direct pro duct of three cycl ic gro ups of o rder p,
respectivel y.
Sec . 11 Co n j u g ac y an d Sy l o w ' s Th e o r e m ( Op t i o n al ) 101
For n = 4 we see the parti ti o ns are 4 = 4, 4 = 3 + 1, 4 = 2 + 2, 4 =
2 + 1 + 1, 4=1 + 1 + 1 + 1, which are five in number. Thus there are five
no ni so mo rphi c groups of o rder p
4
. Can you describe them via the parti ti ons
of 4?
Gi ven an abel ian gro up of o rder n - p
ai
p
a
2
2
' ' ' p"l\ where the p
t
are
distinct pri mes and the A, are al l positive, then G is the direct pro duct of its
so -cal l edp, - Syl ow subgro ups (see, e.g., the Co ro l l ary to L emma 2.10.1). Fo r
each pri me p
t
there are as many gro ups of o rder p"
1
as there are parti ti o ns of
A;. So the number of no ni so mo rphi c abel ian gro ups of o rder n = p\
y
p"
k
k
is f(adf(
a
i) ' ' ' /(
fl
/<r)' where f(m) deno tes the number of partitions of m.
Thus we kno w ho w many no ni so mo rphi c finite abel ian groups there are for
any given o rder.
Fo r instance, ho w many no ni so mo rphi c abel ian groups are there of
o rder 144? Since 144 = 2
4
3
2
, and there are five parti ti o ns of 4, two parti ti ons
of 2, there are 10 no ni so mo rphi c abel i an gro ups of o rder 144.
The materi al treated in this secti on has been hard, the path so mewhat
to rtuo us, and the effort to understand qui te intense. To spare the reader to o
much further agony, we assign onl y three pro bl ems to this section.
PROBL EMS
1. Let A be a no rmal subgro up of a gro up G, and suppo se that b G G is an
el ement of pri me o rder p, and that b A. Sho w that A D (b) = (e).
2. Let G be an abel ian gro up of o rder p", p a pri me, and l et a G G have max-
imal o rder. Show that x
o ( a )
= e for al l x G.
3. Let G be a finite gro up, with N <\ G and a G G. Pro ve that:
(a) The o rder of aN in GIN divides the o rder of a in G, that is,
o(aN) | o(fl ).
(b) If (A) n TV = (e), then o(aN) = ola).
11. CONJ UGA CY A ND SYL OW' S THEOREM ( OPTI ONAL )
In discussing equi val ence rel ations in Section 4 we menti o ned, as an exampl e
of such a rel ati o n in a gro up G, the no ti o n of conjugacy. Recal l that the el e-
ment b in G is said to be conjugate to a G G (o r merel y, a conjugate of a) if
there exists an x G G such that b = x~
l
ax. We sho wed in Section 4 that this
defines an equi val ence rel ati on o n G. The equi val ence cl ass of a, which we
deno te by cl (fl ), is cal l ed the conjugacy class of a.
102 Gr o u p s Ch . 2
Fo r a finite gro up an i mmedi ate questi on presents itsel f: Ho w l arge is
cl (fl )? Of course, this depends strongl y o n the el ement a. Fo r i nstance, if
a Z(G), the center of G, then ax = xa for al l x G, hence x~
x
ax = a; in
o ther wo rds, the conjugacy cl ass of a in this case consists merel y of the el e-
ment a itsel f. On the o ther hand, if cl (a) consists onl y of the el ement a, then
x~
l
ax = a for al l x G. This gives us that xa = ax for al l x G, hence
a Z(G). So Z(G) is characteri zed as the set of tho se el ements a in G
who se conjugacy cl ass has onl y o ne el ement, a itsel f.
Fo r an abel ian gro up G, since G =Z(G), two el ements are conjugate if
and onl y if they are equal . So conjugacy is not an i nteresti ng rel ati on for
abel ian groups; however, for no nabel i an gro ups it is a highl y i nteresti ng no -
tion.
Gi ven a G, cl (fl ) consists of al l x~
x
ax as x runs over G. So to deter-
mi ne which are the distinct conjugates of a, we need to kno w when two con-
j ugates of a coincide, which is the same as asking: When is x~
x
ax = y~
x
ay? I n
this case, transposi ng, we o btai n a(xy~
x
) = (xy~
x
)a; in o ther wo rds, xy~
x
must co mmute with a. This brings us to a concept i ntro duced as Exampl e 10
in Section 3, that of the centralizer of a in G. We repeat so methi ng we did
there.
Definition. If a G, then C(a), the centralizer of a in G, is defined
by C(a) = [x G | xa = ax}.
When C(a) arose in Section 3 we sho wed that it was a subgro up of G.
We reco rd this no w mo re official l y as
L emma 2.11.1. Fo r a G, C{a) is a subgro up of G.
As we saw abo ve, the two conjugates x'
x
ax and y~
x
ay of a are equal
onl y if xy
- 1
C(a), that is, onl y if x and y are in the same right coset of Cia)
in G. On the o ther hand, if x and y are in the same right coset of C{a) in G,
then xy~
x
C{a), hence xy~
l
a = axy~
x
. Thi s yiel ds that x~
x
ax = y~.
x
ay. So x
and y give rise to the same conjugate of a if and onl y if x and y are in the
same right coset of C(a) in G. Thus there are as many conjugates of a in G as
there are right cosets of C(a) in G. This is mo st i nteresti ng when G is a finite
gro up, for in that case the number of right cosets of C(a) in G is what we
cal l ed the index, i
c
(C(a)), of C(a) in G, and is equal to | G | /1 C(a) |.
We have pro ved
Theo rem 2.11.2. Let G be a finite gro up and a G; then the number
of distinct conjugates of a in G equal s the index of C(a) in G.
I n o ther words, the number of el ements in cl () equal s i
G
(C(a)) =
\G\l\C{a)\.
This theo rem, al though it was rel ativel y easy to pro ve, is very i mpo r-
tant and has many co nsequences. We shal l see a few of these here.
One such co nsequence is a ki nd of bo o kkeepi ng resul t. Since conjugacy
is an equi val ence rel ation o n G, G is the uni o n of the disjoint conjugacy
cl asses. Mo reo ver, by Theo rem 2.11.2, we kno w ho w many el ements there
are in each cl ass. Putti ng al l this i nfo rmati o n to gether, we get
Theo rem 2.11.3 (The Cl ass Equati o n). If G is a finite gro up, then
where the sum runs over o ne a from each conjugacy cl ass.
I t is al most a sacred tradi ti o n amo ng mathemati ci ans to give, as the first
appl ication of the cl ass equati o n, a parti cul ar theo rem abo ut gro ups of o rder
p", where p is a pri me. No t wanti ng to be accused of heresy, we fol l ow this
tradi ti on and pro ve the pretty and i mpo rtant
Theo rem 2.11.4. If G is a gro up of o rder p", where p is a pri me, then
Z(G), the center of G, is no t trivial (i.e., there exists an el ement a = e in G
such that ax = xa for al l x G G).
Proof. We shal l expl oit the cl ass equati o n to carry out the proof. L et
z = |Z(G)|; as we po i nted o ut previousl y, z is then the number of el ements
in G who se conjugacy cl ass has onl y o ne el ement. Since e G Z(G), z > 1.
Fo r any el ement b outside Z(G), its conjugacy cl ass contains mo re than o ne
el ement and |C(/3)| < |G|. Mo reo ver, since |C(/>)| divides ]G| by Lagrange's
theo rem, \C(b)\ = p
n{b
\ where 1 < n{b) < n. We divide the pieces of the
cl ass equati o n i nto two parts: that co mi ng from the center, and the rest. We
get, this way,
P" = \G\ = z + 2) r j j yp * + X zSy = ^ 2 P"-
lb)
-
6Z( G) \^K
U
J\ n(b)<nP n(f c)o i
Cl earl y, p divides the l eft-hand side, p", and divides 1
u{b)<n
p"~"
(b)
. The net
resul t of this is that p | z, and since z > 1, we have that z is at least p. So since
z = |Z(G)|, there must be an el ement a # e in Z(G), which pro ves the theo -
rem.
G\=î
G
(C(a)) = ^
M
a a
\C(a)\'
This l ast theo rem has an i nteresti ng appl ication, which so me readers
may have seen in sol ving Pro bl em 45 of Section 5. This is
104 Gr o u p s Ch . 2
Theo rem 2.11.5. If G is a gro up of o rder p
2
, where p is a pri me, then
G is abel ian.
Proof. By Theo rem 2.11.4, Z(G) # (e), so that there is an el ement, a,
of o rder p in Z(G). If A = (a), the subgro up generated by a, then A C Z(G),
hence A C C(x) for al l * G G. Gi ven x G G, x A, then C(x) D A and
.V G C(x); so |C(x)| > p, yet |C(x)| must divide p
2
. The net resul t of this is
that |C(x)| =p
2
, so C(x) = G, whence x G Z(G). Since every el ement of G
is in the center of G, G must be abel ian.
I n the pro bl ems to co me we shal l give many appl i cati ons of the nature
of gro ups of o rder p", where p is a pri me. The natural attack on virtually all
these problems follows the lines of the argument we are about to give. We
choose o ne of a wide possibl e set of choices to il l ustrate this techni que.
Theo rem 2.11.6. If G is a gro up of o rder p", p a pri me, then G con-
tains a no rmal subgro up of o rder p"~
l
.
Proof. We pro ceed by i nducti o n o n n. If n = 1, then G is of o rder p
and (e) is the requi red no rmal subgro up of o rderp
l
~
l
= p
a
=1.
Suppo se that we kno w that for so me k every gro up of o rder p
k
has a
no rmal subgro up of o rder p
k
~
l
. Let G be of o rder p
k+1
; by Theo rem 2.11.4
there exists an el ement a of o rder p in Z(G), the center of G. Thus the sub-
gro up A = (a) generated by a is of o rder p and is no rmal in G. Co nsi der T =
G/A; r is a gro up of o rder \G\I\A\ = p
k+1
lp = p
k
by Theo rem 2.6.3. Since T
has o rder p
k
, we kno w that T has a no rmal subgro up M of o rder p
k
~
1
. Since
T is a ho mo mo rphi c image of G, by the Co rrespo ndence Theo rem (Theo rem
2.7.2) there is a no rmal subgro up N in G, N D A, such that N/A = M. But
then we have
pk-i = |M| = \N/A\ = j ^j ,
that is, p
k
~
x
= \N\/p, l eading us to \N\ = p
k
. Thus TV is o ur requi red no rmal
subgro up in G of o rder p
k
. This compl etes the induction and so pro ves the
theo rem.
By far the mo st i mpo rtant appl ication we make of the cl ass equati o n is
the proof of a far-reaching theo rem due to Syl ow, a No rwegi an mathemati -
cian, who pro ved it in 1871. We al ready sho wed this theo rem to be true for
abel ian groups. We shal l no w pro ve it for any finite gro up. I t is impossibl e to
o verstate the i mpo rtance of Sylow's Theorem in the study of finite gro ups.
Wi tho ut it the subject wo ul d no t get off the gro und.
T heorem 2.11.7 (Sylow's T heorem). Suppo se that G is a gro up of
o rder p"m, where p is a pri me and p \ m. Then G has a subgro up of o rder p".
Proof. If n = 0, this is trivial . We therefo re assume that n > 1. Here,
again, we pro ceed by induction o n \ G\, assumi ng the resul t to be true for al l
gro ups Hsuch that \H\ < \ G\.
Suppo se that the resul t is fal se for G. Then, by o ur induction hypo the-
sis, p" cannot divide \H\ for any subgro up H of G if H # G. In particular, if
a < Z(G), then C(a) + G, hence p"\\C(a)\. Thus p divides \G\l\G{a)\ =
i
G
{C(a))lovaZ(G).
Wri te do wn the cl ass equati o n for G fol l owing the l ines of the argu-
ment in Theo rem 2.11.4. If z = |Z(G)|, then z > 1 and
p"m =\G\=z+ 2 io(C(a)).
it<Z(G)
But p | i
G
(C(a)) if a Z(G), so | 2
e z ( G )
/
G
(C()). Since p |p"m, we get
p ) z. By Cauchy's Theo rem there is an el ement a of o rder p in Z(G). If A is
the subgro up generated by A, then |A| = p and A < G, since A G Z(G). Co n-
sider T = G/A; |T| = |G|/|A| =p"mlp = p"~
l
m. Since |T| < |G|, by o ur in-
duction hypo thesi s T has a subgro up M of o rder p"~
l
. Ho wever, by the Cor-
respo ndence Theo rem there is a subgro up P of G such that PDA and
P/A = M. Therefo re, |P| = \M\ \A\ = p"~
1
p = p" and P is the sought-after
subgroup of G of order p", contradicting our assumption that G had no such
subgroup. This compl etes the induction, and Syl ow's Theo rem is establ ished.
Actual l y, Syl ow's Theo rem consists of three parts, of which we onl y
pro ved the first. The o ther two are (assumi ngp"m = |G|, wherep \ m) :
1. Any two subgro ups of o rder p" i n G are conjugate; that is, if |P| =
\Q\ = p" for subgro ups P, Q of G, then for so me x G G, Q = x~
1
Px.
2. The number of subgroups of o rder p" in G is of the form 1 + kp and di-
vides |G|.
Since these subgro ups of o rder p" po p up al l o ver the pl ace, they are
cal l ed p-Sylow subgroups of G. A n abel i an gro up has o ne p-Syl ow subgro up
for every pri me p dividing its o rder. This is far from true in the general case.
Fo r instance, if G = S
3
, the symmetri c gro up of degree 3, which has o rder 6
= 2- 3, there are three 2-Syl ow subgro ups (of o rder 2) and o ne 3-Syl ow sub-
gro up (or o rder 3).
Fo r tho se who want to see several proofs of that part of Syl ow's Theo -
rem which we pro ved abo ve, and of the o ther two parts, they mi ght l o o k at
the appro pri ate section of o ur bo o k Topics in Algebra.
106 Gr o u p s Ch . 2
PROBL EMS
Easi er Pro bl ems
1. I n 5
3
, the symmetri c gro up of degree 3, find al l the conjugacy cl asses, and
check the val idity of the cl ass equati o n by determi ni ng the o rders of the
central izers of the el ements of S
3
.
2. Do Pro bl em 1 for G the di hedral gro up of o rder 8.
3. If a G G, show that C(x~
l
ax) = x~
1
C(fl )x.
4. If cp is an auto mo rphi sm of G, show that C(cp(a)) = cp(C(a)) for a G.
5. If |G| =p
3
and |Z(G)| >p
2
, pro ve that G is abel ian.
6. If P is a /7-Syl ow subgro up of G and P < G, pro ve that P is the onl y
p-Syl ow subgro up of G.
7. If P < G, P a jD-Syl ow subgro up of G, pro ve that cp(P) = P for every
auto mo rphi sm cp of G.
8. Use the cl ass equati o n to give a pro o f of Cauchy's Theo rem.
If TP is a subgro up of G, l et N(H) = {x G G | x^/ Px = 7/7}. This oes
/7 >r mea/7 that xa = x whenever x G N(H), a G if. Fo r i nstance, if
H<G, then 7V(/7) = G, yet 7/ need no t be in the center of G.
9. Pro ve that N(H) is a subgro up of G, 77 C 7V(77) and in fact 77 < N(H).
10. Pro ve that 7V(x
_ x
Hx) = x "
1
N(77)x.
11. If P is a o-Syl ow subgro up of G, pro ve that P is a /j-Syl ow subgro up of
7V(P) and is the onl y o-Syl ow subgro up of N(P).
12. If P is a p-Syl ow subgro up and a G G is of o rder p'" for so me 7 7 7 , show
that if a~
l
Pa = P then a G P.
13. Pro ve that if G is a finite gro up and 77 is a subgro up of G, then the num-
ber of distinct subgro ups x
_ 1
77x of G equal s i
G
(N(H)).
14. If P is a p-Syl ow subgro up of G, show that the number of distinct x~
l
Px
cannot be a mul ti pl e of p.
15. If TV < G, l et 73(TV) = [x G G | XA = x for al l a G TV}. Pro ve that
B (TV) < G.
16. Show that a gro up of o rder 36 has a no rmal subgro up of o rder 3 or 9.
(Hi nt: See Pro bl em 40 of Section 5.)
17. Show that a gro up of o rder 108 has a no rmal subgro up of o rder 9 o r 27.
18. If P is a p-Syl ow subgro up of G, show that N(N(P)) = 7V(P).
19. If [G\ =p", show that G has a subgro up of o rder p
1
" for al l 1 < m < n.
Sec. 11
CONJUGACY AND SYLOW'S THEOREM (OPTIONAL) 107
20. If p'" divides |G|, show that G has a subgro up of o rder p'".
21. If |G| =p" and H + G is a subgro up of G, show that N(H) J //.
22. Show that any subgro up of o rder p""
1
in a gro up G of o rder p" is no rmal
in G.
Harder PROBLEMS
23. Let G be a gro up, H a subgro up of G. Define for a, b G G, a ~ b if /3 =
/?"
l
ah for so me /z G H. Pro ve that
(a) this defines an equi val ence rel ati on o n G.
(B) If [a] is the equival ence cl ass of a, sho w that if G is a finite gro up,
then [a] has m el ements where m is the i ndex of H fl C(a) in H.
24. If G is a gro up, 7f a subgro up of G, define a rel ation B ~ A for sub-
gro ups A, 5 of G by the condi ti on that B = h~
l
Ah for some h & H.
(a) Prove that this defines an equi val ence rel ati o n on the set of sub-
gro ups of G.
(B) If G is finite, show that the number of distinct subgro ups equival ent
to A equal s the i ndex of N(A) n H in H.
25. If is a p-Syl ow subgro up of G, l et S be the set of al l p-Syl ow subgro ups
of G. Fo r <2
1;
g
2
G S define G^~ g
2
if g
2
= a~
l
Q
x
a with a E. P. Pro ve,
using this rel ation, that if Q #i
3
, then the number of distinct a~
1
Qa, with
a G /\ is a mul tipl e of p.
26. Using the resul t of Pro bl em 25, sho w that the number of p-Syl ow sub-
gro ups of G is of the form 1 + kp. (This is the thi rd part of Syl ow's Theo -
rem.)
27. Let P be a p-Syl ow subgro up of G, and Q ano ther one. Suppo se that
Q + x~
1
Px for any x G G. L et S be the set of al l y~
1
Qy, as _y runs over G.
For g|, <2
2
G S define g
:
~ Q
2
if >
2
= a^gj a, where a G P.
(a) Show that this impl ies that the number of distinct y~
l
Qy is a mul ti pl e
of p.
(B) Usi ng the resul t of Pro bl em 14, show that the resul t of Part (a) can-
no t hol d.
(C) Prove from this that given any two /?-Syl ow subgroups P and Q of G,
then Q = x~
1
Px for so me x G G.
(This is the second part of Syl ow's Theo rem.)
28. If II is a subgro up of G of o rder p'" show that H is co ntai ned in so me
p-Syl ow subgro up of G.
29. If P is a p-Syl ow subgro up of G and a, b G Z(/
J
) are conjugate in G,
prove that they are al ready conjugate in N(P).
1 . PREL I MI NARI ES
Let us recal l a theo rem pro ved in Chapter 2 for abstract gro ups. This resul t,
kno wn as Cayley's Theorem (Theo rem 2.5.1), asserts that any gro up G is iso-
mo rphi c to a subgro up of A (S), the set of 1-1 mappi ngs of the set S o nto it-
sel f, for so me suitabl e S. I n fact, in the proof we gave we used for S the
gro up G itsel f vi ewed merel y as a set.
Historical l y, gro ups aro se this way first, l ong before the no ti o n of an
abstract gro up was defined. We find in the wo rk of Lagrange, Abel , Gal ois,
and o thers, resul ts o n gro ups of permutati o ns pro ved in the l ate ei ghteenth
and earl y ni neteenth centuri es. Y et it was no t unti l the mi d-ni neteenth cen-
tury that Cayl ey mo re o r l ess i ntro duced the abstract co ncept of a gro up.
Since the structure of i so mo rphi c groups is the same, Cayl ey's Theo rem
points out a certai n universal character for the gro ups A (S). If we knew the
structure of al l subgro ups of A (S) for any set S, we wo ul d kno w the structure
of al l groups. This is much to o much to expect. Neverthel ess, o ne coul d try to
expl oit this embeddi ng of an arbi trary gro up G isomorphical l y i nto so me
A(S). This has the advantage of transforming G as an abstract system into
so methi ng mo re co ncrete, namel y a set of nice mappi ngs of so me set o nto
itsel f.
We shal l no t be co ncerned with the subgro ups of A (S) for an arbi trary
set S. If S is infinite, A (S) is a very wil d and compl i cated object. Even if S is
finite, the co mpl ete nature of A (S) is virtual l y impossibl e to determi ne.
1flR
Sec. 1 Pr el i mi n ar i es 109
I n this chapter we consider onl y A (S) for S a finite set. Recal l that if S
has n el ements, then we cal l A (S) the symmetric group of degree n, and de-
no te it by S. The el ements of S are cal l ed permutations; we shal l deno te
them by l owercase Greek l etters.
Since we mul tipl ied two el ements <x, T E A(S) by the rul e (ar)(s) =
(T(T(S)) this wil l have the effect that when we i ntro duce the appro pri -
ate symbol s to represent the el ements of S, these symbol s, or permutati o ns,
wil l mul tipl y from right to left. If the readers l ook at so me o ther bo o k o n al -
gebra, they shoul d make,sure which way the permutati o ns are being mul ti-
pl ied: right to l eft o r l eft to right. Very often, al gebraists mul tipl y permuta-
tions from left to right. To be consistent with o ur definition of the
composition of el ements in 5, we do it from right to l eft.
By Cayl ey's Theo rem we kno w that if G is a finite gro up of o rder n,
then G is i somorphi c to a subgro up of S and 5 has n\ el ements. Speaki ng
l oosel y, we usual l y say that G is a subgro up of S. Since n is so much smal l er
than n\ for n even modestl y l arge, o ur gro up occupies onl y a tiny l ittl e corner
in S. It woul d be desirabl e to embed G in an S for n as smal l as possibl e. For
certain cl asses of finite groups this is achievabl e in a particul arl y nice way.
Let S be a finite set having n el ements; we might as wel l suppose that
S = {x
{
, x
2
,..., x}. Gi ven the permutati o n cr E S = A(S), then
cr(x
k
) E S for k = 1, 2, . . . , n, so cr(x
k
) = x
t
for so me i
k
,l ^ i
k
n. Because
a is 1-1, if j = k, then x
f
= cr(x
;
) i= cr(x
k
) = x
ik
. Therefore, the numbers
i
2
,..., / are merel y the numbers 1,2,... ,n shuffl ed about in some order.
Cl earl y, the action of cr on S is determi ned by what cr do es to the sub-
script of x
;
, so the symbol "x" is real l y excess baggage and, as such, can be
discarded. I n short, we may assume that S = [1, 2,..., n}.
Let's recal l what is meant by the product of two el ements of A (S). If
<r, r E A(S), then we defined err by (OTT){S) = CT(T(S)) for every s L S. We
showed in Section 4 of Chapter 1 that A (S) satisfied four pro perti es that we
used l ater as the mo del to define the no ti o n of an abstract gro up. Thus S, in
particul ar, is a gro up rel ative to the pro duct of mappi ngs.
Our first need is so me handy way of deno ti ng a permutati o n, that is, an
el ement a in S. One cl ear way is to make a tabl e of what a do es to each el e-
ment of S. This mi ght be cal l ed the graph of a. We did this earl ier, writing
out cr, say cr E S
3
, in the fashion: cr: .Vj >x
2
, x
2
>x
3
, x
3
> X ] . But this is
cumberso me and space consuming. We certainl y can make it mo re compact
the second row is the i mage under cr of the number in the first row directl y
above it. There is no thi ng hol y abo ut 3 in al l this; it works equal l y wel l for
any n.
110 Th e Sy mme t r i c Gr o u p Ch . 3
If cr G 5 and cr(l ) = r'
x
, <x(2) = i
2
, ., cr() = /, we use the symbo l
f ?
W
) to represent <x and we write a = ( } ? " ]. No te
'i '2 ' i
n
) \h h h,)
that it is no t necessary to write the first ro w in the usual o rder 1 2 n;
any way we wri te the first row, as l ong as we carry the z-'s al ong accord-
ingl y, we stil l have a. Fo r i nstance, in the exampl e in S
3
cited,
'l 2 3\ /3 1 2\ /2 1 3^
" \2 3 l j U 2 3/ \3 2 1
1 2 w\
If we kno w a = [ . . . , what is cr
- 1
? I t is easy, just fl ip the
h
-
" " l
symbol for <r over to get a = ^ | _ _ _ ^j . (Prove!) I n o ur exampl e
The identity el ement
1 2 3\ _!
=
/ 2 3 1\
=
1 2 3
2 3 1 ' " " 1 2 3 \3 1 2
fl 2
which we shal l wri te as eis merel y e = ( ^ ^
n'
'
n
)
Ho w does the pro duct in S transl ate in terms of these symbol s? Since
err means: "First appl y T and to the resul t of this appl y cr," in forming the
pro duct of the symbol s for a and T we l ook at the number k in the first ro w
of T and see what number i
k
is directl y bel ow k in the seco nd ro w of T. We
then l ook at the spot i
k
in the first row of cr and see what is directl y bel o w it
in the second row of cr. This is the i mage of k under ar. We then run thro ugh
k = 1, 2, . . ., n and get the symbol for CTT. We just do this visual l y.
We il l ustrate this with two permutati o ns
1 2 3 4 5\ , / 1 2 3 4 5
2 3 1 5 4
a n d T =
3 4 5 1 2
in Then < J T -
1 2 3 4 5
1 5 4 2 3
Even the eco no my achieved this way is no t eno ugh. After al l , the first
row is al ways 1 2 n, so we coul d dispense with it, and wri te cr =
( } ? " ) as (z'j, i
2
, , i). This is fine, but in the next section we
\h h ' ' '
l
n)
shal l find a better and briefer way of representi ng permutati o ns.
PROBL EMS
1. Fi nd the pro ducts:
1 2 3 4 5 6\/ l 2 3 4 5 6
( a )
16 4 5 2 1 3A 2 3 4 5 6 1
Sec . 2 Cy c l e Dec o mp o s i t i o n 111
/ l 2 3 4 5\ 1 2 3 4 5
( b )
^2 1 3 4 5j \3 2 1 4 5
(1 2 3 4 5 V
1
/
1
2 3 4 5\ / l 2 3 4 5^
( c )
^4 1 3 2 5/ ^2 1 3 4 5J\4 1 3 2 5
/
2. Eval uate al l the po wers of each permutati o n cr (i.e., find cr* for al l k).
(a)
( b )
( c )
'1 2 3 4 5 6
v
2 3 4 5 6 1
1 2 3 4 5 6 7
2 1 3 4 6 5 7
'l 2 3 4 5 6
,6 4 5 2 1 3
3. Prove that
1 2 n \
1
/ z
t
z
2
z',
1 2 n
4. Fi nd the o rder of each el ement in Pro bl em 2.
5. Find the o rder of the pro ducts yo u o btai ned in Pro bl em 1.
2. CYCL E DECOMPOSI TI ON
We co nti nue the process of simpl ifying the no tati o n used to represent a given
permutati o n. I n doing so, we get so methi ng mo re than just a new symbol ; we
get a device to deco mpo se any permutati o n as a pro duct of particul arl y nice
permutati o ns.
Definition. Let i
x
, i
2
, . . . , i
k
be k distinct integers in S = {1, 2,..., zz).
The symbol (z'j i
2
i
k
) wil l represent the permutati o n cr G S, where
o-(i
L
) = i
2
, cr{i
2
) = z
3
,. . ., cr(z
;
.) = for / < k, a(i
k
) = i
u
and
CJ (S) = s for any s G 5 if s is different from z
l 5
i
2
, , i
k

Thus, in S
7
, the permutati o n (1 3 5 4) is the permutati o n
( 3 2 5 I 4 6 7 ) ' ^
C C a
^
a
P
e r m u t a t
i
n
f
t n e
f
r m
(*i h ' ' ' 'A )
a k-cycle. Fo r the special case k = 2, the permutati o n (z, z
2
) is cal l ed a rraws-
position. No te that if cr = (z\ z'
2
i
k
), then cr is al so (i
k
z, z'
2
i
k
-\),
(i/c-i i
k
i\ ii ' ' ' 4- 2) )
a n
d
s o o n
- (Prove!) Fo r exampl e,
(1 3 5 4) = (4 1 3 5) = (5 4 1 3) =(3 5 4 1).
Two cycl es, say a /ocycl e and an m-cycl e, are said to be disjoint cycles if
112 Th e Sy mmet r i c Gr o u p Ch , 3
they have no i nteger in co mmo n. Whence (1 3 5) and (4 2 6 7) in S
are disjoint cycl es.
Gi ven two disjoint cycl es in S, we cl aim that they co mmute. We l eav<
the proof of this to the reader, with the suggestion that if cr, r are disjoin
cycl es, the reader shoul d verify that (O"T)(7) = (TCT)(Z) for every i G S =
[1,2,... ,n). We state this resul t as
L emma 3.2.1. If cr, T G S arc disjoint cycl es, then ar = rcr.
Let's consider a parti cul ar /ocycl e a = (1 2 k) in S. Cl earl y
cr(l ) = 2 by the definition given abo ve; ho w is 3 rel ated to 1? Since cr(2) = 3
we have cr
2
(l ) = cr(2) = 3. Conti nui ng, we see that cr
;
'(l ) = /' + 1 foi
j < k - 1, whil e a
k
(1) = 1. I n fact, we see that cr
k
= e, where e is the identity
el ement in S
n
.
There are two things to be co ncl uded from the paragraph abo ve.
1. The o rder of a k-cycl e, as an el ement of S, is k. (Pro ve!)
2. If cr = (i
t
i
2
i
k
) is a A>cycl e, then the orbit of i
x
under a (see
Pro bl em 27 in Section 4 of Chapter 1) is {i,, i
2
, , i
k
}. So we can see
that the /c-cycl e cr = (/j i
2
i
k
) is
<r=(i
v
o-ih) a
2
(h) er * -
1
^) ) .
Gi ven any permutati o n Tin S, for / G {1, 2, . . . , }, co nsi der the orbh
of i under r. We have that this orbit is [i, T ( I ) , r
2
(i),. . . , T
s _ 1
(;')}, where
T
S
(Z') = i and s is the smal l est positive i nteger with this pro perty. Co nsi der the
5-cycl e (i r(z') T
2
( Z) r
s _ 1
( / ) ) ; we cal l it the cycle of T determined by i.
We take a specific exampl e and find al l its cycl es. Let
1 2 3 4 5 6 7 8 9\
3 9 4 1 5 6 2 7 8/
what is the cycl e of T determi ned by 1? We cl aim that it is (1 3 4). Why?
T takes 1 into 3, 3 into 4 and 4 i nto 1, and since T(1) = 3, r
2
( l ) = T(3) = 4,
T
3
(1) = T(4) = 1. We can get this visual l y by weaving thro ugh
with the thin path. What is the cycl e of T determi ned by 2? Weavi ng
thro ugh
Sec . 2 Cy c l e Dec o mp o s i t i o n 113
with the thin path, we see that the cycl e of T determi ned by 2 is
(2 9 8 7). The cycl es of T determi ned by 5 and 6 are (5) and (6), respec-
tivel y, since 5 and 6 are l eft fixed by T. SO the cycl es of T are (1 3 4),
(2 9 8 7), (5), and (6). Therefore we have that T = (1 3 4)(2 9 8 7)
(5)(6), where we view these cycl esas defined abo veas permutati o ns in S
9
because every integer in S = {1, 2,. . ., 9} appears in o ne and onl y one cycl e,
and the image of any i under r is read off from the cycl e in which it appears.
There is no thi ng special abo ut the permutati o n T above that made the
argument we gave go thro ugh. The same argument woul d ho l d for any per-
mutati o n in S for any n. We l eave the formal writing do wn of the pro o f to
the reader.
Theo rem 3.2.2. Every permutati o n in S is the product of disjoint
cycl es.
I n wri ti ng a permutati o n cr as a pro duct of disjoint cycl es, we o mi t al l
I-cycl es; that is, we i gno re the z's such that cr(/) = i. Thus we wri te
cr = (1 2 3)(4 5)(6)(7) simpl y as cr = (1 2 3)(4 5). I n o ther wo rds, writing
eras a pro duct of /c-cycl es, with k > 1, we assume that cr l eaves fixed any in-
teger no t present in any of the cycl es. Thus in the gro up S
n
the permuta-
tion r = (1 5 6)(2 3 9 8 7) l eaves fixed 4, 10, and 11.
L emma 3.2.3. If T in S is a fc-cycl e, then the o rder of 7 is k; that is,
r
k
= e and 7' i= e for 0 < j < k.
Co nsi der the permutati o n 7 = (1 2)(3 4 5 6)(7 8 9) in S
9
.
What is its o rder? Since the disjoint cycl es (1 2), (3 4 5 6), (7 8 9)
co mmute, T'" = (1 2)
, n
(3 4 5 6)'"(7 8 9)"'; in o rder that r'" = e we
need (1 2)'" = e, (3 4 5 6)'" = e, (7 8 9)"' = e. (Prove!) To have
(7 8 9)"
!
= e, we must have 3 | m, since (7 8 9) is of o rder 3; to have
(3 4 5 6)'" = e, we must have 4 | m, because (3 4 5 6) is of o rder 4,
and to have (1 2)"' = e, we must have 2 | m, because (1 2) is of o rder 2.
This tel l s us that m must be divisibl e by 12.
On the o ther hand,
T
1 2
= (1 2)
1 2
(3 4 5 6)
1 2
(7 8 9)
1 2
= e.
So T is of o rder 12.
Here, again, the special pro perti es of 7 do no t enter the. picture. What
we did for 7 wo rks for any permutati o n. To fo rmul ate this properl y, recal l
that the l east co mmo n mul ti pl e of m and n is the smal l est positive i nteger v
which is divisibl e by m and by n. (See Pro bl em 7, Chapter 1, Section 5.) Then
we have
114 Th e Sy mmet r i c Gr o u p Ch . 3
Theo rem 3.2.4. Let cr E S have its cycl e deco mpo si ti o n into disjoint
cycl es of l ength m,, m
2
, , m
k
. Then the o rder of cr is the l east co mmo r
mul tipl e of m
u
m
2
, , ni
k
.
Proof. Let cr = T
X
T
2
r
k
, where the T,- are disjoint cycl es of l ength m
t
Since the r, are disjoint cycl es, T,T, = TJ T,-; therefore if M is the l east common
mul ti pl e of m
b
m
2
,. . . , m
k
, then cr'
1
' = (T
X
T
2
r
k
)
M
= T^T
2
r
k
M
= e
(since rf~ e because T, is of o rder m, and m
t
\ M). Therefo re, the o rder ol
cr is at most M. On the o ther hand, if cr
N
= e, then T^T
2
r
k
= e. Thb
forces each T ' / =e, (prove!) because r,- are disjoint permutati o ns, so m-
t
\ A
r
since T, is of o rder m,. Thus N is divisibl e by the l east co mmo n mul tipl e oi
m
l
, m
2
,. . ., m
k
, so M | 7Y. Co nsequentl y, we see that <J is of o rder M as
cl aimed in the theo rem.
No te that the disjointness of the cycl es in the theo rem is i mperati ve
Fo r instance, (1 2) and (1 3), which are not disjoint, are each of o rder 2
but their pro duct (1 2)(1 3) = (1 3 2) is of o rder 3.
Let's consider Theo rem 3.2.4 in the context of a card shuffl e. Suppose
that we shuffl e a deck of 13 cards in such a way that the to p card is put intc
the position of the 3rd card, the second in that of the 4th,.. ., the ith i nto the
i + 2 position, wo rki ng mo d 13. As a permutati o n, cr, of 1, 2, . . . , 13, the
shuffl e beco mes
and o-is merel y the 13-cycl e (1 3 5 7 9 11 13 2 4 6 8 10 12),
so a is of o rder 13. Ho w many ti mes must we repeat this shuffl e to get the
cards back to thei r original o rder? The answer is merel y the o rder of cr, that
is, 13. So it takes 13 repeats of the shuffl e to get the cards back to thei r origi-
nal order.
Let's give a twist to the shuffl e abo ve. Suppo se that we shuffl e the
cards as fol l ows. First take the to p card and put it i nto the second-to-l ast
pl ace, and then fol l ow it by the shuffl e given abo ve. Ho w many repeats are
no w needed to get the cards back to thei r original o rder? The first o perati o n
is the shuffl e given by the permutati o n T = (1 12 11 10 9 8 7 6 5
4 3 2) fol l owed by cr abo ve. So we must co mpute err and find its o rder.
But
1 2 3 4 5 6 7 8 9 10 11 12 13
3 4 5 6 7 8 9 10 11 12 13 1 2
o r = (1 3 5 7 9 11 13 2 4 6 8 10 12)
X (l 12 11 10 9 8 7 6 5 4 3 2)
= (1)(2 3 4 5 6 7 8 9 10 11 12 13),
Sec. 2
Cy c l e Dec o mp o s i t i o n 115
so is of o rder 12. So it woul d take 12 repeats of the shuffl e to get back to the
original order.
Can you find a shuffl e of the 13 cards that woul d requi re 42 repeats? Or
20 repeats? What shuffl e woul d requi re the greatest number of repeats, and
what woul d this number be?
We return to the general discussion. Co nsi der the permutati o n
(1 2 3); we see that ( 1 2 3) = (1 3)(1 2). We can al so see that
(1 2 3) = (2 3)(1 3). So two things are evident. First, we can write
(1 2 3) as the pro duct of two transposi ti ons, and in at l east two distinct
ways. Given the &-cycl e (i
{
i
2
/*), then (i
i
i
2
i
k
) =
(/[ '' ' 0i h)>
s o
every /t-cycl e is a pro duct of k 1 transposi-
tions (if k > 1) and this can be do ne in several ways, so not in a uni que way.
Because every permutati o n is the pro duct of disjoint cycl es and every cycl e is
a pro duct of transposi ti ons we have
Theo rem 3.2.5. Every permutati o n in S is the pro duct of transposi-
tions.
This theo rem is real l y no t surprising for it says, after al l , nothi ng mo re
or l ess than that any permutati o n can be effected by carrying o ut a series of
interchanges of two objects at a time.
We saw that there is a l ack of uni queness in representi ng a given per-
mutation as a pro duct of transposi ti ons. But, as we shal l see in Section 3,
some aspects of this deco mpo si ti o n are i ndeed uni que.
As a final wo rd of this section, we woul d l ike to point o ut the conve-
nience of cycl e no tati o n. When we represent el ements of a permutati o n
group as pro ducts of disjoint cycl es, many things beco me transparentfo r
exampl e, the o rder of the permutati o n is visibl e at a gl ance. To il l ustrate this
point, we now give a few exampl es of certain geo metri c groups, which are in
fact permutati o n gro ups that have al ready appeared in Chapter 2 under dif-
ferent guises.
Examples
1. I nformal l y, a motion of a geo metri c figure is a permutati o n of its vertices
that can be real ized by a rigid mo ti o n in space. Fo r exampl e, there are eight
moti ons of a square, who se vertices are numbered 1, 2, 3, 4 as bel ow.
4 3
1
2
116 The Symmetri c Group Ch. 3
1 2
the refl ections in the two axes of symmetry, ro tati o n by 180 and the identity.
These mo ti o ns can be identified with permutati o ns (1), (14)(23), (12)(34),
(13)(24), and form a subgro up of the gro up o btai ned in Exampl e 1. This sub-
gro up is often cal l ed Klein's 4-group.
3. We l eave it to the reader to verify that the gro up of al l mo ti o ns of an
equi l ateral triangl e is the ful l symmetri c gro up S
3
.
3
1 2
4. The mo ti o ns of a regul ar hexago n form the di hedral gro up of o rder 12,
generated by the permutati o ns a = (15)(24), co rrespo ndi ng to a refl ection
a = (13) is the refl ection abo ut the axis of symmetry j oi ni ng vertices 2 and
4 in the original po si ti o n,
3 =(1234) is the countercl ockwi se ro tati o n by 90,
p
2
=(13)(24) is the countercl ockwi se ro tati o n by 180,
P
3
=(1432) is the countercl ockwi se ro tati o n by 270,
a(i =(12)(34) is the refl ection in the vertical axis of symmetry,
ap
2
=(24) is the refl ection in the o ther diagonal axis,
ap
3
=(14)(23) is the refl ection in the ho ri zo ntal axis, and, of co urse
a
2
=(3
4
= (1) is the "mo ti o n" that l eaves the vertices unchanged.
We al so have the rel ation (3a = a(3
3
.
These mo ti o ns, or symmetries of a square, form a subgro up of S
4
which is
cal l ed the octic group, or the dihedral group of o rder 8. This gro up (or,
strictl y speaking, a gro up i so mo rphi c to it) was i ntro duced in Exampl e 9 of
Section 2.1 wi tho ut menti o n of permutati o ns.
2. There are onl y four symmetri es of a no n-square rectangl e:
4 3
Sec. 2 Cycl e Decomposition 117
abo ut o ne of the axes of symmetry, and p = (123456), correspondi ng to the
countercl ockwi se ro tati o n by 60.
5 4
1 2
I n general , the dihedral gro up of o rder 2n, which was first i ntro duced in Ex-
ampl e 10 of Section 2.1, can be i nterpreted as the gro up of symmetri es of a
regul ar /7-gon (a pol ygon with n edges of equal l ength).
PROBL EMS
Easi er Pro bl ems
1. Show that if cr, T are two disjoint cycl es, then en = TI T.
2. Fi nd the cycl e decomposi ti on and o rder.
7 8 9^
(a)
(b)
(c)
1 2 3 4 5 6
3 1 4 2 7 6
1 2 3 4 5 6
7 6 5 4 3 2
1 2 3 4 5 6
7 6 5 3 4 2
9 8 5
7
1
7 W 1 2 3 4 5 6 7
l / \ 2 3 1 5 6 7 4/ '
3. Express as the pro duct of disjoint cycl es and find the order.
(a) (1 2 3 5 7)(2 4 7 6).
(b) (1 2)(1 3)(1 4).
(c) (1 2 3 4 5)(1 2 3 4 6)(1 2 3 4 7).
(d) (1 2 3)(1 3 2).
(e) (1 2 3) ( 3 5 7 9)(1 2 3)
_ 1
.
(f) (1 2 3 4 5)
3
.
4. Gi ve a co mpl ete proof of Theo rem 3.2.2.
5. Show that a /c-cycl e has o rder k.
6. Fi nd a shuffl e of a deck of 13 cards that requi res 42 repeats to return the
cards to thei r original order.
7. Do Pro bl em 6 for a shuffl e requi ri ng 20 repeats.
8. Express the permutati o ns in Pro bl em 3 as the pro duct of transposi ti ons.
9. Given the two transposi ti ons (1 2) and (1 3), find a permutati o n cr
such that cr(l 2)a~
l
= (1 3).
10. Pro ve that there is no permutati o n crsuch that cr(l 2)cr~
1
= (1 2 3).
11. Pro ve that there is a permutati o n cr such that cr(l 2 3)cr
_ 1
=
(4 5 6).
12. Pro ve that there is no permutati o n cr such that cr(l 2 3)cr~
l
=
(1 2 4)(5 6 7).
13. Pro ve that (1 2) canno t be wri tten as the pro duct of disjoint 3-
cycl es.
14. Pro ve that for any permutati o n cr, cn-cr
- 1
is a transpo si ti o n if r is a trans-
position.
15. Show that if r is a /c-cycl e, then CTTO-^
1
is al so a A;-cycl e, for any permuta-
tion cr.
16. Let $ be an auto mo rphi sm of S
3
. Show that there is an el ement a G S
3
such that <E>(T) = O~~
1
TCT for every T G S
3
.
17. Let (1 2) and (1 2 3 n) be in S. Show that any subgro up of S
that contains bo th of these must be al l of S (so these two permutati o ns
generate S).
18. If T, and T
2
are two transposi ti ons, show that T
X
T
2
can be expressed as the
pro duct of 3-cycl es (not necessaril y disjoint).
19. Pro ve that if T
1 ;
T
2
, and T
3
are transposi ti ons, then T
1
T
2
T
3
J= e, the identity
el ement of S.
20. If T
T
, T
2
are distinct transposi ti ons, show that T
V
T
2
is of o rder 2 o r 3.
21. If cr, T are two permutati o ns that disturb no co mmo n el ement and err =e,
pro ve that cr =T =e.
22. Fi nd an al gorithm for finding crTcr"
1
for any permutati o ns cr, T of 5.
23. Let cr, T be two permutati o ns such that they bo th have deco mpo si ti o ns
into disjoint cycl es of cycl es of l engths m
x
, m
2
,. . . , m
k
. (We say that
they have simil ar deco mpo si ti o ns into disjoint cycl es.) Prove that for
so me permutati o n p, r = pcrp
- 1
.
24. Fi nd the conjugacy cl ass in S of (1 2 n). What is the o rder of the
central izer of (1 2 n) in S?
25. Do Pro bl em 24 for a = (1 2) (3 4).
Sec . 3
Od d an d Ev en Per mu t at i o n s 119
3 . ODD A N D EVEN PERMUTA TI ONS
We noticed in Section 2 that al though every permutati on is the product of trans-
positions, this decomposition is not unique. We did comment, however, that cer-
tain aspects of this kind of decomposition are unique. We go into this now.
Let's consider the special case of S
3
, for here we can see everythi ng ex-
pl icitl y. Let f(x
u
x
2
, x
3
) = (x
x
- x
2
)(x, - x
3
)(x
2
~ x
3
) be an expression in
the three variabl es X , , x
2
, x
3
. We l et S
3
act o n f(x) = f(x
u
x
2
, x
3
) as fol l ows.
If o-G S
3
,then
<r*(f(x)) =(-Vl)
_
^(2))(^(l) " -^(3))(*
CT
(2) - ^(3))-
We consider what cr* does to f(x) for a few of the o-'s in S
3
.
Co nsi der cr = (1 2). Then <x(l ) = 2, o-(2) = 1, and a(3) = 3, so that
o-*(.f(x)) = (-Vt) ~ x
cr(2)
)(x^
V)
- x
a(3)
)(x\
r(2}
- x
( r ( 3 )
)
= ( X
2
- x) (x
2
- x
3
) (X j - x
3
)
= - ( x, - x
2
)(x
1
- x
3
)(x
2
- x
3
)
= - / ( * )
So a'* coming from cr = (1 2) changes the sign of f(x). Let's l ook at the ac-
tion of ano ther el ement, T = (1 2 3), of S
3
o n/ (x). Then
T*(f(x)) = (x
T ( 1 )
- x
T ( 2 )
) ( x
T ( l )
- X
T ( 3 )
) ( X
t ( 2 )
- x
T ( 3 )
)
= (x
2
- x
3
)(x
2
- x
x
)(x
3
- x^
= (x, - X
2
) (X j - x
3
) (x
2
- x
3
)
so f* coming from T = (1 2 3) leaves / (x) unchanged. What abo ut the
other permutati o ns in S
3
; ho w do they affect/(x)? Of course, the identity el -
ement e i nduces a map e* on /(x) which do es not change / (x) at al l . What
does T
2
, for T abo ve, do to /(x)? Since r*f(x) = f(x), we immediatel y see that
(T
2
) * (/(x)) = (xv
( 1 )
- xv
( 2 )
)(x>
( 1 )
- x>
( 3 )
)(xv
( 2 )
- x>
( 3 )
)
= / (x). (Pro ve!)
Now consider err = (1 2)(1 2 3) = (2 3); since T l eaves /(x) al one and cr
changes the sign of f(x), o rmust change the sign of f (x). Simil arl y, (1 3) changes
the sign of f(x). We have accounted for the action of every el ement of S
3
on f(x).
Suppo se that p E S
3
is a pro duct p = T
X
T
2
r
k
of transposi ti ons
T
U
. . . , T
K
; then p acting on /(x) wil l change the sign of f(x) k ti mes, since
each T, changes the sign of fix). So p*(f(x)) = (~l)
k
f(x). If p =o-jcr, o>.
where <j
1 ;
. . ., a, are transposi ti ons, by the same reaso ni ng, p*(/(x)) =
(-1)7W- Therefo re, = (~1)7W. whence ( - 1) ' =(-1)*. This
tel l s us that t and k have thesame parity; that is, if t is o dd, then /<; must be
odd, and if t is even, then k must be even.
This suggests that al though the deco mpo si ti o n of a given permutati o n a
as a pro duct of transpo si ti o n is no t uni que, the parity of the number of trans-
positions in such a decomposition of a might be unique.
We strive for this goal now, suggesting to readers that they carry out
the argument that we do for arbi trary n for the special case n =4.
As we did abo ve, define f(x) = f(x
h
. . . , x) to be
fix) = (x
x
- x
2
) (*! - x) {x
2
- x
3
) (x
2
- x) (,*_! - x)
= EI (*,- - xj),
where in this pro duct i takes o n al l val ues from 1 to n 1 incl usive, and j al l
those from 2 to n incl usive. If <x E 5, define cr* on f(x) by
cx*(/(x)) = ri(^(o - w -
Kj
If cr, T E S, then
(crr)*(/(x)) = n (x
(
m
- x
(!TT)(j)
) = a* ( n (x
T ( / )
- x
T(/ )
)j
= o-* ( T * (x,. -
X]
)Jj =cr*(r*(/(x))) = (cr*r*)(/(x))
So (err)* =cr*T* when appl i ed to / (x).
What does a transpo si ti o n r do to f{x)l We cl aim that T * ( / ( X ) ) =
fix). To pro ve this, assumi ng that T =(/ where r <j , we count up the
number of (x
u
x
v
), with u < v, which get transfo rmed into an (x x
b
) with
a > b. This happens for (x x
;
) if i < u < j , for (x, x) if i < v < j , and fi-
nal l y, for (xj x
;
). Each of these l eads to a change of sign on /(x) and since
there are 2(j - z' 1) +1 such, that is, an o dd number of them, we get an
o dd number of changes of sign o n fix) when acted o n by T*. Thus
T * ( / ( X ) ) = ~fix). Therefo re, o ur cl aim that r*(/(x)) = -fix) for every
transposi ti on r is substanti ated.
Sec . 3 Od d an d Ev en Per mu t at i o n s 121
If cr is any permutati o n in S and cr T
]
T
2
r
k
, where T
{
, T
2
, . . ., r
k
are transpositions, then cr* = (T
X
T
2
' ' '
r
k)*
= T
*
T
* T | as acting on /(x),
and since each T * ( / ( X ) ) = - / ( x) , we see that a*(f(x)) = (~l)
k
f(x). Simi-
l arl y, if cr = i
2
" " ' r> where
1 ;

2
, . . . , f
a r e
transpositions, then
(j *(/(x)) = (l )
f
/(x). Co mpari ng these two eval uations of cr*(/(x)), we
concl ude that (- 1)* = (- 1)'. So these two deco mpo si ti o ns of cr as the pro d-
uct of transposi ti ons are of the same parity. Thus any permutation is either
the product of an odd number of transpositions or the product of cm even
number of transpositions, and no product of an even number of transpositions
zan equal a product of an odd number of transpositions.
This suggests the fol l owing
Definition. The permutati o n cr E S
n
is an odd permutation if cr is the
product of an o dd number of transpo si ti o ns, and is an even permutation if a
is the pro duct of an even number of transpo si ti o ns.
What we have pro ved abo ve is
Theo rem 3.3.1. A permutati o n in S is ei ther an odd or an even per-
mutation, but cannot be both.
Wi th Theo rem 3.3.1 behi nd us we can deduce a number of its conse-
quences.
Let A be the set of al l even permutati o ns; if cr, T G A, then we i mme-
diatel y have that ar G A. Since A is thus a finite cl osed subset of the (fi-
nite) gro up S,A is a subgro up of S, by L emma 2.3.2. A is cal l ed the alter-
nating group of degree n.
We can show that A is a subgro up of S in ano ther way. We al ready
saw that A is cl osed under the pro duct of S, so to kno w that A is a sub-
gro up of S we merel y need show that cr G 5 impl ies that cr
- 1
G S. Fo r any
permutati o n cxwe cl aim that crand a~
]
are of the same parity. Why? Wel l , if
a = T
X
T
2
r
k
, where the r
(
- are transpo si ti o ns, then
O-"
1
= ( 7i T
2
T
k
y
l
= T^T/ T
1
! T
2
L
T
X
L
= T
K
T
K
\ ' T
2
7, ,
since rf
1
= r,. Therefo re, we see that the pari ty of a and a'
1
is (-l)
k
, so
they are of equal parity. This certainl y shows that a G A forces <r
- 1
G A
n
,
whence A is a subgro up of S.
But it shows a l ittl e mo re, namel y that A is a normal subgroup of S.
For suppo se that cr G A and p G 5. What is the parity of p~
l
o-pl By the
above, p and p~
l
are of the same parity and cr is an even permutati o n so p~
l
ap
is an even permutati o n, hence is in A. Thus A is a no rmal subgro up of S.
We summari ze what we have do ne in
Theo rem 3.3.2. A, the al ternati ng group of degree n, is a no rmal
subgro up of S.
We l ook at this in yet ano ther way. Fro m the very definitions invol ved
we have the fol l owing simpl e rul es for the pro duct of permutati o ns:
1. The pro duct of two even permutati o ns is even.
2. The pro duct of two o dd permutati o ns is even.
3. The pro duct of an even permutati o n by an o dd o ne (or of an o dd o ne
by an even o ne) is odd.
If cr is an even permutati o n, l et 8 (a) = 1, and if cr is an o dd permuta-
tion, l et 0(cr) = 1. The foregoing rul es abo ut pro ducts transl ate into
8(O-T) = 9(O-)9(T), SO 9 is a homomorphism of S o nto the gro up E = {1, 1}
of o rder 2 under mul tipl ication. What is the kernel , N, of 91 By the very defi-
nition of A we see that N = A. So by the First Ho mo mo rphi sm Theo rem,
E S/A. Thus 2 = \E\ = \S
n
/A\ = |5|/|A|, if n > 1. Thi s gives us that
Therefo re,
Theo rem 3.3.3. Fo r n > 1, A
n
is a no rmal subgro up of S of o rder 7;!.
Corol l ary. Fo r n > 1, S contai ns \n\ even permutati o ns and \n\ o dd
permutati o ns.
A final few words abo ut the pro o f of Theo rem 3.3.1 before we cl ose
this section. Many different proofs of Theo rem 3.3.1 are kno wn. Qui te
frankl y, we do not particul arl y l ike any of them. So me invol ve what might be
cal l ed a "col l ection pro cess," where one tries to show that e canno t be writ-
ten as the pro duct of an o dd number of transposi ti ons by assumi ng that it is
such a shortest pro duct, and by the appro pri ate finagl ing with this pro duct,
sho rteni ng it to get a contradi cti on. Other proofs use o ther devices. The
pro o f we gave expl oits the gimmick of the function f(x), which, in so me
sense, is extraneo us to the who l e affair. Ho wever, the pro o f given is pro babl y
the most transparent of them al l , which is why we used it.
Final l y, the gro up A, for n > 5, is an extremel y i nteresti ng gro up. We
Sec . 3 Od d an d Ev en Per mu t at i o n s 123
shal l show in Chapter 6 that the onl y no rmal subgro ups of A, for . > 5, are
(e) and A itsel f. A gro up with this pro perty is cal l ed a simple group (not to
be confused with an easy gro up). The abel ian finite simpl e groups are merel y
the groups of pri me order. The A for n > 5 pro vi de us with an infinite fam-
il y of nonabelian finite simpl e groups. There are other infinite famil ies of finite
simpl e groups. I n the l ast 20 years or so the heroic efforts of al gebraists have
determi ned al l finite simpl e groups. The determi nati o n of these simpl e groups
runs about 10,000 printed pages. I nterestingl y eno ugh, any nonabel ian finite
simpl e group must have even order.
PROBL EMS
Easi er Pro bl ems
1. Fi nd the parity of each permutati o n.
/ I 2 3 4 5 6 7 8 9\
( a )
^2 4 5 1 3 7 8 9 6 |
(b) (1 2 3 4 5 6)(7 8 9).
(c) (1 2 3 4 5 6)(1 2 3 4 5 7).
(d) (1 2)(1 2 3 ) ( 4 5)(5 6 8)(1 7 9).
2. If cr is a fc-cycl e, show that cr is an o dd permutati o n if k is even, and is an
even permutati o n if k is o dd.
3. Pro ve that cr and T
_ 1
( X T , for any a, r G S, are of the same parity.
4. If m < n, we can consider S, C 5 by viewing cr G S as acting on
1, 2, . . . , m,.. ., n as it did o n 1, 2, . . . , m and cr l eaves j > m fixed.
Pro ve that the parity of a permutati o n in S,, when viewed this way as an
el ement of S, does not change.
5. Suppo se yo u are tol d that the permutati o n
1 2 3 4 5 6 7 8 9\
3 1 2 7 8 9 6)
in 5
9
, where the images of 5 and 4 have been l ost, is an even permuta-
tion. What must the images of 5 and 4 be?
6. If n > 3, show that every el ement in A is a pro duct of 3-cycl es.
7. Show that every el ement in A is a pro duct of n-cycl es.
8. Fi nd a no rmal subgro up in A
4
of o rder 4.
Hai der Pro bl ems (I n fact, very hard)
9. If n 2 5 and (e) N C A is a no rmal subgro up of A, sho w that 7Y must
contain a 3-cycl e.
10. Using the resul t of Pro bl em 9, show that if n ' 5, the onl y no rmal sub-
groups of A are (e) and A itsel f. (Thus the groups A for n. > 5 give us
an infinite famil y of no nabel i an finite simpl e groups.)
1. DEFI NI TI ONS A ND EXA MPL ES
So far in o ur study of abstract al gebra, we have been i ntro duced to o ne ki nd
of abstract system, which pl ays a central ro l e in the al gebra of today. That
was the no ti o n of a gro up. Because a gro up is an al gebraic system with onl y
o ne o perati o n, and because a gro up need no t satisfy the rul e ab = ba, it ran
so mewhat co unter to o ur pri o r experi ence in al gebra. We were used to sys-
tems where yo u coul d bo th add and mul tipl y el ements and where the el e-
ments did satisfy the co mmutati ve l aw of mul tipl ication ab = ba. Further-
mo re, these systems of o ur acquai ntance usual l y came from sets of
numbersi ntegers, rational , real , and for so me, compl ex.
The next al gebrai c o bj ect we shal l co nsi der is a ring. I n many ways
this system wil l be mo re remi ni scent of what we had previ o usl y kno wn
than were gro ups. Fo r o ne thi ng ri ngs wil l be endo wed with addi ti o n and
mul ti pl i cati o n, and these wil l be subj ected to many of the famil iar rul es
we al l kno w from ari thmeti c. On the o ther hand, rings need not co me
from o ur usual number systems, and, in fact, usual l y have l i ttl e to do wi th
these fami l i ar o nes. Al tho ugh many of the fo rmal rul es of ari thmeti c
ho l d, many strangeo r what may seem as strangepheno mena do take
pl ace. As we pro ceed and see exampl es of ri ngs, we shal l see so me of
these thi ngs o ccur.
Wi th this preambl e over we are ready to begin. Natural l y eno ugh, the
first thing we shoul d do is to define that which we'l l be tal king about.
125
126 Ri ng Th eo r y Ch . 4
Definition. A no nempty set R is said to be a ring if in R there are two
o perati o ns + and such that:
(a) a, b E R impl ies that a + b E R.
(b) a + b = b + a for a, b E R.
(c) (a + b) + c = a + (b + c) for a,b,c<= R.
(d) There exists an el ement 0 E R such that a + 0 = a for every a G R.
(e) Gi ven a E R, there exists ab G R such that a + 6 = 0. (We shal l write
b as a.)
No te that so far al l we have said is that R is an abel ian gro up under +. We
now spel l out the rul es for the mul tipl ication in R.
(f) A, b E R impl ies that a b E R.
(g) a - (b c) = (a b) c for fl , b,c E R.
This is al l that we insist on as far as the mul tipl ication by itsel f is co ncerned.
But the + and are no t al l owed to l ive in sol itary spl endo r. We i nterweave
them by the two distributive laws
(h) (b + c) = a b + a c and
(b + c) a: = b a + c a, for a, b, c E R.
These axioms for a ring l ook famil iar. They shoul d be, for the co ncept
of ring was i ntro duced as a general i zati on of what happens in the i ntegers.
Because of Axi o m (g), the associative l aw of mul tipl ication, the rings we de-
fined are usual l y cal l ed associative rings. Nonassoci ati ve rings do exist, and
so me of these pl ay an i mpo rtant rol e in mathemati cs. But they shal l no t be
o ur concern here. So whenever we use the wo rd "ri ng" we shal l al ways mean
"associative ri ng."
Al tho ugh Axi o ms (a) to (h) are famil iar, there are certain things they
do not say. We l ook at so me of the famil iar rul es that are not insisted upo n
for a general ring.
First, we do no t po stul ate the existence of an el ement I E / ? such that
A 1 = 1 A = A for every a E R. Many of the exampl es we shal l enco unter
wil l have such an el ement, and in that case we say that 7? is' a ring with unit.
I n al l fairness we shoul d po i nt o ut that many al gebraists do demand that a
ri ng have a unit el ement. We do insist that 1^0; that is, the ring consisting
of 0 al one is no t a ring with unit.
Sec . 1 Def i n i t i o n s an d Ex amp l es 127
Second, in o ur previ ous experi ence with things of this sort, whenever
a b = 0 we concl uded that a = 0 or b = 0. Thi s need no t be true, in general ,
in a ring. When it does hol d, the ring is ki nd of nice and is given a special
name; it is cal l ed a domain.
Thi rd, no thi ng is said in the axioms for a ring that wil l impl y the com-
mutative law of mul tipl ication a b = b a. There are no nco mmutati ve rings
where this l aw does not hol d; we shal l see so me soon. Our mai n concern in
this chapter wil l be with commutative rings, but for many of the earl y resul ts
the co mmutati vi ty of the ring studi ed wil l no t be assumed.
As we menti o ned abo ve, so me things make certain rings nicer than o th-
ers, and so beco me wo rthy of having a special name. We quickl y give a l ist of
definitions for so me of these nicer rings.
Definition. A co mmutati ve ring A' is an integral domain if a b = 0 in
R impl ies that a = 0 or b = 0.
I t shoul d be po i nted out that so me al gebra bo o ks insist that an integral
do mai n contai n a unit el ement. I n readi ng ano ther bo o k, the reader shoul d
check if this is the case there. The integers, Z, give us an obvious exampl e of
an integral do mai n. We shal l see o ther, so mewhat l ess obvious o nes.
Definition. A ring R with unit is said to be a division ring if for every
a + 0 in R there is an el ement b G R (usual l y wri tten as a~
x
) such that
a A "
1
= A
-
' A = 1.
The reaso n for cal l ing such a ring a division ring is qui te cl ear, for we
can divide (at l east keepi ng l eft and right sides in mi nd). Al tho ugh no nco m-
mutati ve division rings exist with fair frequency and do pl ay an i mpo rtant
ro l e in no nco mmutati ve al gebra, they are fairl y compl i cated and we shal l
give onl y o ne exampl e of these. This division ring is the great cl assic o ne in-
tro duced by Hami l to n in 1843 and is kno wn as the ring of quaternions. (See
Exampl e 13 bel ow.)
Final l y, we come to perhaps the nicest exampl e of a cl ass of rings, the field.
Definition. A ring R is said to be afield if R is a commutative division
ring.
I n o ther wo rds, a fiel d is a co mmutati ve ring in which we can divide
freel y by no nzero el ements. Otherwi se put, R is a fiel d if the no nzero el e-
ments of R fo rm an abel ian gro up under , the pro duct in R.
128 Ring Theory Ch, 4
Fo r fiel ds we do have so me ready exampl es: the rati o nal numbers, the
real numbers, the co mpl ex numbers. But we shal l see many mo re, perhaps
l ess famil iar, exampl es. Chapter 5 wil l be devo ted to the study of fiel ds.
We spend the rest of the ti me in this section l ooking at so me exampl es
of rings. We shall drop the for the product and shall write a b simply as ab.
Exampl es
1. I t is obvious which ring we shoul d pick as o ur first exampl e, namel y Z, the
ring of integers under the usual addi ti o n and mul tipl ication of i ntegers. Natu-
ral l y eno ugh, Z is an exampl e of an i ntegral do mai n.
2. The second exampl e is equal l y obvi ous as a choice. Let Q be the set of al l
rati onal numbers. As we al l kno w, Q satisfies al l the rul es needed for a fiel d,
so 0 is a fiel d.
3. The real numbers, IR, al so give us an exampl e of a fiel d.
4. The compl ex numbers, C, form a fiel d.
No te that Q C U C C; we descri be this by saying that Q is a subfield of
IR (and of C) and (R is a subfiel d of C.
5. Let R = Z
6
, the integers mo d 6, with the addition and the mul ti pl i cati o n
defined by [a] + [b] = [a + b] and [a][b] = [ab].
No te that [0] is the 0 requi red by o ur axioms for a ring, and [1] is the unit
el ement of R. No te, however, that Z
6
is not an integral domain, for
[2] [3] = [6] = [0], yet [2] * [0] and [3] * [0]. R is a commutati ve ring with unit.
This exampl e suggests the
Definition. An el ement a #0 in a ring R is a zero-divisor in A* if ab - 0
for so me b = 0 in R.
We shoul d real l y cal l what we defined a l eft zero-divisor; ho wever,
since we shal l mainl y tal k abo ut co mmutati ve rings, we shal l no t need any
l eft-right distinction for zero-divisors.
No te that bo th [2] and [3] in Z
6
are zero-divisors. An i ntegral do mai n
is, of course, a co mmutati ve ring wi tho ut zero-divisors.
6. Let R = Z
5
, the ring of i ntegers mo d 5. 7? is, of course, a co mmutati ve ri ng
with unit. But it is mo re; in fact, it is a fiel d. I ts no nzero el ements are [1], [2],
[3], [4] and we no te that [2][3] = [6] = [1], and [1] and [4] are thei r o wn in-
verses. So every no nzero el ement in Z
5
has an inverse in Z
5
.
We general ize this to any pri me p.
7. Let Z
p
be the integers mo d p, where p is a pri me. Agai n Z
p
is cl earl y a
co mmutati ve ring with 1. We cl aim that Z
p
is a fiel d. To see this, no te that if
[a] = [0], then p\a. Therefo re, by Fermat's Theo rem (Corol l ary to Theo rem
2.4.8), a
p
'
x
= l(p). Fo r the cl asses [] this says that [A ""
1
] = [1]. But [a
p
~
l
] =
[ A ] ' ' "
1
, so [a]''"
1
= [1]; therefo re, [a]
p
~
2
is the requi red inverse for [a] in Z
p
,
hence Z
p
is a fiel d.
Because Z
p
has onl y a finite number of el ements, it is cal l ed a finite
field. Later we shal l construct finite fiel ds different from the Z^'s.
8. Let Q be the rati onal numbers; if a G Q, we can wri te a = m/n, where m
and n are rel ativel y pri me integers. Cal l this the reduced form for a. Let R be
the set of al l a G Q in who se reduced form the deno mi nato r is odd. Under
the usual addi ti o n and mul tipl ication in Q the set R forms a ring. I t is an i nte-
gral do mai n with unit but is not a fiel d, for \, the needed inverse of 2, is not
in R. Exactl y which el ements in R do have thei r inverses in R?
9. Let R be the set of al l G <Q>in who se reduced form the deno mi nato r is
no t divisibl e by a fixed pri me p. As in (8), R is a ring under the usual addi -
ti o n and mul ti pl i cati on in Q, is an i ntegral do mai n but is no t a fiel d. What
el ements of R have thei r inverses in R7
Bo th Exampl es 8 and 9 are subri ngs of Q> in the fol l owing sense.
Definition. If R is a ring, then a subring of R is a subset S of R which
is a ring if the o perati o ns ab and a + b are just the o perati o ns of R appl ied to
the el ements a, b G S.
Fo r S to be a subring, it is necessary and sufficient that S be no nempty
and that ab, a b G S for al l a, b G 5. (Prove!)
We give o ne further co mmutati ve exampl e. This o ne co mes from the
cal cul us.
10. Let R be the set of al l real -val ued co nti nuo us functions on the cl osed unit
i nterval [0, 1]. Fo r fgELR and x G [0, 1] define ( / + g)(x) = f(x) + g (x),
and (/ g)(x) = f(x)g(x). Fro m the resul ts in the cal cul us, f+g and / g are
again co nti nuo us functions on [0, 1]. Wi th these o perati o ns R is a co mmuta-
tive ring. I t is not an integral do mai n. Fo r i nstance, if f(x) = x + \ for
0 < x < | and f(x) = 0 for | < x s 1, and if g(x) = 0 for 0 < x < \ and
g{x) = 2x 1 for | < x < 1, then figER and, as is easy to veri fy,/ g = 0.
I t does have a unit el ement, namel y the function e defined by e (x) = 1 for al l
x G [0,1]. What el ements of R have thei r inverses in R?
130 Ring Theory Ch. 4
We shoul d no w l ike to see so me no nco mmutati ve exampl es. These are
no t so easy to co me by, al tho ugh no nco mmutati ve rings exist in abundance,
because we are no t assumi ng any kno wl edge of l inear al gebra on the reader's
part. The easiest and mo st natural first source of such exampl es is the set of
matri ces over a fiel d. So, in o ur first no nco mmutati ve exampl e, we shal l
real l y create the 2X 2 matri ces with real entries.
11. Let F be the fiel d of real numbers and l et R be the set of al l formal square
arrays
a b
c d
where a, b,c, d are any real numbers. Fo r such square arrays we define addi-
tion in a natural way by defining
b\
+
/ 2 b-\
=
(a
x
+ 2 b
x
+ b
2
d
x
J \c
2
d
2
) \c
x
+ c
2
d
x
+ d
2
I t is easy to see that R forms an abel ian gro up under this + with ^ ^ act-
ing as the zero el ement and ^_ " _ the negati ve of ^ ^T o make of
R a ring, we need a mul tipl ication. We define o ne in what may seem a highl y
unnatural way via
a b\(r s\ _ far + bt as + bu
c d)\t uj \cr + dt cs + du
I t may be a l ittl e l aborious, but o ne can check that with these operati ons R is a
noncommutati ve ring with ^ ^ acting as its mul tipl icative uni t el ement.
No te that
1 oVo 0\
=
/0 0
0 O/l l 0/ l O 0
whil e
so
0 OV l 0\
=
0 0
1 0 A 0 0/ \1 0/"
i oyo o\ (o owi o
0 O/ l l 0 1 0 0 0
No te that
1 0
0 0
and
0 0
1 0
are zero-divisors; in fact,
0 0
o o
0 0
so
0 0
1 0
is a no nzero el ement whose square is the 0 el ement of R. This R is kno wn as
the ring of all 2 x 2 matrices over F, the real field.
Fo r tho se unfamil iar with these matri ces, and who see no sense in the
pro duct defined for them, l et's l ook at ho w we do co mpute the pro duct. To
get the to p l eft entry in the pro duct AB, we "mul ti pl y" the first row of A by
the first co l umn of B, where A, B G R. Fo r the top right entry, it is the first
row of A versus the second col umn of B. The bo tto m l eft entry comes from
the seco nd row of A versus the first co l umn of 73, and final l y, the bo tto m
right entry is the second col umn of A versus the second col umn of 73.
We il l ustrate with an exampl e: Let
Then the first row of A is 1, \ and the first co l umn of B is , TT; we "mul ti pl y"
these via 1 + \ TT = 7r/2 + , and so on. So we see that
I n the pro bl ems we shal l have many matri x mul tipl ications, so that the
reader can acqui re some famil iarity with this strange but i mpo rtant exampl e.
12. Let R be any ring and l et
with + and as defined in Exampl e 11. One can verify that S is a ring, al so,
under these o perati o ns. It is cal l ed the ring of 2 X 2 matrices over R.
Our final exampl e is o ne of the great cl assical exampl es, the real qua-
ternions, i ntro duced by Hami l to n (as a no nco mmutati ve paral l el to the co m-
pl ex numbers).
13. The quaternions. Let F be the fiel d of real numbers and consider the set
of al l formal symbol s a
0
+ a
x
i + a
2
j + ct
3
k, where a
0
, a
{
, a
2
, a
3
G F.
Equal i ty and addi ti o n of these symbol s are easy, via the obvious route
and
a
0
+ afi + a
2
j + a
3
k = B
0
+ Bj + B
2
j + B
3
fc
if and onl y if a
0
= B
0
, a
x
= B
x
, a
2
= B
2
and a
3
= B
3
, and
(a
0
+ afi + a
2
j + a
3
k) + (B
0
+ B.i + B
2
j + B
3
k)
= (ao +ft) +(i + ft) ' + ( 2 + ft) j +(a
3
+ B
3
)k.
We no w come to the tricky part, the mul tipl ication. When Hami l to n discov-
ered it on Octo ber 6, 1843, he cut the basic rul es of this pro duct o ut with his
penkni fe on Bro ugham Bri dge in Dubl i n. The pro duct is based on r =j
2
=
k
2
= l,ij = k, jk = i, ki = j and ji = k, kj = i, ik = j. If we go aro und
the circl e cl ockwise
o
the pro duct of any two successive o nes is the next o ne, and goi ng aro und
countercl ockwi se we get the negati ves.
We can write out the pro duct no w of any two quaterni o ns, acco rdi ng to
the rul es abo ve, declaring by definition that
(a
Q
+ a,/ + a
2
j + a
3
k)(B
0
+ BJ + B
2
j + B
3
k)
=To +Jii + J2J + y
3
,
where
To ="oft - ift - a
2
B
2
- a
3
B
3
J\ = oft +"ift +a
2
B
3
- a
3
B
2
^
Ji ="oft - ift +<x
2
B
0
+ a
3
B
x
y
3
=oft +ift
-
2 f t + a
3
B
0
I t l ooks ho rrendo us, do esn't it? But it's no t as bad as al l that. We are
mul tipl ying out formal l y using the distributive l aws and using the pro duct
rul es for the i, j , k abo ve.
If some a, is 0 in x a
0
+ ad + a
2
j + a
3
k, we shal l omi t it in express-
ing x; thus 0 +0/ +0/ + Ok wil l be wri tten simpl y as 0,1 +0; +0/ + 0k as 1,
0 + 3i + 4j + 0k as 3z +4/, and so on.
A cal cul ation reveal s that
(a
0
+ a
x
i + a
2
j + a
3
k)(a
0
- a
x
i - a
2
j - a
3
k)
(I I )
This has a very i mpo rtant co nsequence; for suppo se that x = a
Q
+ a
x
i +
a
2
j + a
3
k i= 0 (so some # 0). Then, since the a's are real , /3 = al + a\ +
a\ + a\ + 0. Then from (I I ) we easil y get
So, if x + 0, then x has an inverse in the quaterni o ns. Thus the quaternions
form a noncommutative division ring.
Al tho ugh, as we menti o ned earl ier, there is no l ack of no nco mmutati ve
division rings, the quaterni o ns abo ve (or so me pi ece of them) are often the
onl y no nco mmutati ve division rings that even many professional mathemati -
cians have ever seen.
We shal l have many pro bl emsso me easy and so me qui te a bit
harderabo ut the two exampl es: the 2 X 2 matri ces and the quaterni o ns.
Thi s way the reader wil l be abl e to acqui re so me skil l with pl aying wi th
no nco mmutati ve rings.
One final co mment in this section: If y
0
, y
x
, y
2
, y
3
are as in (I ), then
This is kno wn as Lagrange's Identity; it expresses the pro duct of two sums of
four squares again as a sum of four squares. I ts verification wil l be one of the
exercises.
Easi er Pro bl ems
* 1. Fi nd al l the el ements in Z
2 4
that are invertible (i.e., have a mul tipl icative
inverse) in Z
2 4
.
2. Show that any fiel d is an integral do mai n.
3. Show that Z is a fiel d if and onl y if /; is a pri me.
4. Verify that Exampl e 8 is a ring. Fi nd al l its invertibl e el ements.
5. Do Pro bl em 4 for Exampl e 9.
6. I n Exampl e 11, the 2 x 2 matri ces o ver the real s, check the associative
l aw of mul tipl ication.
{a
2
+ al+ a
2
2
+ aI )(/3
2
+ [3J + & + f3
2
3
)
(I I I )
To +yl +rl +yl
PROBL EMS
134 Ring T heory Ch. 4
( H )
\4 - 7A 0 1
< (1 1
<
c
>(o o)'
fl 0\ / l 0\/ff /3
( d )
\c dj\0 01 10 0/ \c d
8. Fi nd al l matri ces , such that
: s ) * ( : s $ B ) ( s)
9. Fi nd al l 2 X 2 matri ces | ^) that co mmute with al l 2 X 2 matri ces.
10. Let R be any ring wi th unit, S the ri ng of 2 X 2 matri ces o ver R. (See Ex-
ampl e 12.)
(a) Check the associative l aw of mul tipl ication in S. (Remember: A' need
not be co mmutati ve.)
'a
(b) Sho w that T ,,
,0 c
A, b, c, i? V is a subring of 5.
(c) Show that ^ ^ j has an inverse in T if and onl y if a and c have
A 6
W
inverses in R. I n that case write do wn ^ ^ j expl icitl y.
11. Let F: C - C be defined by P( + W) = A - oz. Sho w that:
(a) F(xy) = F(x)F(y) for x, y C.
(b) F(xx) = |x|
2
.
(c) Usi ng Parts (a) and (b), show that
( A
2
+ b
2
)(c
2
+ d
2
) = (ac - bd)
2
+ (ad + be)
2
.
[Note: F(x) is merel y x.]
12. Verify the identity in Part (c) of Pro bl em 11 directl y.
13. Fi nd the fol l owing pro ducts of quaterni o ns.
(a) (/ +j)(i - j).
(b) (1 - i + 2/ - 2k)(l + 2/ - 4/ + 6k).
(c) (2i - 3; + 4/c)
2
.
(d) i(a
0
+ a
Y
i + a
2
j + a
3
k) (a
0
+ a
x
i + a
2
j + a
3
k)i.
7. Wo rk o ut the fol l owing:
'i 2\ / i r
14. Sho w that the onl y quaterni o ns co mmuti ng with i are of the form a + Bi.
15. Fi nd the quaterni o ns that co mmute with bo th i and j .
16. Verify that
(cv
0
+ afi + a
2
j + a
3
k)(a
0
- af - a
2
j - a
3
k)
= a
2
, + a\ + a\ + a\.
17. Verify Lagrange's I dentity by a direct cal cul ation.
18. I n the quaterni o ns, define
|a
n
+ a
x
i + a
2
j + a
3
k\ = V 5 + i + oi
2
, +i-
Show that \xy \ = \x\\y \ for any two quaterni o ns x and y.
19. Sho w that there is an infinite number of sol utions to x
2
= 1 in the
quaterni o ns.
20. I n the quaterni o ns, consider the fol l owing set G having eight el ements:
G = {1, i, j, k).
(a) Pro ve that G is a gro up (under mul ti pl i cati on).
(b) List al l subgro ups of G.
(c) What is the center of G?
(d) Show that G is a nonabelian group all of whose subgroups are
normal.
21. Show that a division ring is a do mai n.
22. Gi ve an exampl e, in the quaterni o ns, of a no nco mmutati ve do mai n that
is no t a division ri ng.
23. Define the map * in the quaterni o ns by
(a
n
+ af + a
2
j + a
3
k)* = (a
0
a
t
i <x
2
j a
3
k).
Show that:
(a) x** = (x*)* = x.
(b) ( x + y)* = x* + y*.
(c) xx* = x*x is real and no nnegati ve.
(d) (xy)* = y*x*.
[Note the reversal of o rder in Part (d).]
24. Usi ng *, define |x| = V xx* . Show that \xy\ = |x| \y\ for any two qua-
terni o ns x and y, by using Parts (c) and (d) of Pro bl em 23.
136 Ri n g Th eo r y Ch, 4
25. Use the resul t of Pro bl em 24 to pro ve Lagrange's I denti ty.
I n Pro bl ems 26 to 30, l et R be the 2 x 2 matri ces over the real s.
26. If ^ ^ j G R, show that ^ is invertibl e in R if and onl y if ad be = 0.
I n that case find I ^
K
c d i
I a b
s
27. Define detl , I = ad - be. Fo r ,r, y R show that det(xy)
v
c d
(det x)(det y).
28. Show that {x R | det x # 0} forms a gro up, G, under matri x mul tipl ica-
tion and that N = {x R | det x = 1} is a no rmal subgro up of G.
29. If x R is a zero-divisor, sho w that det x = 0, and, conversel y, if x # 0 is
such that det x = 0, then x is a zero-divisor in R.
30. I n P, show that \ I
6
1 1
- o a
Harder Pro bl ems
, /3 real \ is a fiel d.
31. Let R be the ring of al l 2 X 2 matri ces over Z
p
, p a pri me. Sho w that if
32. Let be as in Pro bl em 31. Show that for x, y & R, det(xy) = det(x) det(y).
33. Let G be the set of el ements x in the ring R of Pro bl em 31 such that
det(x) + 0.
(a) Pro ve that G is a gro up.
(b) Fi nd the o rder of G. (Quite hard)
(c) Fi nd the center of G.
(d) Fi nd a p-Syl ow subgro up of G.
34. Let T be the group of matrices A with entries in the fiel d Z
2
such that det A
is no t equal to 0. Pro ve that T is i so mo rphi c to S
3
, the symmetri c gro up
of degree 3.
35. Fo r R as in Exampl e 10, show that S = {/ R | / is differentiabl e on (0,1)}
is a subring of R which is no t an i ntegral do mai n.
If F is a fiel d, l et H(F) be the ring of quaterni ons over F, that is, the
set of al l a
0
+ ad + a
2
j + a
3
k, where a
0
, a
t
, ct
2
, a
3
F and where equal -
ity, addition, and mul tipl ication are defined as for the real quaterni o ns.
Sec . 2 So m e Si mp l e Res ul t s 137
36. If F = C, the compl ex numbers, sho w that H(C) is not a division ring.
37. I n H(C), find an el ement .v #0 such that x
2
= 0.
38. Sho w that H(F) is a division ring if and onl y if a
2
, + a\ + a\ + a
3
= 0
for ai , a
2
, a
3
, a
4
in Z
7
forces a
0
= a
y
= a
2
= a
3
=
0.
39. If Q is the fiel d of rati onal numbers, show that H(Q) is a division ring.
40. Pro ve that a finite do mai n is a division ring.
41. Use Pro bl em 40 to show that Z
p
is a fiel d if p is a pri me.
2. SOME SI MPL E RESUL TS
No w that we have seen so me exampl es of rings and have had some experi -
ence pl aying aro und with them, it wo ul d seem wise to devel op so me co mpu-
tati o nal rul es. These wil l al l ow us to avoid anno yi ng trivial ities that coul d
beset a cal cul ation we mi ght be maki ng.
The resul ts we shal l pro ve in this section are no t very surprising, not
to o i nteresti ng, and certainl y no t at al l exciting. Nei ther was l earning the al -
phabet, but it was so methi ng we had to do befo re going on to bigger and bet-
ter things. The same hol ds for the resul ts we are abo ut to pro ve.
Since a ring R is at l east an abel i an gro up under +, there are
certai n things we kno w from o ur gro up theo ry backgro und, for instance,
-(-) = a, - ( A + b) = (-fl ) + {~b); if a + b = a + c, then b = c, and
so on.
We begin with
L emma 4.2.1. Let R be any ri ng and l et a. b G R. Then
(a) fl O = Ofl = 0.
(b) a(-b) = (-a)b = -(ab).
(c) (~a)(-b) = ab.
(d) If 1 G R, then ( - l ) a = -a.
Proof. We do these in turn.
(a) Since 0 = 0 + 0, aO = a(0 + 0) =fl O + aQ, hence 0 = 0. We have
used the l eft distributive l aw in this proof. The right distributive l aw gives
Ofl = 0.
(b) ab + a(-b) = a(b + (-b)) = fl O = 0 from Part (a). Therefo re,
a(b) = (ab). Simil arl y, (~a)b = -(ab).
(c) By Part (b), (-a)(b) = ((-a)b) = -(-(ab)) = ab, since we are
in an abel ian gro up.
(d) If 1 G R, then ( - l ) a +a = ( - l ) a + (l)a = ( - 1 + 1)A =OA =0. So
( - l ) a = - a by the definition of - A.
Ano ther co mputati o nal resul t.
L emma 42. 2. I n any ring R, (a + bf = a
2
+ b
2
+ ab + ba for a, h G A'.
Proof. This is cl earl y the anal og of (a + B)
2
= a
2
+2a/3 + B
2
in the in-
tegers, say, but keepi ng in mi nd that R may be no nco mmutati ve. So, to it. By
the right distributive l aw (a +b)
2
= (a + b)(a + b) = (a + b)a + (a + b)b =
a
2
+ ba + ab + b
2
, exactl y what was cl aimed.
Can you see the no nco mmutati ve version of the bi nomi al theo rem? Try
it for (a + bf.
One curiosity fol l ows from the two distributive l aws when R has a uni t
el ement. The co mmutati ve l aw of addi ti o n fol l ows from the rest.
L emma 4. 2.3. If R is a system with 1 satisfying al l the axi oms of a
ring, except possibl y a + b = b + a for a, b G R, then R is a ring.
Proof. We must show that a + b = b + a for fl , b G R. By the right dis-
tributive l aw (A +b)(l + 1) = (a +b)l + (A + 6)1 =a + b + a + b. On the
o ther hand, by the l eft distributive l aw (A +b)(l + 1) =a (I + 1) +6(1 + 1)
= a + a + b + b. But then a + b + a + b = a + a + b + b; since we are in a
gro up under +, we can cancel a on the l eft and b on the right to o btai n b + a
= a + 6, as requi red. R is therefo re a ring.
We cl ose this brief section with a resul t that is a l ittl e nicer. We say that
a ring R is a Boolean ring [after the Engl ish mathemati ci an Geo rge Bo o l e
(1815-1864)] if x
2
= x for every x G i?.
We pro ve a nice resul t on Bo o l ean rings.
L emma 4.2.4. A Bo o l ean ring is co mmutati ve.
Proof. Let x, y G K , a Bo o l ean ring. Thus x
2
= x, y
2
= y, (x + y)
2
=
x + y. But (x + y)
2
= x
2
+ xy + yx + y
2
= x + xy + yx + y, by L emma 4.2.2,
so we have (x + y) = (x + y)
2
= x + xy + yx + y, from which we have
xy + yx = 0. Thus 0 = x(xy + yx) = x
2
y + xyx = xy + xyx, whil e 0 =
(xy + yx)x = xyx + yx
2
= xyx + yx. This gives us xy + xyx = xyx + yx, and
so xy - yx. Therefo re, R is co mmutati ve.
Sec . 3 I d eal s , Ho mo mo r p h i s ms , an d Qu o t i en t Ri ngs 13S
PROBL EMS
1. Let R be a ring: since R is an abel ian gro up under +, na has meani ng for
us for n E Z, a E R. Show that (na)(mb) = (nm)(ab) if n, m are integers
and a, b E R.
2. If 7? is an integral domain and ab = ac for a 0, 6, c G R, show that b = c.
3. If A is a finite integral do mai n, show that R is a fiel d.
4. If R is a ring and e G _R is such that e
2
= <?, show that (xe - exe)
2
=
(ex exe)
2
= 0 for every x G A.
5. Let R be a ring in which x
3
= x for every x G R. Pro ve that R is co mmu-
tative.
6. If a
2
= 0 in A, show that ax + xa co mmutes with a.
7. Let R be a ring in which x
4
= x for every x G R. Pro ve that R is co mmu-
tative.
8. If F is a finite fiel d, show that:
(a) There exists a pri me p such that pa = 0 for al l a E F.
(b) If -F has o el ements, then o =p" for so me i nteger n. (Hi nt: Cauchy's
Theo rem)
9. Let p be an o dd pri me and l et 1 + + + ll(p - 1) = />, where A, b
are integers. Show that p \ a. (Hi nt: As a runs thro ugh (7
p
, so does "'.)
10. If p is a pri me and /; > 3, show that i f l +| +
,
- - + l / ( p- l ) =fl /fo, where
a, are integers, then p
2
1 a. (Hint: Consider 1/a
2
as a runs through L^.)
3. I DEA L S, HOMOMORPHI SMS, A N D QUOTI ENT RI NGS
I n studying groups, it turned o ut that ho mo mo rphi sms, and thei r kernel s
the no rmal subgro upspl ayed a central rol e. There is no reaso n to expect
that the same thing shoul d no t be true for rings. As a matter of fact, the
anal ogs, in the setting of rings, of ho mo mo rphi sm and no rmal subgro up do
pl ay a key rol e.
Wi th the backgro und we have acqui red abo ut such things in group the-
ory, the paral l el devel o pment for rings shoul d be easy and quick. And it wil l
be! Wi tho ut any further fuss we make the
Definition. The mappi ng cp : R -> R' of the ring R into the ring A" is a
homomorphism if
(a) cp(a + b) = cp(a) + cp(b) and
(b) cp(ab) = cp(a)cp(b) for al l a, b E R.
140 Ri ng Th eo r y Ch , 4
Since a ring has two o perati o ns, it is onl y natural and just that we de-
mand that bo th these o perati o ns be preserved under what we wo ul d cal l a
ring ho mo mo rphi sm. Furthermore, Property (a) in the Definition tells us that
cp is a homomorphism of R viewed merely as an abelian group under + into
R' (al so viewed as a gro up under its addi ti o n). So we can cal l on, and expect,
certain resul ts from this fact al o ne.
J ust as we saw in Chapter 2, Section 5 for groups, the i mage of R under
a ho mo mo rphi sm from R to R', is a subring of R', as defined in Chapter 4,
Section 1 (Prove!).
Let cp: R R' be a ring ho mo mo rphi sm and l et Ker cp =
[x G R | cp(x) = 0}, the 0 bei ng that of R'. What pro perti es do es Ker cp enjoy?
Cl earl y, from group theo ry Ker cp is an additive subgro up of R. But much
mo re is true. If k G Ker cp and r G R, then cp(k) = 0, so cp(kr) = cp(k)cp(r) =
0cp(r) = 0, and simil arl y, cp(rk) = 0. So Ker cp swal l ows up mul ti pl i cati on
from the l eft and the right by arbi trary ring el ements.
This pro perty of Ker cp is no w abstracted to define the i mpo rtant anal o g
in ring theo ry of the noti on of no rmal subgro up in gro up theo ry.
Definition. Let R be a ring. A no nempty subset / of R is cal l ed an
ideal of R if:
(a) I is an additive subgro up of R.
(b) Given r G R, a G I, then ra G I and ar G L
We shal l so o n see so me exampl es of ho mo mo rphi sms and ideal s. But
first we no te that Part (b) in the definition of ideal real l y has a l eft and a right
part. We coul d spl it it and define a set L of R to be a left ideal of R if L is an
additive subgro up of R and given r G R, a: G L, then ra G L. So we requi re
onl y l eft swal l owing-up for a l eft ideal . We can simil arl y define right ideals.
An ideal as we defined it is bo th a l eft and a right ideal of R. By al l rights we
shoul d then cal l what we cal l ed an i deal a two-sided ideal of R. I ndeed, in
wo rki ng in no nco mmutati ve ring theo ry o ne uses this name; here, by "ideal"
we shall always mean a two-sided ideal. Except for so me of the pro bl ems, we
shal l no t use the no ti o n of one-si ded ideal s in this chapter.
Before going on, we reco rd what was do ne above for Ker cp as
L emma 4.3.1. If cp: R > R' is a ho mo mo rphi sm, then Ker cp is an
ideal of R.
We shal l soon see that every ideal can be made the kernel of a ho mo -
mo rphi sm. Shades of what happens for no rmal subgro ups of gro ups!
Sec . 3 I d eal s , Ho mo mo r p h i s ms , an d Qu o t i en t Ri ngs 141
Final l y, l et K be an ideal of R. Since K is an additive subgro up of R, the
quo ti ent gro up RIK exists; it is merel y the set of al l cosets a + K as a runs
over R. But R is not just a gro up; it is, after al l , a ring. No r is K merel y an ad-
ditive subgro up of R; it is mo re than that, namel y an ideal of R. We shoul d
be abl e to put al l this to gether to make of RIK a ring.
Ho w shoul d we define a pro duct in RIK in a natural way? What do we
want to decl are (a + K)(b + K) to be? The reaso nabl e thing is to define
(a + K)(b + K) = ab + K, which we do . As al ways, the first thing that
comes up is to show that this pro duct is wel l -defined. I s it? We must show
that if a + K = a' + K and b + K = b' + K, then {a + K)(b + K) = ab + K =
a'b' + K=(a' + K)(b' + K). Ho wever, if a + K = a' + K, then a- a' G K,
so (a a')b G K, since K is an ideal of R (in fact, so far, since if is a right
ideal of R). Because b + K = b' + K, we have b - b' G K, so a'(b - b') G K,
since K is an ideal of R (in fact, since i f is a l eft ideal of R). So bo th (a - a')b =
ab - a'b and a'(b - b') = a'b - a'b' are in K. Thus (ab - a'b) + (a'b - a'b') =
ab - a'b' G K.
But this tel l s us (just from gro up theo ry) that ab + K = a'b' + K, ex-
actl y what we needed to have the pro duct wel l -defined.
So RIK is no w endo wed with a sum and a pro duct. Furthermo re, the
mappi ng cp: R RIK defined by cp(a) = a + K for a G R is a ho mo mo r-
phism of R o nto RIK with kernel K. (Prove!) This tel l s us right away that
RIK is a ring, bei ng the ho mo mo rphi c i mage of the ring R.
We summari ze al l this in
Theo rem 4.3.2. Let K be an i deal of A*. Then the quo ti ent group RIK
as an additive group is a ring under the mul tipl ication (a + K)(b + K) =
ab + K. Furthermo re, the map cp: R -* RIK defined by cp(a) = a + K for a G R
is a ho mo mo rphi sm of R o nto RIK havi ng K as its kernel . So RIK is a ho mo -
mo rphi c i mage of R.
J ust from gro up-theo reti c consi derati on of R as an additive gro up, we
have that if cp is a ho mo mo rphi sm of R into A", then it is 1-1 if and onl y if
Ker cp = (0). As in groups, we define a ho mo mo rphi sm to be a monomor-
phism if it is 1-1. A mo no mo rphi sm which is al so o nto is cal l ed an isomor-
phism. We define R and A" to be isomorphic if there is an isomorphism of R
onto R'.
An isomorphism from a ring R o nto itsel f is cal l ed an automorphism of
R. Fo r exampl e, suppo se R is the fiel d C of co mpl ex numbers. Then the map-
pi ng from C to C sendi ng each el ement of C to its compl ex conjugate is an
auto mo rphi sm of C. (Prove!)
One woul d have to be an awful pessimist to expect that the ho mo mo r-
phism theo rems pro ved in Section 7 of Chapter 2 fail in this setting. I n fact
they do hol d, with the sl ightl y obvi ous adaptati o n needed to make the proofs
go thro ugh. We state the ho mo mo rphi sm theo rems wi tho ut any further ado ,
l eaving the few detail s needed to co mpl ete the proofs to the reader.
T heorem 4.3.3 (First Homomorphism T heorem). Let the mappi ng
cp : R > R' be a ho mo mo rphi sm of R onto R' with kernel K. Then R' RIK;
in fact, the mappi ng ip: RIK R' defined by if/(a + K) = cp(a) defines an
i so mo rphi sm of RIK o nto R'.
T heorem 4.3.4 (Correspondence T heorem). Let the mappi ng <p: R > R'
be a ho mo mo rphi sm of R o nto R' with kernel K. If V is an i deal of R \ l et
I = [a E R [ cp(a) E / ' }. Then / i s an ideal of R, I D K and I/K = /'. This sets
up a 1-1 co rrespo ndence between al l the ideal s of R' and tho se ideal s of R
that contain K.
T heorem 4.3.5 (Second Homomorphism T heorem). Let A be a sub-
ring of a ring R and I an ideal of R. Then A + I = {a + i \ a E A, i E 1} is a
subring of i?, / is an ideal of A + /, and (A + / ) / / = yl/(^4 n /).
Theo rem 4.3.6 (T hird Homomorphism T heorem). Let the mappi ng
cp : R > R' be a ho mo mo rphi sm of /? o nto /?' with kernel K. If / ' is an ideal
oiR' and I = {a E R\ cp(a) E / ' }, then RII = /?'//'. Equi val entl y, if K is an
ideal of R and / D /C is an ideal of /?, then RII = (RIK)I(IIK).
We cl ose this section with an inspection of so me of the things we have
discussed in so me exampl es.
Examples
1. As usual we use Z, the ring of i ntegers, for our first exampl e. Let n > 1 be
a fixed integer and l et / be the set of al l mul tipl es of n; then /,, is an ideal of
Z. If Z is the i ntegers mo d , define cp: Z Z by <p(a) = []. As is easil y
seen, <p is a ho mo mo rphi sm of Z o nto Z with kernel / . So by Theo rem
4.3.3, Z Z/I. (This shoul d co me as no surprise, for that is ho w we origi-
nal l y i ntro duced Z.)
2. Let F be a fiel d; what can the ideal s of F be? Suppo se that / + (0) is an
ideal of F; l et a + 0 E /. Then, since / is an ideal of T
7
, 1 = <7~'a E /; but
now, since 1 E /, rl = r E I for every r E F. I n short, / = F. So F has only
the trivial ideals (0) and F itself.
3. Let R be the ring of al l rati onal numbers having o dd deno mi nato rs in thei r
reduced form. Let / be tho se el ements of R which in reduced form have an
Sec. 3 I d eal s , Ho mo mo r p h i s ms , an d Qu o t i en t Ri n g s 143
even numerato r; it is easy to see that / is an ideal of R. Define cp: R - Z
2
, the
integers mo d 2, by cp(a/b) = 0 if a is even (a, b have no co mmo n factor) and
cp(alb) = 1 if a is odd. We l eave it to the reader to verify that cp is a ho mo -
mo rphi sm of R o nto Z
2
with kernel I. Thus Z
2
= RII. Give the expl icit iso-
mo rphi sm of RII o nto Z
2
.
4. Let R be the ring of al l rati onal numbers who se deno mi nato rs (when in re-
duced form) are no t divisibl e by p,p a fixed pri me. Let I be tho se el ements in
R who se numerato r is divisibl e by p; / is an ideal of R and RII 7L
p
, the i nte-
gers mo d p. (Prove!)
5. L et R be the ring of al l real -val ued co nti nuo us functions on the cl osed unit
i nterval where ( / + g)(x) = f(x) + g(x) and (fg)(x) = f(x)g(x) for /, g G R,
x G [0, 1]. Let I = { / G R |/(f ) = 0}. We cl aim that / is an ideal of R.
Cl earl y, it is an additive subgro up. Furthermo re if / G / and g G R, then
/ (f ) = 0, so (/g)(f ) = / (f )g(f ) = 0g(f ) = 0. Thus/g G /; since / is co mmuta-
tive, g/i s al so in /. So / is an ideal of R.
What is RII1 Gi ven any f G R, then
f(x) = (f(x) - / (I)) + / (f ) = g(x) + / (I),
where g(x) = f(x) - / (f ). Because g( |) = / (f ) - / (j ) = 0, g is in /. So
g + 1= I. Thus / + / = ( / ( |) + g) + / =/ ( |) + /. Because / (f ) is just a real
number, A// consists of the cosets a + I for a real . We cl aim that every real
a co mes up. Fo r if / (f ) = B 0, then
/3-
L
/ +/ =(a/T
1
+/)(/+/) =(ar
1
+/)(/(!) +/)
= ( a/ T
1
+ 1)08 + /) = a/r
1
^+ I = a + I.
So A// consists of al l a + I, a real . Thus we have sho wn directl y that RII is
i so mo rphi c to the real fiel d.
We no w use Theo rem 4.3.3 to sho w that RII = real fiel d. Let cp:RÛ
be defined by <p(f) = /(I)- Then it is easy to verify that cp is a ho mo mo r-
phi sm, cp is o nto and Ker cp = { / G R \ / (f ) = 0}; in o ther wo rds, Ker cp = I.
So RII = i mage of cp = IR.
6. Let R be the integral quaterni o ns, in o ther wo rds, R = {a
0
+ a
x
i + a
2
j +
a
3
k | a
Q
, a
x
, a
2
, a
3
G Z}. Fo r a fixed pri me p, l et
=
(
a
o +
a
4 +
A
2 7 + ^1 Pi
a
; f
r
i
=
0>1, 2, 3).
The reader shoul d verify that I
p
is an i deal of R and that RII
p
~ //(Z^) (see
Pro bl em 36 of Section 1 and the paragraph before it).
144 Ring Th eo r y Ch . 4
a, b E IR k R is a subring of the 2 X 2 matri ces over the
' 0 b
Let / = { | q q ] | 6 IR }; / is easil y seen to be an additive subgro up of R. Is it
an ideal of R? Co nsi der
x y\/0 b\
=
(0 xb
0 x\0 0/ \0 0
so it is in /. Simil arl y,
0 b\x y\
=
0 bx\
0 o)\o x) \0 0
y
'
so it, to o , is in /. So / is an ideal of R. What is Rill We appro ach it from two
po i nts of view.
G,ve ( l)
eR
.
, h e
(o )"(o ")
+
(o o)
s 0 , h a
<
since ^ ^ is in I. Thus al l the cosets of / in R l ook l ike ^ ^ + /. If we map
this onto a, that is, i|< ^j + ij = a, we can check that ifi is an isomorphism
o nto the real fiel d. So RII = U.
We see that RII = R ano ther way. Define <p: R ->U. by 
a M / c d ^/ f l + c /j + a"
O a j \0 cj [ 0 a + c
r eal s .
L x
Sec . 3 I d eal s , Ho mo mo r p h i s ms , an d Qu o t i en t Ri ngs 145
^ 0 a
+
0 c
= \ O cr
So cp is indeed a homomorphism of R onto IR. What is Ker 
'
Thus A = 0. Fro m this we see that I = Ker cp. So A// image of cp = IR by
Theo rem 4.3.3.
, b G R [ and l et C be the fiel d of compl ex numbers.
8. Let R = {I 1
b
b a
1 1
Define tf/: R C by <//^_ ^ ^ j =a +6z. We l eave it to the reader to verify that
i/f is an i so mo rphi sm of R o nto C. So R is i so mo rphi c to the fiel d of co mpl ex
numbers.
9. Let R be any co mmutati ve ring with 1. If a G R, l et (A) = [xa \ x G A). We
cl aim that (a) is an ideal of R. To see this, suppo se that u, v G (A); thus
M =xa, v = ya for so me x, y G R, whence
u v = XA yfl = (x y) a G (A).
Al so , if G (fl ) and r G R, then =XA, hence ru = r(xfl ) = (rx)a, so is in (a).
Thus (fl ) is an ideal of R.
No te that if R is no t co mmutati ve, then (a) need no t be an ideal ; but it
is certainl y a l eft ideal of R.
, / a b\f c d\ f ac ad + bc\ ,
a l l d
((J a)\0 c)
=
{0 ac ) '
h e n C C
a b\ , (c d\\ (a + c b + d
146 Ri n g Th eo r y Ch . 4
PROBL EMS
Easi er Problems
1. If R is a co mmutati ve ring and a E R, l et L(a) = [x E R \ xa = 0}. Pro ve
that L(a) is an ideal of R.
2. If 7? is a co mmutati ve ring with 1 and R has no ideal s o ther than (0) and
itsel f, prove that R is a fiel d. (Hi nt: Lo o k at Exampl e 9.)
* 3. If R' is a ho mo mo rphi sm of 7? onto i ?' and 7? has a uni t el ement,
1, show that <p(l) is the unit el ement of R'.
4. If I, J are ideal s of R, define 7 + 7 by I + J = [i + j | / E 7,; E / }. Pro ve
that 7 +7 is an ideal of R.
5. If 7 is an ideal of 7? and A is a subring of 7?, show that I D A is an ideal of A
6. If 7, / are ideal s of R, show that 7 n 7 is an ideal of R.
7. Gi ve a compl ete pro o f of Theo rem 4.3.2.
8. Gi ve a co mpl ete proof of Theo rem 4.3.4.
9. Let cp : R - 7?' be a ho mo mo rphi sm of 7? o nto 7?' with kernel 7v. If A is
a subring of 7?', l et A = [a E R \ cp(a) E /I '}. Show that:
(a) A is a subring of 7\, A D 7C.
(b) A X = A .
(c) If A is a l eft ideal of R', then ^ is a l eft ideal of R.
10. Pro ve Theo rem 4.3.6.
11. I n Exampl e 3, give the expl icit i so mo rphi sm of RII o nto Z
2
.
12. I n Exampl e 4, show that RII Z
p
.
13. I n Exampl e 6, show that R/I
p
= 77(Z
p
).
14. I n Exampl e 8, verify that the mapping i/>given is an isomorphism of 7? onto C.
15. If 7, 7 are ideal s of 7?, l et IJ be the set of al l sums of elements of the form
ij, where i E 7, j E 7. Pro ve that 77 is an ideal of 7?.
16. Show that the ring of 2 X 2 matri ces over the real s has nontri vi al l eft
ideal s (and al so nontrivial right ideal s).
If R, S are rings, define the direct sum of R and S, R S, by
R@S = {(r, s)\ rER,sES]
where (r, s) = (r
u
s
x
) if and onl y if r = r
lt
s = s
] (
and where
(r, s) + (t, u) = (/ + t,s + u), (r, s)(t, u) = (rt, su).
Sec . 3 I deal s , Ho mo mo r p h i s ms , an d Qu o t i en t Ri ngs 147
18. Show that R S is a ring and that the subrings ((/, 0) | / E R] and
((0, s) | s G S] are ideal s of R S i so mo rphi c to R and S, respectivel y.
b real j , show that:
(a) R is a ring.
(b) / is an ideal of R.
(c) RII F F, where F is the fiel d of real numbers.
20. If /, / are ideal s of R, l et R
x
= RII and R
2
= RIJ. Show that <p:R^ A\ R
2
defined by <p(r) = (/- + I, r + J) is a ho mo mo rphi sm of R i nto R
x
R
2
such that Ker <p = I C\ J.
21. L et Zj
5
be the ring of integers mo d 15. Show that Z
1 5
Z
3
Z
5
.
Mi ddl e-Level Problems
22. Let Z be the ring of i ntegers and m, two rel ativel y pri me integers, /,
the mul tipl es of m in Z, and / the mul ti pl es of n in Z.
(a) What is /, n 7?
(b) Use the resul t of Pro bl em 20 to show that there is a o ne-to -o ne
ho mo mo rphi sm from ZII
mn
to Zll
m
ZII
n
.
23. If m, n are rel ativel y pri me, pro ve that Z, Z, Z. (Hint: Use a count-
ing argument to show that the ho mo mo rphi sm of Pro bl em 22(b) is onto.)
*24. Use the resul t of Pro bl em 23 to pro ve the Chinese Remainder Theorem,
which asserts that if rn and n are rel ativel y pri me integers and a, b any in-
tegers, we can find an integer x such that .v = a mo d m and x = b mo d n
simul taneousl y.
25. Let R be the ring of 2 X 2 matri ces over the real numbers; suppose that /
is an ideal of R. Show that / = (0) or I = R. (Contrast this with the resul t of
Pro bl em 16.)
Harder Problems
26. Let R be a ring with 1 and l et S be the ring of 2 X 2 matrices over R. If /
is an ideal of S show that there is an ideal I of R such that I consists of al l
the 2X 2 matri ces over /.
27. If P i , p
2
, , p are distinct o dd pri mes, show that there are exactl y 2"
sol utions of x
l
= x mo d(p! p), where 0 s x < p
x
p.
28. Suppo se that R is a ring who se onl y l eft ideal s are (0) and R itsel f. Pro ve
that ei ther R is a division ring o r R has p el ements, p a pri me, and ab = 0
for every a,b G R.
19. I f R a, b, c real \ and I =
0 0
29. Let R be a ring with 1. An el ement a G R is said to have a /e/r inverse if
/3A = 1 for so me o G R. Show that if the l eft inverse b of a is uni que, then
a/3 = 1 (so b is al so a right inverse of a).
4 . MA X I MA L I DEAL S
This wil l be a section with o ne maj or theo rem. The i mpo rtance of this resul t
wil l onl y beco me ful l y transparent when we discuss fiel ds in Chapter 5. Ho w-
ever, it is a resul t that stands o n its o wn two feet. I t isn't difficul t to pro ve,
but in mathemati cs the co rrel ati o n between difficul t and i mpo rtant isn't al -
ways that high. There are many difficul t resul ts that are of very l ittl e i nterest
and of even l ess i mpo rtance, and so me easy resul ts that are crucial . Of
course, there are so me resul tsmany, manywhi ch are of i ncredi bl e diffi-
cul ty and i mpo rtance.
L emma 4.4.1. Let R be a co mmutati ve ring with unit who se onl y
ideal s are (0) and itsel f. Then R is a fiel d.
Proof. Let a + 0 be in R. Then (a) = [xa | x G R} is an ideal of R, as
we verified in Exampl e 9 in the preceding section. Since a = la G (a), (a) + (0).
Thus, by o ur hypothesi s o n R, (a) = R. But then, by the definition of (a),
every el ement i G R is a mul ti pl e xa of a for so me x G R. I n parti cul ar, be-
cause 1 G R, 1 = ba for so me b G R. This shows that a has the inverse b in R.
So R is a fiel d.
I n Theo rem 4.3.4the Co rrespo ndence Theo remwe saw that if
cp: R R' is a ho mo mo rphi sm of R o nto R' with kernel K, then there is a
1-1 co rrespo ndence between ideal s of R' and ideal s of R that co ntai n K. Sup-
po se that there are no ideal s o ther than K itsel f and R which co ntai n K.
What does this impl y abo ut R"> Since (0) in R' co rrespo nds to K in R, and
R' co rrespo nds to al l of R, we must concl ude that in this case R' has no
ideal s o ther than (0) and itsel f. So if R' is co mmutati ve and has a unit el e-
ment, then, by L emma 4.4.1, R' must be a fiel d.
This pro mpts the fol l owing definition.
Definition. A pro per i deal M o\R is a maximal ideal of R if the onl y
ideal s of R that contai n M are M itsel f and R.
The discussion precedi ng this definition has al ready al most pro ved
for us
Sec . 4 Ma x i ma l I deal s 149
Theo rem 44.2. Let R be a co mmutati ve ring with 1, and l et M be a
maxi mal ideal of R. Then RIM is a fiel d.
Proof. There is a ho mo mo rphi sm of R o nto R' = RIM, and since 1 G 7?
we have that R' has 1 + Af as its unit el ement. (See Pro bl em 3, Section 3).
Because M is a maximal ideal of R, we saw in the discussion abo ve that R'
has no nontrivial ideal s. Thus, by L emma 4.4.1, R' = RIM is a fiel d.
This theo rem wil l be o ur entry into the discussion of fiel ds, for it wil l
enabl e us to construct particul arl y desirabl e fiel ds whenever we shal l need
them.
Theo rem 4.4.2 has a converse. This is
Theo rem 4.4.3. If R is a co mmutati ve ring with 1 and M an ideal of R
such that RIM is a fiel d, then M is a maxi mal ideal of R.
Proof. We saw in Exampl e 2 of Section 3 that the onl y ideal s in a fiel d
F are (0) and F itsel f. Since RIM is a fiel d, it has onl y (0) and itsel f as ideal s.
But then, by the co rrespo ndence given us by Theo rem 4.3.4, there can be no
ideal of R pro perty between M and R. Thus M is a maxi mal ideal of R.
We give a few exampl es of maxi mal ideal s in co mmutati ve rings.
Exampl es
1. Let Z be the integers and M an ideal of Z. As an ideal of Z we certainl y
have that M is an additive subgro up of Z, so must consist of al l mul tipl es of
so me fixed i nteger n. Thus since RIM Z and since Z is a fiel d if and onl y
if n is a pri me, we see that M is a maxi mal ideal of Z if and onl y if M consists
of al l the mul tipl es of so me pri me p. Thus the set of maxi mal ideal s in Z cor-
respo nds to the set of pri me numbers.
2. L et Z be the integers, and l et R = [a + bi\a, b G Z), a subring of
C (j
1
= - 1) . Let M be the set of al l a + bi in R, where 3 | a and 3 | b. We
l eave it to the reader to verify that M is an ideal of R.
We cl aim that M is a maxi mal ideal of R. Fo r suppo se that TV D M and
N M is an ideal of R. So there is an el ement r + si G TV, where 3 does no t
divide r or 3 does no t divide s. Therefo re, 3 \ (r
2
+ s
2
). (Prove using congru-
ences mo d 3 !) But t = r
2
+ s
2
= (r + si)(r - si), so is in TV, since r +si G TV
and TV is an ideal of R. So TV has an i nteger t =r
2
+s
2
not divisibl e by 3. Thus
ut + 3v = 1 for so me integers u, v; but t G TV, hence ut G TV and 3 G M C TV,
so 3L> G TV. Therefo re, 1 = ut + 3v G TV. Therefo re, (A + n')l G TV, since TV is
an ideal of R, for al l a + bi G R. This tel l s us that TV =R. So the onl y ideal of
R abo ve M is i? itsel f. Co nsequentl y, M is a maxi mal ideal of R.
By Theo rem 4.4.2 we kno w that RIM is a fiel d. I t can be shown (see
Pro bl em 2) that RIM is a fiel d havi ng ni ne el ements.
3. Let R be as in Exampl e 2 and l et I = {a + bi \ 5\a and 5|/3}. We assert that
I is not a maxi mal ideal of R.
I n R we can factor 5 = (2 + i)(2 - i). Let M = {x(2 + i ) | x 6 R}. M is
an ideal of R, and since 5 = (2 + i)(2 - i) is in M, we see that / C M . Cl earl y,
/ + M for 2 + i G M and is not in / because 5 j 2. So / + M. Can M = R1
If so, then (2 + /)( + bi) = 1 for so me a, 6. This gives 2a b = \ and
2/3 + A = 0; these two equati o ns impl y that 5a = 2, so a = , b = B u t
| g Z, - | Z; the el ement a + bi = - g/ is no / in R.SoM =t R.
One can show, however, that M is a maximal ideal of R. (See Pro bl em 3.)
4. Let /? = { + /3V5 | A, 6 integers}, which is a subring of the real fiel d under
the sum and pro duct of real numbers. That R is a ring fol l ows from
(A + by/2) + (c + d\/2) = (a + c) + (b + d)V2
and
(A + /3V2)(c + V 2) = (ac + 2bd) + (ad + bc)V2.
Let M = [a + /3V2 G R \ 5\a and 5\b}. M is easil y seen to be an ideal of R.
We l eave it to the reader to show that M is a maxi mal ideal of R and that
RIM is a fiel d having 25 el ements.
5. Let R be the ring of al l real -val ued co nti nuo us functions o n the cl osed
unit interval [0, 1]. We sho wed in Exampl e 5 of Section 3 that if M =
[f&R\ / ( |) = 0}, then M is an ideal of R and RIM is i so mo rphi c to the real
fiel d. Thus, by Theo rem 4.4.3, M is a maxi mal ideal of R.
Of course, if we l et M
y
= {/ G R | f(y) = 0}, where y G [0,1], then M
y
is
al so a maximal ideal . I t can be shown that every maximal ideal in R is of the
formM
y
for some y G [0,1], but to prove it we woul d require some resul ts from
real variabl e theory.
What this exampl e shows is that the maxi mal ideal s in R co rrespo nd to
the poi nts of [0, 1].
PROBL EMS
1. If a, b are integers and 3 \ a or 3 \ b, show that 3 \ (a
2
+ b
2
).
2. Show that in Exampl e 2, RIM is a fiel d having ni ne el ements.
3. I n Exampl e 3, show that M = \x(2 + i) \ x G R] is a maxi mal ideal of R.
Se c . 5 Pol ynomi al Ri ngs 151
4. I n Exampl e 3, show that RIM Z
5
.
5. I n Exampl e 3, show that RII ~ Z
5
Z
5
.
6. I n Exampl e 4, show that M is a maxi mal ideal of R.
7. I n Exampl e 4, show that AVM is a fiel d having 25 el ements.
8. Usi ng Exampl e 2 as a mo del , construct a fiel d having 49 el ements.
We make a short excursion back to congruences mo d p, wherep is an odd
pri me. If a is an integer such that p\a and x
2
= a mo dp has a sol ution x in Z,
we say that a is a quadratic residue mo d p. Otherwise, a is said to be a quadratic
nonresidue mo d p.
9. Show that (p - 1)12 of the numbers 1, 2, .. . , p - 1 are quadrati c
resi dues and (p l )/2 are quadrati c no nresi dues mo d p. [Hint: Sho w
that {x
2
1 x # 0 G Z
p
\ forms a gro up of o rder - l )/2.]
10. Let m > 0 be in Z, and suppo se that m is not a square in Z. Let A =
[a + Vm 6 | a, >G Z }. Pro ve that under the o perati o ns of sum and
pro duct of real numbers i? is a ring.
11. If p is an o dd pri me, l et us set l
p
= [a +Vm b p\a and p\b), where
a + V m b z R, the ring in Pro bl em 10. Sho w that I
p
is an i deal of R.
12. If m is a quadratic nonresidue mod./?, show that the ideal / in Pro bl em
11 is a maxi mal ideal of R.
13. I n Pro bl em 12 show that RII
P
is a fiel d havi ng p
2
el ements.
5. POL YNOMI A L RI NGS
The materi al that we consider in this section invol ves the no ti o n of po l yno -
mial and the set of al l pol ynomi al s o ver a given fiel d. We ho pe that mo st
readers wil l have so me famil iarity with the no ti o n of po l yno mi al from thei r
high school days and wil l have seen so me of the things o ne does with po l yno -
mial s: factoring them, l ooki ng for their ro o ts, dividing o ne by ano ther to get
a remai nder, and so on. The emphasi s we shal l give to the concept and al ge-
brai c object kno wn as a po l yno mi al ring wil l be in a qui te different direction
from that given in high school .
Be that as it may, what we shal l strive to do here is to i ntro duce the
ring of po l yno mi al s over a fiel d and show that this ring is amenabl e to a care-
ful dissection that reveal s its i nnermo st structure. As we shal l see, this ring is
very wel l -behaved. The devel o pment shoul d remi nd us of what was do ne for
the ring of i ntegers in Section 5 of Chapter 1. Thus we shal l run into the ana-
l og of Eucl i d's al gorithm, greatest co mmo n divisor, divisibil ity, and possibl y
mo st i mpo rtant, the appro pri ate anal og of pri me number. This wil l l ead to
uni que factorization of a general po l yno mi al into these "pri me po l yno -
mi al s," and to the nature of the ideal s and the maxi mal ideal s in this new
setting.
But the pol ynomi al ring enjoys o ne feature that the ring of i ntegers did
no t: the notion of a ro o t of a pol ynomi al . The study of the nature of such
rootswhi ch wil l be do ne, for the most part, in the next chapterco nsti -
tutes a l arge and i mpo rtant part of the al gebraic history of the past. I t goes
under the titl e Theory of Equations, and in its ho no rabl e past, a l arge vari ety
of magnificent resul ts have been o btai ned in this area. Hopeful l y, we shal l
see so me of these as o ur devel o pment progresses.
Wi th this sketchy outl i ne of what we i ntend to do o ut of the way, we
no w get do wn to the nitty-gritty of doi ng it.
Let F be a fiel d. By the ring of polynomials in x over F, which we shal l
al ways write as F[x], we mean the set of al l formal expressi ons p(x) =
a
Q
+ a
x
x + + a
n
-ix"~
l
+ ax", n s 0, where the a,, the coefficients of
the polynomial p(x), are in F. We so meti mes empl oy the al ternati ve no ta-
tion: p(x) = a
0
x" + a - [ A' "
- 1
+ + a. I n F[x] we define equal ity, sum, and
pro duct of two pol ynomi al s so as to make of F[x] a co mmutati ve ring as fol -
l ows:
1. Equality. We decl are p(x) = a
0
+ a
x
x + + ax" and q(x) =
b
0
+ b
x
x + + bx" to be equal if and onl y if their co rrespo ndi ng coeffi-
cients are equal , that is, if and onl y if a, = 6
;
for al l / > 0.
We co mbi ne this definition of equal ity of pol ynomi al s p(x) and q(x)
with the convention that if
q(x) = b
0
+ b
x
x + + b
n
x"
and if b
m + :
= = b = 0, then we can dro p the l ast n - m terms and wri te
q (x) as
q(x) = b
0
+ b
lX
+ + b
m
x
m
.
This conventi on is o bserved in the definition of addition that fol l ows,
where s is the l arger of m and n and we add coefficients a
n+1
= = a
s
= 0
if n < s or b
m + l
= = b
s
= 0 if m < s.
2. Addition. If p(x) = a
Q
+ a
x
x + + ax" and q(x) = b
0
+ b
x
x
+ + b,x'", we decl are p(x) + q(x) = c
0
+ cpt + + c
s
x\ where for
each c, = a
{
+ >,-.
So we add pol ynomi al s by addi ng thei r co rrespo ndi ng coefficients.
The definition of mul tipl ication is a l ittl e mo re compl i cated. We define
it l oosel y at first and then mo re precisel y.
Sec. 5 Po l y n o mi al Ri n g s 153
3. Multiplication. If p(x) = a
0
+ a
L
x + + a
n
x " and q(x) =
b
0
+ b
x
x + + b,x'", we decl are p(x)q(x) = c
0
+c
s
x + + c,x', where
the Cj are determi ned by mul tipl ying the expression o ut formal l y, using the
distributive l aws and the rul es of expo nents x " x " = x"
+
", and col l ecting
terms. Mo re formal l y,
Cj = Qjb
Q
+ a
{
_
x
b
x
+ + a
x
bj_
x
+ a
0
bj for every i.
We il l ustrate these o perati o ns with a simpl e exampl e, but first a no ta-
ti onal device: If so me coefficient is 0, we just omi t that term. Thus we wri te
9 +Ox +7x
2
+Ox
3
- I 4x
4
as 9 +7x
2
- 14x
4
.
Let p(x) = 1 +3x
2
, q(x) =4 - 5x +7x
2
- x
3
. Then p(x) +q(x) =
5 - 5x +10x
2
- x
3
whil e
p(x)q(x) = (1 + 3x
2
)(4 - 5x + 7x
2
- x
3
)
= 4 - 5x +7x
2
- x
3
+ 3x
2
(4 - 5x +7x
2
- x
3
)
= 4 - 5x +7x
2
- x
3
+ 12x
2
- 15x
3
+21x
4
- 3x
5
= 4 - 5x +19x
2
- 16x
3
+21x
4
- 3x
5
.
Try this pro duct using the c,- as given abo ve.
I n so me sense this definition of F[x] is no t a definition at al l . We have
indul ged in so me hand waving in it. But it wil l do . We coul d empl oy se-
quences to formal l y define F[x] mo re precisel y, but it woul d merel y cl oud
what to most readers is wel l kno wn.
The first remark that we makeand do no t verifyis that F[x] is a
co mmutati ve ring. To go thro ugh the detail s of checking the axioms for a
co mmutati ve ring is a straightforward but l abori ous task. Ho wever, it is im-
po rtant to no te
L emma 4.5.1. F[x] is a co mmutati ve ring with unit.
Definition. If p(x) = a
0
+ ape + + a x " and a = 0, then the
degree of p(x), deno ted by degp(x), is n.
So the degree of a pol ynomial p(x) is the highest po wer of x that occurs in
the expression for p(x) with a no nzero coefficient. Thus deg(x - x
2
+x
4
) =4,
deg(7x) =1, deg 7 =0. (No te that this definition does not assign a degree to
0. It is, ho wever, so meti mes conveni ent to ado pt the conventi on that the de-
gree of 0 be co, in which case many degree rel ated resul ts wil l hol d in this
extended context.) The pol ynomi al s of degree 0 and the pol ynomi al 0 are
cal l ed the constants; thus the set of constants can be identified with F.
The degree function on F[x] wil l pl ay a simil ar rol e to that pl ayed by
the size of the integers in Z, in that it wil l pro vi de us with a versi o n of Eu-
cl id's Al go ri thm for F[x].
One i mmedi ate and i mpo rtant pro perty of the degree function is that it
behaves wel l for pro ducts.
Lemma 4.5.2. If p(x), q(x) are no nzero el ements of F[x], then
deg(p(x)q(x)) = degp(x) + degq(x).
Proof. Let m = degp(x) and n = deg q(x); thus the po l yno mi al p (x) =
a
0
+ a
t
x + + a
m
x"', where a
m
= 0, and the po l yno mi al q(x) = 6
0
+
b
x
x + + bx", where b # 0. The highest po wer of x that can occur in
p(x)q(x) is x'"
+
", from o ur definition of the pro duct. What is the coefficient
of x"'
+
"? The onl y way that x"'
+
" can occur is from (a
m
x'")(bx") =
a
m
b
n
x"'
+n
. So the coefficient of x"'
+
" i np(x)q(x) is a
m
b, which is no t 0, since
a, + 0, b + 0. Thus deg(p(x)q(x)) = m + n = degp(x) + degq(x), as
cl aimed in the l emma.
One al so has so me i nfo rmati o n abo ut deg(p(.t) + q(x)). This is
Lemma 4.5.3. If p(x), q(x) (E F[x] and p(x) + q(x) # 0, then
deg(p(x) + q(x)) < max(degp(x), deg q(x)).
We l eave the proof of L emma 4.5.3 to the reader. I t wil l pl ay no rol e in
what is to come, whereas L emma 4.5.2 wil l be i mpo rtant. We put it in so that
the " +" shoul d no t feel sl ighted vis-a-vis the pro duct.
An i mmedi ate co nsequence of L emma 4.5.2 is
Lemma 4.5.4. F[x] is an i ntegral do mai n.
Proof If p(x) * 0 and q(x) # 0, then degp{x) > 0, deg q(x) > 0, so
deg(p(x)q(x)) = degp(x) + deg q(x) ^ 0. Therefo re, p(x)q(x) has a degree,
so canno t be 0 (which has no degree assigned to it). Thus F[x] is an i ntegral
do mai n.
One of the thi ngs that we were o nce forced to l earn was to di vi de
o ne po l yno mi al by ano ther. Ho w did we do this? The pro cess was
cal l ed long division. We i l l ustrate with an exampl e ho w thi s was do ne,
for what we do in the exampl e is the mo del of what we shal l do in the gen-
eral case.
We want to divide 2x
2
+ 1 into x
4
- Ix + 1. We do it schematical l y as
fol l ows:
(2x
2
+ 1)1^
-Ix + 1
Ix + 1
i
_ 4
-lx + \\
and we i nterpret this as:
lx + 1 = (2x
2
+ l)(fx
2
- i) + (-7* + f)
and lx + f is cal l ed the remainder in this division.
What exactl y did we do? First, where did the \x
2
come from? I t came
from the fact that when we mul tipl y 2x
2
+ 1 by \x
2
we get x
4
, the highest
po wer occurring in x
4
- lx + 1. So subtracting |x
2
(2x
z
+ 1) from x
4
- lx + 1
gets rid of the x
4
term and we go o n to what is l eft and repeat the pro cedure.
Thi s "repeat the pro cedure" suggests i nducti on, and that is ho w we
shal l carry o ut the proof. But keep in mi nd that al l we shal l be doing is what
we did in the exampl e abo ve.
What this gives us is so methi ng l ike Eucl i d's Al go ri thm, in the integers.
Ho wever, here we cal l it the Division Algorithm.
Theo rem 4.5.5 (Division Al go ri thm). Gi ven the pol ynomial s /(x),
g(x) G F[x], where g(x) = 0, then
where q(x), r(x) G F[x] and r(x) = 0 or deg r(x) < degg(x).
Proof. We go by induction on deg/fx). If either /(x) = 0 or deg/(x) <
degg(x), then /(x) = 0g(x) + /(x), which satisfies the concl usion of the
theo rem.
So suppo se that deg/(x) > degg(x); thus the pol ynomi al /(x) =
a
0
+ a
x
x + + a
m
x"', where a, # 0 and the pol ynomi al g(x) = b
0
+ h
x
x
+ + bx", where b i= 0 and where m > n.
f(x) = q(x)gix) + r(x),
Co nsi der
'g(x) = ^x"'~"(b
0
+ b,x+ + b,,x")
Thus (a,/b)x'"~"g(x) has the same degree and same leading coefficient as does
f(x), so f(x) - (a,lb,
l
)x'"~"g(x) = h(x) is such that the rel ati o n deg ft (x) <
degf(x) hol ds. Thus, by i nducti o n,
h(x) = q
x
(x)g(x) + r(x), where c
h
{x),r(x) GF[x]
and r(x) = 0 or deg r(x) < deg g(x). Rememberi ng what h(x) is, we get
If q(x) = (ajb
n
)x
m
" + q[(x), we have achieved the form cl ai med in the
theo rem.
The Division Al go ri thm has o ne i mmedi ate appl ication: I t al l ows us to
determi ne the nature of al l the ideal s of F[x]. As we see in the next theo rem,
an ideal of F[x] must merel y consist of al l mul tipl es, by el ements of F[x], of
so mefixed pol ynomial .
Theo rem 4.5.6. If / * (0) is an ideal of F[x], then / = [f(x)g(x) | f(x) G
F[x)}; that is, / consists of al l mul tipl es of the fixed po l yno mi al g(x) by the el -
ements of F[x].
Proof. To pro ve the theo rem, we need to pro duce that fixed po l yno -
mial g(x). Where are we going to dig it up? The o ne control we have numeri -
cal l y on a given po l yno mi al is its degree. So why no t use the degree function
as the mechani sm for finding g(x)1
Since / (0) there are el ements in I having no nnegati ve degree. So
there is a pol ynomi al g(x) # 0 in / of mi ni mal degree; that is, g(x) # 0 is in /
and if 0 t(x) G /, then deg t(x) s deg g(x). Thus, by the division al gori thm,
t(x) = q(x)g(x) + r(x), where r(x) = 0 or deg r(x) < degg(.t). But since
g(x) G / and / is an ideal of F[x], we have that q(x)g(x) G /. By assumpti o n,
t(x) G I, thus t(x) - q(x)g(x) is in /, so r(x) = t(x) - q(x)g(x) is in /. Since
g(x) has mi ni mal degree for the el ements of J and r(x) G /, deg r(x) cannot
be l ess than degg(.v). So we are l eft with r(x) = 0. But this says that t(x) =
q(x)g(x). So every el ement in / is a mul ti pl e of g(x). On the o ther hand,
h(x) x"'-"g(x) =
gi
(x)g(x) + r(x)
so
since g(x) G / and / is an ideal of F[x], f(x)g(x) G / for al l f(x) G F[x]. The
net resul t of al l this is that I = [f(x)g(x) | f(x) G F[x]}.
Definition. An integral do mai n R is cal l ed aprincipal ideal domain if
every i deal / in R is of the form I = {xa \ x G R} for so me a G I.
Theo rem 4.5.6 can be stated as: F[x] is a principal ideal domain.
We shal l write the ideal generated by a given polynomial, g(x), namel y
{f(x)g(x)\f(x)EF[x]},
aS
(g(x)).
The proof sho wed that if / is an ideal of F[x], then / = (g(x)), where
g(x) is a pol ynomial of l owest degree contai ned in /. But g(x) is no t unique, for
if a + 0 G F, then ag(x) is in / and has the same degree as g(x), so I = (ag(x)).
To get so me sort of uni queness in al l this, we singl e out a cl ass of pol y-
nomi al s.
Definition. f(x) G F[x] is a monic polynomial if the coefficient of its
highest po wer is 1.
Thus f(x) is moni c means that
f(x) = x" + a,,-!*"-
1
+ + a
x
x + a
0
.
We l eave to the reader to show that if / is an ideal of F[x], then there is
only one mo ni c po l yno mi al of l owest degree in /. Singl ing this o ut as the gen-
erato r of / does give us a "mo ni c" uni queness for the generati o n of /.
Our next step in this paral l el devel o pment with what happens in the in-
tegers is to have the no ti o n of o ne po l yno mi al dividing ano ther.
Definition. Suppo se f(x), g(x) G F[x], with g(x) + 0. We say that
g(x) divides f(x), wri tten as g(x) \f(x), if f(x) = a(x)g(x) for some a(x) G
F[x].
No te that if f(x) + 0 and g(x) \ f(x), then deg g(x) s deg/(x) by L emma
4.5.2. Mo reo ver, the ideal s (/(*)) and (g(x)) of F[x], generated by f(x) and
g(x), respectivel y, satisfy the containing rel ati o n (f(x)) C (g(x)). (Prove!)
We again emphasi ze the paral l el i sm between Z, the integers, and F[x]
by turni ng to the notion of greatest common divisor. I n o rder to get so me sort
of uni queness, we shal l insist that the greatest co mmo n divisor al ways be a
mo ni c po l yno mi al .
Definition. Fo r any two pol ynomial s /(x) and g(x) G F]x] (not bo th 0),
the greatest common divisor of f(x) and g(x) is the mo ni c po l yno mi al d(x)
[someti mes deno ted (f(x), g(x))], such that:
(a) d(x) | f(x) and d(x) | g(x).
(b) If HO |/(x) and /?(x) | g(x), then /7(x) ] d(x).
Al tho ugh we defined the greatest co mmo n divisor of two po l yno mi al s,
we nei ther kno w, as yet, that it exists, no r what its form may be. We coul d
define it in ano ther, and equi val ent, way as the monic polynomial of highest
degree that divides both f(x) and g(x). If we did that, its exi stence wo ul d be
auto mati c, but we woul d not kno w its form.
T heorem 4.5.7. Gi ven f(x) and g(x) + 0 in F[x], then thei r greatest
co mmo n divisor d(x) F[x] exists; mo reo ver, d(x) = a(x)f(x) + b(x)g(x)
for so me a(x), b(x) G F[x\.
Proof. Let / be the set of al l r(x)/(x) + s(x)g(x) as r(x), s(x) run
freel y over F[x]. We cl aim that I is an ideal of R. Fo r,
(r
x
{x)f(x) +
Sl
(x)g(x)) + (r
2
(x)f(x) + s
2
(x)g(x))
= (r
x
(x) + r
2
(x))f(x) + (
Si
(x) + s
2
(x))g(x),
so is again in I, and for r(x) G F[x],
t(x)(r(x)f(x) + s(x)g(x)) = (t(x)r(x))f(x) + (t(x)s(x))g(x),
so it, to o , is again in I. Thus / is an ideal of F[x]. Since g(x) = 0, we kno w
that 7^0, since bo th f(x) and g(x) are in I.
Since 7 ^ 0 is an ideal of F[x], it is generated by a unique monic po l yno -
mial d(x) (Theo rem 4.5.6). Since f(x), g(x) are in 7, they must then be mul ti -
pl es of d(x) by el ements of F[x]. This assures us that d(x) \f(x) and
d(x)\g(x).
Because d(x) G 7 and 7 is the set of al l r(x)f(x) + s(x)g(x), we have
that d(x) = a(x)f(x) + b(x)g(x) for so me appro pri ate a(x), b(x) G F[x].
Thus if h(x) \f(x) and h(x) \g(x), then h(x) | (a(x)f(x) + b{x)g(x)) = d(x).
So d(x) is the greatest co mmo n divisor off(x) and g(x).
This pro ves the theo rem; the uni queness of d(x) is guaranteed by the
demand that we have made that the greatest co mmo n divisor be mo ni c.
Ano ther way to see the uni queness of ^(x) is from
Lemma 4.5.8. If /(x) =0, g(x) * 0 are in F[x] and fix) | g(x) and
g(
x
) I f(x), then f(x) = ag(x), where a G F.
Proof. By the mutual divisibil ity condi ti on o n f(x) and g(x) we have,
by L emma 4.5.2, deg/(x) < degg(x) < deg/f x), so deg/(x) =degg(x). But
f(x) = a(x)g(x), so
degf(x) = dega(x) + degg(x) = deg a(x) + deg/(x),
in co nsequence of which deg a(x) = 0, so a(x) = a, an el ement of F.
We l eave the proof of the uni queness of the greatest co mmo n divisor
via L emma 4.5.8 to the reader.
Definition. The pol ynomi al s /(x), g(x) in F[x] are said to be relatively
prime if thei r greatest co mmo n divisor is 1.
Al tho ugh it is merel y a very special case of Theo rem 4.5.7, to empha-
size it and to have it to refer to, we state:
T heorem 4.5.9. If f(x), g(x) G F[x] are rel ativel y pri me, then
a(x)f(x) + b(x)g(x) = 1 for some- a(x), b(x) G F[x]. Conversel y, if
a(x)f(x) + b(x)g(x) = 1 for so me a(x), b(x) G F[x], then/ (x) and g(x) are
rel ativel y pri me.
Proof. We l eave this "co nversel y" part to the reader as exercise.
As with the integers, we have
T heorem 4.5.10. If q(x) and f(x) are rel ativel y pri me and if
<?(*) I f(x)g(x), then q(x) \ g(x).
Proof. By Theo rem 4.5.9 a(x)f(x) + b(x)q(x) = 1 for so me a(x),
b{x) G F[x]. Therefo re,
a(x)f (x)g(x) + b(x)q(x)g(x) = g(x). (1)
Since q(x) \ b(x)g(x)q(x) and q(x) \f(x)g(x) by hypothesi s, q(x) divides the
l eft-hand side of the rel ation in (1). Thus q(x) divides the right-hand side
of (1), that is, q{x) \ g(x), the desi red concl usion.
We are no w ready to singl e o ut the i mpo rtant cl ass of pol ynomial s that
wil l pl ay the same rol e as pri me objects in F[x] as did the pri me numbers in Z.
Definition. The pol ynomial p(x) G F[x] is irreducible if p(x) is of po s-
itive degree and given any po l yno mi al /(x) in F[x], then either p(x) |/(x) o r
p(x) is rel ativel y pri me to /(x).
We shoul d no te here that the definition impl ies that a polynomial p(x)
of positive degree is irreducible in F[x] if and only ifp(x) cannot be factored
as a product of two polynomials of positive degree. I n o ther wo rds, if p(x) =
a(x)b(x), where a(x) and b(x) are in F[x], then ei ther a(x) is a co nstant (that
is, an el ement of F), o r b(x) is constant. The proof of this fact is very simil ar
to the proof of an anal o go us o bservati o n concerni ng two equi val ent defini-
tions of a pri me number.
First suppo se that p(x) (of positive degree) canno t be facto red as a
pro duct of two no n-co nstant po l yno mi al s. Then, given any f(x) in F[x], we
have onl y two possibil ities for (p(x), f(x)), namel y 1 or a mo ni c po l yno mi al of
the form c-p(x), where c is an el ement of F. Thus (p(x),f(x)) =1 or p(x) | fix),
which shows that p(x) is i rreduci bl e.
No w l et p(x) be i rreduci bl e in F[x] and suppo se p(x) = a(x)b(x) for
so me a(x), b(x) in F[x]. Acco rdi ng to the definition, we must have
p(x) | a(x) or (p(x), a(x)) = 1. If p(x) \ a(x), then b(x) must be a co nstant. If,
o n the o ther hand, p(x) and a(x) are rel ativel y pri me, then by Theo rem
4.5.10, p(x) | b(x), and in this case a(x) must be a constant. This shows that an
irreducibl e pol ynomi al canno t be factored as a pro duct of two no n-co nstant
pol ynomial s.
No te that the irreducibil ity of a po l yno mi al depends o n the fiel d F. Fo r
i nstance, the po l yno mi al x
2
- 2 is i rreduci bl e in Q[x], where Q is the fiel d of
rati o nal numbers, but x
2
- 2 is no t i rreduci bl e in R[x], where U is the fiel d of
real numbers, for in U[x]
x
2
- 2 =(x - V2)(x + V2).
Corol l ary to Theo rem 4.5.10. If p(x) is irreducibl e in F[x] and
p(x) | a
x
(x)a
2
{x) a
k
(x), where a
t
(x),.. ., a
h
(x) are in F[x], thenp( x) \ a
t
(x)
for so me
Proof. We l eave the pro o f to the reader. (See Theo rem 1.5.6.)
Asi de from its o ther pro perti es, an irreducibl e po l yno mi al p (x) in F[x]
enjoys the pro perty that (p(x)), the ideal generated by p(x) in F[x], is a max-
i mal ideal of F[x]. We pro ve this no w.
Theo rem 4.5.11. If p(x) E F[x], then the ideal (p(x)) generated by
p(x) in F[x] is a maxi mal ideal of F[x] if and onl y if p(x) is irreducibl e in F[x].
Proof. We first pro ve that if p(x) is irreducibl e in F[x], then the i deal
M = (p(x)) is a maxi mal i deal of F[x]. Fo r, suppo se that N is an ideal of
F[x], and WDM . B y Theo rem 4.5.6,
N = (f(x)) for so me f(x) G F[x].
Because p(x) G M C N,p{x) = a(x)/(x), since every el ement in TV is of this
form. But/?(x) is irreducibl e in F[x], hence a(x) is a constant o r/ (x) is a con-
stant. If a(x) = a G F, thenp(x) = af(x), so f(x) = Thus f(x) G M,
which says that N G M, hence N = M. On the o ther hand, if /(jc) = b G F]
then 1 = D
_ 1
o G N, since A
7
is an ideal of F[x], thus g(x)l G N for al l
g(x) G F[x]. This says that N = -Ffx]. Therefo re, we have shown M to be a
maxi mal i deal of F[x].
I n the o ther direction, suppo se that M = (p(x)) is a maxi mal ideal of
F[x]. If p(x) is no t irreducibl e, then p(x) = a(x)fc(x), where dega(x) > 1,
deg 6(x) > 1. Let A
7
= (a(x)); then, since p(x) = a(x)b(x), p(x) G N. There-
fore, M C N. Since deg a(x) > 1, A
7
= ((x)) ^ F[x], since every el ement in
(a(x)) has degree at l east that of a(x). By the maxi mal i ty of M we concl ude
that M = N. But then a(x) G N = M, which tel l s us that a(x) = f(x)p(x);
co mbi ned wi th p(x) = a(x)b(x) = b(x)f(x)p(x), we get that b(x)f(x) = 1.
Since deg 1 = 0 < deg b (x) < deg (6 (x)/(x)) = deg 1 = 0, we have reached a
contradi cti on. Thusp( x) is i rreduci bl e.
This theo rem is i mpo rtant because it tel l s us exactl y what the maxi mal
ideal s of F[x] are, namel y the ideal s generated by the irreducibl e po l yno mi -
al s. If M is a maxi mal ideal of F[x], F[x]/M is a fiel d, and this fiel d contains
F (or mo re precisel y, the fiel d {a + M | a G F}, which is i somorphi c to F).
This al l ows us to construct decent fiel ds K D F, the decency of which l ies in
that p (x) has a root in K. The exact statement and expl anati o n of this we
po stpo ne until Chapter 5.
The l ast topi c in this direction that we want to discuss is the factoriza-
tion of a given pol ynomi al as a pro duct of i rreduci bl e ones. No te that if
p(x) = a
0
x" + fl jx"
-1
+ + a^
L
x + a, a
0
i= 0, is irreducibl e in F[x], then
so is a^pix) irreducibl e in F[x]; ho wever, a
()
1
p(x) has the advantage of
bei ng mo ni c. So we have this mo ni c i rreduci bl e po l yno mi al trivial l y o btai n-
abl e fromp (x) itsel f. This wil l al l ow us to make mo re precise the uni queness
part of the next theo rem.
Theo rem 4.5.12. Let /(x) G F[x] be of positive degree. Then ei ther
/(x) is i rreduci bl e in F[x] or f(x) is the pro duct of irreducibl e pol ynomi al s in
F[x]. I n fact, then,
f(
X
) = fl
A
(x)'>
2
(xr-
Pk
(x)'"\
where a is the l eading coefficient of fix), Pi(x),. . ., p
k
(x) are mo ni c and ir-
reduci bl e in F[x], m
x
> 0, . . . , m
k
> 0 and this factorization in this form is
uni que up to the o rder of the p
t
(x).
Proof. We first show the first hal f of the theo rem, namel y that f(x) is
irreducibl e or the pro duct of i rreduci bl es. The proof is exactl y the same as
that of Theo rem 1.5.7, with a sl ight, obvious adjustment.
We go by i nducti o n o n deg/(x). If deg/f x) = 1, then f(x) = ax + b
with a + 0 and is cl earl y i rreduci bl e in F[x]. So the resul t is true in this case.
Suppo se, then, that the theo rem is correct for al l a(x) E F[x] such that
dega(x) < degf(x). If f(x) is irreducibl e, then we have no thi ng to pro ve.
Otherwi se, f(x) = a(x)b(x), a(x) and b(x) E F[x] and degfl (x) < deg/(x)
and deg b(x) < deg/(x). By the i nducti on, a(x) [and b(x)] is i rreduci bl e or is
the pro duct of irreducibl es. But then f(x) is the pro duct of i rreduci bl e pol y-
nomial s in F[x]. This co mpl etes the i nducti o n, and so pro ves the o peni ng hal f
of the theo rem.
Now to the uni queness hal f. Agai n we go by induction on deg/(x). If
deg/(.v) = 1, then/f x) is irreducibl e and the uni queness is cl ear.
Suppo se the resul t true for pol ynomi al s of degree l ess than deg/f x).
Suppo se that
f{x) = a
Pl
(x)>"ip
2
(x)'"
2
-p^x)"* =
aqi
(x)"* q
r
(x)'\
where the p
t
(x), q
t
{x) are mo ni c i rreduci bl e pol ynomi al s and the m
h
n
{
are
al l positive and a is the l eadi ng coefficient of f(x), that is, the coefficient
of the highest po wer term of f(x). Since p
l
(x)\f(x), we have that
Pi(
x
) I q
r
(x)"
r
, so by the corol l ary to Theo rem 4.5.10, p
x
(x) \ q
t
(x)
for so me i. Since q
t
(x) is mo ni c and irreducibl e, as is pi(x), we get p
x
(x) =
q,(x). We can suppo se (on renumberi ng) thatp
x
{ x) = q
x
(x). Thus
= ap^xy^p.ix)"'
2
p
k
(x)"'*
= ap
1
(x)"^
1
q
2
(xr
2
q
r
(x)'\
By i nducti on we have uni que factorization in the requi red fo rm for
f(
x
)/Pi(x), who se degree is l ess than deg/(x). Hence we o btai n that
nil - 1 = i 1 (so nil
= w
i X
m
i
= n
i , , m
k
= n
k
, r = k andp
2
( x ) =
q
2
(x), . . . , p
k
(x) = q
k
(x), on renumberi ng the g's appropriatel y. This com-
pl etes the induction and proves the theo rem.
We have po i nted o ut ho w simil ar the situation is for the i ntegers Z and
the po l yno mi al ring F[x]. This suggests that there shoul d be a wi der cl ass of
rings, of which the two exampl es Z and F[x] are special cases, for which
much of the argumentati o n works. I t wo rked for Z and F[x] because we had
a measure of size in them, ei ther by the size of an integer or the degree of a
pol ynomi al . This measure of size was such that it al l owed a Eucl i d-type al go-
rithm to hol d.
Sec. 5 Pol ynomial s Rings 163
Definition. An integral do mai n R is a Euclidean ring if there is a
function d from the nonzero el ements of R to the no nnegati ve integers that
satisfies:
(a) Fo r a + 0, b * 0 G R, d(a) < d(ab).
(b) Gi ven a 0, b = 0, there exist g and r G R such that b = qa + r, where
r = 0 or < rf(fl ).
The i nterested student shoul d try to see which of the resul ts pro ved for
po l yno mi al rings (and the integers) ho l d in a general Eucl i dean ring. Asi de
from a few pro bl ems invol ving Eucl i dean rings, we shal l no t go any further
with this i nteresti ng cl ass of rings.
The final co mment we make here is that what we did for pol ynomi al s
over a fiel d we coul d try to do for pol ynomi al s over an arbi trary ring. That is,
given any ring R (co mmutati ve or no nco mmutati ve), we coul d define the
po l yno mi al ring R[x] in x over R by defining equal ity, addition, and mul tipl i-
cation exactl y as we did in F[x], for F a fiel d. The ring so constructed, R[x], is
a very i nteresti ng ring, who se structure is tightl y i nterwo ven with that of R it-
sel f. I t wo ul d be to o much to expect that al l , or even any, of the theo rems
pro ved in this section wo ul d carry over to R[x] for a general ring R.
PROBL EMS
I n the fol l owing pro bl ems, F wil l al ways deno te a fiel d.
Easi er Pro bl ems
1. If F is a fiel d, show that the onl y invertibl e el ements in F[x] are the
no nzero el ements of F.
2. If R is a ring, we introduce the ring R[x] of pol ynomial s in x over R, just as
we did F[x]. Defining deg/(x) for f(x) G R[x] as we did in F[x], show that:
(a) deg(/(x)g(x)) < deg/(x) + degg(x) if /(x)g(x) 0.
(b) There is a co mmutati ve ring R such that we can find f(x), g(x) in
R[x] with deg(/(x)g(x)) < deg f(x) + degg(x).
3. Fi nd the greatest co mmo n divisor of the fol l owing pol ynomi al s over Q,
the fiel d of rati o nal numbers.
(a) x
3
- 6x + 7 andx + 4.
(b) x
2
- 1 and 2x
7
- 4x
5
+ 2.
This l eads us to define a cl ass of rings, the Euclidean rings.
(c) 3x
2
+ 1 and x
6
+ x
4
+ x + 1.
(d) x
3
- 1 and x
1
- x
4
+ x
3
- 1.
4. Pro ve L emma 4.5.3.
5. I n Pro bl em 3, l et I = [f(x)a(x) + g(x)b(x)}, where /(x), g(x) run over
Q[x] and a(x) is the first po l yno mi al and b(x) the second o ne in each part
of the pro bl em. Fi nd d(x), so that I = (d(x)) for Parts (a), (b), (c),
and (d).
6. Ug(x),f(x) E F[x] and g(x) \f(x), show that (/(*)) C (g(x)).
7. Pro ve the uni queness of the greatest co mmo n divisor of two po l yno mi al s
in F[x] by using L emma 4.5.8.
8. If f(x), g(x) G F[x] are rel ativel y prime and /(x) | h(x) and g(x) | h(x),
show that/Xx)g(x) | h(x).
9. Pro ve the Corol l ary to Theo rem 4.5.10.
10. Show that the fol l owing po l yno mi al s are irreducibl e over the fiel d F indi-
cated.
(a) x
2
+ 7 over F = real fiel d = U.
(b) x
3
- 3x + 3 over F = rati o nal fiel d =<Q>.
(c) x
2
+ x + 1 over F = Z
2

(d) x
2
+ 1 o ver F = Z
1 9
.
(e) x
3
- 9o ver F = Z
1 3
.
(f) x
4
+ 2x
2
+ 2 o ver F = Q.
11. If p(x) G F[x] is of degree 3 and p{x) = a
Q
x
3
+ a
x
x
2
+ a
2
x + a
3
, show
that p(x) if i rreduci bl e o ver F if there is no el ement r G F such that
p(r) = a
0
r
3
+ a
x
r
2
+ a
2
r + a
3
= 0.
12. If F C K are two fiel ds and /(x), g(x) G Ffx] are rel ativel y pri me in F[x],
show that they are rel ativel y pri me in K[x].
13. Let IR be the fiel d of real numbers and C that of compl ex numbers. Sho w
that R[x]/(x
2
+ 1) = C. [Hint: If A = (R[x]/(x
2
+ 1), l et u be the i mage of
x in A; show that every el ement in A is of the form a + bu, where
A, b G U and u
2
= - 1.]
14. Let F = Z
u
, the integers mo d 11.
(a) Let p(x) = x
2
+ 1; show that p(x) is irreducibl e in F[x] and that
F[x)/(p(x)) is a fiel d having 121 el ements.
(b) Let p(x) = x
3
+ x + 4 G F[x]; show that p(x) is irreducibl e in F[x]
and that F[x]l(p(x)) is a fiel d havi ng l l
3
el ements.
Sec. 5
Po l y n o mi al s Ri ngs 165
15. Let F = T
p
be the fiel d of i ntegers mo d p, where p is a pri me, and l et
q(x) G be irreducibl e of degree n. Sho w that F[x]/(<7(x)) is a fiel d
having at mo st p" el ements. (See Pro bl em 16 for a mo re exact state-
ment.)
16. Let F, q(x) be as in Pro bl em 15; sho w that F[x]/(q(x)) has exactl y p" el e-
ments.
17. Let Pi(x), p
2
(x),... , p
k
{x) G F[x] be distinct irreducibl e pol ynomi al s
and l et q(x) = p
x
(x)p
2
(x) p
k
(x). Show that
(q(x)) (
Pl
(x)) (p
2
(x)) ^(
Pk
(x)y
18. Let F be a finite fiel d. Show that F[x] contai ns irreducibl e pol ynomi al s of
arbitraril y high degree. (Hi nt: Try to i mi tate Eucl i d's proof that there is
an infinity of pri me numbers.)
19. Co nstruct a fiel d havi ngp
2
el ements, for p an o dd pri me.
20. If R is a Eucl i dean ring, show that every ideal of R is principal .
21. If A' is a Eucl i dean ring, show that R has a uni t el ement.
22. If R is the ring of even integers, show that Eucl i d's al gori thm is fal se in R
by exhibiting two even integers for which the al gorithm do es not hol d.
Harder Pro bl ems
23. Let F = Z
7
and l et p(x) = x' - 2 and q{x) = x
3
+ 2 be in F\x\. Show
that p(x) and q(x) are irreducibl e in F[x] and that the fiel ds F[x]/(p(x))
and F[x]/(q(x)) are i somorphi c.
24. Let Q be the fiel d of rati o nal numbers, and l et q(x) = x
2
+ x + 1 be in
Q(x). If a is a compl ex number such that a
2
+ a + 1 = 0, show that the
set [a + ba \ a, b G 0} is a fiel d in two ways; the first by showing it to be
i somorphi c to so methi ng you kno w is a fiel d, the second by showing that
if a + ba + 0, then its inverse is of the same form.
25. If p is a pri me, show that q(x) = 1 + x + x
2
+ x
p
~
l
is irreducibl e in
Q[x].
26. Let R be a co mmutati ve ring in which a
2
= 0 onl y if a = 0. Show that if
q(x) G R[x] is a zero-divisor in R[x], then, if
q(x) = a
0
x" + a
x
x"~
l
++,
there is an el ement b = 0 in R such that ba
0
= ba
x
ba = 0.
27. Let R be a ring and I an ideal of R. If R[x] and I[x] are the po l yno mi al
rings in x over A and I, respectivel y, show that:
(a) I[x] is an ideal of R[x].
(b) R[x]II[x] = (/?//)[*].
*28. Do Pro bl em 26 even if the condi ti on "a
2
= 0 onl y if a = 0" do es no t hol d
in P.
29. Let R = [a + bi | a, b i ntegers} C C. Let d(a + bi) = a
2
+ b
2
. Sho w that
R is a Eucl i dean ring where d is its requi red Eucl i dean function. (R is
kno wn as the ring of Gaussian integers and pl ays an i mpo rtant rol e in
number theory.)
6. POL YNOMI A L S OVER THE RA TI ONA L S
I n o ur consi derati on of the po l yno mi al ring F[x] over a fiel d F, the parti cul ar
nature of F never entered the pi cture. Al l the resul ts ho l d for arbi trary fiel ds.
Ho wever, there are resul ts that expl oit the expl icit character of certai n fiel ds.
One such fiel d is that of the rati o nal numbers.
We shal l present two i mpo rtant theo rems for Q[x], the pol ynomi al ring
over the rational fiel d Q, These resul ts depend heavil y on the fact that we are
deal ing with rational numbers. The first of these, Gauss' Lemma, rel ates the
factorization over the rational s with factorization over the integers. The sec-
o nd o ne, kno wn as the Eisenstein Criterion, gives us a metho d of constructi ng
irreducibl e pol ynomial s of arbitrary degree, at wil l , in Q[x]. I n this the fiel d Q
is highl y particul ar. Fo r instance, there is no easy al gorithm for obtai ni ng i rre-
ducibl e pol ynomial s of arbitrary degree n over the fiel d Z
p
of the integers
mo d p, p a pri me. Even over Z
2
s u c n a n
al gorithm is nonexistent; it woul d be
highl y useful to have, especial l y for coding theory. But it just do esn't exist
so far.
We begin o ur co nsi derati o n with two easy resul ts.
L emma 4.6.1. Let f(x) Q[x]; then
fix) = (a
0
x" + a
lX
"~i + . . . +a)
where u, m, a
0
,. . ., a
n
are integers and the a
0t
a
x
,..., a
n
have no co mmo n
factor greater than 1 (i.e., are rel ativel y pri me) and (w, m) = 1.
Sec. 6 Pol ynomial s Over the Rational s 167
Proof. Si nce/(x) E Q[x], f(x) = q
0
x" + q
x
x"~
l
+ + q
n
, where the
q
t
are rati o nal numbers. So for i = 0, 1, 2, . . . , n, q
(
= 6,/c,, where b
t
, c,- are
i ntegers. Thus
cl earing of deno mi nato rs gives us
where the are integers. If w is the greatest co mmo n divisor of u
0
, u
l t
. . . , u,
then each u, =w<7,, where a
0
, fl j,. . . , a are rel ativel y pri me integers. Then
/ = c,cZ c (
f l
o *" + " i * "
- 1
+ " + ) ;
cancel ing o ut the greatest co mmo n factor of w and CqCj c gives us
f(x) = ^(a
a
x" + + )-
where M, m are rel ativel y pri me integers, as is cl aimed in the l emma.
The next resul t is a resul t abo ut a parti cul ar ho mo mo rphi c i mage of
R[x] for any ring R.
L emma 4.6.2. If R is any ring and / an ideal of R, then I[x], the pol y-
no mi al ring in x over I, is an ideal of R[x]. Furthermo re, P[x]/7[x] (R/I)[x\,
the po l yno mi al ring in x over RII.
Proof. Let R = RII; then there is a ho mo mo rphi sm <p: R> R, defined
by <p(a) = a + I, whose kernel is /. Define 5>: R[x] > R[x] by: If
/(x) = a
0
x" + fl jx"
-1
++ ,
t h e n
<D(/(x)) = <p(oK + pCfli)^"-
1
+ +cp(a).
We l eave it to the reader to pro ve that $ is a ho mo mo rphi sm of R[x] o nto
R[x]. What is the kernel , Ker <&, of <E>? I f /(x) = a
0
x" + + a is in Ker
then 3>(/(x)) =0, the 0 el ement oR[x]. Since
<f>(f(x)) =<p(a
0
)x" +tpiajx"-
1
+ + <pK) = 0,
we concl ude <p(fl
u
) =0, <p(a
L
) = 0, . . . , cp(a) = 0, by the very definition of
what we mean by the 0-pol ynomial in a po l yno mi al ring. Thus each a, is in
the kernel of cp, which happens to be /. Because a
Q
,a
1
,..., a are in /, f(x) =
a
Q
x" + a
x
x"~
l
+ + a is in I[x]. So Ker $ C I[x]. That I[x] C Ker <E>is im-
medi ate from the definition of the mappi ng Hence I[x] = Ker $. By the
First Ho mo mo rphi sm Theo rem (Theo rem 4.3.3), the ring I[x] is then an
ideal of R[x] and R[x] i?[x]/Ker $ =K[x]/J [x]. This pro ves the l emma, re-
memberi ng that R = RII.
As a very special case of the l emma we have the
Corol l ary. Let Z be the ring of integers, p a pri me number in Z, and
I = (p), the ideal of Z generated by p. Then Z[x]//[x] = Z
p
[x\.
Proof. Since Z
p
Zll, the corol l ary fol l ows by appl ying the l emma to
r = z . u
We are ready to pro ve the first of the two major resul ts we seek in this
section.
T heorem 4.6.3 (Gauss' Lemma). Let f(x) G Z[x] be a mo ni c po l yno -
mial and f(x) = a(x)b(x), where a(x) and b(x) are in Q[x]. Then f(x) =
a
1
(x)b
1
(x), where a
x
(x), b
x
(x) are mo ni c pol ynomi al s in Z[x] and deg a
x
(x) =
deg a(x), deg b
x
(x) =deg >(x).
Proof. Suppose f(x) G Z[x] is mo ni c and f(x) = a(x)b(x), where a(x),
b(x) G Q[x], and deg a(x) = s, deg b(x) = r. By L emma 4.6.1, we can ex-
press each of a(x), b(x) as a pro duct of a rati onal number and a po l yno mi al
with i nteger coefficients. Mo re precisel y,
a(x) = (ax' + a{x
s
^ + +;)= âAx),
v
' m
x
' m
x
v
'
where a'
Q
, a{,..., a'
s
are rel ativel y pri me i ntegers and
b(x) = ^ (bo'v" + b{x>-
1
+ + /,/) = ^ ^( x) ,
r/(2 ' ' ' 2
where b^b'
u
,. . ,b'
r
are rel ativel y pri me. Thus
/ (x) = a(x)b(x) = â
x
(x)b
x
(x) = â
x
(x)b
x
(x),
where v and w are rel ativel y pri me, by cancel ing o ut the co mmo n factor of
u
x
ii
2
and m
x
m
2
. Therefo re, wf(x) = va
x
(x)b
x
(x), and/ (x), a
x
(x), b
x
(x) are
all in Z[x], Of course, we may assume with no l oss of general ity that the l ead-
ing coefficients of a
x
(x) and b
x
(x) are positive.
If w = 1, then, si nce/(x) is mo ni c, we get that va'
0
b'
0
= 1 and this l eads
Sec . 6 Po l y n o mi al s Ov er t h e Rat i o n al s 169
easil y to v = 1, a'
0
= b'
Q
= 1 and so f(x) = a
l
(x)b
1
(x), where bo th a
1
(x) and
bi(x) are mo ni c pol ynomi al s with i nteger coefficients. This is precisel y the
cl aim of the theo rem, since deg a
x
(x) = deg A (A:) and deg b
L
(x) = deg b(x).
Suppo se then that w = 1; thus there is a pri me p such that p | w and,
since (i>, vv) = 1, pj f . Al so , since the coefficients a'
0
, a\,.. ., a'
s
of Ai (x)
are rel ativel y pri me, there is an i such that p\a\; simil arl y, there is a
such that p\b'j. Let 7 = (p) be the i deal generated by p in Z; then
Z/I Z
p
and, by the Corol l ary to L emma 4.6.2, Z[x]/7[x] Z
p
[x], so is an in-
tegral do mai n. Ho wever, since p | w, w, the i mage of w in Z[x]/7[x], is 0,
and since p\v, v the i mage of v in Z[x]/7[x] is no t 0. Thus 0/ (x) =
v d
1
(x)b
1
(x), where v 0 and fl ^x) 0, b
x
(x) = 0 becausep \ a \ andp\ b' j for
the given i, j abo ve. This contradicts that Z[x]/7[x] is an integral domai n. So
w 1 is no t possibl e, and the theo rem is pro ved.
I t mi ght be instructive for the reader to try to show directl y that if
x
3
+ 6x - 7 is the pro duct of two po l yno mi al s having rati onal coefficients,
then it is al ready the pro duct of two mo ni c po l yno mi al s with integer coeffi-
cients.
One shoul d say something about C. F. Gauss (1777-1855), considered by many
to be the greatest mathematician ever. His contributions in number theory, al -
gebra, geometry, and so on, are of gigantic proportions. His contributions in
physics and astronomy are al so of such a great proportion that he is considered
by physicists as one of their greats, and by the astronomers as one of the impor-
tant astronomers of the past.
As we i ndi cated at the begi nni ng of this section, irreducibl e pol ynomi -
al s of degree n over a given fiel d F may be very hard to come by. Ho wever,
over the rati onal s, due to the next theo rem, these exist in abundance and are
very easy to construct.
T heorem 4.6.4 (T he Eisenstein Criterion). Let f{x) = x" + a
x
x"~
l
+ + a be a no nco nstant po l yno mi al with i nteger coefficients. Suppo se
that there is so me pri me p such that p \ a
u
p \ a
2
,. , p \ a
n
, but p
2
f a
n
. Then
f(x) is i rreduci bl e in Q[x].
Proof. Suppo se that/(x) = u(x)v(x), where u(x), v(x) are of positive
degree and are pol ynomi al s in Q[x]. By Gauss' L emma we may assume that
bo th u(x) and v(x) are mo ni c po l yno mi al s with i nteger coefficients. Let
I = (p) be the i deal generated by p in Z, and consider Z[x]//[x], which is an
integral do mai n, since we kno w by the Corol l ary to L emma 4.6.2 that
Z[x]//[x] - (Z//)[x] =* Z
p
[x\. The i mage of f(x) = x" + a
L
x
n
^ + + a in
Z[x]//[x] is x", since p\a
x
,... ,p \ a. So if w(x) is the i mage of u(x) and U(x)
that oiv(x) in Z[x]/I[x], then*" =77(X)TJ(X). Since Ti(x)\x", v(x)\x" in Z[x]/I[x],
we must have that u(x) = x
r
, v(x) = x " ~
r
for so me 1 r < n. But then
u(x) = x
r
+ pg(x) and v(x) = x"~
r
+ ph(x), where g(x) and h(x) are po l yno -
mial s with integer coefficients. Since u(x)v(x) = x" + px'hix) + px"~'g(x) +
p
1
g(x)h(x), and since 1 < r < n, the constant term of u(x)v(x) is p
2
st, where
s is the constant term of g(x) and t the constant term of h(x). Because f(x) =
u(x)v(x), their co nstant terms are equal , hence a = p
2
st. Since s and t are in-
tegers, we get that p
2
1 a , a contradi cti on. I n this way we see that/(x) is i rre-
ducibl e.
We give so me exampl es of the use to which the Ei senstei n Criterion
can be put.
1. Let f(x) = x " p , p any pri me. Then o ne sees at a gl ance that f(x) is
irreducibl e in Q[x], for the Ei senstei n Cri teri on appl ies.
2. Let/(j c) = x
5
- 4x + 22. Since 2 | 22, 2
2
|22 and 2 divides the o ther rel -
evant coefficients of f(x), the Ei senstei n Cri teri on tel l s us that f(x) is ir-
reduci bl e in Q[x].
3. L et/f x) = x
n
- 6x
4
+ 12x
3
+ 36x - 6. We see that/(x) is i rreduci bl e
in Q[x] by using ei ther 2 or 3 to check the condi ti ons of the Ei senstei n
Criterion.
4. Let /(x) = 5x
4
- lx + 7; f(x) is no t moni c, but we can modify f(x)
sl ightl y to be in a posi ti on where we can appl y the Ei senstei n Cri teri on.
Let
g(x) = 5
3
/(x) = 5
4
x
4
- 7 5
3
x + 7 5
3
= (5x)
4
- 175(5x) + 875.
If we l et y = 5x, then g(x) = h(y) = V
4
~ 175y + 875. The po l yno mi al
h(y) is irreducibl e in Z[y] by using the pri me 7 and appl ying the Ei sen-
stein Criterion. The irreducibil ity of h(y) impl ies that of g(x), and so
that o f /(x), in Q[x].
This suggests a sl ight general i zati on of the Ei senstei n Cri teri o n to
no nmo ni c pol ynomi al s. (See Pro bl em 4.)
5. Let /(x) = x
4
+ x
3
+ x
2
+ x + 1; as it stands we canno t, of co urse,
appl y the Ei senstei n Criterion to /(x). We pass to a po l yno mi al g(x)
cl osel y rel ated to /(x) who se irreducibil ity in Q[x] wil l ensure that of
f{x). Let g(x) = f(x + 1) = (x + l )
4
+ (x + l )
3
+ (x + I )
2
+ (x + 1) +
1 = x
4
+ 5x
3
+ 10x
2
+ l Ox + 5. The Ei senstei n Cri teri on appl ies to
g{x), using the pri me 5; thus g(x) is irreducibl e in Q[x]. This impl ies
that/(x) is irreducibl e in Q[x]. (See Pro bl em 1.)
Sec . 6 Po l y n o mi al s Ov er t h e Rat i o n al s 171
Gotthol d Eisenstein (1823-1852) in his short l ife made fundamental contribu-
tions in al gebra and anal ysis.
PROBL EMS
1. I n Exampl e 5, show that because g(x) is irreducibl e in Q[x], then so is
fix)-
2. Pro ve that/(x) = x
3
+ 3x + 2 is i rreduci bl e in Q[x].
3. Show that there is an infinite number of integers a such that f(x) =
x
1
+ 15x
2
30x + a is irreducibl e in Q[x]. What a's do you suggest?
4. Pro ve the fol l owing general ization of the Eisenstein Criterion. Let
f(x) = a
0
x" + fl pv"
-1
+ + a have i nteger coefficients and suppo se
that there is a pri me p such that
p\a
a
, p\a
l
,p\a
2
,. . . ,p\a
n
_
u
p\a,
but p
2
1 a; then f(x) is irreducibl e in Q[x].
5. If p is a pri me, show that f(x) = x
p
~
l
+ x
p
~
2
+ + x + 1 is irreducibl e
in Q[x].
6. L et F be the fiel d and cp an auto mo rphi sm of F[x] such that cp(a) = a for
every a G F. If / ( J C ) G 77[x], pro ve that f(x) is irreducibl e in F[x] if and
onl y if g(x) = (p(f(x)) is.
7. Let be a fiel d. Define the mappi ng
p: F [ * ] - >F M by <p(/(x)) = f(x + 1)
for every f(x) G F[x]. Pro ve that cp is an auto mo rphi sm of F[x] such that
F[x] by cp(f(x)) = f(bx) for every /(x) G F[x]. Pro ve that cp is
an auto mo rphi sm of F[x] such that cp{a) = a for every a G F.
9. Let F be a fiel d, o ^ 0, c el ements of F. Define the mappi ng
cp : F[x] ->F[x] by <p(/(x)) =f(bx + c) for every/(x) G F[x]. Pro ve that
cp is an auto mo rphi sm of F[x] such that <p (a) = a for every a G F.
10. Let p be an auto mo rphi sm of F[x], where Fi s a fiel d, such that cp(a) = a
for every a GF. Pro ve that if /(x) G F[x], then deg cp(f(x)) = deg/(x).
11. Let cp be an auto mo rphi sm of F[x], where F is a fiel d, such that cp(a) = a
for every a G F. Pro ve there exist b 0, c in F such that cp(f(x)) =
/(/3x + c ) for every/(x) G F[x].
172 Ring T heory Ch. 4
12. Find a no ni denti ty auto mo rphi sm <p of Q[x] such that cp
2
is the identity
auto mo rphi sm of Q[.v].
13. Show that in Pro bl em 12 yo u do no t need the assumpti on cp(a) = a for
every a G Q because any auto mo rphi sm of Q[x] auto mati cal l y satisfies
cp(a) = a for every a G Q.
14. Let C be the fiel d of co mpl ex numbers. Given an i nteger n > 0, exhibit
an auto mo rphi sm cp of C[x] of o rder n.
7. FI ELD OF QUOTI ENTS OF A N I NTEGRAL DOMA I N
Gi ven the integral do mai n Z, the ring of integers, then intimatel y rel ated to Z
is the fiel d Q of rati onal numbers that consists of al l fractions of i ntegers; that
is, al l quo ti ents m/n, where m, n # 0 are in Z. No te that there is no uni que
way of representi ng f, say, in <Q>because | = f = (7)/(-14) = .
I n o ther wo rds, we are identifying with f, (7)/( 14), and so on. This sug-
gests that what is real l y going on in constructing the rati onal s from the i nte-
gers is so me equi val ence rel ati o n o n so me set based o n the i ntegers.
The rel ation of Q to Z carries over to any integral do mai n D. Gi ven an
integral do mai n D, we shal l construct a fiel d FDD whose el ements wil l be
quo ti ents alb with a, b G D, b # 0. We go thro ugh this constructi on formal l y.
Let D be an integral do mai n, and l et S = {(a, b) \ a, b G D, b = 0}; S is
thus the subset of D X Dthe Cartesi an pro duct of D with itsel fin which
the second co mpo nent is no t al l owed to be 0. Thi nk of (a, b) as alb for a mo -
ment; if so, when woul d we want to decl are that (a, b) = (c, d)? Cl earl y, we
wo ul d want this if alb = c/d, which in D itsel f wo ul d beco me ad = be. Wi th
this as o ur moti vati ng guide we define a rel ati o n ~ o n S by decl aring:
( A , b) ~ (c, d) for (A, b), (c, d) in S, if and onl y if ad = be.
We first assert that this defines an equi val ence rel ati on o n S. We go
thro ugh the three requi rements for an equi val ence rel ati on term by term.
1. (fl , b) ~ (a, b), for cl earl y ab = ba (since D is co mmutati ve). So ~ is
refl exive.
2. ( A , b) ~ (c, d) impl ies that (c, d) ~ (a, b), for (a, b) ~ (c, d) means
fl fl * = be; for (c, d) ~ {a, b) we need cb = da, but this is true, since
cb = be ad = da. So ~ is symmetri c.
3. ( A , b) ~ (c, d), (c, d) ~ (e,f) impl ies that ad = be, cf = de, so adf =
bef = bde; but d = 0 and we are in an integral do mai n, hence af =
be fol l ows. This says that (a, b) ~ (e, / ) . So the rel ati on is transitive.
Sec. 7 Fiel d of Quotients of an Integral Domain 173
We have shown that ~ defines an equi val ence rel ation o n S. Let F be
the set of al l the equi val ence cl asses [a, b] of the el ements (a, b) G S. F is o ur
requi red fiel d.
To show that F is a fiel d, we must endo w it with an addition and mul ti -
pl ication. First the addition; what shoul d it be? Fo rgetti ng al l the fancy tal k
abo ut equi val ence rel ation and the l ike, we real l y want [a, b] to be alb. If so,
what shoul d [a, b] + [c, d] be o ther than the formal l y cal cul ated
a_ c _ ad + be ^
b d~ bd
This mo ti vates us to define
[a,b] + [c,d] = [ad + bc,bd]. (1)
No te that since b j= 0, d = 0 and D is a do mai n, then bd j= 0, hence
[ad + be, bd] is a l egitimate el ement of F.
As usual we are pl agued with having to show that the addition so de-
fined in F is wel l -defined. I n o ther wo rds, we must show that if [a, b] =
[a', b'] and [c, d] = [c', d'], then [a, b] + [c, d] = [a',b'] + [c\ d']. Fro m (1)
we must thus show that [ad + be, bd] = [a'd' + b'c', b'd'], which is to say,
(ad + bc)b'd' = bd(a'd' + b'c'). Since [a, b] = [a', b'] and [c, d] = [c', d'],
ab' = ba' and cd' = dc'. Therefo re, (ad + bc)b'd' = ab'dd' + bb'cd' =
ba'dd' + bb' dc' = (a'd' + b'c')bd, as requi red. Thus " +" is wel l -defined in F.
The cl ass [0, b], b 0, acts as the 0 under " +," we deno te it simpl y as 0,
and the cl ass [-a, b] is the negative of [a, b]. To see that this makes of F an
abel ian gro up is easy, but l abori ous, for al l that real l y needs verification is
the associative l aw.
No w to the mul tipl ication in F. Agai n mo ti vated by thinking of [a, b] as
alb, we define
[a,b][c,d] = [ac,bd]. (2)
Agai n since b 0, d =0, we have bd =fc 0, so the el ement [ac, bd] is
al so a l egi ti mate el ement of F.
As for the " +" we must show that the pro duct so i ntro duced is wel l -de-
fined; that is, if [a, b] = [a',b'], [c, d] = [c', d'], then
[a,b][c,d] = [ac,bd] = [a'c',b'd'\ = [a',b'\c',d'\
We kno w that ab' = ba' and cd' = dc', so acb'd' = ab'cd' = ba'dc' = bda'c',
which is exactl y what we need for [ac, bd] = [a'c', b'd']. Thus the pro duct is
wel l -defined in F.
What acts as 1 in Fl We cl aim that for any a i= 0, b i= 0 in D, [a, a] =
[b, b] (since ab = ba) and [c, d][a, a] = [ca, da] = [c, d], since (ca)d = (da)c.
So [a, a] acts as 1, and we wri te it simpl y as 1 = [a, a] (for al l a # 0 in D).
Gi ven [a, b] # 0, then a 0, so [>, o] is in .F; hence, because [a, b][b, a]
[ab, ba] = [ab, ab] = 1, [a, 6] has an inverse in F. Al l that remai ns to sho w
that the no nzero el ements of F form an abel ian gro up under this pro duct is
the associative l aw and co mmutati ve l aw. We l eave these to the reader.
To cl inch that F is a fiel d, we need onl y now sho w the di stri buti ve l aw.
But [ad + be, bd][e,f] = [(ad + bc)e, bdf], so
([a,b] + [c,d])[e,f] = [ade + bee,bdf],
whil e
[a,b][e,f] + [c,d][e,f]
= [ae, bf] + [ce, df] = [aedf + bfee, bdf
2
]
= [(ade + bee) f bdf
2
] = [ade + bce,bdf][ff]
= [ade + bee, bdf],
which we have seen is ([a, b] + [c, d])[e, / ] . The distributive l aw is no w es-
tabl ished, so F is a fiel d.
Let a + 0 be a fixed el ement in D and consider [da, a] for any d G D.
The map cp: d [da, a] is a mo no mo rphi sm of D into F. I t is certai nl y 1-1,
for if cp(d) = [da, a] = 0, then da = 0, so d = 0, since D is an i ntegral do -
mai n. Al so , cp(d
1
d
2
) = [d
x
d
2
a, a] whil e cp(d
x
)cp(d
2
) = [d
x
a, a][d
2
a, a] =
[d
x
d
2
a
2
, a
2
] = [d
x
d
2
a, a][a, a] = [d
x
d
2
a, a] = cp(d
1
d
z
)- Furthermo re,
[d
x
a,a] + [d
2
a,a] = [d
t
a
2
+ a
2
d
2
, a
2
]
= [d
x
a + d
2
a, a][a, a]
= [(d
x
+ d
2
) a, a]
so cp(d
x
+ d
2
) = [(d
x
+ d
x
)a, a] = [d
x
a, a] + [d
2
a, a] = cp(d
x
) + cp(d
2
). Thus
cp maps D mo no mo rphi cal l y i nto F. So, D is i somorphi c to a subri ng of F, and
we can thus consider D as "embedded" in F. We consi der every el ement
[a, b] of F as the fraction alb.
T heorem 4.7.1. Let D be an i ntegral do mai n. Then there exists a fiel d
FDD which consists of al l fractions alb, as defined abo ve, of el ements in D.
The fiel d F is cal l ed thefield of quotients of D. When D = 7L, then F is
i so mo rphi c to the fiel d Q of rati o nal numbers. Al so , if D is the do mai n of
even integers, then F is al so the enti re fiel d Q.
What we did abo ve in constructi ng the fiel d of quo ti ents of D was a
l ong, formal , wordy, and pro babl y dul l way of doing so methi ng that is by its
Sec. 7 Fiel d of Quotients of an Integral Domain 175
nature so methi ng very simpl e. We real l y are do i ng no thi ng mo re than form-
ing al l formal fractions alb, a, b + 0 in D, where we add and mul tipl y frac-
tions as usual . Ho wever, it is so meti mes necessary to see so methi ng do ne to
its l ast detai l , painful tho ugh it may be. Mo st of us had never seen a real l y
fo rmal and preci se co nstructi o n of the rati o nal s from the i ntegers. No w
that we have co nstructed F fro m D i n this fo rmal manner, forget the de-
tail s and thi nk of F as the set of al l fracti ons of el ements of D.
PROBL EMS
1. Pro ve the associative l aw of addi ti on in F.
2. Pro ve the co mmutati ve l aw of addi ti o n in F.
3. Pro ve that the pro duct in F is co mmutati ve and associative.
4. If K is any fiel d that contains 73, sho w that K D F. (So F is the smal l est
fiel d that contai ns D.)
5
FIELDS
The no ti o n of a ring was unfami l i ar terri to ry for mo st of us; that of a fiel d
to uches mo re cl osel y o n o ur experi ence. Whi l e the onl y ring, o ther than a
fiel d, that we might have seen in o ur earl y training was the ring of i ntegers,
we had a bit mo re experi ence wo rki ng with rati onal numbers, real numbers,
and, so me of us, compl ex numbers, in sol ving l inear and quadrati c equati o ns.
The abil ity to divide by no nzero el ements gave us a bit of l eeway, which we
mi ght no t have had with the integers, in sol ving a variety of pro bl ems.
So at first gl ance, when we start worki ng with fiel ds we feel that we are
on ho me ground. As we penetrate deeper i nto the subject, we start runni ng
across new ideas and new areas of resul ts. Once again we'l l be in unfamil iar
terri tory, but hopeful l y, after so me expo sure to the topic, the no ti o ns wil l be-
co me natural to us.
Fiel ds pl ay an i mpo rtant rol e in geometry, in the theo ry of equati ons,
and in certain very i mpo rtant parts of number theory. We shal l to uch upo n
each of these aspects as we progress. Unfortunatel y, because of the technical
machi nery we woul d need to devel op, we do no t go into Gal ois theory, a very
beautiful part of the subject. We ho pe that many of the readers wil l make con-
tact with Gal ois theory, and beyond, in their subsequent mathemati cal training.
1 . EXA MPL ES OF FI ELDS
Let's recal l that a fiel d F is a commutati ve ring with unit el ement 1 such that for
every nonzero a E F there is an el ement a'
1
G F such that aa~
l
= 1. I n other
words, fiel ds are "something l i ke" the rational s <Q>. But are they real l y? The in-
176
Sec . 1 Ex amp l es of Fi el ds 177
tegers mo d p, Z
p
, wherep is a prime, form a fiel d; in Z
p
we have the rel ation
pi = 1 + 1 + + 1,= 0
(p ti mes)
No thi ng akin to this happens in Q. There are even sharper differences
amo ng fiel dshow pol ynomial s factor over them, special pro perti es of which
we'l l see in so me exampl es, and so on.
We begi n with some famil iar exampl es.
Examples
1. Q, the fiel d of rational numbers.
2. IR, the fiel d of real numbers.
3. C, the fiel d of compl ex numbers.
4. Let F = {a + bi \ a, b G Q} C C. To see that F is a fiel d is rel ativel y easy.
We onl y verify that if a + bi 0 is in F, then (a + bi)~
]
is al so in F. But
what is (fl + biy
1
? I t is merel y
a bi , .
>
2
i ~ / 2 i u2\ (verify!),
( A + b') ( A + b )
and since a
2
+ b
2
i= 0 and is rati onal , therefo re a/(a
2
+ b
2
) and al so
b/(a
2
+ b
2
) are rati onal , hence (a + /3/)
_ 1
is i ndeed in F.
5. Let F = {A + b\/2 \ a, b G 0 } C R. Agai n the verification that Fi s a fiel d
is no t to o hard. Here, to o , we onl y show the existence of inverses in Ff o r
the no nzero el ements of F. Suppo se that a + bVl ^ 0 is in F; then, since
V 2 is irrational, a
2
2b
2
+ 0. Because
( A + bV2)(a - bV2) = A
2
- 2b
2
,
we see that (a + bV2)(alc- V 2 bid) = 1_, where c = a
2
- 2b
2
. The re-
qui red inverse for a + &V2 is a/c - V 2 b/c, which is certainl y an el e-
ment of F, since a/c and b/c are rational .
6. Let F be any fiel d and l et F[x] be the ring of pol ynomi al s in x o ver F.
Since F[x] is an integral do mai n, it has a fiel d of quo ti ents according to
Theo rem 4.7.1, which consists of al l quo ti ents f(x)/g(x), where f(x) and
g(x) are in F[x) and g(x) + 0. This fiel d of quo ti ents of F[x] is deno ted by
F(x) and is cal l ed the field of rational functions in x over F.
7. Z
p
, the integers mo dul o the pri me p, is a (finite) fiel d.
8. I n Exampl e 2 in Section 4 of Chapter 4 we saw ho w to construct a fiel d
having nine el ements.
These eight exampl es are specific o nes. Usi ng the theo rems we
have pro ved earl ier, we have so me general constructions of fiel ds.
178 Fi el ds Ch . 5
9. If D is any integeral do mai n, then it has a fiel d of quo ti ents, by Theo rem
4.7.1, which consists of al l the fractions alb, where a and b are in D and
b #0.
10. If R is a co mmutati ve ring with unit el ement 1 and M is a maxi mal ideal
of R, then Theo rem 4.4.2 tel l s us that RIM is a fiel d.
This l ast exampl e, for parti cul ar R's, wil l pl ay an i mpo rtant ro l e in what
is to fol l ow in this chapter.
We coul d go on, especial l y with special cases of Exampl es 9 and 10, to
see mo re exampl es. But the 10 that we did see abo ve show us a certai n vari-
ety of fiel ds, and we see that it is no t to o hard to run across fiel ds.
I n Exampl es 7 and 8 the fiel ds are finite. If F is a finite fiel d havi ng q el -
ements, viewing F merel y as an abel i an gro up under its addi ti on, " +," we
have, by Theo rem 2.4.5, that qx = 0 for every x F. Thi s is a behavi o r qui te
distinct from that which happens in the fiel ds that we are used to , such as the
rational s and real s.
We singl e o ut this kind of behavi o r in the
Definition. A fiel d /-' is said to have (or, to be of) characteristic p #0
if for so me positive i nteger p, px = 0 for al l x F, and no positive i nteger
smal l er than p enjoys this pro perty.
If a fiel d F is not of characteristic p 0 for any positive i nteger p, we
cal l it a fiel d of characteristic 0. So Q, U, C are fiel ds of characteri sti c 0, whil e
Z
3
is of characteristic 3.
I n the definition given abo ve the use of the l etter p for the characteri s-
tic is highl y suggestive, for we have al ways used p to deno te a pri me. I n fact,
as we see in the next theo rem, this usage of p is consistent.
Theo rem 5.1.1. The characteristic of a fiel d is ei ther 0 o r a pri me
number.
Proof. If the fiel d F has characteristic 0, there is no thi ng mo re to say.
Suppo se then that mx = 0 for al l x F, where m is a positive i nteger. Let p
be the smal l est positive i nteger such that px =0 for al l x F. We cl aim that
p is a pri me. Fo r if p = uv, where u > 1 and v > 1 are i ntegers, then in F,
(ul)(vl) = (uv)l = 0, where 1 is the uni t el ement of F. But F, bei ng a fiel d, is
an integral domai n (Pro bl em 1); therefo re, ul = 0 or vl = 0. I n ei ther case
we get that 0 = (ul){x) = ux [or, simil arl y, 0 = (vl)x = vx] for any x in F.
But this contradicts o ur choice of p as the smal l est integer with this pro perty.
Hence p is a pri me.
Sec. 1 Exampl es of Fiel ds 179
No te that we did no t use the ful l force of the assumpti on that F was a
fiel d. We onl y needed that F was an i ntegral do mai n (with 1). So if we define
the characteristic of an integral do mai n to be 0 o r the smal l est positive i nte-
ger p such that px = 0 for al l i G F . we o btai n the same resul t. Thus the
Co ro l l ary. If /) is an integral do mai n, then its characteristic is ei ther 0
o r a pri me number.
P R O B L E M S
1. Show that a fiel d is an integral do mai n.
2. Pro ve the Corol l ary even if D do es no t have a unit el ement.
3. Gi ven a ring R, l et S = R[x] be the ring of pol ynomi al s in x over R, and
l et T = S[y] be the ring of po l yno mi al s in y o ver S. Show that:
(a) Any el ement f(x, y) in T has the form 22f l
; /
xV , where the a
/ y
are
in R.
(b) I n terms of the fo rm of f(x, y) in T given in Part (a), give the condi -
tion for the equal ity of two el ements f(x, y) and g(x, y) in T.
(c) I n terms of the form for f(x, y) in Part (a), give the formul a for
f(x, y) + g(x, y), for f{x, y), g(x, y) in T.
(d) Gi ve the form for the pro duct of fix, y) and g(x, y) if f(x, y) and
gix, y) are in T. (Ti s cal l ed the ring of polynomials in two variables
over R, and is deno ted by R[x, y].)
4. If D is an integral do mai n, show that D[x, y] is an integral domai n.
5. If F is a fiel d and D = F[x, y], the fiel d of quo ti ents of D is cal l ed the
field of rational functions in two variables over F, and is usual l y deno ted
by F(x, y). Gi ve the form of the typical el ement of F(x, y).
6. Pro ve that F(x, y) is i so mo rphi c to F(y, x).
7. If F is a fiel d of characteristic p # 0, show that (a + b)
p
- a
p
+ b
p
for al l
a,b G F. (Hi nt: Use the bi no mi al theo rem and the fact that p is a pri me.)
8. If F is a fiel d of characteristic p = 0, show that (a + b)
m
= a
1
" + b"\
where m = p", for al l a, b in F and any positive i nteger n.
9. Let F be a fiel d of characteristic p =h 0 and l et cp : F > F be defined by
cpia) = a
p
for al l a E F.
(a) Sho w that <p defines a mo no mo rphi sm of F into itsel f.
(b) Gi ve an exampl e of a fiel d F where cp is not o nto . (Very hard.)
10. If F is a finite fiel d of characteristic p, show that the mappi ng cp defined
abo ve is o nto , hence is an auto mo rphi sm of F.
180 Fi el ds Ch . 5
2. A BRIEF EXCURSI ON I NTO VECTOR SPACES
To get into the things we shoul d l ike to do in fiel d theory, we need so me
technical equi pment that we do no t have as yet. Thi s co ncerns the rel ati on of
two fiel ds K D F and what we wo ul d l ike to consider as so me measure of the
size of K co mpared to that of F. This size is what we shal l cal l the dimension
or degree of K over F.
Ho wever, in these co nsi derati o ns, much l ess is needed of K than that it
be a fiel d. We woul d be remi ss if we pro ved these resul ts onl y for the special
context of two fiel ds K D F because the same ideas, proofs, and spirit hol d in
a far wi der situation. We need the no ti o n of a vector space o ver a fiel d F.
Asi de from the fact that what we do in vector spaces wil l be i mpo rtant in o ur
situation of fiel ds, the ideas devel o ped appear in al l parts of mathemati cs.
Students of al gebra must see these things at so me stage of thei r training. An
appro pri ate pl ace is right here.
Definition. A vector space V over a fiel d ./' is an abel i an gro up under
" +" such that for every a E F and every v E V there is an el ement av E V,
and such that:
1. a(v
{
+ v
2
) = aV[ + av
2
, for a F, u
t
, v
2
V.
2. (a + B)v = av + Bv, for a, B F, v V.
3. a(Bv) = (aB)v, for a, BEF.v E V.
4. lv = v for al l v E V, where 1 is the unit el ement of F.
I n discussing vector spaceswhi ch we wil l do very bri efl ywe shal l
use l owercase Lati n l etters for el ements of V and l owercase Greek l etters for
el ements of F.
Our basic concern here wil l be with onl y one aspect of the theo ry of vec-
tor spaces: the notion of the di mensi on of V over F. We shal l devel op this no -
tion as expeditiousl y as possibl e, no t necessaril y in the best or mo st el egant
way. We woul d strongl y advise the readers to see the other sides of what is
do ne in vector spaces in other bo o ks o n al gebra or l inear al gebra (for instance,
our books A Primer on Linear Algebra and Matrix Theory and Linear Algebra).
Before getting do wn to so me resul ts, we l ook at so me exampl es. We
l eave to the reader the detail s of verifying, in each case, that the exampl e
real l y is an exampl e of a vecto r space.
Exampl es
1. Let F be any fiel d and l et V = {(a,, a
2
,. .., a) | a,- Ff o r al l /} be the set
of /?-tupl es over F, with equal ity and addi ti o n defined co mpo nent-wi se. Fo r
Sec. 2 A Brief Excursion into Vector Spaces 181
v = (c*i, a
2
, , a) and B E F, define Bv by Bv = (Ba
u
Ba
2
,..., Ba
n
).
V is A vecto r space over F.
2. Let F be any fiel d and l et V = F[x] be the ring of pol ynomi al s in x o ver F.
Fo rgetti ng the pro duct of any arbi trary el ements of F[x] but using onl y that
of a po l yno mi al by a constant, for exampl e, we find that
B(a
0
+ a
y
x + + ax") = Ba
0
+ Ba
x
x + + Bax".
I n this way V beco mes a vector space o ver F.
3. Let V be as in Exampl e 2 and l et W = \f(x) E V \ deg(/(x)) < }. Then W
is a vector space over F, and W C V is a subspace of V in the fol l owing sense.
Definition. A subspace of a vecto r space V is a no nempty subset W of
Vsuch that aw Wand w
x
+ w
2
Wf o r al l a in Fand w, w
l
,w
2
E W.
No te that the definition of subspace W of V impl ies that W is a vecto r
space who se o perati o ns are just tho se of V restricted to the el ements of W.
4. Let V be the set of al l real -val ued differentiabl e functions o n [0, 1], the
cl osed unit interval , with the usual addi ti on and mul tipl ication of a function
by a real number. Then V is a vecto r space over R.
5. Let W be al l the real -val ued co nti nuo us functions on [0,1], again with the
usual addi ti o n and mul tipl ication of a function by a real number. W, to o , is a
vector space over R, and the V in Exampl e 4 is a subspace of W.
6. Let F be any fiel d, F[x] the ri ng of po l yno mi al s in x over F. Let f(x)
be in F[x] and / = (/(*)) the ideal of F[x] generated by f(x). Let V =
F[x]/J, where we define a(g(x) + J) = ag(x) + J. V is then a vector space
over F.
7. Let U be the real fiel d and l et V be the set of al l sol utions to the differen-
tial equati o n d
2
yldx
2
+ y = 0. V is a vecto r space over U.
8. Let Vbe any vector space over a fiel d F, v
x
, v
2
,..., v el ements of V. L et
(v
x
, v
2
, . . . , v
n
) = {a
x
v
x
+ a
2
v
2
+ + a
n
v
n
\ a
x
, a
2
, . . . , a F). Then
(v
x
, v
2
,..., v
n
) is a vecto r space o ver Fand is a subspace of V. This subspace
(v
x
, v
2
,..., u) is cal l ed the subspace of V generated or spanned by v
x
,... , v
over F; its el ements are cal l ed linear combinations of v
x
,.. ., v. We shal l
soon have a great deal to say abo ut (v
x
,v
2
,..., v).
9. Let V and W be vector spaces o ver the fiel d F and l et V W =
{(u, vt>) 11 V, w W}, with equal ity and addi ti o n defined co mpo nentwi se,
182 Fields Ch . 5
and where a(v, w) = (av, aw). Then V W is easil y seen to be a vecto r
space over F; it is cal l ed the direct sum of V and W.
10. Let K D Fbe two fiel ds, where the addi ti on " +" is that of K and where
av, for a E F and v E K is the pro duct, as el ements of the fiel d K. Then Co n-
ditions 1 and 2 defining a vecto r space are merel y special cases of the distrib-
utive l aws that ho l d in K, and Co ndi ti o n 3 is merel y a co nsequence of the
associativity of the pro duct in K. Final l y, Condi ti on 4 is just the restate-
ment of the fact that 1 is the uni t el ement of K. So K is a vecto r space
over F.
I n o ne respect there is a sharp difference amo ng these exampl es. We
specify this difference by exami ni ng so me of these exampl es in turn.
1. I n Exampl e 1, if
V l
= (1, 0, . . ., 0), v
2
= (0,1, 0, . . ., 0), . . . , v = (0, 0, . . . , 1),
then every el ement u in V has a unique representati o n in the form v =
a
x
v
{
+ + av, where a
u
<. ., a are in F.
2. I n Exampl e 3,ifv
1
= 1, v
2
=x, . . . , v, = x'^
1
, . . . , v
n + l
= x", then every
v E Vhas a unique representati o n as v = a
x
v
t
+ + av
n
, with the a,-
in F.
3. I n Exampl e 7, every sol ution of d
2
y/dx
2
+ y = 0 is of the uni que fo rm
y = a cos x + 6 sin x, with a and B real .
4. I n Exampl e 8, every v (v
x
,..., v) has a representati o nal bei t not nec-
essarily uniqueas v = a
x
v
x
+ + a
n
v from the very definition of
(v
{
,... , v). Uni queness of this representati o n depends heavil y o n the el e-
ments V i , . . . , V
n
.
5. I n the special case of Exampl e 10, where K = C, the fiel d of co mpl ex
numbers, and F = IR that of the real numbers, then every v E C is of the
unique form v = a + Bi, a, B E U.
6. Co nsi der K = F(x) D F, the fiel d of rati o nal functions in x over F. We
cl ai mand l eave to the readerthat we cannot find any finite set of el e-
ments in K which spans K over F. This pheno meno n was al so true in so me
of the o ther exampl es we gave of vecto r spaces.
The whol e focus of o ur attenti o n here wil l be on this no ti o n of a vecto r
space havi ng so me finite subset that spans it o ver the base fiel d.
Befo re starting this discussion, we must first dispose of a l ist of formal
pro perti es that hol d in a vector space. Y o u, dear reader, are by no w so
Sec. 2 A Brief Excursion into Vector Spaces 183
sophi sti cated in deal ing with these formal , abstract things that we l eave the
proof of the next l emma to you.
L emma 5.2.1. If l* is a vecto r space o ver the fiel d /', then, for every
a E F and every v V:
(a) aO = 0, where 0 is the zero -el ement of V.
(b) Ov = 0, where the first 0 is the zero in F.
(c) av = 0 impl ies that a = 0 o r v = 0.
(d) (-a)v = -(av).
I n view of this l emma we shal l no t run i nto any confusion if we use the
symbol 0 bo th for the zero of F and that of V.
We forget vector spaces for a mo ment and l o o k at sol utions of certai n
systems of l inear equati o ns in fiel ds. Take, for exampl e, the two l inear ho mo -
geneo us equati o ns with real coefficients, x
x
+ x
2
+ x
3
= 0 and 3A'! ~ X
2
+ X
3
= 0. We easil y see that for any A '
1 ;
X
3
such that 4x
t
+ 2x
3
= 0 and x
2
=
- ( A ' I + x
3
), we get a sol ution to this system. I n fact, there exists an infinity of
sol utions to this system o ther than the trivial o ne x
x
= 0, x
2
= 0, x
3
= 0. If we
l ook at this exampl e and ask oursel ves: "Why is there an infinity of sol utions
to this system of l inear equati o ns?", we quickl y co me to the concl usion that,
because there are mo re variabl es than equati o ns, we have ro o m to maneuver
to pro duce sol utions. Thi s is exactl y the si tuati on that hol ds mo re general l y,
as we see bel o w.
Definition. Let /' be a fiel d; then the n-tupl e (B
u
. .., B), where the
Bj are in F, and not all of them are 0, is said to be a nontrivial solution in F to
the system of ho mo geneo us l inear equati o ns
a
n
x
1
+ a
X2
x
2
+ + a
Ul
x
n
= 0
a
2i
x
i +
a
22
x
2 + + a
2ll
x
n
= 0
(*) =
0
a
n
x
x
+ a
i2
x
2
+ + a
in
x
n
= 0
a
rl
x
1
+ a,.
2
x
2
+ + a
rn
x = 0
where the a,
7
are al l in F, if substituting x
x
= B
1
, . . . , x = B satisfies al l
the equati o ns of (*).
Fo r such a system (*) we have the fol l owing
184 Fi el ds Ch . 5
Theo rem 5.2.2. If n > r, that is, if the number of vari abl es (un-
kno wns) exceeds the number of equati o ns in (*), then (*) has a no ntri vi al so-
l ution in F.
Proof. The metho d is that, which so me of us l earned in high school , of
sol ving si mul taneo us equati o ns by el iminating o ne of the unkno wns and at
the same time cutting the number of equati o ns do wn by o ne.
We pro ceed by i nducti o n on r, the number of equati o ns. If r = 1, the
system (*) reduces to a
n
X \ + + a
ln
x = 0, and n > 1. If al l the a
u
= 0,
then X j = x
2
= = x = 1 is a nontri vi al sol ution to (*). So, o n renumber-
ing, we may assume that a
n
# 0; we then have the sol ution to (*), which is
nontrivial : x
2
= =x = 1 andx
x
= ( l / a
n
) ( a
l 2
+ ' ' ' +
a
\n)-
Suppose that the resul t is correct for r = k for so me k and suppo se that
(*) is a system of k + 1 l inear ho mo geneo us equati o ns in n > k + 1 vari-
abl es. As above, we may assume that so me a,
;
- + 0, and, wi tho ut l oss of gen-
eral ity, that a
H
0.
We construct a rel ated system, (**), of k l inear ho mo geneo us equati o ns
in n. 1 variabl es; since n > k + 1, we have that n. 1 > k, so we can appl y
i nducti on to this new system (**). Ho w do we get this new system? We want
to el i mi nate x
x
amo ng the equati o ns. We do so by subtracti ng a
n
/a
u
times
the first equati o n from the ith o ne for each of / = 2, 3, . . ., k + 1. I n do i ng
so, we end up with the new system of k l inear ho mo geneo us equati o ns in
n 1 variabl es:
B
22
x
2
+ + B
2n
x = 0
B
32
x
2
+ + j8
3l I
jr = 0
n
Pk +1,2*2 + ' + Pk+l,nX = ,
where B
u
- a,-
;
- - a
n
l a
l l
for i = 2, 3, . . . , k + 1 and j = 2, 3, . . . , n.
Since (**) is a system of k l inear ho mo geneo us equati o ns in n - 1 vari-
abl es and n - 1 > k, by our induction (**) has a nontrivial sol ution (y
2
,..., y)
in F. Let y
x
= (ot
l2
y
2
+ + a
,
1
y)/a
1 1
; we l eave it to the reader to verify
that the (y
lt
y
2
,..., y) so o btai ned is a requi red nontrivial sol ution to (*).
This compl etes the induction and so pro ves the theo rem.
Wi th this resul t establ ished, we are free to use it in o ur study of vecto r
spaces. We now return to these spaces. We repeat, for emphasi s, so methi ng
we defined earl ier in Exampl e 8.
Sec . 2 A Br i ef Ex c u r s i o n i nt o Vec t o r Sp ac es 185
Definition. Let V be a vector space over /-' and l et u
u
v
2
, , v be in
V. The el ement u G V is said to be a linear combination of u
1 ;
i>
2
,
u
if
y = a
x
v
x
+ + av for so me a
x
, a
n
in P.
As we indicated in Exampl e 8, the set (v
u
u
2
, ., v) of al l l inear co m-
bi nati o ns of v
x
, v
2
, , u is a vecto r space over F, and being contai ned in V,
is a subspace of K Why is it a vector space? If a
x
v
x
+ + au and
Pi
v
i + + Bv are two l inear co mbi nati o ns of v
x
,... , v, then
(a
1
v
1
+ + av) + (B
x
v
1
+ + Bv
n
)
= (! + B
x
)v
x
+ + (Q- + A, ) v
by the axi oms defining a vector space, and so is in (v
x
,.. ., u
n
). If y G F and
+ + au G . . . , u), then
y(a
x
v
x
+ + av) = ya
l
v
l
+ +yav,
and is al so in (v
lt
. .., v). Thus (v
x
,. .., v
n
) is a vector space. As we cal l ed it
earl ier, it is the subspace of V spanned o ver F by v
l t
. . ., v.
This l eads us to the ul tra-i mpo rtant definition.
Definition. The vector space V over F is finite dimensional over F if
V = (vi,.. ., v) for so me v
u
.. . , v in V, that is, if V is spanned over Fby a
finite set of el ements.
Otherwi se, we say that V is infinite dimensional over F if it is not finite
di mensi onal over F. No te that al tho ugh we have defined what is meant by a
finite-dimensional vector space, we stil l have no t defined what is meant by its
di mensi on. That wil l co me in due course.
Suppo se that V is a vector space over F and v
{
,. . ., v in V are such
that every el ement v in (u
1 ?
. . ., v) has a unique representati o n in the form
v = a
x
v
x
+ + av, where a
x
, , a G F. Since
0 G (v
1:
..., v) and 0 = 0v
l
+ + 0v
n
,
by the uni queness we have assumed we obtai n that if a
x
v
x
+ + a
n
v = 0,
then a
x
= a
2
= = a = 0. This pro mpts a second ul tra-i mpo rtant defini-
tion.
Definition. Let V be a vector space over F; then the el ements v
x
,..., v
in V are said to be linearly independent over F if a
x
v
x
+ + av = 0,
where a
x
,.. ., a are in F, impl ies that a
1
= a
2
= = a = 0.
186 Fi el ds Ch . 5
If the el ements v
{
v
n
in V are not l inearl y i ndependent o ver F,
then we say that they are linearly dependent over F. Fo r exampl e, if IR is the
fiel d of real numbers and V is the set of 3-tupl es over IR as defined in Exam-
pl e 1, then (0, 0, 1), (0, 1, 0), and (1, 0, 0) are l inearl y i ndependent over IR
(Prove!) whil e (1, - 2, 7), (0,1, 0), and (1, - 3, 7) are l inearl y dependent over
R, since 1(1, - 2, 7) + (- l )(0, 'l , 0) + (- 1)(1, - 3, 7) = (0, 0, 0) is a nontri vi al
l inear co mbi nati o n of these el ements over R, which is the 0-vector.
No te that l inear i ndependence depends on the fiel d F. If C D U are the
compl ex and real fiel ds, respectivel y, then C is a vector space over R, but it is
al so a vector space over C itsel f. The el ements l,i in C are l inearl y i ndepen-
dent over R but are no t so over C, since il + ( - l ) i = 0 is a nontrivial l inear
combi nati on of 1, i over C.
We pro ve
Lemma 5.2.3. If V is a vecto r space over F and v
x
, . , v in V are l in-
earl y i ndependent o ver F, then every el ement v E (v
x
,. . . , v) has a uni que
representati o n as
v = a
x
v
x
+ + a
n
v
n
with a
x
,. . ., ct in F.
Proof. Suppose that v (v
1
,. . ., v) has the two representati o ns as
v = a
}
v + + av = + + Bv with the as and /3's in F. This
gives us that (a
x
- B
x
)v
x
+ + (a ~ B)v = 0; since v
x
,..., v are l in-
earl y i ndependent over F, we co ncl ude that a
x
- B
x
= 0, . . . , a - B
n
= 0,
yiel ding for us the uni queness of the representati o n.
Ho w finite is a finite-dimensional vecto r space? To measure this, cal l a
subset v i , . . . , v of V a minimal generating set for V over F if V = (v
x
,..., v}
and no set of fewer than n el ements spans V over F.
We no w co me to the third vital l y i mpo rtant definition.
Definition. If V is a finite-dimensional vector space over /'. then the
dimension of V over F, wri tten dim
F
(V), is n, the number of el ements in a
mi ni mal generati ng set for V over F.
I n the exampl es given, di m
H
(C) = 2, since 1, i is a mi ni mal generati ng
set for C over R. Ho wever, di m
c
(C) = 1. I n Exampl e 1, di m
F
(V) = n and in
Exampl e 3, dim
F
(V) = n + 1. I n Exampl e 7 the di mensi on of V o ver F is 2.
Final l y, if(v
x
,..., v) C V, then dim
F
(v
x
,..., v) is at most n.
We no w pro ve
L emma 5.2.4. If V is finite di mensi o nal over F of di mensi on n and if
the el ements v
x
,..., v of V generate V over F, then v
x
,. .. ,v are l inearl y
i ndependent over* F.
Proof. Suppose that v
x
, . . . , u are l inearl y dependent over F; thus there
is a l inear combi nati on a
i
v
1
+ + a
n
v
H
= 0, where no t al l the a
t
are 0. We
may suppose, without l oss of general ity, that a
x
+ 0; then v
x
=
(-l/a
x
)(a
2
v
2
+ + av
n
). Gi ven v E. V, because v
u
. .., v is a generating
set for V over F,
V = ByVy + + BV
(a
2
v
2
+ + av
n
) + B
2
v
2
+ + Bv;
thus v
2
,. .., v span V over F, contradi cti ng that the subset u
l 5
v
2
,..., v is a
mi ni mal generati ng set of V over F.
We no w co me to yet ano ther i mpo rtant definition.
Definition. Let V be a finite-dimensional vector space over F; then
v v is a basis of V over F if the el ements v
L
,...,v span V over T
7
and
are l inearl y i ndependent over F.
By L emma 5.2.4 any mi ni mal generati ng set of V over F is a basis of V
over F. Thus, finite-dimensional vector spaces have bases.
Theo rem 5.2.5. Suppose that V is finite dimensional over /. Then any two
bases of V over F have the same number of el ements, and this number is ex-
actl y dim
F
(V).
Proof. L et u
1 (
. . ., v and w
x
,.. . , w
m
be two bases of V over F. We
want to show that m = n. Suppo se that m > n. Because v
x
,..., v is a basis
of V o ver F, we kno w that every el ement in V is a l inear combi nati on of the
Vj over F. I n particul ar, w
x
,..., w, are each a l inear combi nati on of v
l t
. . . , v
over F. Thus we have
w' i = a u U i + a i
2
y
2
+ + a
h
,v
w
2
= a , ^ ! + a
2 2
i ;
2
+ + a
2ll
v
w, = a
ml
v
x
+ a
m2
v
2
+ + a
m
v
188 Fi el ds Ch . 5
where the a,
;
- are in F.
Co nsi der
ftiVi + + B,w, = (a^By + a
2i
B
2
+ + a
ml
B
m
)v
1
+
+ (a
ln
B
x
+ a
2n
B
2
+ + oi
m
B
m
) v,
The system of l inear ho mo geneo us equati o ns
i , A + a
2i
B
2
+ + a
mi
B, = 0, i = 1, 2 , . . . , ,
has a nontrivial sol ution in T
7
by Theo rem 5.2.2, since the number of vari-
abl es, nu exceeds the number of equati o ns, n. If B
x
,..., B, is such a sol ution
in then, by the abo ve, B
x
w
x
+ + B
m
w, = 0, yet no t al l the /3, are 0. This
contradi cts the l inear i ndependence of w
1 ;
. . ., w
m
over A. Therefo re, m s 77..
Simil arl y, 77 <; 777; hence 777 = 77. The theo rem is no w pro ved, since a mi ni mal
generati ng set of V o ver F is a basis of V over F and the number of el ements
in this minimal generati ng set is di m
F
(V), by definition. Therefo re, by the
abo ve, n = dim
F
(V), co mpl eti ng the proof.
A further resul t, which we shal l use in fiel d theo ry, of a simil ar nature
to the things we have been do i ng is
Theo rem 5.2.6. Let V be a vecto r space over F such that di m
F
(V) =
77. If 777 > 77, then any 777 el ements of V are l inearl y dependent o ver F.
Proof. Let w
u
.. ., w, G V and l et v
x
,..., v be a basis of V o ver F;
here n = dim
F
(V) by Theo rem 5.2.5. Therefo re,
i v 1 = a
n
u
x
+ + a
Ul
v,
n
. . . ,w, = a m l V i + + a
m
v.
The pro o f given in Theo rem 5.2.5, that if 777 > 77 we can find B
u
... , B
m
in F,
and not al l 0, such that B
x
w
x
+ + B
m
w
m
= 0, goes over wo rd for wo rd.
But this establ ishes that w
{
,. .., w, are l inearl y dependent o ver F.
We cl ose this section with a final theo rem of the same fl avor as the pre-
ceding ones.
Theo rem 5.2.7. Let V be a vector space over F wi th dim
F
(V) =77.
Then any 77 l inearl y i ndependent el ements of V form a basis of V o ver F.
Proof. We want to show that if v
x
,..., v G V are l inearl y i ndependent
over F, then they span V over F. Let v G V; then v, v
x
,. . ., v are n + 1 el e-
ments, hence, by Theo rem 5.2.6, they are l inearl y dependent over F. Thus
there exist el ements a, a
1 ;
. . . , a in F, no t al l 0, such that av + a^j + +
a
n
v
n
= 0. The el ement a cannot be 0, o therwi se + + av = 0, and
no t al l the a, are 0. This woul d contradi ct the l i near i ndependence of the el e-
ments v
x
,..., v over F. Thus a #0, and so i; = (l /a)(a
1
i ;
1
+ + ay) =
B
x
Vi + + /3u, where B
t
= -a
i
/a
1
. Therefo re, v
x
,. . ., v span V over F,
and thus must form a basis of V o ver i
7
.
PROBL EMS
Easi er Pro bl ems
1. Determi ne if the fol l owing el ements in V, the vecto r space of 3-tupl es
over R, are l inearl y i ndependent o ver U.
(a) (1, 2, 3), (4, 5, 6), (7, 8, 9).
(b) (1, 0, 1), (0, 1, 2), (0, 0, 1).
(c) (1,2, 3), (0, 4, 5), (1,3,- r).
2. Fi nd a nontrivial sol ution in Z
5
of the system of l inear ho mo geneo us
equati o ns:
x
x
+ x
2
+ -v
3
= 0
XL + 2x
2
+ 3x
3
= 0
3x
x
+ 4x
2
+ 2x
3
= 0
3. If V is a vecto r space of di mensi o n n over Z
p
, p a pri me, show that V has
p" el ements.
4. Pro ve al l of L emma 5.2.1.
5. Let F be a fiel d and V = F[x], the pol ynomi al ring in x over F. Considering
V as a vector space over F, pro ve that V is no t finite dimensional over F.
6. If V is a finite-dimensional vecto r space o ver F and if W is a subspace of
V, pro ve that:
(a) W is finite di mensi onal over F and dim
F
(W) < di m
F
(y).
(b) If di m
F
(W) = di nv(y), then V = W.
* 7. Define what you feel shoul d be a vecto r space ho mo mo rphi sm ip of V
into W, where V
7
and W are vector spaces o ver F. What can yo u say abo ut
the kernel , K, of i/>where 7C = {i>G V | i/>(u) = 0}? What shoul d a vecto r
space i so mo rphi sm be?
190 Fiel ds Ch . 5
8. If V is a vector space over F and W is a subspace of V, define the requi -
site o perati o ns in VIW so that V/W beco mes a vector space over F.
9. If V is a finite-dimensional vecto r space over F and v
x
,. . ., v, in V are
l inearl y i ndependent o ver F, show we can find w
x
,... , w
r
in V, where
m + r = di m
F
( V) , such that v
x
,..., v,, w
x
,..., w
r
fo rm a basis of V
over F.
10. If i/>: F V" is a ho mo mo rphi sm of V onto V with kernel if, show that
V VIK (as vecto r spaces over F). (See Pro bl em 7).
11. Show that if di m
F
(y) = n and W is a subspace of V with di m
f
(W) = m,
then di m
F
(V/W) = n - m.
12. If V is a vector space over F of di mensi o n n, pro ve that V is i so mo rphi c
to the vector space of /z-tupl es over F (Exampl e 1). (See Pro bl em 7).
13. Let if D F be two fiel ds; suppo se that K, as a vector space over F, has fi-
nite di mensi on n. Show that if a E if, then there exist a
0
, cvj ,... , a in F,
no t al l 0, such that
a
0
+ a
x
a + a
2
a
2
+ + aa" = 0.
14. Let F be a fiel d, F[x] the po l yno mi al ring in x over F, and fix) 0 in
F[x]. Consi der V = F[x]// as a vecto r space over F, where J is the ideal
of F[x] generated by fix). Pro ve that
di m
F
( V ) = deg fix).
15. If V and W are two finite-dimensional vecto r spaces over F, pro ve that
V W is finite di mensi onal over F and that dim
F
iV W) =
di m
f
(V) + di m
F
(W).
16. Let V be a vector space over F and suppo se that U and TV are subspaces
of V. Define / + W = {u + w | u E [7, w E W). Pro ve that:
(a) U + W is a subspace of V".
(b) 7 + W is finite di mensi o nal o ver Fi f bo th U and W are.
(c) 7 n W is a subspace of V.
id) 7 + W is a ho mo mo rphi c i mage of 7 W.
(e) If 7 and W are finite di mensi o nal over F, then
dim
F
( 7 + W) = dim
F
( 7) + di m
F
(I ) - di m
F
( 7n W).
Sec . 3 Fi el d Ex t en s i o n s 191
Harder Pro bl ems
17. Let K D Fbe two fiel ds such that dim
F
(K) = m. Suppose that Vi s a vec-
to r space over K. Pro ve that:
(a) V is a vector space over F.
(b) If V is finite di mensi onal o ver K, then it is finite di mensi onal over F.
(c) If dim
K
(V) = n, then dim
F
(V) = mn [i.e., di m
F
(V) =
dimjcCV) dim
F
(K)].
18. Let K D Fbe fiel ds and suppo se that Vis a vector space over 7CT such that
dim
F
(V) is finite. If dim
F
(K) is finite, show that dim
K
(V) if finite and de-
termi ne its val ue in terms of dim
F
(V) and di m
f
(K).
19. Let D be an integral do mai n with 1, which happens to be a finite-dimen-
sional vector space over a fiel d F. Pro ve that D is a fiel d. (No te: Since F\,
which we can identify with F, is in D, the ring structure of D and the vec-
to r space structure of D over F are in harmo ny with each other.)
20. Let V be a vector space over an infinite fiel d F. Show that V cannot be the
set-theoretic union of a finite number of proper subspaces of V. (Very hard)
3 . FIELD EXT ENSI ONS
Our attenti o n no w turns to a rel ati o nshi p between two fiel ds K and F, where
K D F. We cal l K an extension (or extension field) of F, and cal l F a sub field
of The o perati o ns in F are just tho se of K restri cted to the el ements of F.
In all that follows in this section it will be understood that F C K.
We say that K is a finite extension of F if, vi ewed as a vector space over
F, d'\m
F
(K) is finite. We shal l write di m
f
(K) as [K: F] and cal l it the degree
of K over F.
We begi n o ur discussion with what is usual l y the first resul t one pro ves
in tal king abo ut finite extensions.
Theo rem 5.3.1. Let L D K D F be three fiel ds such that both [L : K]
and [A^: F] are finite. Then L is a finite extension of F and [L : F] =
[L:K][K:F].
Proof. We shal l pro ve that L is a finite extensi o n of F by expl icitl y ex-
hibiting a finite basis of L over F. I n doi ng so, we shal l o btai n the stronger re-
sul t asserted in the theo rem, namel y that [L : F] = [L : K][K: F].
Suppose that [L : K] = m and [K:F]= n; then L has a basis v
x
, v
2
,..., v,
over A^, and K has a basis w
t
, w
2
,..., w over i
7
. We shal l prove that the mn
192 Fi el ds Ch . 5
el ements v-Wj, where / =1, 2, . . . , m and / =1, 2, . . . , n, co nsti tute a basis of
L over F.
We begin by showing that, at l east, these el ements span L o ver F; this
wil l , of course, show that L is a finite extensi on of F. L et a G L; since the el e-
ments v
x
, . . . , u, form a basis of L o ver K, we have a =k
x
v
x
+ + k
m
v
where k
x
, k
2
,. .., k, are in K. Since w
x
,..., w is a basis of K over F, we
can express each /c, as
= fnWi + f
i2
w
2
+.- +//w,
where the are in F. Substi tuti ng these expressions for the k
t
in the forego-
ing expression of a, we o btai n
a = ( / n W i + f
12
w
2
+ + f
ln
w) vy
+ + (fmlWi + f
m2
W
2
+ + f
mn
W
n
) V
m
.
Therefo re, on unscrambl i ng this sum expl icitl y, we o btai n
a = fyyWyVy + f
12
W
2
V
1
+ + fijWjVj + + f
mn
WV
m
.
Thus the mn el ements v
t
Wj in L span L over F; therefo re, [L : F] is finite and,
in fact, [L : F] <m.
To show that [L : F] = mn, we need onl y show that the mn el ements
vjWj abo ve are l inearl y i ndependent over F, for thento gether with the fact
that they span L over Fwe wo ul d have that they form a basis of L o ver F.
By Theo rem 5.2.5 we wo ul d have the desired resul t [L : F] = mn =
[L:K][K:F].
Suppo se then that for so me b
u
in F.we have the rel ati o n
0 = b
u
v
1
w
1
+ b
12
v
1
w
2
+ + b
1
v
1
w
n
+ b^v^y
+ + b
2
v
2
w
n
+ + b
mX
v
m
Wy + b
m
v,w.
Reassembl i ng this sum, we o btai n c
x
v
x
+ c
2
v
2
+ + c,v
m
= 0, where c
x
=
b \ i
w
i + +b
ln
w, . . . , c
m
= >,i Wi + + b
mn
w
n
. Since the c,- are el e-
ments of K and the el ements v
x
, . .., v in L are l inearl y i ndependent o ver K,
we o btai n c
t
c
2
= = c
m
= 0.
Recal l ing that q =b
n
w
x
+ +b
ln
w, where the b
tj
are in F and
where w
x
,..., w
n
in K are l inearl y i ndependent over F, we deduce from the
fact that c
x
= c
2
= c, = 0 that every b^ =0. Thus onl y the trivial l i near
co mbi nati o n, with each coefficient 0, of the el ements y,-iv
;
- over F can be 0.
Hence the v
t
Wj are l inearl y i ndependent o ver F. We saw abo ve that this was
eno ugh to pro ve the theo rem.
The reader shoul d co mpare Theo rem 5.3.1 with the sl ightl y mo re gen-
eral resul t in Pro bl em 17 of Section 2. The reader shoul d no w be abl e to
sol ve Pro bl em 17.
As a co nsequence of the theo rem we have the
Corol l ary. If L D K D F are three fiel ds such that [L : F] is finite,
then [K : F] is finite and divides [L : F].
Proof. Since L D K, K canno t have mo re l inearl y i ndependent el e-
ments over F than does L. Because, by Theo rem 5.2.6, [L : F] is the size of
the l argest set of l inearl y i ndependent el ements in L over F, we therefo re get
that [K: F] < [L : F], so must be finite. Since L is finite dimensional o ver F
and since K contains F, L must be finite di mensi o nal over K. Thus al l the
condi ti ons of Theo rem 5.3.1 are ful fil l ed, whence [L : F] = [L : K][K : F].
Co nsequentl y, [K : F] divides [L : F], as is asserted in the Corol l ary.
If K is a finite extensi on of F, we can say qui te a bit abo ut the behavi o r
of the el ements of K vis-a-vis F.
Theo rem 5.3.2. Suppo se that K is a finite extension of F of degree n.
Then, given any el ement u in K there exist el ements a
0
, a
x
,.. . , a
n
in F, not
al l zero , such that
a
0
+ a
x
ti + + a
n
u" = 0.
Proof. Since [K: F] = dim
F
(K) = n and the el ements 1, u, u
1
,. . . , u"
are n +1 in number, by Theo rem 5.2.6 they must be l inearl y dependent over
F. Thus we can find a
Q
, a
1 ;
. . . , a in F, no t al l 0, such that a
n
+ a
x
ii +
a
2
u
2
+ + au" = 0, provi ng the theo rem.
The concl usion of the theo rem suggests that we singl e o ut el ements in
an extensi o n fiel d that satisfy a nontri vi al po l yno mi al .
Definition. If K D F are fiel ds, then a G K is said to be algebraic over
F if there exists a po l yno mi al p (x) ^0 in F[x] such that p (a) = 0.
By p(a) we shal l mean the el ement a
Q
a" + ata"^
1
+ + a in X,
where p(x) = a
0
x" + a
x
x
n
~
l
+ + a
n
.
If K is an extensi on of F such that every el ement of K is al gebraic over
F, we cal l K an algebraic extension of F. I n these terms Theo rem 5.3.2 can be
restated as: IfK is a finite extension ofF, then K is an algebraic extension ofF.
The co nverse of this is no t true; an al gebraic extension of F need not be
of finite degree over F. Can yo u co me up with an exampl e of this situation?
194 Fi el ds Ch . 5
An el ement of K that is no t al gebraic over F is said to be transcendental
o ver F.
Let's see so me exampl es of al gebraic el ements in a concrete context.
Consider C 73 <Q, the compl ex fiel d as an extension of the rational one. The com-
pl ex number a = 1 + i is al gebraic over 0 , since it satisfies a
2
- 2a + 2 = 0
over Q. Simil arl y, the real number b = V1 +V l + \/2 is al gebraic o ver Q>,
since b
2
= 1 + VT T vl , so (6
2
- l)-
1
= 1 + V 2 , and therefo re
((b
2
l )
3
I )
2
= 2. Expandi ng this out, we get a nontrivial po l yno mi al ex-
pression in b with rati o nal coefficients, which is 0. Thus b is al gebraic over Q.
I t is possi bl e to co nstruct real numbers that are transcendental o ver Q
fairl y easil y (see Secti on 6 of Chapter 6). Ho wever, it takes so me real effort
to establ ish the transcendence of certai n famil iar numbers. The two famil -
i ar numbers e and IT can be sho wn to be transcendental o ver Q. That e is
such was pro ved by Hermi te in 1873; the proof that TT is transcendental
o ver Q is much harder and was first carri ed o ut by L i ndemann in 1882. We
shal l no t go i nto the pro o f here that any parti cul ar number is transcenden-
tal over Q. Ho wever, in Section.,7 of Chapter 6 we shal l at l east show that
7r is i rrati o nal . This makes it a po ssi bl e candi date for a transcendental num-
ber of Q, for cl earl y any rati o nal number b is al gebrai c o ver <Q because it
satisfies the po l yno mi al p (x) = x - b, which has rati o nal coefficients.
Definition. A co mpl ex number is said to be an algebraic number if it
is al gebraic over <Q.
As we shal l soon see, the al gebraic numbers form a fiel d, which is a
subfiel d of C.
We return to the general devel o pment of the theo ry of fiel ds. We have
seen in Theo rem 5.3.2 that if K is a finite extension of F, then every el ement
of K is al gebraic over F. We turn this matter aro und by asking: If K is an ex-
tension of F and a G K is al gebraic over F, can we so meho w pro duce a finite
extensi on of Fusi ng al The answer is yes. This wil l co me as a co nsequence of
the next theo remwhi ch we pro ve in a context a l ittl e mo re general than
what we real l y need.
Theo rem 5.3.3. Let D be an integral do mai n with 1 which is a finite-
di mensi onal vector space over a fiel d F. Then D is a fiel d.
Proof. To pro ve the theo rem, we must pro duce for a 0 in D an in-
verse, a~
l
, in D such that aaT
1
= 1.
As in the proof of Theo rem 5.3.2, if dim
F
(D) = n, then 1, a, cr,
in D are l inearl y dependent over F. Thus for so me appro pri ate a
0
, a
u
. . . , a
in F, not al l of which are 0,
a
0
a" + aid"'
1
+ + = ().
Let p(x) = B
0
x
T
+ B
x
x
r
~
1
+ + B,. = 0 be a po l yno mi al in F[x] of l owest
degree such that p (a) = 0. We assert that B
r
+ 0. Fo r if B
r
= 0, then
0 = B
0
a
r
+ B
x
cr
l
+ + B
r
_
x
a
= (Bâ'-
1
+ + + B
r
_
L
) a.
Since D is an integral do mai n and a = 0, we concl ude that B
a
a'~
l
+
Bya'
2
+ + B
r
-1 = 0, hence q(a) = 0, where q(x) = Bf)
x
'
1
+
B
x
x
r
~
2
+ + /3
r
_ ! in F[x] is of l ower degree than p(x), a contradiction.
Thus B
r
i= 0, hence B~
1
is in Fand
Pr
giving us that (p
Q
a
r
~
l
+ + /3
r
_i )//3
r
, which is in D, is the a~
l
in 73 that
we requi red. This proves the theo rem.
Havi ng Theo rem 5.3.3 in hand, we want to make use of it. So, ho w do
we pro duce subrings of a fiel d K that co ntai n F and are finite di mensi onal
over Fl Such subrings, as subrings of a fiel d, are automatical l y integral do-
mai ns, and woul d satisfy the hypo thesi s of Theo rem 5.3.3. The means to this
end wil l be the el ements in K that are al gebraic over F.
But first a definition.
Definition. The el ement a in the extensi on K of F is said to be alge-
braic of degree n if there is a po l yno mi al p (x) in F[x] of degree n such that
p (a) = 0, and no no nzero po l yno mi al of l ower degree in F[x] has this pro p-
erty.
We may assume that the po l yno mi al p (x) in this definition is mo ni c, for
we coul d divide this pol ynomi al by its l eadi ng coefficient to obtain a mo ni c
po l yno mi al q(x) in F[x], of the same degree as p(x), and such that q(a) = 0.
We hencefo rth assume that this po l yno mi al p(x) is mo ni c; we cal l it the min-
imal polynomial for a over F.
L emma 5.3.4. Let a G K be al gebraic o ver F with mi ni mal pol ynomial
p(x) in F[x]. Then p(x) is i rreduci bl e in F[x].
196 Fi el ds Ch . 5
Proof. Suppo se that p(x) is no t irreducibl e in F[x]; then p(x) =
f(x)g(x) where f(x) and g(x) are in F[x] and each has positive degree. Since
0 =p(a) =/(a)g(a), and since f(a) and g(o ) are in the fiel d K, we co ncl ude
that f(a) = 0 or g(a) = 0, bo th of which are impossibl e, since bo th f(x) and
g(x) are of l ower degree than/f x). T her ef o r e, i s i rreduci bl e in F[x].
Let a G K be al gebraic of degree n over F and l et p (x) G F[.r] be its
mi ni mal pol ynomi al over F. Given f(x) G F[x], then/(x) = q(x)p(x) + r(x),
where q(x) and r(x) are in F[x] and = 0 or deg r(x) < deg/?(x) fol l ows
from the division al gorithm. Theref o re,/(a) =q(a)p(a) + r(a) = r(a), since
p(a) =0. I n short, any po l yno mi al expression in a o ver F can be expressed as
a pol ynomi al expression in a of degree at most n - 1.
Let F[a] = {/(a) | f(x) G F[x]). We cl aim that F[a] is a subfiel d of # that
contains both F and a, and that [F[a]: F] = n. By the remark made abo ve,
F[a] is spanned over F by 1, a, a
2
, . . . , A "
- 1
, so is finite di mensi onal over F.
Mo reo ver, as is easil y verified, F[a] is a subring of K; as a subring of K, F[a] is
an integral domai n. Thus, by Theo rem 5.3.3, F[a] is a fiel d. Since it is spanned
over F by 1, a, a
2
,.. ., a"
- 1
, we have that [F[a] :F] < 72. To sho w that
[F[a]: F] = n we must merel y show that 1, , a
2
, . . . , a"~
l
are l inearl y i nde-
pendent over F. But if cr
0
+ a
x
a + + evâ"
- 1
= 0, with the a,- in F, then
q(a) = 0, where q(x) = a
Q
+ ape + + a
n
_
x
x"~
x
is in F[x]. Since q(x) is of
l ower degree than p(x), which is the minimal pol ynomi al for a in F[x], we are
forced to concl ude that q(x) = 0. This impl ies that a
0
= a
x
= = _! =0.
Therefo re, 1, a, a
2
, . . . , a"
- 1
are l inearl y i ndependent over F and fo rm a basis
of F[a] over F. Thus [F[a]: F] = n. Since F[a] is a fiel d, not merel y just a set
of pol ynomial expressions in a, weshall denote F[a] by F(a). No te al so that if
M is any fiel d that contains bo th F and a, then M contains al l po l yno mi al ex-
pressions in a over F, hence M D F(a). So F(a) is f/ze smallest subfield of K
containing both F and a.
Definition. F(a) is cal l ed the fiel d or extensi on o btai ned by adjoining
a to F.
We no w summari ze.
Theo rem 5.3.5. Let K D F and suppo se that a in K is al gebraic over F
of degree n. Then F(a), the fiel d o btai ned by adjoining a to F, is a finite ex-
tensi o n of F, and
[F(a): F] = n.
Befo re l eaving Theo rem 5.3.5, l et's l ook at it in a sl ightl y different way.
Let F[x] be the po l yno mi al ring in x o ver F, and l et M = (p(x)) be the ideal
Sec. 3 Field Extensions 197
of F[x] generated by p(x), the mi ni mal po l yno mi al for a in K over F.
By L emma 5.3.4, p(x) is irreducibl e in F[x]; hence, by Theo rem 4.5.11,
M is a maxi mal ideal of F[x]. Therefo re, F[x]/(p(x)) is a fiel d by Theo rem
4.4.2.
Defi ne the mappi ng i/>: F[x] > by \f>(f(x)) = / (a). The mappi ng i/>is
a ho mo mo rphi sm of F[x] into /C, and the i mage of F[x] in /C is merel y 7(a)
by the definition of 7(a). What is the kernel of (/>? I t is by definition / =
{f{x) G F[x] | />(/(x)) = 0}, and since we kno w i{,(f(x)) = /(a), / = {fix) G
7[x] | f(a) = 0}. Since p(x) is in / and p(x) is the mi ni mal pol ynomi al for a
over F, p (x) is of the l owest possibl e degree amo ng the el ements of 7. Thus
/ = ip(x)) by the proof of Theo rem 4.5.6, and so / = M. By the First Ho mo -
mo rphi sm Theo rem for rings, F[x]IM i mage of F[x] under if/ Fid), and
since F[x]/M is a fiel d, we have that F{a) is a fiel d. We l eave the proof, from
this po i nt of view, of [7(a): 7] = degp(x) to the reader.
PROBL EMS
1. Sho w that the fol l owing numbers in C are al gebraic numbers.
(a) V 2 + V 3.
(b) V 7 + #12.
(c) 2 + i V5.
(d) cos(27r//c) + i sin(2ir/k), k a positive integer.
2. Determi ne the degrees over <Q of the numbers given in Parts (a) and (c)
of Pro bl em 1.
3. What is the degree of cos(27r/3) + i sin(27r/3) over Q?
4. What is the degree of COS(2T T /8) + i sin(27r/8) over Q?
5. If p is a pri me number, pro ve that the degree of cos(27r/p) + / sin(2- 7 r /p)
o ver Q is p 1 and that
fix) = 1 + x + x
2
+ + x ^
1
is its mi ni mal pol ynomi al over Q.
6. (For tho se who have had cal cul us) Show that
1 1 1
e = 1 + + +
. . . + +. . .
1! 2! m
is i rrati onal .
7. If a in K is such that a
2
is al gebraic over the subfiel d 7 of K, show that a
is al gebraic over 7.
198 Melds Ch . 5
8. M F C i f and f(a) is al gebraic over F, where /(x) is of positive degree in
F[x] and a G if, pro ve that a is al gebraic over F.
9. I n the discussion fol l owing Theo rem 5.3.5, show that F[x]/M is of degree
n = deg TJ>(X) over F, and so [F(a): F] = n = deg/?(*)
10. Pro ve that cos 1 is al gebraic o ver Q. (1 = o ne degree.)
11. If a E if is transcendental over F, l et F(a) = {f(a)/g(a) \ f(x), g(x) + 0 G
F[x]}. Show that F(a) is a fiel d and is the smal l est subfiel d of if contai n-
ing bo th F and a.
12. If a is as in Pro bl em 11, show that F(a) F(x), where F(x) is the fiel d of
rati onal functions in x over F.
13. Let if be a finite fiel d and F a subfiel d of if. lf[K:F] = n and F has q
el ements, show that i f has q" el ements.
14. Using the resul t of Pro bl em 13, show that a finite fiel d has p" el ements
for so me pri me p and so me positive integer n.
15. Construct two fiel ds i f and F such that if is an al gebraic extensi o n of F
but is no t a finite extensi o n of F.
4. FI NI TE EXTENSI ONS
We co nti nue in the vein of the precedi ng section. Agai n K D F wil l al ways
deno te two fiel ds.
Let F(i f ) be the set of al l el ements in i f that are al gebraic o ver F. Cer-
tainl y, F C E(K). Our objective is to pro ve that E(K) is a fiel d. Once this is
do ne, we'l l see a l ittl e of ho w E(K) sits in if.
Wi tho ut further ado we pro ceed to
Theo rem 5.4.1. F(i f ) is a subfiel d of if.
Proof. What we must sho w is that if a, b G i f are al gebraic o ver F, then
a b, ab, and alb (if 0) are al l al gebraic over F. This wil l assure us that
F(i f ) is a subfiel d of if. We'l l do al l of a b, ab, and alb in o ne shot.
Let if o = F{a) be the subfiel d of K o btai ned by adjoining a to F. Since
a is al gebraic over F, say of degree m, then, by Theo rem 5.3.5, [K
0
: F] = m.
Since Z? is al gebraic over F and since i f
0
contai ns F, we certainl y have that b
is al gebraic over K
0
. If 6 is al gebraic o ver F of degree n, then it is al gebraic
over if
0
of degree at most n. Thus if! = K
0
(b), the subfiel d of i f o btai ned by
adjoining b to K
0
, is a finite extensi o n of K
0
and [i f
:
: i f
0
] < n.
Thus, by Theo rem 5.3.1, [if, : F] = [if
2
: i f
0
][i f
0
: F] < 777/7; that is, if, is
a finite extension of F. As such, by Theo rem 5.3.2, K
t
is an al gebraic exten-
Sec. 4 Fi ni t e Ex t en s i o n s 199
sion of F, so al l its el ements are al gebrai c over F. Since a E K
0
C K
x
and 6
K
x
, then al l of the el ements a b, ab, alb are in K
x
, hence are al gebraic o ver
F. This is exactl y what we wanted. The theo rem is pro ved.
If we l o o k at the pro o f a l ittl e mo re careful l y, we see that we have actu-
al l y pro ved a l ittl e mo re, namel y the
Corol l ary. If a and b in K are al gebraic over F of degrees m and n,
respectivel y, then a b, ab, and alb (if b = 0) are al gebraic over F of degree
at mo st mn.
A special case, but o ne wo rth no ti ng and recordi ng, is the case K = C
and F = <Q. I n that case we cal l ed the al gebraic el ements in C over Q the al-
gebraic numbers. So Theo rem 5.4.1 in this case beco mes
Theo rem 5.4.2. The al gebraic numbers form a subl iel d of C.
Fo r al l we kno w at the mo ment, the set of al gebraic numbers may very
wel l be al l of C. This is no t the case, for transcendental numbers do exist; we
show this to be true in Section 6 of Chapter 6.
We return to a general fiel d K. I ts subfiel d E(K) has a very parti cul ar
qual ity, which we pro ve next. This pro perty is that any el ement in K which is
al gebraic over E(K) must al ready be in ELK).
I n o rder no t to digress in the co urse of the proof we are abo ut to give,
we i ntro duce the fol l owing no tati o n. If a
x
, a
2
, . . . , a
n
are in K, then
F(a
x
,.. ., a) wil l be the fiel d o btai ned as fol l ows: K
x
= F(a
x
), K
2
=
Ki(a
2
) = F(a
x
, a
2
), K
3
= K
2
(a
3
) = F(a
x
, a
2
, a
3
), ... , K = K
n
_
x
(a
n
) =
F(a
x
, a
2
,. .., a).
We no w pro ve
Theo rem 5.4.3. If u in K is al gebraic over E(K), then u is in E(K).
Proof. To pro ve the theo rem, al l we must do is show that u is al gebraic
over F; this wil l put u in E(K), and we wil l be do ne.
Since u is al gebraic over E(K), there is a nontri vi al po l yno mi al /(x) =
x" + a
x
x"~
l
+ a
2
x"~
2
+ + a, where a
t
, a
2
,. . ., a are in E(K), such that
/ () = 0. Since a
x
, a
2
,. .., a are in E(K), they are al gebraic over F of de-
grees, say, m
x
,
m
i , >
W7
>respectivel y. We cl aim that [F(a
x
,. .., a): F] is
at mo st m
l
m
2
m. To see this, merel y carry o ut n successive appl ications
of Theo rem 5.3.1 to the sequence K
x
, K
2
,..., K of fiel ds defined above. We
l eave its pro o f to the reader. Thus, since u is al gebraic over the fiel d K
n
=
F(a
x
, a
2
, . . . , a
n
) [after al l , the po l yno mi al satisfied by u is f(x) =
200 Fi el ds Ch . 5
x" + a
x
x
n
~
l
+ + , which has al l its coefficients in F(a
x
, a
2
,. ., a,,)], the
fiel d K
n
(it) is a finite extensi o n of K, and since K
n
is a finite extensi o n of F,
we have, again by Theo rem 5.3.1, that K(u) is a finite extensi o n of F. Be-
cause u G K(u), we o btai n from Theo rem 5.3.2 that u is al gebraic over F.
This puts u in E(K) by the very definition of E(K), thereby pro vi ng the theo -
rem.
There is a famo us theo rem due to Gauss, often referred to as the Fun-
damental Theorem of Algebra, which asserts (in terms of extensi o n) that the
onl y finite extension of C, the fiel d of co mpl ex numbers, is C itsel f. I n real ity
this resul t is no t a purel y al gebrai c o ne, its val idity dependi ng heavil y o n
topol ogical pro perti es of the fiel d of real numbers. Be that as it may, it is an
extremel y i mpo rtant theo rem in al gebra and in many o ther parts of mathe-
matics.
The formul ation of the Fundamental Theo rem of Al gebra in terms of
the nonexi stence of finite extensi ons of C is a l ittl e different from that whi ch
is usual l y given. The mo st frequent fo rm in which this famous resul t is stated
invol ves the concept of a ro o t of a po l yno mi al , a concept we shal l discuss at
so me l ength l ater. I n these terms the Fundamental Theo rem of Al gebra be-
co mes: A pol ynomi al of positive degree having coefficients in C has at l east
o ne ro o t in C. The exact meani ng of this statement and its equi val ence wi th
the o ther form of the theo rem stated abo ve wil l beco me cl earer l ater, after
the devel o pment of the materi al o n ro o ts.
A fiel d L with the pro perty of C descri bed in the paragraphs abo ve is
said to be algebraically closed. If we grant that C is al gebraical l y cl osed
(Gauss' Theo rem), then, by Theo rem 5.4.3, we al so have
The field of algebraic numbers is algebraically closed.
PROBLEMS
1. Sho w that a = V2 V3 is al gebraic o ver <Q of degree at mo st 4 by ex-
hibiting a pol ynomi al f(x) of degree 4 over Q such that f(a) = 0.
2. If a and b in K are al gebraic over F of degrees m and n, respectivel y, and
if m and n are rel ativel y pri me, show that [F(a, b): F] = mn.
3. If a G C is such that p (a) = 0, where
p(x) = x
5
+ Vix
3
+ V5x
2
+ Vlx + VLT,
show that a is al gebraic over <Q of degree at mo st 80.
Sec . 5 Co n s t r u c t i b i l i t y 201
4. If K D F is such that [K: F] = p, p a pri me, show that K = F(a) for every
a in K that is not in F.
5. If [K:F] = 2" and T is a subfiel d of K contai ni ng F, show that [T:F]= 2"'
for so me m n.
6. Give an exampl e of two al gebraic numbers a and >of degrees 2 and 3, re-
spectivel y, such that ab is of degree less than 6 over Q.
7. I f/v D F are fiel ds and a
x
, . . . , a are in show that F(a
ly
. . ., a) equal s
F(f l
t T ( 1 )
,.. . , a()) for any permutati o n cr of 1, 2, . . . , .
5. CONSTRUCTI B I L I TY
I n ancient Greece, unl ike in the o ther cul tures of the ti me, the Greek mathe-
maticians were i nterested in mathemati cs as an abstract discipl ine rather
than as a pragmati c bag of tricks to do accounts or to carry o ut measure-
ments. They devel o ped stro ng i nterests and resul ts in number theory and,
most especial l y, in geo metry. I n these areas they po sed penetrati ng ques-
tions. The questi o ns they asked in geo metrytwo of which wil l make up the
topic treated hereare stil l of i nterest and substance. The Engl ish mathe-
mati ci an G. H. Hardy, in his sad but charmi ng l ittl e bo o k A Mathematician's
Apology, describes the ancient Greek mathemati ci ans as "col l eagues from
ano ther col l ege."
Two of these Greek questi ons wil l be o ur co ncern in this section. But,
as a matter of fact, the answer to bo th wil l emerge as a consequence of the
criterion for constructibil ity, which we wil l o btai n. We state these questions
no w and wil l expl ain a l ittl e l ater what is entai l ed in them.
QUESTI ON 1
Can o ne dupl i cate a cube using just strai ght-edge and compass? (By dupl icat-
ing a cube, we mean do ubl i ng its vo l ume.)
QUESTI ON 2
Can o ne trisect an arbi trary angl e using just strai ght-edge and compass?
Despi te the seemingl y infinite number of angl e-trisectors that cro p up
every year, the answer to bo th questi o ns is "no ." As we shal l see, it is i mpos-
sibl e to trisect 60 using just straight-edge and compass. Of course, so me
angl es are trisectabl e, for i nstance, 0, 90, 145, 180,. . . , but most angl es
(in a very precise meani ng of "mo st") are not.
202 Fi el ds Ch . 5
Befo re getti ng to the exact meani ng of the questi o ns themsel ves, we
want to spel l o ut in expl icit terms exactl y what the rul es of the game are.
By a straight-edge we do not mean a riderthat is, an i nstrument for mea-
suring arbi trary l engths. No! A strai ght-edge is merel y a strai ght l ine,
with no quanti tati ve o r metri c pro perti es attri buted to it. We are given a
l ine segmentto whi ch we assign l ength 1and al l o ther l engths that we
get from this must be o btai nabl e merel y empl o yi ng a strai ght-edge and
co mpass.
Let us cal l a no nnegati ve real number, b, a constructible length if, by
a finite number of appl i cati o ns of the strai ght-edge and co mpass and the
po i nts of i ntersecti o n o btai ned between l ines and circl es so co nstructed,
we can co nstruct a l ine segment of l ength b, starti ng o ut from the l ine seg-
ment we have assi gned l ength 1.
Fro m o ur high school geo metry we recal l so me things we can do in this
framework.
1. Whatever l ength we construct o n o ne l ine can be co nstructed o n any
o ther l ine by use of the co mpass acting as a transfer agent.
2. We can draw a l ine paral l el to a given l ine that goes thro ugh a given
point.
3. We can construct a l ength n for any no nnegati ve i nteger n.
Fro m these and by using resul ts abo ut the simil arity of triangl es, we can
construct any no nnegati ve rati o nal l ength. We do n't do that at this mo ment
for it wil l co me out as a special case of what we are abo ut to do .
We cl aim the fol l owing pro perti es:
1. If a and b are constructibl e l engths, then so is a + b. Fo r if AB is a
l ength segment of l ength a and CD is o ne of l ength b, we can transfer this
l ine segment CD, by means of a co mpass, to obtai n the l ine ABE, where AB
is of l ength a and BE is of l ength b. Thus the l ine segment AE is of l ength
a + b. If b > a, ho w woul d you construct b al
2. If a and 6 are constructibl e l engths, then so is ab. We may assume
that < 7 ^0 and b 0, otherwi se, the statement is trivial . Co nsi der the fol l ow-
ing di agram
L
j
B
Where L
x
and L
2
are two distinct l ines intersecting at P, and such that PA
has l ength a, PB has l ength b, and PJ has l ength 1. Let L
3
be the straight l ine
thro ugh ./ and A and L
4
the l ine paral l el to L
3
passing thro ugh B. If C is the
po i nt of intersection of L
x
and L
4
, we have the di agram
Al l of these constructions can be carri ed out by straight-edge and compass.
Fro m el ementary geo metry the l ength of PC is ab. Therefo re, ab is con-
structibl e.
3. If a and b are constructibl e and b + 0, then alb is constructibl e. Co n-
sider the di agram
where P, A, B, J, L
u
and L
2
are as in Pro perty 2 abo ve. Let L
5
be the l ine
thro ugh A and B and l et L
6
be the l ine thro ugh / paral l el to L
s
. If D is the
po i nt of intersection of L
x
and L
6
, then, again by el ementary geo metry, the
l ength of PD is alb. We stress again that al l the constructions made can be
carried out by straight-edge and compass.
Of course, this shows that the no nnegati ve rati o nal numbers are con-
structibl e l engths, since they are quo ti ents of no nnegati ve integers, which we
kno w to be constructibl e l engths. But there are o ther constructibl e l engths,
for i nstance, the irrational number V 2 . Because we can construct by straight-
edge and co mpass the right-angl e triangl e
L
L
204 Fi el ds Ch . 5
with sides AB and BC of l ength 1, we know, by the Pythago rean Theo rem,
that AC is of l ength V 2 . So V 2 is a constructibl e l ength.
I n Pro perti es 1 to 3 we sho wed that the constructibl e l engths al most
form a fiel d. What is l acking is the negatives. To get aro und this, we make
the
Definition. The real number a is said to be a constructible number if
\a\, the absol ute val ue of a, is a constructibl e l ength.
As far as we can say at the mo ment, any real number mi ght be a con-
structibl e o ne. We shal l soon have a criterion which wil l tel l us that certai n
real numbers are no t constructibl e. Fo r instance, we shal l be abl e to deduce
from this criterion that bo th V 2 and cos 20 are not constructibl e. This in
turn wil l al l ow us to show that the answer to bo th Questi o ns 1 and 2 is "no ."
But first we state
Theo rem 5.5.1. The constructi bl e numbers form a subl iel d of the fiel d
of real numbers.
Proof. Pro perti es 1 to 3 al most do the trick; we must adapt Pro perty 1
sl ightl y to al l ow for negatives. We l eave the few detail s to the reader.
Our next goal is to show that a constructi bl e number must be an al ge-
brai c numberno t any ol d al gebraic number, but o ne satisfying a rather
stringent condition.
No te, first, that if a s 0 is a constructi bl e number, then so is \fa. Co n-
sider the di agram
I t is of a semicircl e of radi us (a + 1)12, center at C, AD is of l ength a, DB is
of l ength 1, and DE is the perpendi cul ar to AB at D, intersecting the circl e at
E. Al l this is constructibl e by strai ght-edge and compass. Fro m el ementary
geo metry we have that DE is of l ength V a, hence \/~a is constructibl e.
We no w head for the necessary co ndi ti o n that a real number be con-
structibl e. Let K be the fiel d of constructi bl e numbers, and l et K
0
be a sub-
fiel d of K. By theplane of K
Q
we shal l mean the set of al l poi nts (a, b) in the
real Eucl i dean pl ane whose co o rdi nates a and b are in K
0
. If (a, b) and (c, d)
are in the pl ane of i f
0
, then the straight l ine joining them has the equati o n
(y - b)/(x - a) = (b - d)/(a - c), so is of the form ux + vy + w = 0, where
u, v, and w are in i f
0
. Gi ven two such l ines upc + v
x
y + w
x
= 0 and
u
2
x + v
2
y + M
;
2 = 0, where u
x
, v
x
, w
x
and u
2
, v
2
, w
2
are al l in if
0
, either they
are paral l el or their po i nt of i ntersecti on is a po i nt in i f
0
. (Prove!)
Gi ven a circl e who se radi us r is in if
0
and who se center (a, b) is in the
pl ane of i f
0
, then its equati o n is (x a)
2
+ (y b)
2
= r
2
, which we see, on
expanding, is of the formx
2
+ y
2
+ dx + ey +/ = 0, where d, e, and / are in if
0
.
To see where this circl e intersects a l ine in the pl ane of if
0
, ux + vy + w = 0,
we sol ve si mul taneousl y the equati o ns of the l ine and of the circl e. Fo r in-
stance, if v # 0, then y = -(ux + w)/v; substituting this for y in the equati o n
of the circl e x
2
+ y
2
+ dx + ey + f = 0 l eads us to a quadrati c equati o n for
the x-co o rdi nate, c, of this intersection poi nt, of the form c
2
+ s
x
c + s
2
= 0,
with s
x
and s
2
in K
0
. By the quadrati c formul a, c = ( s
x
Vsf - 4s
2
)/2,
and if the l ine and circl e intersect in the real pl ane, then s\ - 4s
2
> 0. If
s = s\ 4s
2
> 0 and if K
x
= if
()
(Vi'), then we see that the x-coordi nate, c,
l ies in i f j . If Vs G K
0
, then K
x
= K
0
; o therwi se, [K
x
: K
0
] = 2. Since the
y-co o rdi nate d = (uc + w)/v, we have that d is al so in K
x
. Thus the i nter-
section po i nt (c, d) l ies in the pl ane of K
x
where [K
x
: K
0
] = 1 o r 2. The story
is simil ar if v = 0 and u i= 0.
Final l y, to get the i ntersecti on of two circl es x
2
+ y
2
+ dx + ey + f = 0
and x
2
+ y
2
+ gx + hy + k = 0 in the pl ane of i f
0
, subtracting o ne of these
equati o ns from the o ther gives us the equati o n of the l ine in the pl ane of i f
0
,
(d - g)x + (e - h)y + (f - k) = 0. So to find the po i nts of i ntersecti on of
two circl es in the pl ane of K
0
is the same as finding the points of intersection
of a l ine in the pl ane of if
0
with a circl e in that pl ane. This is precisel y the situa-
tion we disposed of above. So if the two circl es intersect in the real pl ane, thei r
points of intersection l ie in the pl ane of an extension of i f
0
of degree 1 or 2.
To construct a constructibl e l ength, a, we start in the pl ane of <Q, the ra-
ti o nal ; the strai ght-edge gives us l ines in the pl ane of Q, and the co mpass
circl es in the pl ane of Q. By the abo ve, these i ntersect at a po i nt in the pl ane
of an extensi on of degree 1 or 2 of Q. To get to a, we go by this pro cedure
from the pl ane of Q to that of L
u
say, where [L
t
:Q] = 1 or 2, then to that
of L
2
, where [L
2
: L
x
] = 1 or 2, and co nti nue a finite number of times. We
get, this way, a finite sequence <Q = L
0
C L
x
C C L
n
of fiel ds, where each
[Lj: L,_!] = 1 o r 2 and where a is in L.
By Theo rem 5.3.1, [L : Q] = [L : L ^j j f L ^j : L _
2
] [L
x
: Q] and
since each of [L ,: L ^] = 1 o r 2, we see that [L : Q] is a power of 2. Since
a G L, we have that Q(a) is a subfiel d of L, hence by the Corollary to Theo-
rem 5.3.1, [Q(a): Q] must divide a power of 2, hence [Q(a): <Q>] = 2"' for
some nonnegative integer m. Equi val entl y, by Theo rem 5.3.5, the mi ni mal
206 Fi el ds Ch . 5
pol ynomi al for a over Q must have degree a po wer of 2. Thi s is a necessary
condition that a be constructibl e. We have pro ved the i mpo rtant criterion for
constructibil ity, namel y
Theo rem 5.5.2. I n o rder that the real number a be constructi bl e, it is
necessary that [Q(a) : <Q>] be a po wer of 2. Equi val entl y, the mi ni mal po l yno -
mial of a over 0 must have degree a po wer of 2.
To dupl icate a cube of sides 1, so of vo l ume 1, by strai ght-edge and
compass woul d requi re us to construct a cube of sides of l ength 6 who se vol -
ume woul d be 2. But the vo l ume of this cube woul d be 6
3
, so we wo ul d have
to be abl e to find a constructi bl e number b such that b
3
= 2.
Gi ven a real number b such that b
1
= 2, then its mi ni mal po l yno mi al
over Q is p(x) =x
3
- 2, for this pol ynomi al is monic and irreducibl e over Q (if
yo u want, by the Eisenstein Criterion), and p(b) = 0. Al so , as is cl ear to the
eye, p(x) is of degree 3. Since 3 is no t a po wer of 2, by Theo rem 5.5.2, there is
no such constructibl e b. Therefo re, the question of the dupl ication of the cube
by straight-edge and compass has a negative answer. We summari ze this in
Theo rem 5.5.3. I t is i mpossi bl e to dupl icate a cube of vo l ume 1 by
straight-edge and compass.
We no w have di sposed of the cl assical Questi o n 1, so we turn o ur atten-
tion to Questi o n 2, the trisection of an arbi trary angl e by strai ght-edge and
compass.
If we coul d trisect the parti cul ar angl e 60, we wo ul d be abl e to con-
struct the triangl e ABC in the di agram
where 0 = 20 and AC is of l ength 1, by straight-edge and compass. Since AB
is of l ength cos 20, we wo ul d have that b = cos 20 is a constructibl e number.
We want to show that b = cos 20 is no t a constructibl e number by pro -
ducing its minimal po l yno mi al o ver <Q, and showing that this po l yno mi al is of
degree 3. To this end we recal l the tripl e-angl e formul a from tri go no metry,
namel y that cos 3(f> = 4 co s
3
<j> 3 cos <j>. If 6 = cos 20, then, since
co s(3 20) = cos 60 = |, this tri go no metri c formul a beco mes Ab
3
- 3b = ,
and so 86
3
- 6b - 1 = 0. If c = 26, this beco mes c
3
- 3c - 1 = 0. If b is con-
structibl e, then so is c. But p(c) = 0, where p(x) = x
3
- 3x - 1, and this
c
Sec. 6
Ro o t s of Po l y n o mi al s 207
po l yno mi al is irreducibl e over . (Pro ve!) So p(x) is the minimal pol ynomi al
for c over Q. Because p(x) is of degree 3, and 3 is no t a po wer of 2, by Theo -
rem 5.5.2 we have that c is no t constructi bl e. So we cannot trisect 60 by
straight-edge and compass. This answers Questi o n 2 in the negative.
Theo rem 5.5.4. I t is impossibl e to trisect 60 by straight-edge and
co mpass.
We ho pe that this theo rem wil l di ssuade any reader from joining the
ho rdes of angl e-trisectors. There are mo re profitabl e and pl easanter ways of
wasting o ne's ti me.
There is yet ano ther cl assical pro bl em of this kind to which the answer
is "no ." This is the questi on of squari ng the circl e. Thi s question asks: Can
we construct a square who se area is that of a circl e of radius 1 by straight-
edge and co mpass? This is equi val ent to asking whether x/TT is a con-
structibl e number. If this were the case, then since TT = (VT T )
2
, the number it
woul d be constructibl e. But Li ndemann pro ved in 1882 that TT is in fact tran-
scendental , so certainl y is not al gebraic, and so cannot be constructibl e. There-
fore, the circl e of radius 1 canno t be squared by straight-edge and compass.
Of course, what we did abo ve do es no t consti tute a proof of the i mpo s-
sibil ity of squari ng the circl e, since we have presuppo sed Li ndemann's resul t
wi tho ut pro vi ng it. To pro ve that TT is transcendental woul d take us to o far
afiel d. One mi ght expect that it wo ul d be easier to pro ve that TT is not con-
structibl e than to prove that it is not al gebraic. This do es not seem to be the
case. Unti l no w al l proofs that TT is no t constructi bl e go via the route of ex-
pl oiting the transcendence of TT.
PROBL EMS
1. Co mpl ete the proof of Theo rem 5.5.1.
2. Pro ve that x
3
- 3x - 1 is irreducibl e o ver Q.
3. Show that the construction given for \/a, a&O do es i ndeed give us V a.
4. Pro ve that the regul ar heptago n (seven-sided pol ygon with sides of equal
l ength) is not constructibl e by strai ght-edge and compass.
6. ROOTS OF POL YNOMI A L S
Let F[x], as usual , be the pol ynomi al ring in x over the fiel d F and l et K be an
extension fiel d of F. If a E if and
f(x) = <x
Q
+ ctyX + + ax'\
208 Fi el ds Ch . 5
then by f(a) we understand the el ement
f(a) = a
0
+ ajfl + + aa"
in K. This is the usage we have made of this no tati o n thro ugho ut this chap-
ter. We wil l no w be i nterested in tho se a's in K such that f(a) = 0.
Definition. The el ement A if is a root of the pol ynomi al f(x) F[x]
iff (a) = 0.
I n what we have do ne up until no w we have al ways had an extensi o n
fiel d K of F given to us and we co nsi dered the el ements in K al gebrai c over
F, that is, those el ements of K that are ro o ts of no nzero po l yno mi al s in F[x].
We saw that if a K is al gebraic o ver F of degree nthat is, if the mi ni mal
po l yno mi al for a o ver F is of degree nthen [F(a): F] = n, where F(a) is
the subfiel d of K o btai ned by adjoining a to F.
What we do no w is turn the pro bl em aro und. We no l o nger wil l have
the extensi on K of F at o ur disposal . I n fact, o ur principal task wil l be to pro -
duce it al most from scratch. We start with so me po l yno mi al f(x) of positive
degree in F[x] as o ur onl y bit of data; our goal is to construct an extensi on
fiel d K of F in which f(x) wil l have a ro o t. Once we have this constructi on of
K under control , we shal l el abo rate on the general theme, thereby o btai ni ng
a series of interesting co nsequences.
Befo re setting off o n this search for the appro pri ate K, we must get
so me information abo ut the rel ati o n between the ro o ts of a given po l yno mi al
and the factorization of that pol ynomi al .
L emma 5.6.1. If a /. is a ro o t of the pol ynomi al f(x) F[x] of de-
gree n, where L is an extensi on fiel d of F, then f(x) factors in L[x] as fix) =
(x - a)q(x), where q(x) is of degree n - 1 in L[x]. Conversel y, if f(x) =
(x - a)q(x), wi th/(x), q(x), and a as abo ve, then a is a root of f(x) in L.
Proof. Since F C L, F[x] is co ntai ned in L[x]. Because a L, x - a is
in L[x]; by the Division Al go ri thm for pol ynomial s, we have f(x) =
(x a)q(x) + r(x), where q(x) and r(x) are in L[x] and where r(x) = 0 or
degr(x) < deg(x - a) = 1. This yiel ds that r(x) = b, so me el ement of L.
Substituting a for x in the rel ati on abo ve, and using the fact that/ (G) = 0, we
o btai n 0 = (a - a)q(a) + b = 0 + b = b; thus 6 = 0. Since r(x) = b = 0, we
have what we wanted, namel y/(x) = (x ~ a)q(x).
Fo r the statement that dagq(x) = n 1 we no te that since f(x) =
(x - a)q(x), then, by L emma 4.5.2, n = deg/(x) = deg(.v - a) + degq(x) =
1 + deg q(x). This gives us the requi red resul t, deg q(x) = n 1.
The converse is co mpl etel y trivial .
Sec. 6 Ro o t s o f Po l y n o mi al s 209
One i mmedi ate co nsequence of L emma 5.6.1 is
Theo rem 5.6.2. Let f(x) in F[x] have degree n; then f(x) can have at
mo st n ro o ts in any extension, if, of F.
Proof. We go by induction o n n. If n = 1, then f(x) = ax + b, where a
and b are in F and where a = 0. Thus the onl y ro o t of f(x) is ->/A, an el e-
ment of F.
Suppose that the theo rem is correct for al l pol ynomial s of degree k - 1
o ver any fiel d. Suppose that f(x) in F[x] is of degree k. If /(x) has no ro o ts in
if, then the theo rem is certainl y correct. Suppo se, then, that a E if is a ro o t of
/(x). By L emma 5.6.1, /(x) = (x - a)q(x), where g(x) is of degree k 1 in
i f [x]. Any ro o t i? in i f o f /(x) is ei ther A or is a ro o t of q(x), since 0 =f(b) =
(7J> a)q(b). By induction, o(x) has at mo st k 1 roots in if, hence /(x) has at
most k ro o ts in if. This compl etes the induction and proves the theo rem.
Actual l y, the proof yiel ds a l ittl e mo re. To expl ain this "l ittl e mo re," we
need the no ti o n of the mul tipl icity of a ro o t.
Definition. If if is an extensi on of /'. then the el ement a in K is a root
of multiplicity k > 0 of fix), where f(x) is in F[x], if f(x) = (x - a)
k
q(x) for
so me q(x) in ifjx] and x - a do es not divide q(x) (or, equival entl y, where
q(a) * 0).
The same pro o f as that given for Theo rem 5.6.2 yiel ds the sharpened
version:
Let f(x) be a polynomial of degree n in F[x]; then f(x) can have at most n
roots in any extension field K of F, counting a root of multiplicity k as k roots.
Theo rem 5.6.3. Let f(x) in F[x] be mo ni c of degree n and suppo se
that if is an extension of F in which /(x) has n ro o ts, counting a root of mul ti-
pl icity k as k ro o ts. If these ro o ts in if are a
u
a
2
, , a,, each having
mul tipl icity k
u
k
2
, ., k
m
respectivel y, then f(x) factors in if[x] as / (x) =
(x-a
l
)
k
'(x-a
2
)^---(x-a
m
)
k
-.
Proof. The proof is easy by maki ng use of L emma 5.6.1 and of induc-
tion on n. We l eave the carrying o ut of the pro o f to the reader.
Definition. We say that /(x) in F[x) splits into linear factors over (or
in) if if f(x) has the factorization in K[x] given in Theo rem 5.6.3.
There is a nice appl ication of Theo rem 5.6.3 to finite fiel ds. Let F be a
finite fiel d havi ng q el ements, and l et a,, a
2
,. . ., be the nonzero el e-
210 Fi el ds Ch . 5
ments of F. Since these form a gro up of o rder q 1 under the mul tipl ication
in F, by Theo rem 2.4.5 (pro ved ever so l ong ago), a
q
~
l
= 1 for any a = 0 in
F. Thus the po l yno mi al x''~
]
- 1 in F[x] has q - 1 distinct ro o ts in F. By
Theo rem 5.6.3, the po l yno mi al x
q
~
l
- 1 = (.r - a,)(x a
2
) (x a
q
^
l
). If
we al so consider 0, then every el ement a in F satisfies a'
1
= a, so that the
pol ynomi al x
q
x has the q el ements of F as its distinct ro o ts. By Theo rem
5.6.3 we have
Theo rem 5.6.4. Let F be a finite fiel d having q el ements. Then x
q
x
factors in F[x] as
x'i - x = x(x - a
x
)(x -,) (x - A ,
7
_
1
) >
where a
u
a
2
,. .., a
q
- \ are the no nzero el ements of F, and
x"~' - 1 = (.v - fl ,)(x - a
2
) (x - a
q
_
y
).
A very special case of this theo rem is that in which F = Z
p
, the integers
mo dul o the pri me p. Here q = p and ay,a
2
,..., a
p
-\ are just 1, 2, . . ., p - 1
in so me o rder.
Corol l ary. I n Z
p
[x], the po l yno mi al x
p
~
1
- 1 factors as
- 1 = (
X
- l )(x _
2
) ( * - (p - 1)).
Try this o ut for p = 5,7, and 11.
As a corol l ary to the corol l ary, we have a resul t in number theo ry,
kno wn as Wilson's Theorem, which we assigned as Pro bl em 18 in Section 4 of
Chapter 2.
Corol l ary. If p is a pri me, then (p 1)1 = 1 mo d p.
Proof. By the Co ro l l ary abo ve,
x"-
1
- 1 = 0 - 1)(* - 2) (x - (p - 1));
substituting x = 0 in this gives us
- 1 = ( - l ) ( - 2) - ( - ( > - 1)) = ( - l ) ' -
l
l -2 - -(p - 1)
= ( - l ) ' -
1
^ - 1)!
in Z
p
. I n the integers this transl ates i nto "co ngruent mo d p." Thus
( - l ) " "
1
^ - 1)! = - 1 mo dp,
and so (p - 1)1 = (~l)
p
mo dp. But (- 1)'' = - 1 moAp; hence we have
pro ved Wil son's Theo rem.
Sec. 6 Ro o t s of Po l y n o mi al s 211
We change direction to consider the pro bl em menti o ned at the begin-
ning of this section: gi ven/'(x) G F[x], to construct a finite extension K of F
in which /(x) has a root. As we shal l see in a mo ment, this construction of K
wil l be qui te easy when we bring the resul ts abo ut po l yno mi al rings pro ved in
Chapter 4 into pl ay. Ho wever, to verify that this constructi on wo rks wil l take
a bit of wo rk.
Theo rem 5.6.5. Let ./" be a fiel d and fix) a pol ynomi al of positive de-
gree n in F[x]. Then there exists a finite extensi o n K of F, with [K : F] < n, in
which f(x) has a root.
Proof. By Theo rem 4.5.12, f(x) is divisibl e in F[x] by so me irreducibl e
po l yno mi al p(x) in F[x]. Since p(x) divides fix), degp(x) < deg/(x) = n,
and/ (x) = p(x)q(x) for so me po l yno mi al q(x) in F[x]. If 6 is a root of p(x)
in so me extensi on fiel d, then b is auto mati cal l y a ro o t of f(x), since f(b) =
p(b)q(b) = Oq(b) = 0. So to pro ve the theo rem it is eno ugh to find an exten-
sion of F in which p (x) has a root.
Because p(x) is irreducibl e in F[x], the ideal M = (p(x)) of F[x] gener-
ated by p(x) is a maxi mal ideal of F[x] by Theo rem 4.5.11. Thus by Theo rem
4.4.2, K = F[x]/M is a fiel d. We cl aim that this is the fiel d that we are seeking.
Strictl y speaki ng, K does not contain F; as we no w show, however, K
does contain a fiel d isomorphic to F. Since every el ement in M is a mul tipl e in
F[x] of p(x), every such no nzero el ement must have degree at l east that of
p(x). Therefore, M fl F = (0). Thus the ho mo mo rphi sm i/>: F[x] >AT defined
by <A(g(*))
=
g(
x
) + M for every g(x) in F[x], when restricted to F, is 1 1 on
F. Therefore, the image F of F in K is a fiel d isomorphic to F. We can identify
F, via ifj, with F and so, in this way, we can consider K an extension of F.
Deno te x + M G K by a, so that ip(x) = a, a G K. We l eave it to the
reader to show, from the fact that ifi is a ho mo mo rphi sm of F[x] onto A" with
kernel M, that i//(g(x)) = g(a) for every g(x) in F[x]. What is i(/(p(x))7 On
the o ne hand, sincep(x) is in i Tx], ip(p(x)) = On the o ther hand, since
p(x) is in M, the kernel of izV, i]/(p(x)) = 0. Equati ng these two eval uati ons of
i/f(p(x)), we get that p(a) = 0. In other words, the element a = i//(x) in K is a
root ofp(x).
To finish the proof, al l we need is to sho w that [K : F] = deg p (x) < n.
This came up earl ier, in the al ternati ve pro o f we gave of Theo rem 5.3.5.
There we l eft this point to be pro ved by the reader. We shal l be a l ittl e mo re
genero us here and carry o ut the proof in detail .
Gi ven h(x) in F[x], then, by the Division Al go ri thm, h(x) = p(x)q(x) +
r(x) where q(x) and r(x) are in F[x], and r(x) = 0 or degr(x) < degp(x).
Go i ng mo dul o M, we obtai n that
212 Fi el ds Ch . 5
ijj(h(x)) =4>(p(x)q(x) +r(x)) =Mp(x)q(x)) + <A(r(x))
= MPtoMqix)) + H'ix))
= H'ix)) = r(a)
[since if/(p(x)) =p(a) =0].
So, since every el ement in K =F[x]/M is (/)(/? (x)) for so me /r(x) in F[x]
and <//(h(x)) =r(a), we see that every el ement of K is of the form r(a), where
r(x) is in F[x] and deg r(x) < d&gp(x). If deg/?(x) =m, the discussion just
made tel l s us that 1, a, a
2
,... , a'"
- 1
span K over F. Mo reo ver, these el ements
are l inearl y i ndependent o ver F, since a rel ation of the type a
Q
+ a
x
a +
+ a,,,- ^'"
- 1
=0 wo ul d impl y that g(a) =0 where g(x) = a
0
+ a
x
x + +
a
m
- \ X
m
~
x
is in F[x]. This puts g(x) in M, which is impossibl e since g(x) is of
l ower degree than p(x), unl ess g(x) =0. I n o ther wo rds, we get a co ntradi c-
tion unl ess a
0
=a\ = = =0. So the el ements 1, a, a
2
, . . . , a"'~
l
are
l inearl y i ndependent over F. Since they al so span K over F, they form a basis
of K over F. Co nsequentl y,
dim
F
AT =[K: F] =m =degj? (.v) <n = deg/ (x).
The theo rem is pro ved.
We carry out an i terati o n of the argument used in the l ast pro o f to
pro ve the i mpo rtant
Theo rem 5.6.6. Let f(x) G F[x] be of degree n. Then there exists an
extension K of F of degree at most nl over F such that f(x) has n ro o ts, count-
ing mul tipl icities, in K. Equival entl y, f(x) spl its into l inear factors o ver K.
Proof. We go by induction on n. If n =1, then f(x) = a + Bx, where a,
B G F and where B ^0. The onl y ro o t of f(x) is a/B, which is in F. Thus
K = F and [K : F) = 1.
Suppo se that the resul t is true for al l fiel ds for pol ynomi al s of degree k,
and suppo se that/f x) G F[x] is of degree k + 1. By Theo rem 5.6.5 there ex-
ists an extension K
x
of F with [/Cj : F] <k +1 in which /'(x) has a ro o t a^
Thus in K
x
[x], f(x) factors as/ (x) =(x - a
1
)o (x), where q(x) G K
x
[x] is of
degree k. By i nducti on there exists an extension K of K j of degree at mo st /c!
over Zfi over which q(x) spl its into l i near factors. But then/(x) spl its i nto l in-
ear factors over K. Since [K: F] =[K: KJK, : F] <(Ar +l )/c! =(k +1)!,
the induction is co mpl eted and the theo rem is pro ved.
We l eave the subject of fiel d extensi ons at this point. We are exactl y at
what might be described as the begi nni ng of Gal ois theory. Havi ng an exten-
Sec. 6 Ro o t s of Po l y n o mi al s 213
sion K of F of finite degree over which a given po l yno mi al f(x) spl its into l in-
ear factors, there exists an extension of least degree enjoying this pro perty.
Such an extensi on is cal l ed a splitting field off(x) over F. One then pro ceeds
to pro ve that such a spl itting fiel d is uni que up to i somorphi sm. Once this is
in hand the Gal oi s theo ry goes into ful l swing, studying the rel ationship be-
tween the gro up of auto mo rphi sms of this spl itting fiel d and its subfiel d struc-
ture. Eventual l y, it l eads to showing, amo ng many o ther things, that there
exist po l yno mi al s over the rati onal s of al l degrees 5 o r higher whose ro o ts
canno t be expressed nicel y in terms of the coefficients of these po l yno mi al s.
This is a brief and very sketchy descri pti on of where we can go fro m
here in fiel d theo ry. But there is no hurry. The readers shoul d assimil ate the
materi al we have presented; this wil l put them in a go o d position to l earn
Gal oi s theo ry if they are so incl ined.
PROBL EMS
2. If F is a finite fiel d havi ng the q 1 no nzero el ements a
x
, a
2
, .., a
q
-i,
pro ve that a
x
a
2
a
q
^
x
= (l )
9
.
3. Let Q be the rati onal fiel d and l et p(x) = x
4
+ x
3
+ x
2
+ x + 1. Sho w
that there is an extension K of <Q>wi th [K: Q] = 4 o ver which p (x) spl its
into l inear factors. [Hi nt: Fi nd the ro o ts ofp(x).]
4. If q(x) = x" + a
x
x
n
~
l
+ + a, a = 0, is a po l yno mi al with i nteger
coefficients and if the rati onal number r is a ro o t of q (x), pro ve that ;is
an integer and r\a.
5. Show that q(x) = x
3
- Ix + 11 is i rreduci bl e over Q.
6. If F is a fiel d of characteristic p # 0, show that (a + b)
p
= a
p
+ b
p
for al l
a and b in F.
1. Extend the resul t of Pro bl em 6 by showing that (a + b)
m
= a'" + b'",
where m = p".
8. Let F = Z
p
, p a pri me, and consi der the po l yno mi al x'" x in Z
p
[x],
where m = p". Let K be a finite extensi o n of Z
p
o ver which x'" x spl its
into l inear factors. I n K l et K
Q
be the set of al l ro o ts of x'
n
- x. Show that
Ko is a fiel d having at most p" el ements.
9. I n Pro bl em 8 show that K
0
has exactly p" el ements. (Hint: See Pro bl em
14.)
10. Co nstruct an extension fiel d K of <Q such that [K : Q] = n, for any
n > 1.
214 Fi el ds Ch . 5
11. Define the mappi ng 8 : F[x] F[x] by
S(a
a
+ ape + a
2
x
2
+ + ax")
= a
y
+ 2a
2
x + + iaiX
1
"
1
+ + na,,x"-\
Pro ve that:
(a) S(f(x) + g(x)) = 8(f(x)) + 8{g(x)).
(b) 8(f(x)g(x)) = f(x)8(g(x)) + 8(f(x))g(x) for al l / O) and g(x) in F[x}.
12. If F is of characteristic p j= 0, characteri ze al l f(x) in F[x] such that
S(f(x)) = 0.
13. Show that if f(x) in F[x] has a ro o t of mul tipl icity greater than 1 in so me
extension fiel d of F, then f(x) and 8(/(x)) are not rel ativel y pri me in
F[x].
14. If F is of characteristic p 0, show that al l the ro o ts of x'" x, where
m = p", are distinct.
15. If f(x) in F[x] is i rreduci bl e and has a ro o t of mul tipl icity greater than 1
in so me extension of F, show that:
(a) F must be of characteristic p for so me pri me p.
(b) f(x) = g(x
p
) for so me po l yno mi al g(x) in F[x].
I n this final chapter we treat several unrel ated topics. One of these co mes
from gro up theo ry, and al l the rest from the theo ry of fiel ds. I n handl ing
these special topics, we draw from many of the resul ts and ideas devel o ped
earl ier in the bo o k. Al tho ugh these topics are so mewhat special , each of
them has resul ts that are trul y i mpo rtant in thei r respective areas.
The readers who have managed to survive so far shoul d have pi cked up
a certai n set of techni ques, experi ence, and al gebraic kno w-ho w to be abl e to
fol l ow the materi al with a certai n degree of ease. We no w feel free to treat
the vari ous matters at hand in a so mewhat sketchi er fashion than we have
hereto fo re, l eaving a few mo re detail s to the reader to fil l in.
The materi al we shal l handl e do es no t l end itsel f readil y to pro bl ems, at
l east no t to pro bl ems of a reaso nabl e degree of difficul ty. Accordingl y, we
wil l assign rel ativel y few exercises. Thi s sho ul d co me as a rel ief to tho se
wanti ng to assimil ate the materi al in this chapter.
1. THE SI MPL I CI TY OF A
I n Chapter 3, where we discussed S, the symmetri c group of degree n, we
sho wed that if n > 2, then S has a no rmal subgro up A, which we cal l ed the
alternating group of degree n, whi ch is a gro up of o rder n\l2. I n fact, A was
merel y the set of al l even permutati o ns in S.
215
216 Sp ec i al To p i c s ( Op t i o n al ) Ch . 6
I n discussing A
n
, we said that A, for n 5, was a si mpl e gro up, that is,
that A has no no rmal subgro ups o ther than (e) and itsel f. We pro mi sed
there that we wo ul d pro ve this fact in Chapter 6. We no w make go o d on this
pro mi se.
To make cl ear what it is that we are abo ut to pro ve, we shoul d perhaps
repeat what we said abo ve and formal l y define what is meant by a simpl e
gro up.
Definition. A no nabel i an gro up is said to besimple if its onl y no rmal
subgro ups are (e) and itsel f.
We i mpose the pro vi so that G be no nabel i an to excl ude the trivial ex-
ampl es of cycl ic gro ups of pri me o rder from the designation "si mpl e." These
cycl ic groups of pri me o rder have no nontri vi al subgro ups at al l , so, perfo rce,
they have no pro per no rmal subgro ups. An abel ian gro up with no pro per
subgro ups is easil y seen to be cycl ic of pri me o rder.
We begin with the very easy
L emma 6.1.1. If n s- 3 andT
U
T
2
are two transpositions in S, then T
X
T
2
is ei ther a 3-cycl e or the pro duct of two 3-cycl es.
Proof. If Tj = T
2
, then T
1
T
2
= r\ = e and e is certainl y the pro duct of
two 3-cycl es, for instance as e =(123)(132).
If r
x
#T
2
, then they ei ther have o ne l etter in co mmo n or no ne. If they
have o ne l etter in co mmo n, we may suppo se, on a suitabl e renumberi ng, that
T , =(12) and T
2
=(13). But then T , T
2
=(12)(13) =(132), which is al ready a
3-cycl e.
Final l y, if T, andT
2
have no l etter in co mmo n, we may suppose, wi tho ut
l oss of general ity, that r
x
=(12) and T
2
=(34), in which case T I T
2
=(12)(34) =
(142)(143), which is i ndeed the pro duct of two 3-cycl es. The l emma is no w
pro ved.
An i mmedi ate co nsequence of L emma 6.1.1 is that for /z > 3 the
3-cycl es generate A, the al ternati ng gro up of degree n.
Theo rem 6.1.2. If a is an even permutati o n in S, where n >3, then a
is a pro duct of 3-cycl es. I n o ther wo rds, the 3-cycl es in S generate A
n
.
Proof. Let cr G S be an even permutati o n. By the definition of the par-
ity of a permutati o n, a is a pro duct of an even number of transpo si ti o ns.
Thus cr = TJTJ T
2
, ^[ T
2
, - T
2 m
~ i T
2 m
is a pro duct of 2m transposi ti ons
Ty, T
2
, . . . , T
2 M
. By L emma 6.1.1, each T O / ^T ^ is ei ther a 3-cycl e o r a pro duct
Sec . 1 Th e Si mp l i c i t y of A 217
of two 3-cycl es. So we get that a is ei ther a 3-cycl e or the pro duct of at mo st
2m 3-cycl es. This pro ves the theo rem.
We no w give an al gorithm for co mputi ng the conjugate of any permu-
tati o n in S. Let a G S, and suppo se that a(i) = /'. What does TCTT~
1
l o o k
l ike if T G 5? Suppose that T ( Z) = s and T ( / ) = f; then TO~T~ *(s) =
T O- ( T
_ 1
( . V ) ) = Tir(i) = r(/') = r. I n o ther words, to co mpute T C T T
- 1
repl ace
every symbol in a by its image under T.
Fo r i nstance, if a = (123) and T = (143), then, since T(1) = 4, T(2) = 2,
T(3) = 1, and T(4) = 3, we see that T CT T "
1
= (421) = (142).
Gi ven two /c-cycl es, say (12 k) and (i
1
i
2
' 4)>then they are conju-
gate in S because if r is a permutati o n that sends 1 into ^, 2 into i
2
, , k
i nto i
k
, then T(12 / C ) T ~
!
= (iii
2
, , h)- Since every permutati o n is the
pro duct of disjoint cycl es and conjugation is an auto mo rphi sm, we get, from
the resul t for /c-cycl es, that to co mpute T a r '
1
for any permutati o n cr, repl ace
every symbol in a by its i mage under T. I n this way we see that it is extremel y
easy to co mpute the conjugate of any permutati o n.
Gi ven two permutati o ns a
x
and o~
2
in S, then they are conjugate in S,
using the observati on abo ve, if in thei r deco mpo si ti o ns into pro ducts of dis-
j oi nt cycl es they have the same cycl e l engths and each cycl e l ength with the
same mul tipl icity. Thus, for i nstance, (12)(34)(567) and (37)(24)(568) are
conjugate in S
8
, but (12)(34)(567) and (37)(568) are not.
Recal l that by a partition of the positive i nteger n, we mean a deco mpo -
sition of n as n = n
x
+ n
2
+ + n
k
, where 0 s n
1
< n
2
S ' " < n
k
. If cr in
S is the disjoint pro duct of an j -cycl e, an n
2
-cycle,..., an 7r
/c
-cycl e, then
n
\ +
n
i + + n
k
= n, and a permutati o n Ti s conjugate to cr if and onl y if T
is the disjoint pro duct of cycl es in the same way. Therefo re, the number of
conjugacy cl asses in S is equal to the number of parti ti ons of n.
Fo r i nstance, if n = 4, then the parti ti o ns of 4 are 4 = 4, 4 = 1 + 3,
4 = 1 + 1 + 2, 4 = 1 + 1 + 1 + 1, and 4 = 2 + 2, which are five in num-
ber. Thus S
4
has five conjugacy cl asses, namel y the cl asses of (1234), (123),
(12), e, and (12)(34), respectivel y.
We summari ze everything we said abo ve in three distinct statements.
L emma 6.1.3. To find nrr ' in S,
n
repl ace every symbol in .the cycl e
structure of a by its i mage under T.
L emma 6.1.4. Two el ements in S
n
are conjugate if they have simil ar
deco mpo si ti o ns as the pro duct of disjoint cycl es.
L emma 6.1.5. The number of conjugacy cl asses in S is equal to the
number of parti ti o ns of n.
218 Sp ec i al To p i c s ( Op t i o n al ) Ch. 6
Cl earl y, from the resul ts abo ve, any two 3-cycl es in S are conjugate in
S. A 3-cycl e is an even permutati o n, so is in A. One might wo nder if any
two 3-cycl es are actual l y conj ugate in the smal l er gro up A. Fo r n "s 5 the
answer is "yes," and is qui te easy to pro ve.
L emma 6.1.6. If n ~- 5, then any two 3-cycl es in S are al ready conju-
gate mA.
Proof. Let CT\ and r/
2
be two 3-cycl es in S; by L emma 6.1.4 they are
conjugate in S. By renumberi ng, we may assume that a
x
= (123) and a
2
=
T ( 123) T
_ 1
for so me r G S. If T is even, then we are do ne. If r is o dd, then
p = T(45) is even and p(123)p"
1
= T(45)(123)(45)-
1
T~
1
= T ( 123) T
_ 1
= cr
2
.
Therefo re, <Tj and a
2
are conjugate in A. We thus see that the l emma is cor-
rect.
I n S
3
the two 3-cycl es (123) and (132) are conjugate in 5
3
but are no t
conjugate in A
3
, which is a cycl ic gro up of o rder 3.
We no w prove a resul t that is no t onl y i mpo rtant in gro up theo ry, but
al so pl ays a key rol e in fiel d theo ry and the theo ry of equati o ns.
Theo rem 6.1.7. If n > 5. then the onl y nontrivial pro per no rmal sub-
gro up of S is A.
Proof. Suppose that /V is a no rmal subgro up of S and N is nei ther (e)
no r S. Let cr #e be in N. Since the center of S is just (e) (See Pro bl em 1)
and the transpositions generate S, there is a transposi ti on r such that
err i= ra. By L emma 6.1.4, r
x
= CT T O- "
1
is a transposi ti on, so TT
X
= TCTTO-~
1
ê
is in N, since cr G N and T U T = rcrr
- 1
G N because N is no rmal in S
n
. So N
contains an el ement that is the pro duct of two transposi ti ons, namel y TT
X
.
If T and Tj have a l etter in co mmo n, then, as we saw in the pro o f of
L emma 6.1.1, TT
X
is a 3-cycl e, hence TV contai ns a 3-cycl e. By L emma 6.1.4 al l
3-cycl es in S are conjugate to TT
X
SO must fal l in N, by the no rmal i ty of N in
S. Thus the subgro up of S generated by the 3-cycl es, which, accordi ng to
Theo rem 6.1.2, is al l of A, l ies in N. No te that up to this point we have not
used that n >5.
We may thus assume that T and T
X
have no l etter in co mmo n. Wi tho ut
l oss of general ity we may assume that r = (12) and T
2
= (34); therefo re,
(12)(34) is in N. Since n > 5, (15) is in S hence (15)(12)(34)(15)^
1
=
(25)(34) is al so in JV; thus (12)(34)(25)(34) = (125) is in N. Thus in this case
al so, /Vmust contain a 3-cycl e. The argument abo ve then shows that N D A.
We have shown that in bo th cases rV must contain A. Since there are
no subgro ups strictl y between A
n
and S
n
and N = S, we o btai n the desi red
resul t that N =A.
Sec . 1 Th e Si mp l i c i t y of A
N
219
The resul t is fal se for n = 4; the subgro up
TV = {e, (12)(34), (13)(24), (14)(23)}
is a pro per no rmal subgro up of S
4
and is not A
4
.
We no w kno w al l the no rmal subgro ups of S when n a 5. Can we de-
termi ne from this al l the no rmal subgro ups of A for n > 5? The answer is
"yes"; as we shal l soon see, Ais a simpl e gro up if n a 5 , The proof we give
may strike many as strange, for it hinges on the fact that 60, the o rder of A
5
,
is no t a perfect square.
Theo rem 6.1.8, The gro up A
5
is a simpl e group of o rder 60.
Proof. Suppo se that A
5
is no t simpl e; then it has a pro per no rmal
subgro up TV whose order is as small as possible. Let the subset T =
[a E S
5
\ T V cr
- 1
C TV}, the no rmal i zer of TV in S
5
. Since TV is no rmal in A
5
, we
kno w that T D A
s
. T is a subgro up of S
s
, so if T + A
5
, we wo ul d have that
T = S
5
. But this woul d tel l us that TV is no rmal in 5
5
, which, by Theo rem
6.1.7, woul d impl y that N D A
5
, giving us that TV = A
5
, contrary to our suppo -
sition that TV is a pro per subgro up of A
5
. So we must have T = A
s
. Since (12)
is odd, it is no t in A
5
, hence is not in T. Therefo re, M = (12)7V(12)
- 1
+ TV.
Since TV < A
5
, we al so have that M <1 A
5
(Pro ve!), thus both M D TV
and MT V = [mn j m G M,n G TV} are no rmal in A
5
. (See Pro bl em 9.) Because
M + TV we have that M D TV + TV, and since TV is a mi ni mal pro per no rmal
subgro up of A
5
, it fol l ows that M f l TV = (e). On the other hand,
(12)M/v(12)
_ 1
= ( 12) M( 12) -
1
( 12) M12)
_ 1
= T V M (since (12)TV(12)"
1
= M
and ( 12) M( 12)
_ 1
= TV) = MT V by the no rmal i ty of M and TV in A
5
. There-
fore, the el ement (12) is in the no rmal i zer of MT V in S
5
, and since MN is no r-
mal in A
5
, we get, as we did abo ve, the MT V is no rmal in S
5
, and so MTV = A
5
by Theo rem 6.1.7.
Co nsi der what we no w have. Bo th M and TV are no rmal subgro ups of
A
5
, each of o rder |TV|, and MT V = A
5
and M D TV = (<?). We cl aim, and l eave
to the reader, that MN must then have o rder |TV|
2
. Since MTV = A
5
, we o b-
tain that 60 = |yl
5
1 = |MT V | = |TV |
2
. But this is sheer nonsense, since 60 is no t
the square of any integer. This establ ishes Theo rem 6.1.8.
To go from the simpl icity of A
s
to that of A for n s 5 is no t too hard.
No te that the argument we gave for A
s
did no t depend o n 5 until the punch
l ine "60 is no t the square of any i nteger." I n fact, the reaso ni ng is val id as
l ong as we kno w that n\!2 is no t a perfect square. Thus, for exampl e, if n = 6,
then 6!/2 = 360 is no t a square, hence A
6
is a simpl e group. Since we shal l
need this fact in the subsequent discussion, we reco rd it before going on.
Corollary to the Pro o f of Theo rem 6.1.8. A
6
is a simpl e gro up.
We return to the questi o n of whether or no t n\l2 is a square. As a mat-
ter of fact, it is not if n > 2. This can be shown as a co nsequence of the beau-
tiful theo rem in number theo ry (the so-cal l ed Bertrand Po stul ate), which as-
serts that for m > 1 there is al ways a pri me between m and 2m. Since we do
not have this resul t at o ur disposal , we fol l ow ano ther ro ad to show the sim-
pl icity of A for al l n s 5.
We no w pro ve this i mpo rtant theo rem.
T heorem 6.1.9. Fo r al l n > 5 the gro up A is simpl e.
Proof. By Theo rem 6.1.8 we may assume that n > 6. The center of A
for n > 3 is merel y (e). (Prove!) Since A is generated by the 3-cycl es, if
cr + e is in A. then, for so me 3-cycl e T, err TCT.
Suppo se that N # (e) is a no rmal subgro up of A and that cr ^ e is in TV.
Thus, for so me 3-cycl e r, err rcr, which is to say, crrcr
_ 1
T
_ 1
i= e. Because
N is no rmal in A, the el ement < TCT~
1
T~
[
is in N, hence O-TO-~
X
T~
1
is al so in N.
Since T is a 3-cycl e, so must crrcr
- 1
al so be a 3-cycl e. Thus N co ntai ns the
pro duct of two 3-cycl es, and this pro duct is no t e. These two 3-cycl es invol ve
at mo st six l etters, so can be consi dered as sitting in A
6
which, since n ^ 6,
can be consi dered embedded isomorphical l y in A. (Pro ve!) But then
N D A
6
j= (e) is a no rmal subgro up of A
6
, so by the Corol l ary abo ve,
N C\ A
6
= A
6
. Therefo re, Nmust contai n a 3-cycl e, and since al l 3-cycl es are
conjugate in A (L emma 6.1.6), N must contai n al l the 3-cycl es in S. Since
these 3-cycl es generate A, we o btai n that N is al l of A, thereby pro vi ng the
theo rem.
There are many different proofs of Theo rem 6.1.9they usual l y in-
vol ve showing that a no rmal subgro up of A must contai n a 3-cycl ewhich
are sho rter and possibl y easier than the o ne we gave. Ho wever, we l ike the
bi zarre twist in the pro o f given in that the who l e affair boil s do wn to the fact
that 60 is no t a square. We reco mmend to the reader to l ook at so me o ther
proofs of this very i mpo rtant theo rem, especial l y in a bo o k on gro up theo ry.
The A pro vi de us with an infinite famil y of finite simpl e gro ups. There
are several o ther infinite famil ies of finite simpl e groups and 26 parti cul ar
o nes that do not bel o ng to any infinite famil y. This determi nati o n of al l
finite simpl e groups, carried out in the 1960s and 1970s by a l arge number of
gro up theorists, is o ne of the maj or achi evements of twenti eth-century math-
ematics.
Sec . 2 Fi ni t e Fi el ds I 221
PROB L EMS
* 1. Pro ve that if n > 2, the center of S is (e).
*2 . Pro ve that if n > 3, the center of A
n
is (e).
3. What can yo u say abo ut the cycl e structure of the pro duct of two
3-cycl es?
4. If m < /?, show that there is a subgro up of S i somorphi c to S,.
5. Show that an abel ian gro up havi ng no pro per subgro ups is cycl ic of
pri me o rder.
6. Ho w many conjugacy cl asses are there in S
6
?
7. If the el ements a
u
a
2
,..., a generate the gro up G and b is a no ncentral
el ement of G, pro ve that ba-, # a
{
b for so me i.
8. If M < N and N < G, show that aMa~
1
is no rmal in TV for every a G G.
9. If M <\ G and N < G, show that MN is a no rmal subgro up of G.
10. If 77 > 5 is odd, show that the n-cycl es generate A.
11. Show that the central izer of (12 k) in 5 has o rder k(n k)\ and that
(12 k) has n\l(k(n - k)\) conjugates in S
n
.
12. I n the proof of Theo rem 6.1.8, sho w that \MN\ = \N\
2
.
2. FI NI TE FI EL DS I
Our goal in this section and the next two is to get a co mpl ete description of
al l finite fiel ds. What we shal l sho w is that the mul tipl icative gro up of
no nzero el ements of a finite fiel d is a cycl ic gro up. This we do in this section.
I n the next two, the objectives wil l be to establ ish the existence and uni que-
ness of finite fiel ds having p" el ements for any pri me p and any positive i nte-
ger 77.
So me of the things we are abo ut to do al ready came up in the pro bl em
sets in gro up theo ry and fiel d theo ry as hard pro bl ems. The techni ques that
we use co me from gro up theo ry and fiel d theory, with a l ittl e number theo ry
thro wn in.
We recal l what the Eider <p-function is. We define the Eul er ^-function
by: cp(l) = 1 and, for 77 > 1, cp(n) is the number of positive integers l ess than
77 and rel ativel y pri me to 77.
We begi n with a resul t in number theo ry who se proof, however, wil l ex-
pl oit gro up theo ry. Before doi ng the general case, we do an exampl e.
Let 77 = 12; then <p(12) = 4, for onl y 1, 5, 7, and 11 are l ess than 12 and
rel ativel y pri me to 12. We co mpute <p(d) for al l the divisors of 12. We have:
(p(l ) = 1, ip(2) = 1, (f(3) = 2, <p(4) = 2, 9( 6) = 2, and cp{\2) = 4. No te that
the sum of al l <p(d) over al l the divisors of 12 is 12. This is no fl uke but is a
special case of
Theo rem 6. 2. 1. If n >1, then 2(p(d) =77, where this sum runs o ver al l
divisors d of 77.
Proof. Let G be a cycl ic gro up of o rder 77 generated by the el ement a.
If d | n, ho w many el ements of G have o rder dl lib = a"
ld
, then al l the sol u-
tions in G of .t'' = e are the po wers e, b, b
2
,. . . , /3
r f
~' of o. Ho w many of
these have o rder dl We cl aim, and l eave to the reader, that b" has o rder d if
and onl y if r is rel ativel y pri me to d. So the number of el ements of o rder d in
G, for every divisor d of n, is (p(d). Every el ement in G has o rder so me divi-
sor of / I , so if we sum up the number of el ements of o rder dnamel y ip(d)
over al l d dividing 77, we account for each el ement of G once and onl y o nce.
Hence 2^>(d) = n if we run o ver al l the divisors d of 77. The theo rem is no w
pro ved.
I n a finite cycl ic gro up of o rder 77 the number of sol utions of x
d
= e, the
unit el ement of G, is exactl y d for every d that divides 77. We used this fact in
the proof of Theo rem 6.2.1. We no w pro ve a converse to this, getting thereby
a criterion for cycl icity of a finite gro up.
Theo rem 6. 2. 2. Let G be a finite gro up of o rder 77 with the pro perty
that for every d that divides 77 there are at most d sol utions of x
d
= em G.
Then G is a cycl ic gro up.
i
Proof. Let 41(d) be the number of el ements of G of o rder d. By hy-
pothesi s, if a 6 G is of o rder d, then al l the sol utions of x
d
- e are the dis-
tinct po wers e, a, a
2
...., a
d
~
l
, of which number, <p(d) are of o rder d. So if
there is an el ement of o rder d in G, then tzV(d) = <p(d). On the o ther hand, if
there is no el ement in G of o rder d, then (/7(d) = 0. So for al l d | n we have
that ip(d) s <p(d). Ho wever, since every el ement of G has so me o rder d that
divides TJ we have that 2i//(d) = n, where this sum runs o ver al l divisors d of
77. But
77 = 2^(d) <Xcp(d) =77
since each i/V(d) s <p(d). This gives us that Ei/f(d) = X<p(d), which, to gether
with (/'(d) < <p(d), forces <//(d) = cp(d) for every d that divides n. Thus, in
particul ar, i/>(77) = 9( 77) s 1. What does this tel l us? After all, ijj(n) is the
number of elements in G of order n, and since tjj(n) ' 1 there must be an ele-
ment 'a- in G of order n. Therefo re, the el ements e, a, a
2
,. . . , a"~
l
are al l dis-
Sec . 2 Fi ni t e Fi el ds I 223
tinct and are n in number, so they must give al l of G. Thus G is cycl ic with a
as generato r, proving the theo rem.
I s there any situation where we can be sure that the equati o n x
d
= e
has at most d sol utions in a given gro up? Certai nl y. If K* is the group of
no nzero el ements of a fiel d under mul tipl ication, then the pol ynomi al x" ~ 1
has at most n ro o ts in K* by Theo rem 5.6.2. So, if G C K* is a finite mul ti -
pl icative subgro up of K*, then the number of sol utions of x
d
= 1 in G is at
mo st d for any positive i nteger d, so certainl y for al l d that divide the o rder
of G. By Theo rem 6.2.2 G must be a cycl ic gro up. We have pro ved
T heorem 6.2.3. If K is a fiel d and A'* is the gro up of no nzero el e-
ments of K under mul tipl ication, then any finite subgro up of K* is cycl ic.
A very special case of Theo rem 6.2.3, but at the mo ment the most im-
po rtant case for us, is
T heorem 6.2.4. If AT is a finite fiel d, then K* is a cycl ic gro up.
Proof. K* is a finite subgro up of itsel f, so, by Theo rem 6.2.3, K* is
cycl ic.
A parti cul ar instance of Theo rem 6.2.4 is of great i mpo rtance in num-
ber theo ry, where it is kno wn as the existence of primitive roots mod p for p a
pri me.
T heorem 6.2.5. If p is a pri me, then Z* is a cycl ic group.
PROBL EMS
1. If A G G has o rder d, prove that a' al so has o rder d if and onl y if r and d
are rel ativel y pri me.
2. Fi nd a cycl ic generato r (primitive ro o t) for 2 . \ \ .
3. Do Pro bl em 2 for Zfv
4. Co nstruct a fiel d K having ni ne el ements and find a cycl ic generato r for
the gro up K*.
5. If p is a pri me and m = p
2
, then Z
m
is no t a fiel d but the el ements
{[A ] | ( A , p) = 1} form a gro up under the mul tipl ication in Z,. Prove that
this gro up is cycl ic of o rder p(p - 1).
6. Determi ne al l the finite subgro ups of C*, where C is the fiel d of co mpl ex
numbers.
I n the rest of the pro bl ems here cp wil l be the Eul er ^-function.
7. If p is a pri me, show that <p(p") = p"^
1
(p 1).
8. If m and n are rel ativel y pri me positive integers, pro ve that
<p(mn) = <p(rri)cp(ri).
9. Using the resul t of Pro bl ems 7 and 8, find cp(n) in terms of the factoriza-
tion of n into pri me po wer factors.
10. Pro ve that l im cp(n) = .
3. FI NI TE FI ELDS I I : EXI STENCE
Let if be a finite fiel d. Then i f must be of characteristic p, p a pri me, and K con-
tains 0, 1, 2, . . . , p - 1, the p mul tipl es of the unit el ement 1 of if. So i f D Z
; ;
,
or, mo re precisel y, i f contains a fiel d isomorphic to Z
p
. Since i f is a vector
space over Z
p
and cl earl y is of finite dimension over Z
p
, if [if: Z
p
] = n, then
i f hasp" el ements. To see this, l et v
h
v
2
, , v be a basis of if over Z
p
. Then
for every distinct choice of ( a
b
a
2
, . . . , a
n
), where the a,- are in Z
p
, the el ements
a
L
Vi + a
2
v
2
+ + av
n
are distinct. Thus, since we can pi ck ( a
x
, a
2
, . . . , a) in p" ways, i f has p"
el ements.
The mul tipl icative gro up K* of no nzero el ements of i f is a gro up of
o rder p" 1. So, we have that a'"
- 1
= 1, where w = p", for every a in if,
hence a"
1
= a. Since this is al so obviousl y true for a = 0, we have that a'" = a
for every a in K. Therefo re, the po l yno mi al x
m
- x in Z
p
[x] has m = /?" dis-
tinct ro o ts in if, namel y al l the el ements of K. Thus x'" - x factors in K[x] as
x'" - x = (x - a
L
)(x - a
2
) (x - a,),
where a,, a
2
, . . . , a, are the el ements of if.
Everythi ng we just said we al ready said, in mo re or l ess the same way,
in Section 6 of Chapter 5. Since we wanted these resul ts to be fresh in the
reader's mi nd, we repeated this materi al here.
We summari ze what we just did in
Theo rem 6.3.1. Let i f be a finite fiel d of characteristic p, p a pri me.
Then i f contains m = p" el ements where n = [i f: Z
p
], and the po l yno mi al
x'" - x in Z
p
[x] spl its into l inear factors in K[x] as
Sec, 3 Finite Fields II: Existence 225
x'" - x = (x -
ai
)(x - a
2
) (x - a,),
where a
x
, a
2
,..., a
m
are the el ements of K.
Two natural questi ons present themsel ves:
1. Fo r what pri mes p and what i ntegers n do es there exist a fiel d havi ngp"
el ements?
2. Ho w many no ni so mo rphi c fiel ds are there having p" el ements?
We shal l answer bo th questi ons in this section and the next. The an-
swers wil l be
1. Fo r any pri me p and any positive i nteger n there exists a finite fiel d
having p" el ements.
2. Two finite fiel ds having the same number of el ements are i somorphi c.
I t is to these two resul ts that we no w address oursel ves. First, we settl e
the questi on of the existence of finite fiel ds. We begin with a general remark
abo ut i rreduci bl e pol ynomial s.
L emma 6.3.2. Let F be any fiel d and suppo se that p(x) is an i rre-
ducibl e po l yno mi al in F[x]. Suppo se that q(x) in F[x] is such that in so me ex-
tensi on fiel d of F, p (x) and q (x) have a co mmo n root. Then p (x) divides q (x)
in F[x].
Proof. Suppo se that p(x) do es no t divide q(x); since p(x) is irreducibl e
in F[x], p(x) and q(x) must therefo re be rel ativel y pri me in F[x]. Thus there
are po l yno mi al s u (x) and v (x) in F[x] such that
u(x)p(x) + v(x)q(x) = 1.
Suppo se that the el ement a in so me extensi o n K of F is a ro o t of bo th p (x)
and q(x); thus p(a) = q(a) = 0. But then 1 = u(a)p(a) + v(a)q(a) = 0, a
contradi cti on. So we get that p(x) divides q(x) in F[x].
No te that we can actual l y pro ve a l ittl e mo re, namel y
Corol l ary. If /(x) and g(x) in F(x) are no t rel ativel y pri me in K[x],
where K is an extension of F, then they are no t rel ativel y pri me in F[x].
Let F be a fiel d of characteristic p + 0. We cl aim that the pol ynomi al
fix) = x'" - x, where m = p", cannot have a mul ti pl e ro o t in any extension
fiel d K of F. Do you remember what a mul tipl e ro o t of a po l yno mi al is? We
refresh yo ur memo ry. If g(x) is in F[x] and if K is an extension fiel d of F, then
a in K is a multiple root of g(x) if g(x) = (x - afq(x) for so me q(x) in AT[.v].
We return to the po l yno mi al f(x) = x'" - x abo ve. Since f(x) =
x(x'"~
[
- 1) and 0 is not a root of x' "
- 1
1, it is cl earl y true that 0 is a simpl e
(i.e., no t mul tipl e) ro o t o f /(x). Suppo se that a G K, K D F, is a ro o t o f /(x);
thus a'" = a. If y = x a, then
fiy) = )'"' ~ y = (
x
~
a
)'" ~ (
x
- a) = x'
n
- a'" - (x - a)
(since we are in characteri sti cp 0 and m = p")
= x'" - x (because a'" = a) = fix).
So
fix) = f(y) = y"
1
- y = (x- )"' - (x - a)
= (x - a)((x - a)'""
1
- 1),
and cl earl y this is divisibl e by x - a onl y to the first po wer, since x - a do es
no t divide (x - a)"
1
'
1
- 1. So a is no t a mul tipl e ro o t of f(x).
We have pro ved
Theo rem 6.3.3. If n > 0, then/ (x) = x'" x, where m = p", has no
mul ti pl e roots in any fiel d of characteristic p.
We shoul d add a wo rd to the proof abo ve to nail do wn the statement of
Theo rem 6.3.3 as we gave it. Any fiel d of characteristic p =#=0 is an extensi o n
of Z
p
, and the pol ynomi al f(x) is in Z
p
[x]. So the argument abo ve, with K
any fiel d of characteristic p and F = Z
p
, pro ves the theo rem in its given form.
We have exactl y what we need to pro ve the i mpo rtant
Theo rem 6.3.4. Fo r any pri me p and any positive i nteger n there ex-
ists a finite fiel d having p" el ements.
Proof. Consi der the po l yno mi al x'" - x in Z
p
[x], where m = p". By
Theo rem 5.6.6 there exists a finite extensi on K of Z
p
such that in K[x] the
po l yno mi al x'" x factors as
x'" - x = (x - fl j)(x - a
2
) (x - a,),
where a
lt
a
2
,.. ., a
m
are in K. By Theo rem 6.3.3, x'" - x do es no t have any
mul ti pl e roots in K, hence the el ements a
u
a
2
,..., a
m
are m = p" distinct el -
ements. We al so kno w that a
, a
2
, , a, are all the roots of x'" - x in K,
since x'" x is of degree m.
Sec. 4 Finite Fiel ds III: Uni queness 227
Let A = {a G K ] a"
1
= a}; as we j ust saw, A has m distinct el ements. We
cl aim that A is a fiel d. If a, b G A, then a"' = a and >'" = b, hence (a/?)'" =
fl "'/3"' = ab, thus A/> G A. Because we are in characteristic p # 0 and m = /?",
( A + b)'" = fl '" +fo'" = a + b, hence a + 6 is in A.
Since A is a finite subset of a fiel d and is cl osed with respect to sum and
pro duct, A must be a subfiel d of K. Since A has m = p" el ements, A is thus
the fiel d who se existence was asserted in the statement of the theo rem. Wi th
this the theo rem is proved.
PROBL EMS
* 1 . Gi ve the detail s of the pro o f of the Co ro l l ary to L emma 6.3.2.
The next two pro bl ems are a repeat of o nes given earl ier in the bo o k.
2. If f(x) = a
0
x" + fl ,*""
1
+ + a is in F[x], l et f'(x) be the formal
derivative of fix) defined by the fol l owing equati o n: f'(x) = a
0
x"
_ 1
+ in - l ^x "
- 2
+ + ( - i)a
i
x"~'~
i
+ + a_
y
. Pro ve that:
(a) ifix) + gix))' =f\x)+ g'{x)
(b) (/(x)g(x))' = f\x)g(x) + fix)g'ix) for al l / 0: ) and g(x) in F[x].
*3. Pro ve that f(x) in F[x] has a mul ti pl e ro o t in so me extension of F if and
onl y if fix) and/ '(x) are no t rel ativel y pri me.
4. If fix) = x" x is in F[x], pro ve that/(x) do es no t have a mul tipl e ro o t in
any extensi on of F if F is ei ther of characteristic 0 or of characteristic
p # 0, where p does not divide 1.
5. Use the resul t of Pro bl em 4 to give ano ther pro o f of Theo rem 6.3.3.
6. If F is a fiel d of characteristic p i= 0, construct a pol ynomi al with mul tipl e
ro o ts of the formx" - x, where p \ {n 1).
7. If K is a fiel d having p" el ements, show that for every m that divides n
there is a subfiel d of K havi ngp' " el ements.
4. FI NI TE FI EL DS I I I : UNI QUENESS
No w that we kno w that finite fiel ds exist havi ng p" el ements, for any pri me p
and any positive integer n, we might ask: Ho w many finite fiel ds are there
with p" el ements? Fo r this to make any sense at al l , what we are real l y asking
is: Ho w many distinct no ni so mo rphi c fiel ds are there with p" el ements? The
answer to this is short and sweet: o ne. We shal l show here that any two finite
fiel ds having the same number of el ements are i somorphi c.
L et K and L be two finite fiel ds having p" el ements. Thus K and L are
bo th vector spaces of di mensi o n n over Z
p
. As such, K and L are i so mo rphi c
as vector spaces. On the o ther hand, K* and L * are bo th cycl ic gro ups of
o rder p" 1 by Theo rem 6.2.4; hence K* and L* are i so mo rphi c as multi-
plicative groups. One wo ul d i magi ne that o ne coul d put these two i somor-
phisms to gether to pro ve that K and L are i somorphi c as fields. But it just
isn't so. The pro o f does not take this di recti on at al l . But the finiteness of K
and L to gether with these two i so mo rphi sms (of two structures carri ed by K
and L) do suggest that, perhaps, K and L are i so mo rphi c as fiel ds. This is in-
deed the case, as we no w pro ceed to show.
We begin with
L emma 6.4.1. If q(x) in Z
p
[x] is irreducibl e of degree n, then
q(x) | (x
m
- x), where m = p".
Proof. By Theo rem 4.5.11 the ideal (q(x)) of Z
p
[x] generated by q(x) is
a maxi mal ideal of Z
p
[x] since q(x) is irreducibl e in Z
p
[x]. Let A =
Z
p
[x)l(q(x)); by Theo rem 4.4.3, A is a fiel d of degree n over Z
p
, hence hasp"
el ements. Therefo re, u'" = u for every el ement u in A.
Let a = x + (q(x)) be the coset of x in A = Z
p
[x\l(q(x))\ thus q{a) = 0
and q(x) is the minimal pol ynomi al for a over Z
p
. Since a is in A, a
1
" = a, so a
is seen as a root of the pol ynomial x'" - x, where m = p". Thus x
m
- x and
q(x) have a co mmo n root in A. By Lemma 6.3.2 we have that q(x) \ (x'" - x).
We are no w in a position to pro ve the main resul t of this section.
Theo rem 6.4.2. ff K and L are finite fiel ds having the same number of
el ements, then K and L are i so mo rphi c fiel ds.
Proof. Suppo se that K and L have p" el ements. By Theo rem 6.2.4, L*
is a cycl ic gro up generated, say, by the el ement b in L. Then, certainl y,
Z
p
(b)the fiel d o btai ned by adjoining b to Z
; J
i s al l of L. Since
[L : Z
P
] = n, by Theo rem 5.3.2 b is al gebraic over Z
p
of degree n, with
n = deg(g(x)), where q(x) is the mi ni mal po l yno mi al in Z
p
[x] for b, and is ir-
reducibl e in Z
p
[x].
The mappi ng i/>: Z
p
[x] - L = Z
p
(/b) defined by i/>(/(x)) = /(&) is a
ho mo mo rphi sm of Z
p
[x] o nto L with kernel (o (x)), the ideal of Z
p
[x] gener-
ated by q(x). So L = Z
p
[x]/(?(x)).
Because o (x) is irreducibl e in Z
p
[x] of degree , by L emma 6.4.1 q(x)
must divide x'" - x, where m =/J ". Ho wever, by L emma 6.3.1, the po l yno -
mial x'" x factors in K[x) as
x
m
- x = (x - a
x
)(x - , ) (x - ,),
Sec . 5 Cy c l o t o mi c Po l y n o mi al s 229
where a
x
, a
2
, , a, are al l the el ements of K. Therefo re, q(x) divides
(x - a
x
)(x a
2
) (x - a,). By the Co ro l l ary to Theo rem 4.5.10, q(x) can-
no t be rel ativel y pri me to al l the .v a
t
in K[x], hence for so me 7, q(x) and
x - cij have a co mmo n factor of degree at l east 1. I n short, x - a, must
divide q(x) in K[x], so q(x) = (x - cij)h(x) for so me h(x) in K[x). There-
fore, q(cij) = 0.
Since q(x) is irreducibl e in Z
p
[x] and a
y
is a root of q(x), q(x) must be
the mi ni mal po l yno mi al for aj in Z
p
[x]. Thus Z
p
(aj) Z
p
[x]/(q(x)) = L. Thi s
tel l s us, amo ng o ther things, that we have [Z
p
(aj) :Z
p
] = n, and since
Zp(fl y) C K and [K:Z
p
] = n we concl ude that Z
p
(aj) = if. Therefo re,
K = Zp(fl y) L. Thus we get the resul t that we are after, namel y, that K and
L are i so mo rphi c fiel ds. This pro ves the theo rem.
Co mbi ni ng Theo rems 6.3.4 and 6.4.2. we have
Theo rem 6.4.3. Fo r any pri me p and any positive integer n there ex-
ists, up to i somorphi sm, o ne and onl y o ne fiel d having p" el ements.
5. CYCL OTOMI C POL YNOMI A L S
Let C be the fiel d of compl ex numbers. As a co nsequence of De Mo i vre's
Theo rem the compl ex number B = cos 27r/ + i sin Irrln satisfies 6" = 1
and 6'" + 1 if 0 < m < n. We cal l ed 0 a primitive nth root of unity. The o ther
primitive nth ro o ts of unity are
where (/c, n) = 1 and 1 < /<: < n.
Cl earl y, d
n
satisfies the po l yno mi al x" - 1 in Q[x], where Q is the fiel d
of rati o nal numbers. We want to find the mi ni mal (monic) pol ynomi al for 6
over Q.
We define a sequence of pol ynomi al s inductivel y. At first gl ance they
might no t seem rel evant to the questi o n of finding the minimal pol ynomial
for 9 o ver Q. It wil l turn out that these po l yno mi al s are highl y rel evant to
that questi o n for, as we shal l pro ve l ater, the po l yno mi al <j>(x) that we are
abo ut to i ntro duce is a mo ni c po l yno mi al wi th i nteger coefficients, is irre-
ducibl e o ver Q, and, mo reo ver, c6(#) = 0. This wil l tel l us that (p(
x
) i
s
*
n e
desired monic minimal pol ynomial for 8
n
over Q.
We no w go abo ut the business of defining these pol ynomial s.
230 Sp ec i al To p i c s ( Op t i o n al ) Ch. 6
Definition. The po l yno mi al s <p(x) are defined inductivel y by:
(a) cp^x) = x - 1.
(b) If n > 1, then <p,,(x) = (x" - l )/TIcfc
rf
(x), where in the pro duct in the
deno mi nato r d runs over al l the divisors of n except for n itsel f.
These pol ynomi al s are cal l ed the cyclotomic polynomials and c6(x) is
cal l ed the nth cyclotomic polynomial.
At the mo ment it is no t o bvi o us that the <p(x) so defined are even
pol ynomial s, no r do we, as yet, have a cl ue as to the nature of the coefficients
of these pol ynomi al s 4>
n
(x). Al l this wil l co me in due ti me. But first we want
to l ook at so me earl y exampl es.
Exampl es
1. 
2
(x)) =
(x
4
- l )/(x
2
- 1) =x
2
+1.
4.
9 5
(x) = (x
5
-l )/cfe
I
(x) = ( x
5
- l ) / ( x
x
6
1
5
-
U x ) =
<t>
x
(x)tex)<h(x)
=
x
6
- 1
(x - l )(x + l )(x
2
+ X + 1)
X
3
+ 1
2
, .
= , . = x
z
- X + 1.
X + 1
- 1) = X + 1.
- 1) = X
2
+ X + 1.
(x
4
- l )/((x - l )(x + 1)) =
- 1) =x
4
+x
3
+x
2
+ X + 1.
We noti ce a few things abo ut the po l yno mi al s above:
1. They are al l mo ni c po l yno mi al s with integer coefficients.
2. The degree of <f>(x) is <p(n), where <p is the Eul er ^-functi o n, for
1 < n < 6. (Check this out.)
3. Each cfe(x), for 1 < n ^ 6, is i rreduci bl e in Q(x). (Verify!)
4. Fo r 1 < n < 6, 6 is a ro o t of <p(x). (Verify!)
These few cases give us a hi nt as to what the general story mi ght be for
al l <f>(x). A hint, yes, but onl y a hint. To establ ish these desi red pro perti es
for c6 (x) wil l take so me wo rk.
To gain so me further insight i nto these pol ynomi al s, we consi der a par-
ticul ar case, o ne in which n p " \ where p is a pri me. To avoid cumberso me
subscripts, we shal l deno te </>(x) by i/>
(m)
(x), where n = p ' " . The pri me p wil l
be kept fixed in the discussion. We shal l o btai n expl icit formul as for the
tl/('"\x)'s and determi ne thei r basic pro perti es. Ho wever, the metho d we use
wil l no t be appl icabl e to the general case of <p
n
(x)- TO study the general sit-
uati o n wil l requi re a wider and deeper set of techni ques than tho se needed
for 4i
(l
"\x).
We no te o ne simpl e exampl e. If p is a pri me, the onl y divisor of p that
is no t p itsel f is 1. Fro m the definition of 4>
p
(x) = i//
(1)
(x) we have that
^\x) = <f
p
(x) = = x"~
l
+ + X + 1.
No te that in studying the Ei senstei n Cri teri o n we sho wed that this po l yno -
mial is irreducibl e in Q(x).
What can we say for the hi gher i/>
(m)
(x)'s?
L emma 6.5.1. Fo r al l m > 1,
<l>
(
"'\x) =
x
^"-~-i =
1
+
xP
"'~'
+ x2
"""
i
+ + j ^-
1
* " ' "
1
.
Proof. We go by induction on m.
If m = 1, we sho wed above that <//
(1)
(x) = (x
p
- l )/(x - 1) = 1 + x +
x
1
+ + x
p
~\ so the l emma is true in this case.
Suppo se that >p
(r)
= (x'"' - l )/(x
p r
"
L
- 1) for al l r < m. Co nsi der
Since the onl y pro per divisors of p ' " are 1, p , p
2
, . .., p " ' ~ \ from the
definition of i//"
0
(x) we have that
X
P" ' _ i
^'"
}
^
=
(x - 1)I A
0 )
(X) ^ " ' -^ (j t )"
By i nducti on, ip
(,)
(x) = (x>'
r
- l ) / ( x
p r _ 1
- 1) for r < m, hence
(x - l )i //"(x) ' ' ' <!>
{
'"~
l
\x)
yl> 1 yP
2
1 y P ' " '
1
_ 1
= (r - U- -~ - = xP'"-
1
- i
{ X i j
x - 1 x" - 1 x>""~
2
- 1
But then
x
p
"' - 1
!//<"'(X) =
compl eti ng the i nducti on and provi ng the l emma.
No te here that
^'\x) =
x P
m
_ r \ = 1 + x
p
'"~* + + x
(
"-
r)
'""~
l
232 Special Topics (Optional )
Ch. 6
is a monic pol ynomial with integer coefficients. Its degree is cl earl y p'"
l
(p 1),
which is i ndeed <p(p'"). Final l y, if 9 is a primitive p"'th ro o t of unity, then
0
pl
" = 1, but d
p
"'~
l
+ 1, hence i\i
(m)
(0) = 0; so 0 is a root of <//
f f l )
(x). The final
thing we want to kno w is: Is <//'"'(x) irreducibl e over 0 ?
ip
im
\x) = 1 + x
p
"'~' + + xO-
1
^' "
- 1
= ^(x
1
""'
1
)
and we kno w that i/y
(1)
(x) is i rreduci bl e in <G>[x]. We shal l use the Ei senstei n
Cri teri on to pro ve that ip
(
"'\x) is irreducibl e in <Q[x].
We digress for a mo ment. If f(x) and g(x) are two po l yno mi al s with in-
teger coefficients, we defi ne/(x) = g(x) mo d p if /(x) =g(x) +pr(x), where
r(x) is a po l yno mi al with i nteger coefficients. This is equi val ent to saying that
the co rrespo ndi ng coefficients of /(x) and g(x) are co ngruent mo d p. Ex-
pandi ng (f(x) + g(x))
p
by the bi no mi al theo rem, and using that al l the bi no -
mial coefficients are divisibl e by p, since p is a pri me, we arrive at
(f{x) + g(x)Y - f(x)" + g(x)" mod p.
Gi ven/(x) = a
0
x" + ciix"~\ + + , where the ,are i ntegers, then,
by the abo ve,
f(x)
p
= (
0
x" + a.x"-
1
++ ) " = fl x"
p
+ of x
<
"-
1
>" + + a
p
= G
0
X "
p
+ fl !^""
1
^ + + a mo d p,
the l atter co ngruence being a co nsequence of Fermat's Theo rem (the Co ro l -
l ary to Theo rem 2.4.8). Since f(x
p
) = a
0
x"
p
+ a1
t
"-V>' + + a
n
, we o bai n
that
for al l no nnegati ve k.
We return to o ur ip
(m)
(x). Since i //
m)
(x) = i}/
{l)
(x
p
"'~
l
) we have, from the
discussion abo ve, that i//"
!)
(x) iiV
(1)
(x
p
'"~') mo d p. Therefo re,
No te that
f(x
p
)^f(xY modp.
I terati ng what we just did, we arrive at
f(x
pk
) = f(x)
pk
modp
2
x + p
)
.p'
= <jj
(1)
(x)
p
'"~
l(p
-
l)
mo d p = ijj
(
'"\x + 1) mo d p.
This tel l s us that
i/>
(
"0 + 1) =
x
p
'"'~
l(p
-
l)
+ pr{x),
where r(x) is a po l yno mi al with i nteger coefficients. So al l the coefficients of
i //
m)
(x + 1), wi th the excepti on of its l eadi ng coefficient 1, are divisibl e by p.
If we knew for so me reaso n that the co nstant term of h(x) = (//
m)
(x + 1) was
no t divisibl e by p
2
, we coul d appl y the Ei senstei n Cri teri o n to show that h(x)
is i rreduci bl e. But what is the co nstant term of h(x) = ip
(m)
(x + 1)? I t is
merel y n(0) = (//'"'(l ), which, from the expl icit form of i//"(x + 1) that we
fo und four paragraphs earl ier, is exactl y p. Thus h(x) is irreducibl e in Q[x],
that is, (//'"'(x + 1) is irreducibl e in Q[x]. But this i mmedi atel y impl ies that
t//
m)
0') is i rreduci bl e in Q[x).
Summari zi ng, we have pro ved
Theo rem 6.5.2. Fo r n = p"\ where p is any pri me and m any no nneg-
ative integer, the pol ynomi al cp
n
(x) is i rreduci bl e in Q[x].
As we po i nted o ut earl ier, this is a very special case of the theo rem we
shal l so o n pro ve; namel y, that <p(x) is i rreduci bl e for al l positive integers n.
Mo reo ver, the resul t and pro o f of Theo rem 6.5.2 pl ay no rol e in the proof of
the general pro po si ti o n that <f>(x) is i rreduci bl e in Q[x]. But because of the
resul t in Theo rem 6.5.2 and the expl icit form of <p(x) when n = p'", we do
get a pretty go o d idea of what shoul d be true in general . We no w pro ceed to
the discussion of the irreducibil ity of cp(x) for general n.
Theo rem 6.5.3. Fo r every integer h > 1,
<p{x) = (x - 6
m
)- - fx - 0<*<"),
where 6
(i>
, 9
(2)
,... , are the ip(n) distinct primitive nth roots of unity.
Proof. We pro ceed by i nducti o n o n n.
If n = 1, then cpi(x) = x - 1, and since 1 is the onl y first ro o t of unity,
the resul t is certainl y correct in this case.
Suppo se that resul t is true for al l m < n. Thus, if d | n and d + n, then,
by the i nducti on, 0
d
(x) = (x - 0
(l
/) (x - 6
(
f
d)]
), where the fl âre the
primitive t/th ro o ts of unity. No w
x" - 1 = 0 - Q { x - & ) (x - Q ,
where the , run over all nth ro o ts of unity. Separati ng o ut the primitive nth
roots of unity in this pro duct, we o btai n
x" - 1 =( J E - 6
(1)
) o - ^""M*).
where u(x) is the pro duct of al l the o ther x thus by o ur i nducti o n hy-
pothesis v(x) is the pro duct of the 4>
c
i{x) over al l divisors d of n with the ex-
ception of d = n. Thus, since
(x" - 1)
=
(x - ( * - fl <*")i;fr)
0 n W
u(x) u(x)
= (x - e
m
)(x - 6
(2)
) (x - 0
M
"),
we have pro ved the resul t cl ai med in the theo rem.
Fro m the form of 4>
n
(x) in Theo rem 6.5.3 we i mmedi atel y see that
4>
n
{x) is a mo ni c po l yno mi al in C[x] of degree cp(n). Kno wi ng this, we pro ve
that, in fact, the coefficients of 4>(x) are integers. Why is this true? Pro ceed-
ing by induction o n n, we may assume this to be the case if d \ n and d = n.
Therefo re, if v(x) deno tes the po l yno mi al used in the pro o f of Theo rem
6.5.3, then (x" - l)/v(x) = <f>,,(x) G C[x], hence v(x) \x" - 1 in C[x]. But, by
the l ong-division process, dividing the mo ni c po l yno mi al v(x) with i nteger
coefficients into x" 1 l eads us to a mo ni c po l yno mi al with i nteger coeffi-
cients (and no remai nder, since rj(x) | (x" - 1) in C[x]). Thus (x" - l)lv(x) =
(j>(x) is a mo ni c po l yno mi al with i nteger coefficients. As we saw, its degree is
cp(n). Thus
Theo rem 6.5.4 Fo r every positive integer n the po l yno mi al (f>(
x
) is a
moni c pol ynomi al with i nteger coefficients of degree cp(n), where cp is the
Eul er <p-function.
Kno wi ng that (fi
n
(x) is a pol ynomi al , we can see that its degree is cp(n)
in yet ano ther way. Fro m cp(x) = (x" - 1)1 v(x), using i nducti o n on n,
deg(0(x)) = n - deg(u(x)) = n 1cp{d), the sum over al l divisors d of n
o ther than d = , from the form of u(x). I nvo ki ng the resul t of Theo rem
6.2.1, n - 1cp(d) = cp(n), where again this sum is over al l d \ n, d = n. We
thus obtai n that deg(</>(*))
=
<P('
7
)-
The resul t we are abo ut to pro ve is wi thout questi on o ne of the mo st
basic ones abo ut cycl otomic pol ynomi al s.
Theo rem 6.5.5. Fo r every positive integer n the po l yno mi al </> (x) is
i rreduci bl e in Q[x].
Proof. Let f(x) in Q[x] be an i rreduci bl e po l yno mi al such that
/(x) | <p(x). Thus 4>(x) = f(x)g(x) for so me g(x) in Q[x]. By Gauss' L emma
we may assume that bo th /(x) and g(x) are mo ni c po l yno mi al s with i nteger
coefficients, thus are in Zj x]. Our objective is to show that <p(x) = f(x); if
this were the case, then, since /(x) is i rreduci bl e in Q[x], we woul d have that
</>(x) is irreducibl e in Q[x].
Since c/>(x) has no mul tipl e ro o ts, f(x) and g(x) must be rel ativel y
pri me. Let p be a pri me number such that p do es no t divide n. If 6 is a ro o t
of fix), it is then a ro o t of <j>
n
(x), hence by Theo rem 6.5.3 9 is a primitive nth
ro o t of unity. Because p is rel ativel y pri me to n, 9
P
is al so a primitive nth
root of unity, thus, by Theo rem 6.5.3, 9
P
is a ro o t of <p
n
{x). We therefo re
have that 0 = <p
n
(6
p
) =f(O
p
)g(d
p
), from which we deduce that either f(8
p
) =
0 or g(9
p
) = 0.
Our aim is to show that f(9
p
) = 0. Suppo se no t; then g(9
p
) = 0, hence
9 is a ro o t of g(x
p
). Because 9 is al so a ro o t of the irreducibl e pol ynomi al
f(x), by L emma 6.3.2 we o btai n that/ ( A ) | g{x
p
). As we saw in the course of
the proof of Theo rem 6.5.2, g(x
p
) = g(x)
p
mo d p.
L et / be the ideal in Z generated by p; by the Corol l ary to Theo rem
4.6.2, Z[x]/7[x] Z
p
[x], which means that reduci ng the coefficients of any
po l yno mi al mo d p is a ho mo mo rphi sm of Z[x] o nto Z
p
[x].
Since al l the pol ynomi al s <p(x), v(x), fix), and g(x) are in Z[x], if
4>(x),v(x),f(x),andg(x) are thei r i mages in Z
p
[x], al l the rel ations amo ng
them are preserved going mo d p. Thus we have the rel ations x" - 1 =
(x)rj(x), (x)_ = f(x)g (x) and f(x) \ g(x
p
) = g(x)" .
Therefo re, f(x) and g(x) have a co mmo n root, A , in so me extension K
of Z
P
. No w x" 1 = (p (x)tT(x) =/ ( x) g( x) , hence a, as a root of both f(x)
and g(x), is a mul ti pl e ro o t of x" - 1. But the formal derivative (x" - 1)' of
x" - 1 is nx"
_ 1
0, since p does no t divide n; therefo re, (x" 1)' is rel a-
tivel y pri me to x" - 1. By the resul t of Pro bl em 3 of Section 3 the po l yno -
mi al x" 1 cannot have a mul ti pl e ro o t. Wi th this contradi cti on arrived at
from the assumpti on that 8
P
was no t a ro o t of f(x), we concl ude that when-
ever 9 is a ro o t of fix), then so must 9
P
be o ne, for any pri me p that does no t
divide n.
Repeati ng this argument, we arrive at: 9' is a ro o t of /(x) for every in-
teger ;that is rel ativel y pri me to n. But 9, as a ro o t of fix), is a ro o t of <p
n
(x),
so is a pri mi ti ve nth ro o t of unity. Thus 9
r
is al so a primitive nth root of unity
for every r rel ativel y pri me to n. By runni ng over al l r that are rel ativel y
pri me to n, we pick up every pri mi ti ve nth ro o t of unity as so me such 9
r
.
Thus all the primitive nth roots of unity are roots of f{x). By Theo rem 6.5.3
we see that c6(x) =fix), hence c/J(x) is i rreduci bl e in Q[x].
I t may strike the reader as artificial and unnatural to have resorted to
the passage mo d p to carry out the pro o f of the irreducibil ity of a pol ynomial
with rati o nal coefficients. I n fact, it may very wel l be artificial and unnatural .
As far as we kno w, no proof of the irreducibil ity of c6 (x) has ever been given
staying compl etel y in Q[x] and no t going mo d p. I t woul d be esthetical l y sat-
isfying to have such a proof. On the o ther hand, this is no t the onl y i nstance
where a resul t is pro ved by passing to a rel ated subsidiary system. Many the-
o rems in number theo ryabo ut the o rdi nary i ntegershave proofs that ex-
pl oit the integers mo d p.
Because 4>(x) is a mo ni c po l yno mi al with i nteger coefficients which is
irreducibl e in Q[x], and since 8, the primitive nth root of unity, is a ro o t of
<p(x), we have
Theo rem 6.5.6. ep(x) is the mi ni mal po l yno mi al in Q[x] for the pri mi -
tive nth ro o ts of unity.
P R O B L E M S
1. Verify that the first six cycl otomic pol ynomi al s are irreducibl e in Q[x] by a
direct frontal attack.
2. Wri te do wn the expl icit forms of:
(a) 1 is an integer and (fl '" - 1) | (a" - 1), pro ve that m \ n.
5. If K is a finite extensi on of Q>, the fiel d of rati onal numbers, pro ve that
there is onl y a finite number of ro o ts of unity in K. (Hi nt: Use the resul t of
Pro bl em 10 of Section 2, to gether with Theo rem 6.5.6.)
6. L I OUVI L L E' S CRI TERI ON
Recal l that a compl ex number is said to be algebraic of degree n if it is the
ro o t of a pol ynomi al of degree n o ver Q, the fiel d of rational numbers, and is
no t the root of any such po l yno mi al of degree l ower than n. I n the terms used
in Chapter 5, an al gebraic number is a compl ex number al gebraic over Q.
A compl ex number that is no t al gebraic is cal l ed transcendental. So me
famil iar numbers, such as e, TT, e"*, and many o thers, are kno wn to be tran-
scendental . Others, equal l y famil iar, such as e + TT, eir, and TT
1
', are suspected
of bei ng transcendental but, to date, this aspect of thei r nature is stil l o pen.
The French mathemati ci an J o seph Liouvil l e (1809-1882) gave a crite-
rion that any al gebraic number of degree n must satisfy. Thi s criterion gives
us a condition that l imits the extent to which a real al gebraic number of de-
gree n
(
.can be appro xi mated by rati o nal numbers. This criterion is of such a
Sec. 6 L i o u v i l l e' s Cr i t er i o n 237
nature that we can easil y construct real numbers that viol ate it for every
n > 1. Any such number wil l then have to be transcendental . I n this way we
shal l be abl e to pro duce transcendental numbers at wil l . Ho wever, no ne of
the famil iar numbers is such that its transcendence can be pro ved using Li o u-
vil l e's Cri teri on.
I n this section of the bo o k we present this resul t of Liouvil l e. I t is a sur-
prisingl y simpl e and el ementary resul t to pro ve. This takes no thi ng away
from the resul t; in o ur o pi ni o n it greatl y enhances it.
Theo rem 6.6.1 (Liouvil l e). Let a be an al gebraic number of degree
n > 2 (i.e., a is al gebraic but not rati o nal ). Then there exists a positive con-
stant c (which depends onl y on a) such that for al l integers u, v with
v > 0, | a u/v | > civ".
Proof. Let A be a ro o t of the po l yno mi al f(x) of degree n in Q[x],
where <Q is the fiel d of rati onal numbers. By cl earing of deno mi nato rs in the
coefficients of f(x), we may assume that f(x) r
0
x" + r^x"
- 1
+ + r,
where al l the r, are integers and where r
0
> 0.
Since the po l yno mi al fix) is i rreduci bl e of degree n it has n distinct
ro o ts a = a, a
2
, . . . , a in C, the fiel d of co mpl ex numbers. Therefo re, f(x)
factors o ver C as /Yx) = r
0
(x - a)(x ~ a
2
) (x - a). Let u, v be i ntegers
with v > 0; then
u\ _ i\
)U
"_ , riu"'
1
r_
T
M
hence
V
"
f
\i)
= +
' ' i
M
" "
L u
+ + I 'n-iuv"'
1
+ rv"
is an integer. Mo reo ver, since f(x) is irreducibl e in Q[x] of degree n s 2, fix)
has no rati o nal roots, so v"f(u/v) is a no nzero integer, whence | v"f(u/v) | > 1.
Usi ng the facto red form o f /(x), we have that
hence
u
- | - a
v
\f(ulv)\
r
{)
\(ulv) - a
2
\ \{ulv) - a\
:
v"\f(u/v)\
r
0
v"\(ulv) - a
2
\ \(u/v) - q,\
1
r
0
v"\(ulv) - a
2
\ - \{ulv) - a\ '
Let s be the l argest of \a\, \a
2
\, , \a\. We divide the argument accord-
ing as \u/v\ > 2s or \it/v\ s 2s. If \u/v\ > 2s, then, by the triangl e inequal ity,
\a - (u/v)\ > \u/v\ \a\ > 2s s = s, and, since v > 1, \a - (u/v)\ > s/v".
On the o ther hand, if \u/v\ ^ 2s, then, again by the triangl e inequal ity,
- (u/v) \ < |a,| +|/rj | <5 +2s =3s. Therefo re,
t =
(u\
a
n
- ~
W
a
n
-
W
(3s)"
so that lit > l / O) "
- 1
= l / ( 3" ~y
, M
) . Go i ng back to the i nequal i ty abo ve
that \a - (u/v)\ > l/[r
0
v"\a
2
- (u/v)\ \a - (u/v)\], we have that
\a - (u/v)\ 5: l/(r
0
3"~
1
s"~
1
v"). These numbers r
0
, 3"'\ s"~
x
ate determi ned
once and for al l by a and its mi ni mal po l yno mi al f(x) and do not depend on u
or v. If we l et b = 1/(7-
0
3""
1
J """
1
)
;
T H E N
b > 0 and \a ~ (u/v)\ > b/v". This
covers the second case, where \ u/v\ s 2s.
ff c is a positive number smal l er than bo th b and s, we have from the
discussion that \a u/v\ > c/v" for al l integers u, v, where v > 0, thereby
pro vi ng the theo rem.
Let's see the parti cul ars of the proof for the parti cul ar case a = V2.
The mi ni mal pol ynomi al for a in Q[x] isf(x) = (x - a)(x + a), so a = a
x
and
a = a
2
. So if u and v are i ntegers, and v > 0, then
- / ( ^ <
2
- "
;
) ^ T
2
-
2
) = "
S
-
2
^
0
'
an integer. So |u
2
/(w/u)| > 1 > llv
2
. The âbove is the l arger of |V2|
and | - V2); that is, s = V5. Al so , the above is l /(3
2
"
1
(V
/
2)
2
-
1
) = 1/(3V2),
so if c is any posi ti ve number l ess than 1/(3V2), then |V2 - ulv\ > c/v
2
.
What the theo rem says is the fol l owing: Any al gebraic real number has
rati onal numbers as cl ose as we l ike to it (this is true for al l numbers), but if
this al gebraic real number a is of degree n s 2, there are restrictions o n ho w
finel y we can appro xi mate a by rati o nal numbers. These restrictions are the
o nes i mpo sed by Liouvil l e's Theo rem.
Ho w do we use this resul t to pro duce transcendental numbers? Al l we
need do is to pro duce a real number T, say, such that whatever positive i nte-
ger n may be, and whatever positive c we choose, we can find a pair of inte-
gers u, v, with v >0 such that | T - u/v\ < c/v". We can find such a r easil y by
writing down an infinite decimal invol ving 0's and l 's, where we make the 0's
spread o ut between the l 's very rapidl y. Fo r i nstance, if T =
0.10100100000010 . . . 010 . . . , where the 0's between successive l 's go l ike
ml, then T is a number that viol ates Liouvil l e's Cri teri on for every n > 0.
Thus this number r is transcendental .
Sec. 7 The I r r at i o n al i t y of IT 239
We coul d, of course, use o ther wi de spreads of O's between the l 's
m'", (ml)
2
, and so o nto pro duce ho rdes of transcendental numbers. Al so ,
i nstead of using just l 's, we coul d use any of the ni ne no nzero digits to o btai n
mo re transcendental numbers. We l eave to the reader the verification that
the numbers of the sort we descri bed do not satisfy Liouvil l e's Criterion for
any positive i nteger n and any positive c.
We can use the transcendental number r and the variants of it we de-
scribed to pro ve a famous resul t due to Canto r. This resul t says that there is
a 1-1 co rrespo ndence between al l the real numbers and its subset of real
transcendental numbers. I n o ther wo rds, in so me sense, there are as many
transcendental real s as there are real s. We give a brief sketch of ho w we
carry it out, l eaving the detail s to the reader.
First, it is easy to construct a 1-1 mappi ng of the real s o nto those real s
strictl y between 0 and 1 (try to find such a mappi ng). Thi s is al so true for the
real transcendental numbers and tho se of them strictl y between 0 and 1. Let
the first set be A and the second o ne B, so B C A. Then, by a theo rem in set
theo ry, it suffices to construct a 1-1 mappi ng of A into B.
Gi ven any number in A, we can represent it as an infinite decimal
0.0 ^2 . a .. . , where the a
t
fal l between 0 and 9. (We now wave our hands
a l ittl e, bei ng a l ittl e bit i naccurate. The reader shoul d try to tighten up
the argument.) Define the mappi ng / from A to B by /(0.a
1
rt
2
.. a . . .) =
O.fl
1
0a
2
00a
3
000000fl
4
. .. ; by the Liouvil l e Cri teri on, except for a smal l set of
a
u
a
2
, , a
n
,. .. , the numbers 0.a
1
0a
2
00fl
3
000000a
4
. . . are transcendental .
The / we wro te do wn then pro vi des us wi th the requi red mappi ng.
One final wo rd abo ut the ki nd of appro xi mati o n of al gebraic numbers
by rati onal s expressed in Theo rem 6.6.1. There we have that if a is real al ge-
braic of degree n s 2, then \a u/v\ > c/v" for so me appro pri ate positive c.
If we coul d decrease the n to \a u/v\ > civ'" for m < n and so me suitabl e c
(dependi ng on a and m), we woul d get an even sharper resul t. I n 1955 the
(then) yo ung Engl ish mathemati ci an K. F. Ro th pro ved the powerful resul t
that effectivel y we coul d cut the n do wn to 2. Hi s exact resul t is: If a is al ge-
brai c of degree n "s 2, then for every real number r > 2 there exists a positive
co nstant c, dependi ng on a and r, such that \a ujv\ > c/v
r
for al l but a finite
number of fractions u/v.
7. THE I RRATI ONAL I TY OF TT
As we i ndi cated earl ier, Li ndemann in 1882 pro ved that TT is a transcendental
number. I n parti cul ar, from this resul t of Li ndemann it fol l ows that TT is irra-
tional . We shal l no t pro ve the transcendence of TT herei t wo ul d requi re a
rather l ong deto urbut we wil l , at l east, pro ve that TT is irrational . The very
nice proof that we give of this fact is due to I. Niven; it appeared in his paper
"A Simpl e Proof That TT I s I rrati o nal ," which was publ i shed in the Bulletin
of the American Mathematical Society, vol . 53 (1947), p. 509. To fol l ow
Ni ven's proof onl y requi res so me materi al from a standard first-year cal cul us
course.
We begin with
L emma 6.7.1. If u is a real number, then l im u"/n! = 0.
Proof. If u is any real number, then e" is a wel l -defined real number
and e" = l + u +
2
/2! + M
3
/3! + + u"ln\ + . The series 1 + u + u
2
/2l +
+ u"ln\ + converges to e"; since this series converges, its nth term
must go to 0. Thus l imu"lnl = 0.
We no w present Ni ven's pro o f of the irrational ity of TT.
Theo rem 6.7.2. TT is an irrational number.
Proof. Suppose that TT is rati onal ; then TT = alb, where a and b are po s-
itive integers.
Fo r every integer n > 0, we i ntro duce a pol ynomi al , who se pro perti es
wil l l ead us to the desired concl usion. The basic pro perti es of this po l yno mi al
wil l hol d for al l positive n. The strategy of the proof is to make a judicious
choice of n at the appro pri ate stage of the proof.
L et/ (x) = x"(a - bx)"ln\, where TT = alb. This is a po l yno mi al of de-
gree 2/7 with rational coefficients. Expandi ng it out, we o btai n
_ a
0
x" + a
Y
x"^ + + a
n
x
2
"
TW
n
\
where
a
n
= a", a, = -na'^b, . . . , , - = ^' " j , a""b', ... ,a = (- 1)"6
are integers.
We deno te the ith derivative of f(x) with respect to x by the usual no -
tati o n f
(
'\x), understandi ng f
{0
\x) to mean/ (x) itsel f.
We first no te a symmetry pro perty of f(x), namel y, that f(x) =
/(T T - x). To see this, no te that/f x) = (b"ln\)x"(TT - x)", from who se form it
is cl ear that/f x) =/(T T - x). Since this hol ds for /(x), it is easy to see, from
the chain rul e for differentiation, that/
( ; )
(x) = ( - l ) ' /
( / )
( 7 r - x).
Sec. 7 The Irrational ity of TT 241
This statement about f(x) and all its derivatives allows us to conclude
that for the statements that we make about the nature of all f /;e/
w
(0), there are
appropriate statements about all the f('\ir).
We shal l be i nterested in the val ue o f /
( , )
(0), and/
w
( 7 r ) , for al l no nneg-
ative i. No te that from the expanded fo rm of fix) given above we easil y ob-
tain that/
w
( 0) is merel y /'! times the coefficient of .v' of the po l yno mi al f(x).
This i mmedi atel y impl ies, since the l owest po wer of x appeari ng in f(x) is
the nth, that/
( , )
(0) = 0 if / < n. Fo r i > n we o btai n that/
( ; )
(0) = ila^Jnl;
since i s n, i\ln\ is an integer, and as we po i nted out above, a,-_ is al so an
integer; therefo re /
( , )
(0) is an integer for all nonnegative integers i. Since
f^(ir) = ( - l ) ' f ( O) , we have that /
( , )
(7r) is an i nteger for al l no nnegati ve
i ntegers i.
We i ntro duce an auxil iary function
F(x) = f{x) - f
2
\x) + + i-l)J
(2
"\x).
Sincef
(
'"
]
{x) = 0 ifm > In, we see that
g = F\x) = fV\x) - fîx) + + ( - 1) " /
( 2
' M
= -F{x) + f(x).
Therefo re,
F"(x) sin x + F'ix) cos x
F\x) cos x + Fix) sin x
(F"(x) + Fix)) si n* = fix) smx.
Fro m this we concl ude that
I fix) sin x dx = [F'ix) sin x - Fix) cos X]Q
= (F'(T T ) sin TT - Fin) cos TT) - (F'(0) sin 0 - 7(0) cos 0)
= Fin) + F(0).
But from the form of Fix) abo ve and the fact that al l f
( , )
(0) and/
( 0
(7r) are
i ntegers, we concl ude that F(T T ) + -F(O) is an integer. Thus JZf(
x
)
s m x
dx is
an integer. This statement about JQ fix) sin * dx is true for any integer n > 0
whatsoever. We no w want to cho o se cl everl y eno ugh to make sure that the
statement "/ J fix) sin x dx is an i nteger" canno t possibl y be true.
dx
iF\x) sin x - Fix) cos x)
We carry out an esti mate o n Sof(
x
) sin
x
dx. Fo r 0 < x < TT the func-
ti o n/(.v) = x"(a - bx)"ln\ ^ ir"a"ln\ (since a > 0), and al so 0 < sin.v < 1.
Thus 0 < JZf(x) sin x dx < /ST T V/H! dx = T T "
+
V/7?!.
Let u = TTCI; then, by L emma 6.7.1, lirnû"ln\ = 0, so if we pick n large
enough, we can make sure that u"ln\ < IITT, hence 7 r "
+
V 7! = mi"ln\ < 1.
But then Jof(x) sin x dx is trapped strictl y between 0 and 1. But, by what we
have shown, Jaf(x) sin .v dx is an integer. Since there is no i nteger strictl y be-
tween 0 and 1, we have reached a co ntradi cti o n. Thus the premi se that TT is
rati o nal was fal se. Therefo re, TT is i rrati o nal . Thi s co mpl etes the pro o f of
the theo rem.
i ;
I n d ex 245
A(S), 16
Abel , 43
Abel ian group, 43
Al gebraic
el ement, 193
extension, 193
number, 194
Al gebraical l y cl osed fiel d, 200
Al ternating group, 121, 215
Associative l aw, 12, 41
Associative ring, 126
Automorphism of a group, 68
inner, 69
Automorphism of a ring, 141
Basis, 187
Bertrand Postul ate, 220
Bijection, 11
Bool e, 138
Bool ean ring, 138
Cantor, 239
Carrol l , 7
Cartesian product, 5
Cauchy, 80
Cauchy's Theorem, 80, 89
Cayl ey, 69
Cayl ey's Theorem, 69
Center, 53
Central izer, 53,102
Characteristic subgroup, 76
Characteristic of a fiel d, 178
Chinese Remainder Theorem, 147
Cl ass Equation, 103
Commutative ring, 127
Commuting el ements, 53
Commuting mappings, 21
Compl ement of a set, 5
Compl ex number, 32
absol ute val ue of, 34
argument of, 35
compl ex conjugate of, 32
imaginary part of, 32
pol ar form of, 35
purel y imaginary, 32
real part of, 32
Composition of mappings, 11
Congruence, 57
cl ass, 60
Conjugacy, 58
cl ass, 58
Conjugate
el ements, 58
of a compl ex number, 32
Constant
function, 9
pol ynomial , 153
Constructibl e
l ength, 202
number, 204
Correspondence Theorem
for groups, 86
for rings, 142
Coset
l eft, 64
right, 58
Cycl e of a permutation, 111
Cycl ic group, 53, 55, 60
generator of, 53
Cycl otomic pol ynomial , 230
Degree
of an al gebraic el ement, 195
of a fiel d extension, 191
of a pol ynomial , 153
De Moivre's Theorem, 36
De Morgan Rul es, 6
Dihedral group, 45,116,117
Dimension of a vector space, 186
Direct product of groups, 93
Direct sum
of rings, 146
of vector spaces, 181, 182
Divides, 22,157
Division Al gorithm, 155
Division ring, 127
Divisor, 22
Domain
integral , 127
principal ideal , 157
Dupl ication of a cube, 201
246 I n d ex
Eisenstein Criterion, 169
El ement, 3
al gebraic, 193
identity, 41
invertibl e, 133
orbit of, 21,65
transcendental , 194
unit, 41
Empty set. 4
Equal ity
of mappings, 9
of sets, 4
Equival ence cl ass, 58
Equival ence rel ation, 57
Eucl id, 27
Eucl idean ring, 163
Eucl id's Al gorithm, 22
Eul er, 63
Eul er cp-function, 62
Eul er's Theorem, 63
Extension fiel d, 191
Factor, 22
Factor group, 78
Factorial , 17
Fermat, 63
Fermat's Theorem, 63
Fiel d, 127
al gebraical l y cl osed, 200
characteristic of, 178
extension, 191
of al gebraic numbers, 199, 200
of quotients, 172-175
of rational functions, 177,179
spl itting, 213
Fiel d extension, 191
al gebraic, 193
degree of, 191
finite, 191
Finite abel ian groups, 96
Finite dimensional vector space, 185
First Homomorphism Theorem
for groups, 85
for rings, 142
Formal derivative, 227
Function, 8
constant, 9
Eul er, 62
identity, 9
injective, 10
one-to-one, 10
onto, 10
surjective, 10
Fundamental Theorem
of Al gebra, 200
on Finite Abel ian Groups, 100
Gal ois theory, 212
Gauss, 169
Gauss' Lemma, 168
Gaussian integers, 38,166
Greatest common divisor
of integers, 23
of pol ynomial s, 158
Group, 41
abel ian, 43
al ternating, 121, 215
axioms, 41
center of, 53
cycl ic, 53, 55, 60
dihedral , 45,116,117
factor, 78
finite, 42
Hamil tonian, 72
homomorphism, 67
Kl ein's, 116
nonabel ian, 43
octic, 116
order of, 42
quotient, 78
simpl e, 123, 216
Hamil ton, 72,132
Hamil tonian group, 72
Hardy, 201
Hermite, 194
Homomorphism of groups, 67
image of, 70
kernel of, 70
trivial , 67
Homomorphism of rings, 139
kernel of, 140
Homomorphism Theorems
for groups, 84-87
for rings, 142
I n d ex 247
Ideal , 140
l eft, 140
maximal , 148
right, 140
trivial , 142
two-sided, 140
Identity el ement, 41
Identity function, 9
Image, 8
inverse, 10
Index of a subgroup, 59
Induction, 29
Induction step, 31
Inductive hypothesis, 30
Injective mapping, 10
Integers, 21
Gaussian, 38,166
rel ativel y prime, 25
Integral domain, 127
Intersection of sets, 4
Invariants of abel ian groups, 100
Inverse
of a group el ement, 41
of a mapping, 12
Invertibl e el ement, 133
Irreducibl e pol ynomial , 159
Isomorphic groups, 68
Isomorphic rings, 141
Isomorphism
of groups, 68
of rings, 141
Kernel of a homomorphism
for groups, 70
for rings, 140
Kl ein's 4-group, 116
Lagrange, 59
Lagrange's Identity, 133
Lagrange's Theorem, 59
Leading coefficient, 162
Least common mul tipl e, 28
Lindemann, 194
Linear combination, 185
Linear dependence, 186
Linear independence, 185
Liouvil l e, 236
Liouvil l e's Criterion, 236
Mappings, 8
commuting, 21
composition of, 11
identity, 9
injective, 10
one-to-one, 10
onto, 10
product of, 11
surjective, 10
Mathematical induction, 29
Matrices
real 2X2,130,131
2X 2 over a ring, 131
Maximal ideal , 148
McKay, 89
Minimal generating set, 186
Minimal pol ynomial , 195
Monomorphism
of groups, 68
of rings, 141
Motion of a figure, 115
Mul tipl e, 22
Mul tipl icity, 209
Niven, 240
Normal subgroup, 71
index of, 59
Nul l set, 4
Number
al gebraic, 194
compl ex, 32
prime, 21, 26
purel y imaginary, 32
transcendental , 194, 236
Octic group, 116
One-to-one
correspondence, 11
mapping, 10
Orbit, 21, 65
Order
of an el ement, 60
of a group, 42
Partition
of a set, 58
of a positive integer, 100
248 I ndex
Permutation, 16,109
even, 121
odd, 121
Pol ynomial , 152,179
coefficients of, 152
constant, 153
cycl otomic, 230
degree of, 153
irreducibl e, 159
l eading coefficient of, 162
minimal , 195
monic, 157
rel ativel y prime, 159
root of, 208
Pol ynomial ring, 152
Prime number, 21, 26
Primitive root of unity, 37, 39, 229
Primitive root mod p, 66
Principal ideal domain, 157
Principl e of Mathematical Induction, 29
Product of mappings, 11
Projection, 9
Quadratic nonresidue, 151
Quadratic residue, 151
Quaternions, 131,136
Quotient group, 78
Rational functions, 177,179
Reftexivity, 57
Rel ation, 56
congruence rel ation, 57
conjugacy rel ation, 58
equival ence rel ation, 57
Rel ativel y prime
integers, 25
pol ynomial s, 159
Ring, 126
associative, 126
Bool ean, 138
commutative, 127
division, 127
Eucl idean, 166
homomorphism, 139
noncommutative, 127
of pol ynomial s, 152,179
with unit, 126
Roots of unity, 42
primitive, 37, 39, 229
Root of a pol ynomial , 208
mul tipl icity of, 209
Roth, 239
Second Homomorphism Theorem
for groups, 86
for rings, 142
Sets, 3
Cartesian product of, 5
difference of, 4
equal ity of, 4
intersection of, 4
union of, 4
Simpl e group, 123, 216
Spl itting fiel d, 213
Squaring of a circl e, 207
Subfiel d, 128,191
Subgroup, 51
characteristic, 76
cycl ic, 53
index of, 59
normal , 71
proper, 51
Syl ow, 101
trivial , 51
Subring, 129
Subset, 3
Subspace of a Vector space, 181
spanned by el ements, 181
Surjective mapping, 10
Syl ow, 104
Syl ow subgroup, 101,105
Syl ow's Theorem, 105
Symmetric group, 16,109
Symmetry, 57
Third Homomorphism Theorem
for groups, 87
for rings, 142
Transcendental
el ement, 194
number, 194, 236
Transitivity, 57
Transposition, 20, 111
Triangl e Inequal ity, 34
Index 249
Trisection of an angl e, 201
Union of sets, 4
Unit el ement
of a group, 41
in a ring, 126
Vector space, 180
basis of, 187
dimension of, 186
finite dimensional , 185
infinite dimensional , 185
minimal generating set for, 186
Wel l -Ordering Principl e, 22
Wil es, 63
Wil son's Theorem, 65, 210
Zero divisor, 128

Abstract Algebra - 3rd Edition by I.N. Herstein

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Abstract Algebra - 3rd Edition by I.N. Herstein

Uploaded by

Copyright:

Available Formats

Third Edition

,..., a are al l its el ements, sho w

You might also like