Thetheorv

of groups
Ian D._,Macdona|d
The theory of groups
IAN D. MACDONALD

The theory of groups

OXFORD ' AT THE CLARENDON PRESS

1968
Oxford University Press, Ely House, London W. 1
GLASGOW NEW YORK TORONTO MELBOURNE WELLINGTON
CAPE TOWN SALISBURY IBADAN NAIROBI LUSAKA ADDIS ABABA

BOMBAY CALCUTTA MADRAS KARACHI LAHOEE DACGA

KUALA LUMPUR HONG KONG TOKYO

© OXFORD UNIVERSITY PRESS 1968

PRINTED IN GREAT BRITAIN

Preface

THE present work sprang from a course of lectures given to an

Australian pass class, when I found that no existing elementary
text really met my needs. It bears many marks of its genesis in
that fiery crucible. I refer first of all to the presentation of even
the most elementary ideas in detail and with the support of
many illustrations and examples. Next, I have in mind the
single-minded concentration on groups, to the exclusion of
links With other mathematics and of applications; experience
shows that a course of this kind has to be self—contained as well
as simple, and a glance at the Appendix will show how little
material is assumed known: even Zorn’s lemma and its cognates
are passed over in silence and never used. The third point is
the careful collection of problems for the reader to solve, some
invitingly easy, others difficult or worse. Naturally my pass stu-
dents reached their depth after only a few chapters. The rest of
the book followed on logically, and the subject—matter in
particular selected itself. No doubt the outcome is open to many
criticisms, but I hope that they will be tempered by the experience
of giving a course such as mine.
It seems to me evident, if not self-evident, that even the hum-
blest university student should be exposed to a little abstract
algebra. The problem of where to break into the subject has an
attractive answer in the theory of groups. But this belief, and
the consequent existence of this book, must not be construed as
a disparagement of any other branch of modern algebra. I feel
that, in particular, the theory of group representations is of the
utmost importance in developing the theory of groups. This
book contains as much group theory as I consider it advis-
able to study in isolation; after mastering it the enterprising
reader is urged to enrich his knowledge by broadening his
algebraic horizons. It may be well to note here that a modern
vi Preface

honours course must contain much other abstract algebra of

importance.
A few comments on the mathematics of the book are addres-
sed to the expert. An honest attempt has been made to place
equal emphasis on finite and on infinite groups. Minimal axioms
for groups have been taken, both to give the novice exercise
and because they have a place in a work of this type. The
main text does not contain the new proofs of the structure
theorem for finitely generated abelian groups and of the three
Sylow theorems: my feeling is, on pedagogic grounds, that to the
beginner they are hardly easier and probably less enlightening
than the classical proofs. The problems are usually common
currency among group theorists when they are not standard.
Finally, some words of gratitude. A number of my colleagues
and students have read the manuscript in part and made
suggestions concerning it. During the summer of 1967—8 the
University of Queensland made available a Junior Research
Assistantship, and Mr. J. R. Ellwood, who occupied this posi-
tion, read the manuscript with care and solved all the problems.
The typists were Mrs. Margaret F. Boden of the University of
Newcastle, New South Wales and Miss Pauline C. Baker of the
University of Queensland; their efi‘orts were considerable,
skilful, and always cheerful, and Mrs. Boden in particular pro-
duced many fine stencils. In the last stage of the work I have
become much aware of the high professional standards and
competence of the Clarendon Press. My expectations in this
respect were more than fulfilled, and the officers and staff have
been kind, helpful, and courteous Without fail. To all these
good people I ofl'er my thanks.
I. D. M.

The University of Queensland

Australia
April 1968
 Contents

Some examples of groups

l—I

Basic theorems and concepts 16

©9°S®Plrhww

. Complexes and subgroups 34

Normal subgroups and factor groups 53

Finitely generated abelian groups 82

. Normal structure 102

Sylow theorems 127

Generators and relations 149

. Nilpotent groups 172

P

Soluble groups 196

l—l

11. Survey of examples 213

REVISION PROBLEMS 230

APPENDIX 239

NOTE ON FURTHER READING 249

INDEX 251
Some examples of groups

SINCE examples are of great importance in the theory of groups

we devote this chapter to a survey of the examples that are
commonly encountered and used. We must of course indicate
which of the several possible definitions of ‘group’ we have in
mind, and the reason for our particular choice is simplicity in
proving that certain systems we shall present are in fact groups.
If the reader is familiar with most of the facts in this chapter
he may yet find it of value as a systematic revision of his know-
ledge.
DEFINITION. A group is a set and a binary operationi' in
that set satisfying certain conditions;;t that is, the elements
a, b (in that order) in the group G define an element usually
written as ab and usually called the product of a and b, the
binary operation being (conventionally) called multiplication.
The conditions are:
(i) G is closed under multiplication.
(ii) The associative law for multiplication holds, that is
a(bc) = (ab)c
for all elements a, b, c in G.
(iii) There is at least one identity element in G', that is an
element (3 such that we = a for all a in G.

1' For more details about binary operations see the Appendix.
I The reader who Wishes to assume more, for instance that identities and
inverses have some of the two-sided and uniqueness properties enumerated in
Theorems 2.01—2.05, may do so. He will not then have to prove the corre-
sponding theorems, but he will have the additional task of verifying that all
the examples presented in this and subsequent chapters are groups in the sense
of his definition.
853137 B
2 Some examples of groups

(iv) Some identity element e has the following property: given

any element a in G there is at least one solution x to the
equation ax = e, each solution being called an inverse
to a and written as ail.
It is important to note that we are not assuming uniqueness of
identities and inverses in a given group. Another fact that
requires careful consideration is that we are using what might be
called right identities and right inverses. Suppose we make just
one change in the above axioms, replacing the equation ax = e
in (iv) by xa = 6, so that (in an obvious sense) we are dealing
with right identities and left inverses; then many systems that
are far from being groups satisfy the new axioms. We may for
instance take any set S, such as the set of all teapots, and define
products by putting xy = a: for each pair as, y of elements of S.
Axioms (i) and (ii) are readily verified, and every element 3/ of 8
serves as an identity as required in (iii). To solve the equation
xa. = y for x with a given 3/, we simply take a: = y. The axioms
are therefore satisfied if S non-empty. But 8 does not constitute
a group if it has more than one element, for the equation
was = y has no solution x if 3/ 3A a.
We pass on to examples of systems that are groups. The most
familiar and readily available are groups of numbers.
Example 1.01. The set Z of all integers with addition as its
operation forms a group. To hold fast to the conventions in our
definition would involve writing 3+7 as 3. 7 and calling 10
the ‘product’ of 3 and 7. We therefore have every excuse to
call the binary operation addition and to use the familiar nota-
tion in this special case. The group axioms are easily verified:
(i) is simply that the sum of two integers is an integer, and (ii) is
well known, plausible, and a consequence of any axiom system
for integers. We can take the number 0 as an identity, for surely
(1+0 = a for all elements a of Z. As for (iv), an inverse of a is
—a since we have, just as simply, a+(——a) = 0.
Example 1.02. The set Q of all rational numbers forms a group
under addition, the proof being similar to that for Z.
 Some examples of groups 3

Example 1.03. The set R of all real numbers forms a group

under addition. Again the reader should be able to find a proof
quite like that given for Z.

Example 1.04. The set 0 of all complex numbers forms a group

under addition.
It may be well to mention some sets of numbers that do not
form groups. The odd integers do not form a group under
addition (though the even integers do); a sufficient reason for
this is that (i) does not hold; another reason is that there is no
identity element, which implies that (iii) and (iv) cannot hold.
The non-negative integers do not form a group under addition
for in their case (iv) does not hold (it is an empirical fact that
many naturally occurring sets with multiplication are not groups
because (iv) fails). The integers with the binary operation of
subtraction do not form a group because the associative law is
not valid; for instance 5—(4—1) = 2 and (5—4)—1 = 0.
Since multiplication is the other familiar operation on (real
and complex) numbers, it is reasonable to ask what sets of
numbers form groups when the group multiplication is taken
to be the ordinary multiplication of numbers. The set of all
real numbers is not a candidate for group status in this sense,
for it contains the number 0 and (since the number 1 is clearly
the identity element) we require for axiom (iv) a solution a: to
the equation «:0 = 1; and because there is no such 90 we do not
have a group.

Example 1.05. The set of positive rational numbers forms a

group under (ordinary) multiplication. To see this, note that
(i) and (ii) are easily checked, take 6 = 1 in (iii), and take
a—1 : l/a in (iv); l/a exists since a 72 0. We remark that the
non—zero rationals form a group for similar reasons.

Example 1.06. The set of positive (or of non—zero) real

numbers forms a group under multiplication of numbers. As
the reader should have by now acquired some facility in proving
such a statement there is no need to give further details.
4 Some examples of groups

Example 1.07. The non—zero complex numbers form a group

under the usual multiplication.
Example 1.08. The complex numbers of unit modulus form
a group under the usual multiplication.
Example 1.09. For each prime number p there is a subset of
the complex numbers that forms a multiplicative group of
some interest. Let Z; denote the following set:
{zzz E 0 and z?" = 1 for some integer n}.
(Note that the positive integer n depends on the element 2.)
We verify the axioms individually for this important group. Let
21, 22 be elements of Z; with z?” = 1, 2329’" = 1. Then (2122?" = 1
where r is the larger of m and n, so Z; is closed under multiplica—
tion. The associative law holds simply because it is always valid
for multiplication of complex numbers. Clearly Z1? contains
the complex number 1, which will serve as an identity element.
If z?" = 1 then (1 /z)1’” = 1, so if z is in Z; so is l/z—note that
z 72 O—and l/z will act as an inverse of 2. Therefore Z; is a
multiplicative group.
A slight knowledge of complex numbers reveals that the
elements of Z; are precisely all those numbers of the form

cos ii: +i sin 2%?

for n = 0, l, 2,..., k on integer, and 0 g k < p”.

Further important types of groups have residue classes
instead of numbers as elements. A residue class modulo n is a
certain equivalence class of integers;1‘ thus the residue class
[a] modulo n is the set of all solutions as to the congruence
x E a modulo n,

where n is a fixed positive integer. More explicitly but less

precisely (in the strict mathematical sense) we have
[a] = {..., a—n, a, a+n, a—l—2n,...}.
1‘ As explained in the Appendix.
 Some examples of groups 5

The set of residue classes modulo n is denoted by Z”; it contains

91, elements, each being an infinite set of integers.

Example 1.10. We define a binary operation called addition

on residue classes by the following equation:

[a+b] = [use].
This merely says that the sum of the residue classes containing
the integers a and b respectively is to be that containing a+b.
We ought first, however, to have verified that the result [a—l—b]
is the same whatever representatives we choose in [a] and [b], so
that the definition depends only on the classes containing a and b,
not on these numbers themselves. This we proceed to do. Take
any element of [a] ; it will be a—Hm for some integer 16. Similarly
an arbitrary element of [b] will be b—l—ln. Thus
[a] = [a+kn], [b] = [b—l—ln].
We require [a+kn]+[b+ln] = [a+b]
if we are to have a proper definition of sum. But

[a+kn]+[b+ln] = [(a+kn)+(b+ln)],
and since (a+kn)+(b+ln) E a—l—b modulon

we have [(a+lm)+(b+ln)] = [a+b],

and so all is well.
The group axioms are easily verified now that the delicate
matter of definition has been dealt with. Closure is obvious, the
associative law is well—known or alternatively follows from the
corresponding axiom for integers, [0] is an identity element,
[—0,] is an inverse for [(2]. Therefore we have a group of residue
classes, denoted by Zn, with addition as its binary operation.
Notice that it contains precisely n elements Where n is an arbi—
trary positive integer, We therefore have a solution to the
problem of finding a group with a given finite number of ele-
ments.
6 Some examples of groups

The subject of multiplication of residue classes and the forma-

tion of multiplicative groups requires some care. The usual
definition of multiplication is the obvious one:
[ab] = [0]l-
It is essential to verify that this is a proper definition. This may
be carried out in a way similar to that explained in the case of
addition, the key step being the remark that
(a—l—Imxb—I—ln) = ab+(la+kb+kln)n
and so [(a+kn)(b+ln)] = [ab].
The reader ought to fill in all the details for himself.
Unfortunately this multiplication of residue classes will not
generally yield groups. Just as the number 0 cannot occur in
multiplicative groups of numbers so the residue class [0] cannot
occur in multiplicative groups of residue classes when n > 1,
for how can we find an inverse element for [0]? Remember that
[x][0] = [O] 75 [1] for all 3:. There is another complication.
Consider residue classes modulo 4, for which
[2ll2] = [4] = [0];
and so the product of two non-zero classes may be [0]. We must
exclude [0] from any group, and having done that there are still
cases when we do not have closure.
We can, however, salvage something.
Example 1.11. The set of non—zero residue classes modulo 1)
forms a group under multiplication, 10 being any prime number.
The associative law and the existence of an identity element [1]
are obvious facts, but axioms (i) and (iv) require care in their
consideration. Let [a] be a non—zero residue class modulo 1).
This is precisely the same as supposing that 19 does not divide
the integer a. An elementary and well—known result of number
theory]L states that for suitable integers u, u we have the equality
on +102: = 1.
Thus in terms of residue classes modulo 1) we have
[001%] = [W] = [W+1Wl= [1]-
1' See the Appendix.
 Some examples of groups 7

Now consider (i), the closure requirement, and suppose that [a]
and [b] are non-zero residue classes for which [a][b] = [0]. It
follows from the existence of u and from the commutative and
associative laws that
[b] = [allullbl = lullallbl = MN] = [0],
and we arrive at a contradiction. Therefore (i) holds. Since we
have found an inverse [u] to [a], we have also verified (iv). This
shows that we have a multiplicative group whose elements are
the 19—1 non—zero residue classes modulo p.
All our examples so far have been groups with the property
that every pair of elements commutes, that is ab is always equal
to ha, and this, of course, has been a consequence of the com-
mutative laws for addition and multiplication of real and
complex numbers. We omit detailed verification.

DEFINITION. The elements a, b of the group G commute if

ab = be. The group G is commutative or abelicm if every pair of
its elements commutes.
A pair of elements of the form x, xx in any group commutes
because of the associative law.
There are many groups that are non-abelian, and they are
often best presented as groups ofmappings of one sort or another.
This seems to be their natural context. To make this remark
precise, we recall that there is a natural multiplication of
mappings and that this multiplication is associative; details
are to be found in the Appendix. The name semigroup is given
to any non-empty set with an associative multiplication under
which it is closed. Thus every group is a semigroup with certain
additional properties; note that axiom (iii) implies that no group
is empty. On the other hand many semigroups (for example,
the integers with the usual binary operation of multiplication)
fail to satisfy group axioms (iii) and (iv).
These remarks, together with the facts presented in the
Appendix, should make the following theorem and its proof
apparent. ‘
8 Some examples of groups

THEOREM 1.12. The set of all mappings of a fixed (non-empty)

set into itselfforms a semlgroup. E]
To be quite explicit about the multiplication of mappings
used here, let «)5 and 9!: be two mappings of our fixed set X into
itself; then q/J is defined by putting x(<;[>z/1) = (q)1/1 for each
element x in X.
We now have a means of constructing many semigroups. For
groups of mappings, however, we require inverses of mappings,
and we recall that a mapping from a set X into X has an inverse
if and only if it is both one—one and onto.T This explains our
interest in the concept of ‘permutation’.
DEFINITION. A permutation of a set is a one—one mapping of
the set onto itself.
Example 1.13. The set SX of all permutations of a (non-empty)
set X forms a group. It is easy to see that 8X is closed, and
(since 8X is a subset of the set of all mappings of X into X)
associativity is a consequence of Theorem 1.12. An identity
element of SK is the permutation L which leaves every member
of X fixed, and which therefore has the property that p1. = p
for each element p of SX. Finally, since p is one—one and onto X,
each permutation p has an inverse mapping p—l, with pp_1 = L,
and it is easily verified that p‘1 is one—one and onto X, that is
p—1 is a member of SX.
This group SX is known as the symmetric group on the set X.
In a sense that will later be made precise, the structure of the
group 8X depends only on the number of elements in X. Since
there are n! distinct permutations of a finite set of n elements,
we see that 8X has n! elements if X has n elements.
When X is finite and small it is practicable to write down the
elements of SX explicitly and to carry out exploratory calcula-
tions. If X = {1, 2, 3} then SX is
{a (23), (31), (12), (123), (132)}
in the usual permutation notation. This particular SX is non-
T See the Appendix.
 Some examples of groups 9

abelian, since we have, using the usual multiplication of map-

Pings’ (23x31) = (132),
(31x23) = (123),
so that (23)(31) 7e (31)(23).
Example 1.14. Let RX be the set of permutations p of the
set X such that each p leaves all but a finite number of elements
of X fixed. It turns out that RX is an interesting group. To
prove closure the reader should show that if up is the set of
elements of X that are displaced by the permutation p then
0P1 p, S ”p1 U ”pr The other axioms are readily verified. This
group RX is known as the restricted symmetric group on the set X.
Example 1.15. Let AX be the set of permutations p of X such
that each p leaves all but a finite number of elements of X fixed,
and has even parity (that is, p is the product of an even number
of transpositions). Once again the reader is invited to prove
the group properties ofAx under the appropriate multiplication.
The group Ax is called the alternating group on the set X.
We next consider certain permutations of rather special sets,
namely the sets ofpoints making up Euclidean space of one, two,
or three dimensions. Initially our account of these groups of
mappings will be geometrical and tentative, and it should not
be regarded as rigorous. After these intuitive considerations
we shall examine groups of matrices, both those that arise when
coordinates are introduced in Euclidean geometry and those
that result from generalization.
Let E be a line—a one-dimensional Euclidean space. It is
obvious in the intuitive sense that two sorts of congruence (that
is, distance-preserving) mappings of 3 onto if exist:
(i) Translations—a mapping qS of .5,” into ,7 is a translation if
(,5 moves every point a fixed distance to the right or left.
(ii) Reflections in a fixed point 0—-a mapping 75 of E into E
is a reflection if 95 maps each point P into the point P95 which
is on the other side of 0 but at the same distance from 0.
Clearly there is a translation L which acts as an identity
10 Some examples of groups

mapping; clearly each translation has an inverse, which is again

a translation; clearly each reflection 95 has ¢ as an inverse,
because 75¢ = 1..
Hence we find two more groups.
Example 1.16. The set of all translations of .5,” (with the usual
multiplication) forms an abelian group. It is not hard to visual-
ize the group properties for this set.
Example 1.17. The set of all congruence mappings of the line
3 forms a group. A congruence mapping may be presented as
a combination of suitable translations and reflections, which
should enable the reader to see how the effect of a congruence
mapping may be ‘undone’ and an inverse constructed. The other
laws are easier to grasp. Among points that the reader might
ponder on is the fact that the product of two reflections is a
translation.
The congruence mappings of the plane 9 into itself are a
little more complicated to describe. There are:
(i) Translations—a mapping 96 of 9’ into Q is a translation
if the segment joining P to P95 is of constant length and
direction for all points P.
(ii) Rotations about a fixed point O—a mapping qS of 9’
into Q is a rotation if P and P96 are at the same distance
from 0 and if the angle subtended at 0 by P and Pzfi is
constant for all points P.
(iii) Reflections in a fixed line g—a mapping 45 of W into 9
is a reflection in .5? if each point P maps onto its mirror
image in 3.
We note several of the numerous groups formed by sets of
congruence mappings of 9.
Example 1.18. The set of all translations of9’ forms an abelian
group.

Example 1.19. The set of all rotations of 9 about a fixed point

0 forms an abelian group.
 Some examples of groups 11

Example 1.20. The set of all congruence mappings of .95 forms

a group. Note that this group includes all rotations about every
point 0 of 9.
In the case of congruence mappings of three-dimensional
space 5” into 5’ we indicate some familiar types:
(i) translations;
(ii) rotations about a fixed line;
(iii) reflections in a fixed plane.
It is a remarkable fact that any congruence mapping of y which
is a combination of translations and rotations and which leaves
one point fixed is equivalent to a rotation about a line through
that fixed pointxl'
The following groups of congruence mappings of 5’ are of
interest.
Example 1.21. The set of all translations of .57 forms an abelian
group.
Example 1.22. The set of all rotations of .5” about a fixed line
forms an abelian group.
Example 1.23. The set of all rotations of .5? leaving a given
point fixed forms a group.
Example 1.24. The set of all congruence mappings of .5” forms
a group.
Many other important groups arise as sets of mappings which
leave some geometrical figure invariant. We shall give one case
which gives a group that we shall mention several times, refer-
ring the interested reader elsewherei for a fuller account.
Example 1.25. The set of all congruence mappings of 9‘ which
leave fixed a regular n-sided polygon is a group. We mean that
the polygon is unchanged as a whole, but not necessarily
unchanged point by point; and of course n must be an integer
greater than 2. If the centre of the polygon is 0 then the required
1‘ A proof may be found in J. L. Synge and B. A. Griffith, Principles of
mechanics, p. 280 (3rd ed., McGraw-Hill, 1959).
i For instance to P. S. Alexandrofi, Introduction to the theory of groups,
Chapter 5 (Blackie, 1959).
12 Some examples of groups

mappings are rotations of 9’ about 0 through the angle 2k7r/n

for k = O, 1, 2,...,n~1, and reflections in the lines joining 0 to
each vertex and to the mid-point of each side. It will be found
that there are 2n such mappings and that these form a group,
aso-called dihedral group.
We shall now try to introduce some mathematical rigour into
our geometrical considerations. We take coordinates in the
usual fashion, though we shall not try to relate all the groups
that now force themselves on our attention to Examples
1.16—1.24.
Coordinates for E arise from the real numbers. Groups
corresponding to the congruence groups 1.16 and 1.17 are:
Example 1.26. The set of all mappings 95,1,Witl1 (1 real, such that
qd = x—l—d
for all real numbers 90, forms an abelian group. The verification
is omitted.
Example 1.27. The set of all mappings of the previous example
together with all mappings (/51 of the type xl’ba = 2a—x forms
a group. The verification is omitted.
In higher dimensions points are regarded as ordered sets of
real numbers when coordinates are introduced, and we write
such ordered sets as row vectors. Now an nxn matrix is
essentially a mapping of the set of row vectors into itself, the
row vector being multiplied by the matrix in the usual way.
Let therefore fin denote the set of all row vectors such as
x = (x1,x2,...,xn),
where x1,x2,...,xn are real numbers; and let all” be the set of
all n X n matrices, so that an element A of £7, has the form
all a12 . . . am
“21 “22 n n n “2,",

A :

an1 an2 . . . am
with the entries a“ real.
 Some examples of groups 13

Then A defines a mapping of 9” into 9?”, namely the mapping

which takes each a: into xA where
9511 = (a11x1+a21x2+...+amxm ..., alnx1+a2nx2+...+amxn).

Example 1.28. The set din of all non-singular nX n matrices

with real entries forms a group under matrix multiplication. The
closure law is obvious. The associative law follows, for example,
from the facts that a matrix represents a mapping and that the
product of two matrices represents the product of the associated
mappings. The unit matrix serves as an identity element, and
an inverse of the element A is the inverse matrix A-l. So we
have a group.

Example 1.29. The set of all real a matrices With deter-

minant 1 forms a group under matrix multiplication. The least
obvious point in the proof of this statement is the closure law.
Many variations on these examples are possible. In addition
to having infinitely many possibilities for n, we can choose
the rational numbers or the complex numbers or other sets
of numbers for entries. We can even choose residue classes
modulo some positive integer.
Example 1.30. The following set of matrices, in Which i
denotes a complex number for which 132 = —1, forms a group
under matrix multiplication:

{(5 ‘1’), (‘3 if)’ (—3 3), (‘1’ ‘3), (2’ 12), (1’.- 7f),
(13" 2’), (Z 2-)}-
The verification is omitted. This group is called the quaternion
group.
The multiplicative semigroup of all nx n matrices with (say)
real entries contains many subsets which form multiplicative
groups not containing the unit matrix. We exhibit one non—
trivial specimen.
14: Some examples of groups

Example 1.31. The set of matrices

{(33 :3) :00 real and non-zero}

forms an abelian group under matrix multiplication. The proof

depends on the fact that
x x y _ my gay
(0 0)(0 0) — (0 0),
which proves closure, shows that an identity element is the
matrix in which so = 1, and indicates how to find inverses since
as x 1/90 1/90 _ 1 1
0 o 0 0 "‘ 0 0'
Our final example is of lesser interest than multiplicative
groups of matrices.
Example 1.32. The set of all mx 7?. matrices forms an abelian
group under matrix addition. The entries may be rational num—
bers, real numbers, complex numbers, etc.

Problems
(Harder problems are starred)
1. Discuss the following systems with a View to determining which of
them are groups.
(i) The positive real numbers, with the binary operation of division.
(ii) The integers that are divisible by 10, With addition.
(iii) The non-zero rational numbers, with (1.” taken as the ‘product’
of a and. b.
(iv) {a+b«/2: a and b rational, not both zero} with the binary opera-
tion of number multiplication.
(v) The integers, with 211+ 3b taken as the ‘product’ of a and b.
(Vi) The rationals, with a—I—b—ab taken as the ‘product’ of a and b.
(vii) The set of all 2 X2 singular matrices with real entries, under
matrix addition.
(viii) The set of all vectors in three-dimensional Euclidean space,
with the binary operation of vector product.
(ix) The set {¢k: 70 real} of all mappings 45,0, Where
{13¢}: = kw:
of 3 into 9.
(x) {a+b«/2+c«/3: a, b, c rational, not all zero} With the binary
operation of number multiplication.
 Some examples of groups 15

n? Which of
2' WhiC h 0f the groups in the PreViOIIS question are abelia
the rest are semigroups?
3. Show that the non-zero residue classes modulo 6 do not form a group
under multiplication.
Prove that if the non-zero residue classes modulo 7:. form a group under
multiplication and n > 1 then 11, is prime.
4. A set of matrices forms a group under matrix multiplication. Show
that either every member of the set is singular or every member is non-
Singular.
5. Let 10 be any prime number. Show that the following set forms an
additive group: {m/p": m and n integers}.
Prove that the mapping 9!) for which 90¢ = poo, where ac is an arbitrary
element of the set, is a permutation.
6. Let G be a group of permutations on a set X. Show that those per-
mutations which leave fixed a given element x of X form a group.
7. Let T” be the set of n X n unitriangular matrices with real entries (that
is matrices with l for each entry on the principal diagonal and 0 every-
where below this diagonal). Show that T” is a multiplicative group.
8. Let T; be the set of n X n unitriangular matrices with residue classes
modulo 1) for entries (10 is a fixed prime). Show that T; is a multiplicative
group, and find the number of elements in it.
*9. Let S be any set and G be any group. Prove that if multiplication of
the mappings is suitably defined then the set of all mappings of S into G
forms a group.
10. The set of all ordered triples of integers is denoted by 0; thus a typical
element of G’ is (oz, [3, 'y) where oz, ,3, y are integers. Multiplication in G’ is
defined as follows:
(“3397)(65 7], D = (“+(_ 1)B§’ [3+(_ 1)y7]a (—1)£y+€)'
Show that G with this multiplication is a group.
*11. Show that there is a largest multiplicative group of 2 X 2 matrices
with real entries containing (0 0), and find this group.
5 —5
*12. A semigroup S has the property that for each element a: in S there
is an element denoted by 96—1 in S such that yam—1 = y for all y in S. Is
S a group?
“‘13. Let S be a semigroup with the following properties:
(i) There is an element 6 such that ex = x for all a: in S.
(ii) For each such element 6 and for each x in S there is an element
arl for which was—1 = e.
Prove that the set {86: s e S}, where e is fixed, forms a group.
2

Basic theorems and concepts

THE theorems in this chapter are of two sorts. First, we must

deduce some facts from the axioms. Second, we have to discuss
the important concepts of homomorphism and isomorphism.
The facts appear rather pedantic unless they are viewed against
a background of systems that are not groups, while the con-
cepts are of an importance that cannot be over-emphasized.
We recall our definition of a group as a semigroup with right
identities and right inverses, in a sense made precise in Chapter 1 .
Sometimes a definition is given that states that each right
identity is also a left identity and that each right inverse is also
a left inverse; but we can deduce these facts from our axioms.

THEOREM 2.01. If the element x of (my group G has 92—1 as a

(right) inverse, so that mac—1 = e for some (right) identity element 6,
then 95—15:; = 6.
Proof. Let e be a (right) identity element of G of the kind
mentioned in axiom (iv). We have to use the associative law in

The left-hand side of this equation reduces to x-le and so to

904, by axioms (iii) and (iv); therefore
x4 = (x—lx)x—1.
But, by axiom (iv), 96—1 itself has an inverse, denoted by (x—1)—1,
such that x—1(x—1)'1 = e; and we have
96"1(96'1)‘1 = {(w‘lx)x‘1}(x‘1)”1-
The left-hand side is e by (iv); and the right-hand side is
 Basic theorems and concepts 17

(x—lx){x-1(x—1)‘1} by the associative law, and clearly this equals

x—lx. Thus we have proved that x‘lx is the identity element e,
as required. [I
The above theorem implies that a and 9&4 always commute.
This is perhaps not surprising to us, for it happened in all the
examples of the previous chapter. The reader should consider
whether there is a corresponding theorem in a semigroup with
right identities and left inverses, such as a set with my defined
to be m. (We discussed such objects in Chapter 1. Here is a
specific instance. It will be found that if S is

:<.: 2): m:
then matrix multiplication in S is such that the product of two
matrices is the left-hand factor.)
Another readily acceptable fact follows.
THEOREM 2.02. If x is any element of any group G with identity
element e, such that sex—1 = e for all x in G, then ex = a.
Proof. The previous theorem and the axioms are used as
fOHOWS: ex = (xx-Um = x(x—1x) = me = 37.
Thus the result is proved; a and e always commute. |:|

THEOREM 2.03. If a and b are elements of any group G, and if

there is an element a“ for which xa = ab or an element y for which
an = by, then a = 6.
Proof. Suppose ma = xb. The existence of inverses gives
x‘1(xa) = x—1(xb).
Use of the associative law and Theorem 2.01 gives
ea = eb.
Theorem 2.02 now gives a = b. The deduction from ay = by
needs only the axioms and being thus easier is omitted.]]
This theorem gives us left and right cancellation laws, in an
obvious sense. The reader should consider whether there are
such laws in the set S of matrices mentioned above.
853137 0
18 Basic theorems and concepts

In all our examples, there was only one identity element and
each element had only one inverse.

THEOREM 2.04. In any group G there is only one identity

element, and each element of G has only one inverse.
Proof. Suppose we had two identity elements, say e1 and e2.
Since e2 is a right identity e1 e2 = el, and since (by Theorem 2.02)
e1 is a left identity e1 e2 = e2. Therefore e1 = e2, and G cannot
have more than one identity element.
Suppose that the element a of G’ has both as and y as inverses.
Then are = ay, and by left cancellation x = y; therefore a—1 is
unique. [I

THEOREM 2.05. In any group, a: = (x—1)—1.

Proof. We have defined (96"1)‘1 as a solution y of the equation
x—ly = e; we know that each of e, 31—1, y is unique. But clearly
y = x is one solution to x—ly = e, by Theorem 2.01. Since it
is the only solution, we have x = (w-1)'1.[]
The next result is of a different character. It is a generaliza-
tion of the associative law that appeared as axiom (ii) and it
states, roughly, that a product of group elements can be
bracketed in any way one pleases.

THEOREM 2.06. Let fixed elements a1, a2,..., an be taken in order

in any group G. Then the value of the product of all these elements
in the given order is unafiected by the sequence in which products
are formed.
Proof. The meaning of this theorem is subtle. For the purpose
of illustration let n = 3. The statement then is that the elements
a1(a2 a3) and (a1 a2)a3 are the same, for there are only two possible
ways of forming products; either form a2 a3 first or form a1 a2
first. The statement does not suggest that (a1a2)a3 = (a2 a1)a3,
which is certainly false in general. When n = 4 the possible
products ofa1, a2, a3, a4 in that order are a1{a2(a3 a4)}, (a1 ag)(a3 a4),
a1{(a2 a3)a4}, {a1(a2 a3)}a4, {(a1a2)a3}a4, and no others. The
theorem implies that these products are all equal.
 Basic theorems and concepts 19

We prove the theorem by induction on n. There is nothing

to do when n is 1 or 2. Let n > 2. The inductive hypothesis is
that products of fewer than n elements are well-defined, since
the method of bracketing is then immaterial. We consider an
element formed by taking products in the ordered set
{(11, a2,...,an}. We may write a typical product of a1,a2,...,an
in order as
(“I ar)(ar+1 “11)
where 1 < r < n.
Take two such products:
P1=(a1"'ar)(ar+1 an)»

192 = (a1"' as)(a’s+1"'a’n)°

We may assume that r < s; for if r = s then certainly p1 = 102.
Thus we have 1 g r < s < n and we seek to prove that 101 = 102.

NOW 201 =(a1awxam aa)“

= (“1m ar){:ar+1 a's+1 "ann’

= {(“1-H ar+1' :::)}(as+1-

by the associative law. Since
(a1...a,.)(ar+1...as) = a1...a,a,+1...as,
the latter being well-defined by the inductive hypothesis, we
have 121 = 202. u
The above result holds equally well for semigroups, as we
have nowhere used group axioms (iii) and (iv). We shall use it
continually in future without specific reference.
THEOREM 2.07. In any group G the inverse of alman is
agl...af1.
Proof. Clearly use of axioms (iii) and (iv) (that is
cancellatlon ) 1n (a1---an)(a;1"- (11.1)

gives e; note the tacit use of the previous theorem. If it is so

desired, this statement can be refined in rigour by the use of
induction, but this is unnecessary except in the strictest sense.
Since inverses are unique by Theorem 2.04, a;1...af1 is the
inverse of a1 an. [I
20 Basic theorems and concepts

We now have to consider the question of indices. Already we

know what 00—1 means if a is any element of a group G. The
general associative law allows us to give a reasonable meaning to
a” Where n is a positive integer; we may formally define a1 = a,
and a” = aim—la for n > 1, for instance. Theorem 2.07, with
a1 = = an = (1, yields in this notation that (an)—1 =(a“1)”.
It therefore seems right and proper to define or" as (MO—1, or as
M4)", Where n is a positive integer. This train of thought is
similar to that which makes corresponding definitions for powers
of real numbers.
We clearly have amen = am” when m and n are positive
integers. We can extend this to non—negative integers by defining
a0 to be e. Then we have:

THEOREM 2.08. If a is any element of a group G and m, n are

any integers then aman = am”.
Proof. We may assume at this stage that at least one of m, n
is negative and that neither is 0.
Suppose that both are negative, and put m = —r, n = —s.
Then using Theorem 2.07 we find
(aman)—1 = (an)—1(am)—1 = a—na—m = “Bar = “7+8,

for (x‘lr1 = x as in Theorem 2.05 implies that (a”)-1 = or”, and

taking inverses we have
aman = (ar-I—s)—1 = a—r—s = am+n

as required.
The only other case we need consider in detail is that in which
m > O and n < 0; put n = —s, then emu," = ama—s.
(i) If m > s then amen-3 = aim—Susan‘s = am‘s.
(ii) If m = s then ulna—3 :2 e = am—s.
(iii) If m < s then (emu—s)*1 = aka—7“ = as‘m by (i), and so
“ma—s = (as—m)—1 = am—s_

Therefore in each possible case ama-s = am—B, and so

“man = “ma—s = am—s ___. am+n,

as required. [|
 Basic theorems and concepts 21

We note without proof another result that can be established

by a similar consideration of special cases.
THEOREM 2.09. If a is any element of a group G and if m, n are
any integers then mm)" = am”. [1
Our next task is to equip ourselves with concepts that enable
us to start constructing a theory of groups. So far we have merely
surveyed examples and drawn rather mechanical inferences
from axioms. The idea of ‘homomorphism’ is our starting-point.
DEFINITION. A homomorphism of a group G into a group H
is a mapping ¢ from G to H for which
(mg/MS = WNW)
for all x, y in G.
Note that the product try is formed by means of the multiplica-
tion in G, whereas (x¢)(y¢) is formed in H. It is possible to
illustrate mappings from G to H in diagrams of the following sort.
(x, y) -———> 902/

(96¢, 21¢) —-——> (iZ/H’

WNW)
Here (x, y) is the ordered pair of elements x, y in G, and (96¢, yzfi)
is similarly interpreted. The upper and lower arrows pointing
right represent multiplication in G and in H respectively; thus
multiplication in G maps (as, y) onto mg. The mapping 95 clearly
induces a mapping of GX G that takes (as, y) onto (M), 3145); this
is represented on the left. The right-hand arrow represents 95
itself. Now 95 is a homomorphism if and only if (wy)¢ = (96¢)(y95),
that is if and only if the two paths from (x, y) to the opposite
corner of the diagram have the same ending. This is what is
intended by such statements as that ‘¢ is a homomorphism if it
commutes with group multiplication’.
Example 2.10. Let Z and Zn denote the additive groups of
integers and of residue classes modulo n respectively. Define
22 Basic theorems and concepts

a mapping 4; from Z into Zn by setting 045 = [a] for each a in Z.

Since <a+b>¢ = [a+b] = [a]+[b] = a¢+b¢
by the way in which addition of residue classes was defined, 95
is a homomorphism.
Example 2.11. Let .112 be the multiplicative group of non-
singular 2 X 2 matrices with real entries. We define a mapping 95
fromJ/2 to the multiplicative group R* of non-zero real numbers
by setting Agb = |A|; here A is an element of .112 and lA| is
its determinant. It follows from the well—known fact
IABI = IAIIB I
that 95 is a homomorphism, for

(ABM = IABI = IAHBI = (A¢)(B¢).

THEOREM 2.12. The product of two homomorphism is a homo-
morphism.
Proof. This means that if 95 is a homomorphism from G into H,
and 0 is a homomorphism from H into K, Where G, H, K are of
course groups, then the mapping 5&0 from G into K is a homo-
morphism. Let x, y be arbitrary elements of G. Then
(WNW) = {(wy)¢}9
by the definition of 430;
(903/)95 = (w¢)(y¢)
since 45 is a homomorphism from G into H; and
{(x¢)(y¢)}9 = {(x¢)9}{(3/¢)9}
since 0 is a homomorphism from H into K. Hence we have
(xii/MW) = {(x¢)0}{(?/¢)9} = {w(¢9)}{y(¢9)},
which means that M is a homomorphism as required. I]

THEOREM 2.13. Let ¢ be a homomorphism from the group G

into the group H. If e denotes the identity element in G then e95
is the identity element in H, and if x is an arbitrary element of G
then x—q is the inverse of 90¢ in H.
 Basic theorems and concepts 23

Proof. Note that x = e is the only solution to the equation

x2 = x in G. Since e2 = c, we have (9919)2 = eqS. It follows that
(345 is the identity element in H. Similarly aux—1 = e in G implies
(x¢)(x—1¢) = 695 in H. Since inverses are unique in H we now
have 96‘q = (x¢)—1.|:]
Thus, in a sense, a homomorphism ‘preserves identities and
inverses’. Certainly some (but not all) of the group structure
is preserved; our examples show that a homomorphic image of
an infinite group may be finite, and a homomorphic image of a
non—abelian group may be abelian. Even a homomorphism
from one group onto another may involve some loss; see Examples
2.10 and 2.11. Hence the need for a stronger concept.
DEFINITION. An isomorphism from a group G to a group H is
a homomorphism qS from G to H for which there exists an inverse
homomorphism 5!: from H to G.
These homomorphisms are inverse as mappings, that is, 451/; is
the identity mapping on G and 1]n is the identity mapping on H.
The groups G and H are said to be isomorphic, and we use the
notation G2H

to mean ‘G is isomorphic to H’.

Of course the theorems on homomorphisms above apply to
the special case of isomorphisms. Before further elucidating
the nature of isomorphism, we study some examples.
Example 2.14. Let G be the multiplicative group of complex
numbers {zkzo g k g n—l},

where 2k = cos(2k7r/n)+i sin(2k77/n) and n is an integer greater

than 1, k being an arbitrary integer. (It can be verified that G
is a group.) Let H be Z1, (Example 1.10). Define 96, a mapping
from G to H, by setting 2”) = [10]. Since zkz, = zk+1 by de
Moivre’s theorem, we have
(21021)?5 = (21m)?S = [Ia—H] = [kl—I’ll],
and so qS is a homomorphism; in this calculation k+l may have
to be reduced modulo it but this makes no difference to zk+1-
Next we define 5b, a mapping from H to G, by putting [k]:// = 2k;
24 Basic theorems and concepts

by the previous statement this is a good definition. It is easily

shown that 31/ is a homomorphism. It is plain that 954: and 50¢ are
both identity mappings, and so ¢ is an isomorphism.
Example 2.15. Let G be the multiplicative group of non—zero
complexnumbers and let H be the multiplicative group ofmatrices
of the form ( ab 2), where a, b are real numbers not both zero;

it should be verified that H is in fact a group. Define 9!) by

. a b
(a+rb)¢ = (—b a)’

where a, b are of course real. Then

(a+ib)(c+r£d) = (ac—bd)+t(ad+bc),

{(a+tb)(c+id)}¢ = (32::
333%)

{<a+éb>¢}{<c+td>¢} = (fb :)(_‘fd ‘2) = (3;:d 32%;)-

Hence {(a+ib)(c—|—id)}¢ = {(a+ib)¢}{(c—l—id)¢}, and so gt is a
homomorphism. If 5!: maps (3b 2) onto a+ib then one may
similarly show that it is a homomorphism too, and since 95¢ and
SM are identity mappings it follows that 96 is an isomorphism.
The following result will be of much use, and could indeed
have been used in the study of the examples above.
THEOREM 2.16. The mapping qS from the group G to the group
H is an isomorphism if and only if it is a homomorphism which is
one—one and onto.
Proof. If cl: is an isomorphism then 91> is a homomorphism which
has an inverse. By a general result on mappings,T a given
mapping has an inverse if and only if it is one—one and onto.
Therefore qS is one—one and onto.
Suppose conversely that (t is a one—one homomorphism from
0 onto H. Then 915, as a mapping, has an inverse (/1, by the general
1‘ See the Appendix.
 Basic theorems and concepts 25

result quoted above. If 9!: is a homomorphism, then 96 is an iso-

morphism, so we wish to prove that (mm/1 = (uzfi)(v¢) for any
elements u, v in H. There are unique elements av, 31 in G for which
n = 9695, v = 3/95
since 95 is one—one and onto. Then
(WM/I = {(x¢)(y¢)}¢' = {(xy)¢}¢
since 95 is a homomorphism; next

{(xy)¢}¢l = (WW) = my
since 96¢ is the identity mapping on G. However,
33 = “50: y = ”‘15”
and so we have (marl: = (uz/1)(vgl). Therefore at is a homomor-
phism, and 95 is an isomorphism, as required.[|
At this stage there is an easy proof that neither of the homo-
morphisms in Examples 2.10 and 2.11 is an isomorphism. It is
sufficient to show that neither is one—one. In the case of Z and Zn
we have that (kn)95 = [0] for all integers k, and so 95 is certainly
not one—one. In the case of .112 and R* we have A95 = (—A)95,
a statement which is adequate to show that 95 is again not
one—one.

THEOREM 2.17. Isomorphisrn between groups is an equivalence

relation.
Proof. We refer to the Appendix for a discussion of equivalence
relations in general. It is necessary here to demonstrate reflexive, .
symmetric, and transitive properties for isomorphism.
(i) Clearly the group G is isomorphic to G; the required
homomorphisms 95 and ¢ may be taken as identity mappings
from G to G.
(ii) Suppose 95 is an isomorphism from G to H and let 1/1 be
its inverse homomorphism, then 3!: is an isomorphism from H
to G (note that 95 is one—one and onto). If we interchange G
and H, and interchange 96 and zl, we find that H is isomorphic
to G.
26 Basic theorems and concepts

(iii) Suppose that 451 is an isomorphism from G’ to H and 452

is an isomorphism from'H to K. Then by Theorem 2.12 461452
is a homomorphism from G to K. Further, ¢1 952 is one-one and
onto since both 951 and 962 have these properties. Therefore 451 452
is an isomorphism, and G is isomorphic to K.
Because the relation of isomorphism is reflexive, symmetric,
and transitive, it is an equivalence relation. [:1
The objects of study in a theory of groups are the equivalence
classes of isomorphic groups, and the relations between these
classes. Such a class, being all those groups isomorphic to a
given group, is called an abstract group. Since the isomorphism
relation preserves structure in a certain strong sense, though it
destroys individuality, an abstract group embodies the structure
common to many groups, which may, of course, be quite
different in the nature of their elements. For instance, we have
studied (Example 2.15) a group of complex numbers which is
isomorphic to a group of matrices.
Now many of the examples in Chapter 1 were groups of
mappings. Not every group is a group of mappings—a complex
number is not a mapping, for instance—but it is natural to
wonder if every group is isomorphic to a group of mappings. The
additive group R of reals is isomorphic, in a very natural way,
to the group of mappings described in Example 1.26; the corre-
spondence that maps the real number (1 onto the mapping 95,,
is an isomorphism, as one may verify. This is a special case of
the next result, which is due to Cayley.

THEOREM 2.18. Every group is isomorphic to a group of

permutations of a suitable set.
Proof. The group G’ is given, and we have to produce a suit-
able set. We take the set of elements in G. Next we have to
define permutations of this set, one corresponding to each
element of G. Let a be a fixed element of G’, and define the
mapping pa of G’ into G by setting
xpa = m
for all a; in G.
 Basic theorems and concepts 27

We show that pa is a permutation. Firstly, pa is onto G, for

given b in G we find that pa maps burl onto b. Secondly, pa is
one—one; for if xpa = ypm then xa = get and so at = y.
Now we prove that P 2 {pa : a E G}, with the usual multiplica-
tion of mappings, is a group. -
90(Papb) = (go/rapt = (5mm; = (Mb,
xpab = $011));
hence the associative law in G gives pa pb = pub for all a, b in G.
It follows that P is a group with identity element pe and with
pa"1 = pa_1 for all pa in P.
Lastly we show that G is isomorphic to P. Let qS be the
mapping from G to P such that cat = pa for all a in G. Then 95
is a homomorphism since
(wbfit = Pub = Pa = (a¢)(b¢>);
95 is one—one since if pa = pb then epa = epb, and so a = b; and
clearly 45 is onto P. Hence 96 is the required isomorphism.
Therefore G is isomorphic to a group of permutations of a
suitable set. D
The permutation group P, which has just been put in corre-
spondence with the group G, is called the right regular representa-
tion of G.
We mention without proof that a similar theorem holds for
semigroups—every semigroup with an identity element is iso-
morphic to a semigroup of mappings. (Of course, isomorphism
of semigroups has to be defined formally to make this statement
meaningful.)
Theorem 2.18 shows that a finite group (that is, a group with
a finite number of elements) is isomorphic to a group of permuta-
tions of a finite set. It is an interesting fact that such a group of
permutations is isomorphic to a multiplicative group of matrices.
A minor result is needed in preparation for the proof of this
theorem.
LEMMA 2.19. If the sets X and Y are in one—one correspondence
and G is a, group of permutations of X then G is isomorphic to a
group of permutations of Y.
28 Basic theorems and concepts

Proof. It is intuitively clear that the elements of X may be

replaced by the corresponding elements of Y. It is left to the
reader to supply a formal proof. [I
It follows in particular that if X is a finite set of n elements
and if G’ is a particular group of permutations, then G, regarded
as an abstract group, is independent of the nature of the elements
of X. It is convenient and usual to take X = {1, 2,...,n}, and
to denote the groups 8X, AX of Examples 1.13, 1.15 by Sn, An
respectively in this case. We call 8,, the symmetric group of
degree n, and An the alternating group of degree n.

THEOREM 2.20. Every group of permutations on a finite set of n

elements is isomorphic to a group of n X n non-singular matrices.
Proof. The lemma allows us to choose the finite set of n
elements. For our purpose we choose the set V of row vectors,
or 1 ><n matrices, which have a single entry 1 and the other
entries 0; thus V has n elements. Let these elements be rpm, in,
where v; has 1 in the jth column.
Our given group G of permutations is isomorphic to a group
of permutations of V, by Lemma 2.19. Let p be an arbitrary
element of G; then we may and shall denote the image of ”1
under p as vjp forj = 1,..., n; here we are thinking ofp as acting on
V and equally as acting on {1,..., n}. We now define A(p) to be
the matrix whose jth row is vjp. Thus A(p) is an n>< n matrix
for which
viP) = vjp
for j = 1,...,n. Clearly A(p) is non-singular because its deter-
minant is $1.
Next we prove that the set {A(p):p e G} of matrices forms
a group A under matrix multiplication. For each j we have
vj{A(p)A(a)} = {vjA(p)}A(0‘) = iA(O') = ”WW

0 being an element of G; but further

vjA(PU) : j(pa')’

and since (jp)a = j(pa) for each j we have A(p)A(o) = A(pa).

 Basic theorems and concepts 29

So we have a group of matrices with identity element A(l) and

with A(p-1) = A(p)"1, where L is the identity permutation.
Finally we prove that the mapping which takes p in G to
A(p) in A is an isomorphism. We have already shown that it
is a homomorphism. It is one—one. For suppose thatA(p) = A(a),
then by looking at the defined structure of A(p) we find vjp = v7.0
for each j and it follows that p = a. It is also onto A, by defini—
tion of A.
It follows that G is isomorphic to A, as required. l]
The matrices involved in the proof had an entry 1 occurring
just once in each row and in each column, with 0 elsewhere.
Such matrices are called permutation matrices.
COROLLARY 2.21. Every finite group is isomorphic to a suitable
group of n X n non-singular matrices.
Proof. This follows at once from Theorems 2.18, 2.20, which
ensure that a finite group is isomorphic to a group of permuta-
tions and that the latter is in turn isomorphic to a group of
matrices. [I
Example 2.22. As an illustration of these results we produce
a matrix group isomorphic to 8'3. We regard 83 as the set of
all permutations of V where V = {01, oz, vs} and
01 = (1, O, 0), v2 = (O, 1, 0), v3 = (O, O, 1).
Now apply the method of the previous theorem. The permuta-
tion (12) of {1, 2, 3} corresponds to

()’02 = (10 01 00).

v3 0 0 1
Similarly the permutations (23) and (3 1) correspond respectively
to
01 1

(”3) = (0
l—‘OO

122 0
O

(”2) =
”3
on—ao

0
CO
A

”1 1
 30 Basic theorems and concepts

The permutations (123) and (321) correspond respectively to

02 0 0

COD—I
v3 = 1 ,

l—‘O
01 0
v3 1

l—‘O

l—‘OO
01 = 0 .

v2 0 0
Finally the identity permutation L corresponds to
01 1 0 0
112 = 0 1 0 -
123 0 O 1

We have reached the most appropriate point at which to

introduce another fundamental concept concerning mappings
of groups, namely that of ‘automorphism’.

DEFINITION. An automorphism of the group G is an iso-

morphism of G onto G.
To the uninitiated it may seem paradoxical that a group G
could possess any automorphism other than the trivial one
which is the identity mapping of G onto itself. We therefore
illustrate.
Example 2.23. Let G be any abelian group. Define a mapping
()5 of G into G by putting 9095 = x-l, for all x in G. The abelian
nature of G ensures that 45 is a homomorphism:

(WW = (96:11)“1 = 3/”156‘1 = x‘ly‘l = (90¢)(y95)

by Theorem 2.07. Next we observe that xgbz = x, for all a: in G,
by Theorem 2.05; this means that gt itself is an inverse homo-
morphism to 9b. It follows that St is an isomorphism, and there-
fore an automorphism of G.
Example 2.24. For a more subtle example, we take for G the
symmetric group on some set X; for the sake of definiteness and
simplicity let X = {1, 2, 3} and G 2 8'3. Choose one permutation
p of the set X ; for instance, [3 = (12). A mapping 95 of G into G
 Basie theorems and concepts 31

is defined as follows. If x is an element of G, express it as ((1 I2) 3;)

where {a, b, c} = {1, 2, 3}; then apply p to all the symbols in

((1 i i), obtaining

33¢ = (1p 2p 3p).
ap bp Op
The reader should verify
(i) that 90¢ e G;
(ii) that 95 is a homomorphism; and
(iii) that 95 has an inverse homomorphism.
We omit these details, merely asserting that they prove 95 to
be an automorphism of G. This example might repay some study
and reflection.
Further examples will appear in context later. At present we
prove an important property of sets of automorphisms:
THEOREM 2.25. The set of all automorphisms of a group forms
a group.
Proof. Let G be a group and let A( G) be the set of all auto-
morphisms of G. Multiplication in A(G) is assumed to be the
natural binary operation for mappings; if a, B e A(G) then
90(093) = (96005
for all x in G. Clearly 043 is an automorphism of G, because (as
proved in Theorem 2.17) the product of two isomorphisms is an
isomorphism. The associative law holds, for we are dealing with
products of mappings. The identity element is that which leaves
every element of G fixed, and the inverse of ca is the mapping
a—l, which is meaningful because cc is an isomorphism. [j
We call A(G) the (full) group of automorphisms of G.

Problems
1 . Show that if k and u are positive integers then there is a homomorphism
from Z7”, onto Zn. For What values of k is this an isomorphism?
2. Prove that the symmetric group 8,, on n symbols (n > 2) is homo-
morphic to a group with 2 elements.
32 Basic theorems and concepts

3. The set Hom(G’, A) is defined to be the set of all homomorphisms from

the group G into the abelian group A. Show that if multiplication of
homomorphisms is suitably defined then Hom(G‘, A) is an abelian group.

4. Prove that the set {[1], [3], [5], [7]} of residue classes modulo 8 forms
a multiplicative abelian group.
Find the number of elements in Hom(G’, A) when G is the symmetric
group Sa of degree 3 and A is the above group of residue classes.

5. Let Jig be the multiplicative group of all non-singular 2 x 2 matrices

with real entries. Let T be the group of all mappings 454 of the form
_ a¢x+bi
x95) 0)- x + di ’
where at), b;, 6;, d; are real numbers for which ai d1. 7k b5 05; multiplication
in T IS defined by 906151-951) = ($951092:

for all x. Show that there is a homomorphism from .12 to T. Is it an

isomorphism Y

6. Let G be the multiplicative group {1, —— l, i, —i} of complex numbers,

and let H be the multiplicative group {[1], [3], [5], [7]} of residue classes
modulo 8. Show that G is not isomorphic to H although they are both
abelian and have the same number of elements.

7. Show that the group of congruence mappings of 97’ into itself which
leaves fixed an equilateral triangle (see Example 1.25) is isomorphic to
the symmetric group on the vertices of the triangle.

*8. Prove that the set of all non-singular 2 X 2 matrices with residue
classes modulo 2 for entries forms a group isomorphic to the symmetric
group of degree 3.

9. Show that the additive group O’ of complex numbers is isomorphic to

the group formed by all matrices of the type (fl) 2) with the operation
of addition.

10. Let Sbe the set {x :1: real, 0 < a: < 277-}, and define a: 6—) y to be x+y if
0 g x+y < 277, andx+y—27rif27r < x—l—y < 477-. Prove thatanabelian
group results, and that this group is isomorphic to the multiplicative
group of complex numbers of unit modulus.

11. Find a group of matrices isomorphic to the alternating group A4 on

4 symbols.

*12. Find the full group of automorphisms of the symmetric group 83.
 Basic theorems and concepts 33

13. A one-one mapping 4’ of the group G onto the group G’ is an anti-

automorphz'sm if (xy)q$ = (y¢)(x¢) for all x, y in G. Prove that ¢ is an
anti-automorphism of G if and only if 95 = cult, where am = x‘1 for all m
in G', and ill is an automorphism of G.
*14. Is the multiplicative group of positive rationals isomorphic to the
additive group of rationals?

*15. Show that if A is any abelian group then the mapping 4),, taking
each element of A onto its nth power, for a fixed integer n, is a homo-
morphism.
Discuss Whether (fin is an automorphism or not in the following cases:
mA=a
(ii)A=Z andn=p;
(iii)A=Q.
***16. Show that in any semigroup S the following conditions are
equivalent.
(a) There is an element a such that me = x for all x in S, and for each
such element 6 and for each :1: in S there is an element r1 such that
x’lx = e.
(b) For each a: in 8' there is an element 96—1 such that yam—1 = y for all
y in 8.

858137 D
3

Complexes and subgroups

THE purpose of this chapter is to consider subsets of elements,

in a fixed group, and relations among subsets. As we shall
deal mainly with abstract groups from now on, it is desirable to
make a small change in notation; we shall use the symbol 1
uniformly for the identity element (or unit element, as it may
sometimes be called) of any group in which the binary operation
is named multiplication.
We have already met situations in which one group is con-
tained, as a set, in another. For instance, the integers are a
subset of the additive group of rationals. But not every subset
of the rationals forms a group with the original multiplication.
For instance, the odd integers are a subset of the additive group
of rationals but do not form an additive group in themselves.
DEFINITION. A complex of a group is any subset of elements
of the group. If G is a complex of G we write 0’ E G, in the
usual set notation.
DEFINITION. A subgroup of a group is a complex which itself
forms a group under the original group multiplication. If S is
a subgroup of G we write 8 g G.
There are two obvious subgroups of the group G. There is
G itself, and there is the subgroup (it clearly is a subgroup)
which contains only the identity element 1 of G and which will
also be written as 1. Any other subgroup of G is called proper.
We shall write 8’ < G to mean that S is a subgroup of G and
8 7E G; in other words, either 8 is a proper subgroup of G, or
S = 1 and G 7i 1.
 Complexes and subgroups 35

Example 3.01. As a further illustration we list all the subgroups

of the symmetric group SS on {1, 2, 3}. There are, in addition to
‘33 and 1‘ {1, (123), (132)},
{1, (12)},
{1: (13)},
{1, (23)}-
These are, it will be found, the only complexes closed under
multiplication; and it may be verified that they are in fact
subgroups.

The positive rationals with multiplication are not regarded

as a subgroup of the additive rationals, since they do not form
a group under addition of rationals.
Since we shall find important complexes which themselves
are elements of groups, some consideration must be given to
complex multiplication and its properties.

DEFINITION. The product CD of the complexes 0, D (in that

order) in a given group is {cdzc e 0', d e D}.
Example 3.02. As an illustration we calculate 0'102 and 0103
where 0,- are the following complexes in 6’3:
01 = {(12), (13)}, 02 = {(12), (23)}: 03 = {(13), (23)}-
We find that
0'102 = {(12)(
)(12)),(12)(23), (l3)(12), (13)(23)}
= {1, (,123) (132)},
0103 = {((1)2 (1)3), (12)(23), (l3)(13), (13)(23)}
= {1, (123), (132)}.
The meaning of equations like CD = D0 should be carefully
noted; it is not meant that each element of 0 commutes with
each element of D. The meaning of 0D 9 D0 is that for each
6 in 0 and each d in D there is an element 0’in C’ and an element
01’ in D for which col = d’o’. It will be found that 0102: 0201
36 Complexes and subgroups

in our example above, although (13) in 01 does not commute

with (23) in 02.
The fact that a given complex 0 of a group G is closed under
the multiplication in the group G is neatly expressed by 0'2 E 0.
Note that if 0 contains 1 then both D 9 0D and. D 9 D0 hold
for all complexes D; so if 0 is multiplicatively closed and
contains 1 then 02 = 0. In particular S2 = S whenever S is
a subgroup of G.

THEOREM 3.03. Multiplication of complexes in a given group

is associative.
Proof. Take complexes 01, 02, 03 in a given group. We are
required to prove that
01(02 03) = (0102)03-
Take a: E 01(0'2 0'3), then x = 01(62 (:3) where at 6 0'; for i = l, 2, 3.
By the associative law in the group,
01(02 03) = (010963-
Hence a: e (0102)03, that is,
01(02 03) E (010903-
But since a very similar proof shows that
01(0203) 2 (010903,
we have 01(02 03) = (0102)03, as requiredfl
It follows that a multiplicatively closed set of complexes in a
given group forms a semigroup. Such a set of complexes will
not usually form a group. The fact that the remaining group
axioms cannot all hold in general is shown by Example 3.02, in
Which the complexes 01, 0'2, 03 of 83 are such that 0102 = 0103
and (3'2 75 03.
It is now possible to use expressions like 010203 Without
ambiguity. '
There are special sets of complexes in the group G that do
form groups under complex multiplication. One is the set {G}
consisting of the one complex G. Another is the set of one—
element complexes of G. Any complex 0 in a group G With the
 Complexes and subgroups 37

property that 0'2 = 0 gives a one-element group {0}. A less

trivial example may be constructed in 8'3. Let
0 = {1, (123), (132)},
D = {(12), (13), (23)}.
A little calculation gives
02=D2= O, OD=D0=D.
It follows that {0, D} is a group under complex multiplication,
the identity element being 0.
We mention a theorem that relates the operation of complex
multiplication with the set operations of taking unions and
intersections.

THEOREM 3.04. If 01, 0'2, 03 are complexes of a given group then

(i) 01(02 U 03) = 0102 U 0103;
(ii) (02 U 0:001 = 02 01 U 03 01;
(iii) 01(02 0 03) E 01 02 n 01 03;
(iv) (02 fl 0'3)C'1 9 0'2 01 n 03 01.
Proof. In View of the small use we intend to make of this
theorem we offer only a specimen by way of proof. Take (iii).
Let x e 01(0'2 n 03), so that x = 3/2 where y e 01, z e 02, z e 03. It
follows that yz e 01 02 and 3/2 6 0'1 03, and so at = 3/2 E 01 0'2 n 01 0’3;
that is 01((72 n 03) E 01020 0103, as required. D
Our next theme, a criterion for a complex to be a subgroup,
needs some preparation.
THEOREM 3.05. If S is a subgroup of the group G, then the
identity element of G is the same as that of S, and the inverse of
each element of S is the same in G as in 8'.
Proof. In any group the identity element is the only solution
to the equation x2 = 90. Since the multiplications in S and G
are the same and since the identity element of S is an element
of G for which 962 = x, the identity elements of S and G coincide.
For each x in S the inverse x‘1 in S is clearly the inverse in G,
therefore; so inverses in S and G coincide. [I
38 Complexes and subgroups

If S is a subgroup of G then we could of course construct a

homomorphism (namely the inclusion mapping) from 8, con—
sidered as a group in its own right, into G. In that case Theorems
2.13 and 3.05 become very similar.

THEOREM 3.06. A non-empty complex 0 of a group G is a sub-

group of G if and only if any—1 lies in G for each x, y in 0'.
Proof. It is clear that if 0' is a subgroup containing :3 and y
then y‘1 and so any-1 lie in 0. Note that any subgroup contains
an identity element (in view of the definition of a group) and
is therefore non-empty.
Suppose conversely that 0 is a non-empty complex contain-
ing xy‘l whenever it contains at: and y. In particular, for each
x in 0', the element 9093—1 = 1 lies in 0. Taking x = 1, we
next have ly—l = y—1 lying in 0’. And then, for each x and y
in 0, we have x(y—1)—1 = xy also a member of 0. Therefore 0 is
closed under the multiplication of G, which we are, of course,
using as the multiplication in 0 also.
The associative law for multiplication of a, b, o in 0' holds
because a, b, c are elements of G, in which the associative law
certainly holds. We have shown that if 0 is non-empty then
0' contains 1, which also acts as an identity element in 0.
Finally 96—1, the inverse of x in G, lies in 0 whenever a: does, and
serves as a required inverse in 0. Therefore 0 is a subgroup of
G. [I
Note that if we define 0-1 to be {aszar1 e 0} then the theorem
may be stated succinctly in this way: 0 g G if and only if 0' ¢ 6,
0 S G, and 00—1 E 0.
This theorem makes it easy to verify the following statement:
if g is any element of a group G then the complex
{g": n = 0, i1,...}
is a subgroup of G. We thus have many subgroups, one deter-
mined by each element g of G, though these need not all be
distinct. They are abelian, and we shall have more to say about
their nature later on.
 Complexes amt subgroups 39

Two lines of thought lead to our next batch of results about

relations between complexes and subgroups. In the first place
one may ask whether the usual set operations can be performed
on subgroups with pleasing results; in other words, is the inter-
section or the union of a set of subgroups again a subgroup?
The union of subgroups certainly need not be a subgroup—for
example, {1, (12)} U { 1, (13)} is not a subgroup of the symmetric
group on {1, 2, 3}. This focuses attention on another problem——
what can be said about subgroups containing a given complex ?
When this is answered satisfactorily we can consider subgroups
containing the union of given subgroups, which is itself a
complex, and so hope to find a group—theoretic analogue of
set-theoretic union.

THEOREM 3.07. Let {SAM EA} be a set of subgroups of the

group G, where A138 some index set. Then n SA is a subgroup of G.
AeA

Proof. Let 8' denote fl SA. Note that S is non-empty since

AEA
1 6 SA for each A and so 1 e 8. Take arbitrary elements 91:, g in 8.
Since :13, g belong to each subgroup SA and so ((1—1 belongs to each
SA, we find that cry—1 belongs to each SA. Hence ray-1 is an
element of 8. By Theorem 3.06 it follows that S is a subgroup
of GED
Example 3.08. The additive group Z provides an illustration
ofthis. It may be verified that the complexes {7621' : Io = 0, j: 1,...}
form subgroups, for 2' = 1, 2,... . Their intersection is {0},
again a subgroup.
If 81, 82 are subgroups of a group G then it rarely happens that
81 U 82 is a subgroup of G. In fact this happens if and only if
one of SI, 82 is contained in the other. We prove briefly part
of this statement. Suppose that G has an element a which lies
in 8'1 but not in 82, and take any element a: in 6'2. If ax lies in
82 then a lies in 82, a contradiction; therefore ax lies in 8'1, which
implies that no lies in S1. We have now shown that 8'2 is a subset
of SI, and the required result follows.
40 Complexes and subgroups

It is also exceptional to have the complex 8182 a subgroup

when 8'1, 82 are subgroups of G. Examples to substantiate this
assertion may again be found in the symmetric group on {1, 2, 3}.
However, there is at least one subgroup of G that contains
a given complex 0, for G itself is such a subgroup. Further, the
intersection of all subgroups of G containing 0 is another such
subgroup, for this intersection clearly contains 0 and by
Theorem 3.07 is itself a subgroup. Clearly the intersection is
the smallest subgroup containing 0, in the sense that it is con-
tained in any subgroup that contains 0. This leads to a definition
and a theorem.
DEFINITION. The subgroup generated by a non-empty com-
plex 0 in the group G is the intersection of all subgroups of G
containing 0. It will be denoted by gp{0}. When 0 is finite,
0 = {61"":cn}5

we shall write gp{0} as gp{cl,..., on}, and we shall say that this
subgroup isfinitely generated. There are finitely generated groups
which are not finite; the infinite group Z is generated by the
integer 1, for instance.

Example 3.09. Consider the group Z; (Example 1.09) and

its subgroup generated by 0' where
0' = {61,...,Cn ,
2 . . 2
cm = cosfin—f—rsmp—Z (1 g m g rt),

for some fixed positive integer rt. De Moivre’s theorem gives

05» = cm_1 for each m; thus cm_1 egp{cm}, and by induction
cm egp{cn} for each m. It follows that 0’ S gp{cn}. Then
gp{0} is contained in gp{cn}, by definition of the former. It
should now be clear that gp{0} = gp{cn}.
The above definition does not provide a very satisfactory
practical method of determining the subgroup generated by a
complex. If we are asked to find the subgroup of S3 generated
by {(12), (13)}, the natural approach is to add to this complex
 Complexes and subgroups 41

elements found by multiplication and inversion until a subgroup

1s formed. Thus (12)(12) = 1,

(12x13) = (123),
(13)<12) = (132),
<12)(13><12) = (23);
so we have proved that 83 = gp{(l2), (13)}. We next make this
idea precise.

THEOREM 3.10. The subgroup generated by the non-empty

complex 0 of the group G is the complex containing 1 and all
elements of the form 61 02 on for each n 2 1, where each 0,; lies in
C’ u 0-1.
Proof. By 0‘1 we mean {razor1 e 0}.
Let 0* = {1} U {0102...cnzn 2 1, am. e 0 U 0‘1}. Clearly
0 S 0*, for each element of 0 has the form required of elements
in 0* with n = 1. Further 0* is a subgroup, for it is evidently
a non-empty complex with the property that if x, y lie in 0* then
so do y‘1 and icy—1. Therefore 0* is a subgroup containing 0,
and by the definition of gp{0} we have
gp{0} E 0*-
Now take an arbitrary element a: of 0*. If a: = 1 then cer-
tainly a“ lies in gp{0}. If x = 61 on with each 0,- in 0 U 0’1 then
the fact that gp{0} is closed under the operations of inversion
and multiplication shows that x e gp{0} again. (To be strict,
induction on n is required to show that cl, 6162,..., 61620.6”
all lie in gp{0}.) Hence
0* E gp{0}.
It follows that gp{0} = 0*, as required. I]
We ought formally to note that our labours have shown this:
the group—theoretic analogue of SI U 82, where 8’1 and 6'2 are sub-
groups of a given group, is gp{8’1, 82}.
If we look at the idea of generators for a group from another
angle, a further problem arises. Given a group G, can we find
42 Complexes and subgroups

a complex 0 for which gp{0} = G? We certainly can, merely by

taking 0' = G; that is, by saying that G is generated by the set
of all its elements. But in practice we usually want generating
sets with special properties, such as being finite or not containing
a proper subset that generates G’ (though neither of these
properties can generally be arranged for). We shall do no more
than note that such problems are difficult or impossible in
general, and indicate (without proof) two of the several possible
generating sets for S3:
{(12), (13)},
{(23), (123)}.
The central part in our study of complexes has been played
by subgroups, and it is now time to begin using an additional
tool to investigate group structure.

THEOREM 3.11. If the relation :6 ~ y is defined between the pair

of elements 3:, y of the group G to mean that say-1 lies in the fixed
subgroup S, then x ~ y is an equivalence relation on G.

Proof. (i) Since now—1 = 1 lies in the subgroup S, we have

90 ~ x.
(ii) Supposex ~ 3/; that is, nay—1 lies in S. Then (xy‘lrl = 3/964
lies in S, and so 3/ ~ 90.
(iii) Suppose (I: ~ g and y ~ 2, and so avg/‘1 and (1/24 lie in S.
Then 5024 = (my—1)(yz‘1) lies in S, and we have 2: ~ 2.
Since the relation is therefore reflexive, symmetric, and
transitive, it is an equivalence relation. [I

DEFINITION. A right coset of the subgroup S in the group G

is an equivalence class resulting from the equivalence relation
defined above.
Since this set of right cosets provides a partition of 0, it is
natural to seek an explicit description of right cosets. The
equivalence class containing a is {aura—1 e S}, which may be
written as {90:17 6 8a} or simply as Sa, if the latter is taken to
mean S{a} or {95:96 = sa for some .9 in S}.
 Complexes and subgroups 43

Left cosets may be defined in a similar but obvious way. One

starts with the relation 35 ~ g defined as 27—13] e S, proves this
to be an equivalence relation, defines a left coset as an equi-
valence class, and can show that the coset containing a is aS
with suitable interpretation.
We shall see in Chapter 4 that certain sets of cosets in any
group themselves form a group under complex multiplication.
Let us calculate the right and left cosets of the subgroup S
generated by (12) in the symmetric group on {1, 2, 3}. The right
cosets include the subgroup itself, {1, (12)}; next, S(13) is {(13),
(123)}; and S(23) is {(23), (132)}. We have arrived at a partition
of the group, and so have a complete set of right cosets. It will
similarly be found that the left cosets are

S = {1, (12)},
(13)S = {(13), (132)},
(23)8 = {(23), (123)}-
Note that S(l3) = S(123); any element chosen from a coset
determines that coset.
THEOREM 3.12. Let S be a subgroup, and a and b be elements,
of the group G. Then Set = Sb if and only if a and b lie in the same
right coset of S in G.
Proof. Suppose, in the first place, that Set = Sb. This implies
that a = sub for a suitable element x in S; and so ub—1 e S. By
definition of a right coset, a and b lie in the same right coset of
S in G.
To prove the converse implication, suppose that a and b lie
in the same right coset. We find that orb—1 e S, and a = 90b for
some at: in S. It follows that Su 2 Saab. Therefore we have to
prove no more than Sac = S. Because S is a subgroup we have
S96 9 S; and if y e S then gas—1 E S, as in Theorem 3.06, and so
3/ 6 Sin, proving that S E Sx. Therefore Sm = S.|:]
COROLLARY 3.13. Let S be a subgroup, and a an element, of
the group G. Then S0, = S if and only if a lies in S.
Proof. Take b = 1 in the theorem above. [I
44 Complexes cmol subgroups

Another aspect of Theorem 3.12 should be carefully noted.

Suppose Ed is a right coset of S in G, and suppose that an
arbitrary element 2 is chosen in Sa; then the cosets Sc and Se
coincide. It is sometimes useful to be able to rename cosets
in this way.
We remark that everything above, after the statement of
Theorem 3.12, can be rewritten for left cosets.

DEFINITION. A right transversal of the group G with respect

to the subgroup S is any complex of G containing one and only
one element from each right coset of S in G. Left transversals
are defined similarly. Sometimes a right (or left) transversal is
called a complete set of right (left) coset representatives.

In our example above, one right transversal is {1, (13), (23)},

which is also a left transversal. Another is {1, (13), (132)}, but
this is not a left transversal.
The final aim of this chapter is to study some important
arithmetical conditions satisfied by finite groups. We need some
definitions; these will also be very useful in the study of infinite
groups.

DEFINITION. The order of a, group is the number of elements

it contains. The order of G will be denoted by [G].

The order may be a positive integer or may be infinite. For

example, the order of 8,, is m, while the orders of Z, Q, R are all
infinite.

DEFINITION. The order of an element 9 in the group G is the

least positive integer n such that g” = 1, if n exists; 9 has
infinite order if such an integer 71, does not exist. The order of y
will be denoted by |g|.

For instance, the order of the element (123) in 8'3 is 3 ; the order
of the number 2 in the multiplicative group of positive rationals
is infinite. An element of a group G has order 1 if and only if it
is the identity element, clearly.
 Complexes and subgroups 45

DEFINITION. A periodic group is a group in which every

element has finite order.
Every finite group is clearly periodic, but there are also
infinite examples. The reader will find it easy to show that the
permutation group (on the positive integers) generated by
{(12), (34), (56),...} is infinite and periodic.

DEFINITION. A torsionfree group is a group in which every

element, except the identity, has infinite order.

It is easy to verify that Z, Q, R are all torsionfree.

The next result shows that the two meanings given to ‘order’
by our definitions are sensibly related.
THEOREM 3.14. Let g be an element of a fixed group. Then
(i) gm = l, where m is a positive integer, if and only if the
order of g divides m;
(ii) the order of g equals the order of the subgroup generated by 9.
Proof. (i) Suppose that gm = 1 and that g has order n. By
the division algorithmT there are integers u, v for which
m = un+v and 0 < v < n. Therefore, by Theorems 2.08 and

2'09’ 1 = gm = we = (WWW-
Since g has order n we conclude that g” = 1. But since u is the
minimal positive integer for which 9” = 1, and 0 g 1) < n, we
have 2; = 0; that is, n divides m.
Now suppose, conversely, that n divides m, with n the order
of 9. Then m = un for some integer u, and

9’" = 9“” = W)“ = 1;

this gives gm = 1 as required.
(ii) In this part of the proof it is possible for orders to be
infinite. Suppose first that g has finite order n. The elements
g0, g1, g2,..., 9”"1 are all distinct, for ifg7 = g8 withn > r > s 2 0
then 97—3 = 1, and by (i) n divides r—s, which is impossible.
1' See the Appendix.
46 Complexes and subgroups

N0W take any power gm of g. Since by the division algorithm

there exist integers u, v With m = un+v and 0 < v < n, we

have gm = 9W = (W9. = g".

Therefore {gmzm = 0, 31:1,...} = {gmz0 < m < it}.
But since this set is the subgroup generated by g, we have proved
that the subgroup has precisely rt distinct elements, as required.
Finally suppose that g has infinite order. If gr 2 g3 with r > s
then an argument used above shows that g has finite order
dividing r—s, a contradiction. Therefore gp{g} is infinite,
as required. [I
It is important to note that g” = 1 does not imply that g has
order u. If g = (123) in the group 8'3 then g6 = 1, but g has
order 3, not 6.
Theorem 3.14 is often used to prove that two elements have
the same order. Take inverse elements a and a—1 in any group,
for instance. If a has order n then a" = 1, and so (em—1 = 1, and
so (u—1 n = 1; all we can conclude at this stage is that a—1 has
finite order dividing rt. But another application of the same
argument shows that the order of a divides the order of 05-1,
and it follows that they are equal. It should be clear too that a
has infinite order if and only if (171 has infinite order.
The theorem that follows is fundamental in the theory of
finite groups. It is due to Lagrange.
THEOREM 3.15. The order of a subgroup of a finite group divides
the order of the group.
Proof. Let S be a subgroup of the finite group G'. We have
seen that the right cosets of S in G give a partition of G. But
each right coset contains rt elements, if n is the order of S; for
there is a one—one correspondence between S and 8a. (This may
be set up by making you in So correspond to x in S; if xu = ya
then clearly a: = y.) Thus if there are k right cosets then the
order of G’ is 16ml]
COROLLARY 3.16. The order of an element of a finite group
divides the order of the group.
 Complexes and subgroups 47

Proof. We need remark only that the order of an element

equals the order of the subgroup that it generates, by Theorem
3.14. I]
The following converse of Lagrange’s theorem is false: a
group of finite order n contains a subgroup of order m for each
factor m of n. It will be found, for instance, that the alternating
group A4 on four symbols has no subgroup of order 6.
Another consequence of Lagrange’s theorem is that the
number of left cosets of S in G is the same as the number of
right cosets when G is finite. A little preparation is needed to
generalize this satisfactorily to infinite groups.

THEOREM 3.17. If T is a right transversal for the group G with

respect to the subgroup S then T‘1 is a left transversal in the same
situation.
Proof. By T"1 we mean, as usual, {xzx—l e T}.
Let x be an arbitrary element of G. Then 90—1 may be expressed
as st, with s e S, t = T. Therefore a: = t—ls‘1 e t‘lS, since S
is a subgroup. Each element of G then belongs to some left
coset t—lS where t e T. Now suppose that tl‘lS = tg‘lS where
t1, t2 lie in T. Then tfl = tz‘ls for some 8 e S, and tltz‘1 E S.
This implies that t1 = t2 because T was a right transversal. So
we have a partition of G into left cosets, and T—1 is a left trans-
versal.[:|

COROLLARY 3.18. The right cosets of S in G are in one—one

correspondence with the left cosets of S in G. D

DEFINITION. The index of the subgroup S in G is the number of

(right or left) cosets of S in G. This index will be denoted by I G: S | .

Note that the last result shows that in calculating an index

it makes no difference whether we work with right cosets or left
cosets. If G has finite order then its order is the product of the
order of S and the index of S in G.
Some facts about subgroups with finite index will now concern
us. These facts may be of no great significance in themselves,
48 Complexes and subgroups

but they are interesting exercises on the previous theorems and

they will be very helpful in the later chapters.

LEMMA 3.19. Let A and B be subgroups of the group G. If a:

and y are elements of G for which Ax n By is non-empty, then
Ax n By is a right coset of A n B in G.

Proof. Let 2 E Am 0 By. Since .2 G Am we have Ax = Az, by

Theorem 3.12 and a remark following; similarly By = Bz.
Therefore Ax 0 By = A2 n B2. It is easily verified that
Azn B2 2 (A n B)z, and we prove the reverse inclusion as
follows. If 20 E A2 0 B2 then 20 e Az and 20 e Bz. Therefore
zo = az, 20 = bz
for suitable a in A and b in B. Clearly a = b, and clearly this
element lies in A n B. Therefore 20 e (A n B)z, as required.
It follows that
Aay=Aaz= (AnB)z.[|

THEOREM 3.20. Let A and B be subgroups of the group G with

finite indices m, n respectively. Then
(i) A n B has finite index in G', bounded by mn;
(ii) if m and n are coprime then A n B has index run in G.

Proof. (i) There are in right cosets of the form A91: and n of
the form By. It follows that there are at most mn non-empty
intersections Ax n By. The lemma above shows that the set of
such non—empty intersections is just the set of right cosets of
A n B in G. Therefore A n B has finite index in G, and mu is
an upper bound for this index.
(ii) When m and n are coprime we denote the index of A n B
in G by k. We know that h g inn. Since
AnAgG, AnB<B<G,
the following equalities are obvious:
[GzA n B] = |G:A]|A:A n B],
lG:A n Bl = IG:B][B:A 0 Bl.
 Complexes and subgroups 49

The former implies that m divides lo, and the latter that n
divides k. We conclude that mn divides k because m and n are
coprime. This fact together with k g mn shows that k = mn. [I
Note that in case (ii) of the above theorem Ax n By is never
empty. The reader might like to prove that in general Ax n By
may well be empty, implying that the coprime condition in
case (ii) cannot be abandoned.
The determination of all subgroups of a given group G, or
even some description of their properties, is in general an
unreasonably difficult problem. This is so even when G is
generated by two elements. But the case when G is generated
by one element is surprisingly easy to discuss.

DEFINITION. A cyclic group is a group generated by one of its

elements.
Thus if g generates G then G = {9”e = 0, il,...}. Clearly
all cyclic groups are abelian. There are cyclic groups of every
finite order and of infinite order, for Zn and Z are cyclic. Each
element of an arbitrary group G clearly lies in, indeed generates,
a cyclic subgroup of G.
THEOREM 3.21. Every subgroup of a cyclic group is cyclic.
Proof. Let S be a subgroup of the cyclic group G with genera-
tor g. The elements of S are certain powers of g. If S is not the
subgroup containing only one element then g)‘ e S where A 32$ 0;
since 8 is a subgroup we may take A > 0. Let A be the minimal
positive integer such that g" E S, and let 9" be an arbitrary
element of S. By the division algorithm p. = uA—I—v Where u, v
are integers and 0 g o < A. Therefore
9" = 9"(9")-“,
and it follows that g” e S. The minimal property of A now shows
that '0 = 0. Therefore A divides [.L, and gl‘ is a power of 9"; this
proves that S is cyclic with generator 9". E]
There is usually more than one generator of a cyclic group,
and we offer some examples of this situation in place of an
exhaustive discussion. The additive group of integers is
853137 E
50 Complexes and subgroups

generated by 1 or by —1, but by no other integer. The group

Zn is generated by [1]. The group Z12 is generated by [1] or
[5] or [7] or [11]—it is clearly generated by [1] and hence by
the others as
5[5] = [1], 7[7] = [1], 11[11] = [1].
However, [3] is not a generator, for there is no integer Ic for which
k[3] = [1]; if there were, then 370—1 would be divisible by 12
and so by 3, a contradiction.
Further properties of cyclic groups are presented among the
problems which follow.

Problems
1. Show that the residue classes [2n+ 1] modulo 16, Where 0 < n g 7,
form a multiplicative group, and find all its subgroups.
2. Find the order of the subgroup generated by the following complex
of the symmetric group So:
{(1234), (14x32), (56)}-
3. Prove that a group is finite if and only if it contains only finitely many
subgroups.

4. Show that the alternating group on five symbols has no subgroup of

index 2, 3, or 4, and find a subgroup of index 5.

5. Let 9 be the multiplicative group generated by the matrices A , B, Where

1 0 1 1
A — (0 —1)’ B — (0 -1)-
(i) Show that A and B have order 2.
(ii) Prove that every element of 9 except I can be written in the form
01...0,, for some n 9 1, with each 0.; = A or B, and 0,- ;E Oi+1 for
13 = 1, 2,..., n— l. Prove that no such element is the identity, and deduce
that AB has infinite order.

6. Find examples to prove that each of the following statements about

the group G' is false.
(i) If every element of G has order dividing n then G has order
dividing n.
(ii) If G is finite and every proper subgroup is cyclic then G' is cyclic.
(iii) If G has finite order divisible by n then G contains an element of
order n.
 Complexes and subgroups 51

(iv) If G contains a complex 0 satisfying the conditions 02 = 0' and

1 e 0, then 0 is a subgroup of G.
(V) If 81, 5'2, 8'3 are subgroups of G then 8105’2 n SS) = 8182 n 3183.

7. Let 0 be a non-empty complex of the periodic group G. Prove that

the followi11g conditions on 0 are equivalent:
(i) 00—1 E 0;
(ii) 02 = 0';
(iii) 02 E C'.
8. Which of the following groups are cyclic? Which are finitely
generated?
(i) The multiplicative group {[n] :n is not divisible by 5} of residue
classes modulo 25.
(ii) The additive group of rationals.
(iii) The additive group of reals.
(iv) The multiplicative group of positive rationals.
(v) The multiplicative group Z? of complex numbers (Example 1.09)

9. Let S be a non-empty subset of the group G, and define the relation

as ~ 3/ between elements as, y to mean that nay—1 e S. Prove that if this is
an equivalence relation then S is a subgroup of G.
Another relation in G is obtained by defining :76 ~ 3/ to mean that
Sm = Sy. Show that although this is an equivalence relation there are
cases in which S is not a subgroup of G.

10. Let a, b denote the permutations (12), (12...n) respectively in the

symmetric group Sn on n symbols; let n > 2. Prove that
b—‘abi = (i+1,i+2) for «I = O, l,...,n—2.
By showing that every transposition can be expressed as a suitable
product of such transpositions (75+ 1,12—l— 2), prove that S" = gp{a, b}.
*11. Let g be the multiplicative group generated by the matrices
_ 1 1) _ (2 0)
A — (0 1 ’ B — 0 1 '
Prove that BAB‘1 = A2, and deduce that every element of g may be
written in the form BuABBv for suitable integers oz, ,8, y.
*12. Let G be a finite cyclic group of order n. Prove that if m is any
factor of n then G contains one and only one subgroup of order m.
Deduce that a group has no proper subgroup if and only if its order is
prime or 1.

*13. The complex 0,, where r is rational, in the (additive) group R of

reals, is defined by 0 __ {was > 1.}
T — ' '
52 Complexes and subgroups

Prove that 0, 03 = Cr+m Where 1', s are rational, and complex multiplica-
tion is used. Deduce that the set {0,: 'r e R} forms a group under complex
multiplication.

**14. The finite group G contains a set of complexes that forms a group
under complex multiplication. Prove that each complex contains the
same number of elements of G. Prove further that, if the complex 0'1
acts as the unit element of the complex group and 2: lies in the complex
0, then 0' = 0001 = 01x.
Deduce that if G‘ is cyclic then the complex group is cyclic.
4

Normal subgroups and factor groups

IN this chapter we continue to assemble the stock of elementary

concepts and tools necessary for deep penetration into the theory
of groups. The key idea is that of a normal subgroup, whose
immediate justification is that it enables us to solve the following
obviously important problem: given a group G, what groups can
occur as homomorphic images of G? That is to say: what de-
scription can be found for each group H for Which there exists
a homomorphism from G onto H ?
DEFINITION. The subgroup N of the group G is normal if
xN = N96 for each element x of G. The fact that N is a normal
subgroup of G is expressed symbolically thus: N <1 G. If
Na! GandN¢G,writeN<l G.

It should be noted with considerable care that the equation

xN = Nx does not assert that every element of N commutes
with 91:. What is intended is that {x} and N shall commute as
complexes. In other words, the left coset xN is identical with
the right coset Nx. This condition is equivalent to xN g Na: and
xN 2 N9:: simultaneously; the former of this pair means that,
given 3/ in N, an element 2 (depending on 3;) exists in N such that
xy = 22:, and the latter is to be similarly interpreted.
An immediate consequence of the definition is the fact that
SN = NS for any subgroup S of the group G. (Indeed this
equation holds if S denotes any complex of G.)
It is often helpful to use the fact that the subgroup N is normal
in G if112‘l = N for all x in G; this is another one-step corollary
of the above definition.
54 Normal subgroups and factor groups

The normal subgroups 1 and G are present in every group G,

for which reason they are often called the trivial normal sub-
groups. It should be clear that wary subgroup of any abelian
group is normal. Some less trivial examples follow.

Example 4.01. Consider the symmetric group 83 on {1, 2, 3}

once again. We assert that {1, (123), (132)} is a normal subgroup.
The non—trivial part of the proof of this is to verify the following
statements:

(12){1, (123), (132)} = {(12), (13), (23)} = {1, (132), (123)}(12),

(13m, (123), (132)} = {(13), (23), (12)} = {1, (132), (123)}(13),
(23){1, (123), (132)} = {(23), (12), (13)} = {1, (132), (123)}(23).
There is a more conceptual proof that the subgroup in question
is normal, and it can be put in a quite general setting. Suppose
that G is any group with a subgroup S of index 2. Take any
element x in G but not in S. Then S r) Sm = 9 while G = {8, 89;}
by properties of cosets, and it follows that the elements of 890 are
precisely the elements of G which are not in 8. But a quite
similar argument shows that these elements are just the set 968.
Therefore Sm = .768. That is to say, any subgroup of index 2
is normal in G. In particular, the alternating group An is normal
in the symmetric group 8,, because it has the required index.
Our example above was the case n = 3.
It may be easily shown that S3 has no normal subgroup of
index 3.

Example 4.02. Consider .112, the multiplicative group of all

non-singular 2X 2 matrices with real entries. Let JV be the
subset of all matrices of unit determinant. It should be clear
that JV is a subgroup of .112, and we give a few words of proof to
indicate that it is normal. Let A be a matrix in JV, and let X
be any matrix in .112; since

IX‘IAXI = [XI—IIAHXI = IAI = 1

it follows that X—IAX is in JV. Hence X—lJVX 9 JV. This
 Normul subgroups and factor groups 55

argument, with X replaced by X—l, gives X./VX—1 S JV. We

now have the relations
X‘IJVXEJV, MEX—luVX;
it follows that X—1./VX = ./V for all X in M2. Hence ./V is a
normal subgroup of Jig.
It is perhaps worth stating and noting carefully that, if the
subgroup S of the group G satisfies the condition 96—1895 9 S for
all x in G, then S is normal. The useful trick (exhibited in
Example 4.02) of replacing an by 96—1 will establish this assertion.
DEFINITION. The kernel K of the homomorphism 95 from the
group G into the group H is the set {xzx E G, 9095 = 1}.
As an illustration we indicate the kernel of the homomorphism
96 from .112 to R* described in Example 2.11. If A e .112 then
A95 = |A[; therefore the kernel consists of those matrices A
which have determinant 1. As shown above in Example 4.02,
such matrices form a normal subgroup of .112.
THEOREM 4.03. The kernel of a homomorphism from a group G
is a normal subgroup of G.
Proof. Let 96 be a homomorphism from the group G into the
group H. We show first that the kernel K is a subgroup. As
1 6K, clearly, K is non-empty. In order to apply the usual
subgroup criterion we take arbitrary elements as, y in K. Thus
$95 = l, q = 1. Then
(9631—99ls = (x¢)(y¢)‘1 = :
since St is a homomorphism. The consequent fact that 9631—1 e K
implies that K is a subgroup of G, by Theorem 3.06.
Next we have to show that K is a normal subgroup. Let k, g
be arbitrary elements of K, G respectively. Then, since 915 is a
homomorphism,
(9‘1k9)¢ = (9¢)‘1(k¢~)(9¢) = (9¢)‘11(g¢) = 1.
Thus grllcg e K. We have proved that g—lKg E K for all g in G,
and this suffices to establish that K is a normal subgroup of G. 1:]
We now investigate cosets of a normal subgroup.
56 Normal subgroups and factor groups

THEOREM 4.04. The right cosets of the subgroup S in the group G

form a group under complex multiplication if and only if S is
normal in G.

Proof. Suppose S is normal in G. What is the product of the

cosets S2: and Sy? It is the complex SxSy. But note that the
relation xS = Sx allows us to write this as S2953], or as Sony (by
the subgroup property of S). Therefore if S is normal then we
have SxSy = n
for all x and g in G.
Verification of the group axioms for the set of cosets is now
easy. Proof of associativity is standard. The rest follows from
the above equation—closure is obvious, the identity element
is the coset S, and the inverse of the coset S95 is Sx—l. Thus the
cosets of S in G form a complex group.
To prove the theorem fully we have to deduce normality of S
in G from the fact that the cosets of S form a complex group.
This fact, though not essential to the development of the subject,
is none the less of some interest.
We are supposing that corresponding to each pair as, y of
elements in G there is an element 2 for which
SxSy = Sz.
Since 1 lies in S there is an element sin S for which
my = lxly = sz.
Therefore SxSy = Say,
since Ss = S, and so Ssz = S2. Take 3/ = 1:
SxS = Sac.
We may deduce from this (since 1 e S) that 90S E Sx for all x
in G. At this point the customary substitution of x—1 for ac
gives Sm S xS, and it follows that S is a normal subgroup of G. [I

DEFINITION. The factor group (or quotient group) of the group

G with respect to its normal subgroup N is the group formed by
the cosets ofN in G with the operation of complex multiplication.
 Normal subgroups and factor groups 57

We denote this factor group by the symbol G/N. A factor group

G/N is called proper if N 7E G and N 75 1.
Note that if G is any group then the order of G/N is the index
of N in G, by the definitions. We stress again the nature of the
factor group G/N of G by pointing out that G/N is a set of sets
of elements of G, and not a subset of G.
Example 4.05. Let Z be the additive group of integers and let
N” be the subset consisting of all multiples of the integer n,
which by the way we shall suppose to be greater than 1. It is
easy to see that N” is a subgroup, and since Z is abelian N” is
normal. The coset of NW containing the integer a, will be denoted
(in additive notation) by Nn—l—a. It is, explicitly, the set
{a+kn:k = 0, il,...}, or
{..., 00—12, a, a—l—u, a+2n,...}.
This coset is a typical element of the factor group Z/Nn. A part
of the definition of factor group is that the sum of Nn—l—a and
Nn+b is (Nn+u)+(Nn—|—b), which is, of course, Nn—l—(a-l—b).
This description is reminiscent of the group of residue classes
modulo n, earlier denoted by Zn. It is plain from the definitions
that Z/Nn is indeed isomorphic to Zn. This throws some light
on the reasons why we attached importance to Zn as a group
example.
Example 4.06. What is the group G/N when G = 83 and
N = {1, (123), (132)}, in the usual notation? We proved
earlier that N is a normal subgroup. The elements of G/N are
N and Na: where N90 = {(12), (13), (23)} and x is any one of the
elements (12), (13), (23). The multiplication in G/N is easily
described since N2 = N;
N.N=N, N.Nx=Nx,
Nx.N=Nx, Nx.Nx=Nx2=N.
It follows that the group G/N has order 2. We mentioned this
very example in Chapter 3, in connection with groups resulting
from multiplication of complexes, though it could not then be
called a factor group.
58 Normal subgroups and factor groups

We have now reached a position in which we can relate

homomorphisms and normal subgroups by means of factor
groups. We remarked in the course of the proof of the previous
theorem that if N is a normal subgroup of G then
NxNy = n
for all x, y in G; in terms of homomorphisms, this is just the
statement that the mapping d), defined by $45 = Nx, is a homo-
morphism from G onto G/N. The kernel is {sx = N}, since
N is the unit element of G/N; in other words, the kernel is N.
This mapping is called the natural homomorphism from G onto
G/N. The fact that a factor group defines a homomorphism has
the following significant converse.

THEOREM 4.07. (First isomorphism theorem). Let K be the

kernel of the homomorphism qS from the group G onto the group H.
Then H is isomorphic to G/K.

Proof. We remark first that the factor group G/K exists

since the kernel K is a normal subgroup of G, as shown in
Theorem 4.03. Note that the homomorphism ¢ is required to
be onto H.
We try to define a mapping ¢ from G/K to H by specifying cat
as the image of the element Kw of G/K, where x is an arbitrary
element of G. (w = 95¢.

It is of course necessary to verify that (Koch/I is independent of

the choice of the representative at: for the coset Kw. Suppose
then that Karo: Km, that is, 960 = km for some k GK. The
proposed definition would give

(Kxofifi = “’0 ‘15-

But since It is in the kernel K of the homomorphism 95 we have

$095 = (709% = (7695) (96919) = 33¢,

(K9085: ()Kxo 81‘-
Thus the proposed definition of (fl is acceptable.
 Normal subgroups and factor groups 59

Next we show that zfi is a homomorphism. Take 91:, y in G. Then

(KxKykl' = (KM/W = (my)?S
since K is normal and xK = Kw. On the other hand

(Kath/«KW = (x¢)(y¢) = WINS

since 95 is by hypothesis a homomorphism. Therefore
(KxKyM = (K96)¢(Ky)sln
and ¢ is a homomorphism from G/K into H.
We hope to show, of course, that 1/: is an isomorphism from
G/K onto H. To this end it is enough to prove that 1/1 is both
one—one and onto, by Theorem 2.16. Suppose
(K9030 = (IQ/NI,
and let us deduce that Kw = Ky. We have at once $95 = W.
Since 45 is a homomorphism

(mg/“U95 = (x¢)(y¢ ‘1 = 1,
which implies that Qty—1 6 K. It follows that Kx = Ky, and so x];
is one—one. Finally we show that zfi is a mapping onto H. Since
95 is known to be onto H we see that an arbitrary element of H
has the form 9695, for a suitable element x of G. Clearly K96 maps
onto and) under 1/}. Hence 1/; is onto H.
This proves that 1/1 is an isomorphism from G/K onto H, and
the theorem is proved. I]
COROLLARY 4.08. The homomorphism qrom G to H is one—one
o'fand only ifK = Ll]
This remarkable theorem allows us to determine all the
homomorphic images H of G, up to isomorphism. A typical
image H is found by taking a normal subgroup N of G and
forming G/N, since we have displayed a one—one correspondence
between the homomorphic images of G and the factor groups of G.
In the most general situation of this sort we have a homo—
morphism 95 of G into (but not necessarily onto) H. The image
GqS of G under ct is, of course, a subgroup of H. Suppose we
regard G as given and we seek information about H. Then, on
60 Normal subgroups and factor groups

the basis of the results we already have, we can say that H has
some subgroup which is isomorphic to some factor group of G'.
Example 4.09. There is a homomorphism ¢ from the additive
group Z of integers onto the additive group Zn of residue
classes modulo n; it is defined by (145 = [a] for all a e Z. The
kernel is {a : [a] = [0]}, that is, the set of all multiples of n. This
is a normal subgroup Nn of Z, and we have already seen that
Z/Nn g Zn. The existence of this isomorphism is, of course,
just what we would expect from the first isomorphism theorem.
The remaining ideas of this chapter will be arranged rather
haphazardly around the ideas of normal subgroup and homo-
morphism. They are still ofimportance, but not ofthe same order
of difficulty as some fundamental ideas yet to appear.
One natural line of inquiry is concerned with how far from
being normal a subgroup can be. The subgroup N of the group
G is normal if and only if 96-l takes only one value, namely N,
as x varies in G. We define a ‘normal complex’ in a similar way.
For a general complex we have the concept of conjugates,
defined as follows.
DEFINITION. A conjugate of the complex 0' in the group G is
any complex of the form $4096 for some element x in G. We
shall write 90-1096 as 0’”.
For the most part we shall consider either one-element
complexes or subgroups. In the former case we identify the
complex {3/} with the group element 3/ which it contains, and we
write x-lyx as y”. The number of conjugates of a given subgroup
is a rough measure of how near to or far from being normal that
subgroup is.
Example 4.10. Take 8'3, as usual, and ask what the conjugates
of the element (123) are. It will be seen that
(12)—1(123)(12) = (132),
so (132) is certainly one such conjugate. Since conjugate ele-
ments (clearly) have the same order, and since 83 has only two
elements of order three, (123) has no further conjugates. The
 Normal subgroups and factor groups 61

subgroup gp{(123)}, being normal, has only one conjugate,

namely itself. It will be found, however, that the subgroups
generated by the elements (12), (13), (23) respectively are
conjugate in pairs. The sort of calculation needed to prove this is:
(123)—1{1,(12)}(123) = {1, (23)}.
Suppose we define the relation 0' ~ D between complexes 0'
and D in a fixed group G to hold if and only if 0' is conjugate to D.
The proof that this is an equivalence relation is surely easy
enough for the reader to supply. The resulting equivalence
classes are called conjugacy classes. Such classes of elements and
subgroups may be investigated and exploited by means of
normalizers.
DEFLN’ITION. The normalizer of the (non—empty) complex 0'
in the group G is the complex {acme-1095 = 0'} of G. This nor-
malizer will be denotedT by N( 0’).
We note that it is not necessary for 0 and N( 0) to commute
element by element—only commutation of these as complexes
is required. The normalizer of a one-element complex is often
called the centralizer of that complex.
Example 4.1 1. The normalizer of the subgroup of S3 generated
by (123) is 8'3 itself, since the subgroup is normal. Some calcula-
tion will show that the subgroup generated by (12) is its own
normalizer. The centralizer of the element (123) can easily be
shown to be {1, (123), (132)}, and the centralizer of ( 12) is {1, (12)}.
THEOREM 4.12. The normalizer N( 0) of the (non-empty)
complex 0 in the group G is a subgroup of G. If T is any right
transversal of N( 0') in G then the distinct conjugates of 0 in G are
{t-lOt: t e T}.
Proof. It is easy to prove this theorem by routine arguments,
and we leave the reader to do so in detail. The proof that we
1‘ In this book mappings are written on the right, as a rule. Consistency
therefore suggests UN for the normalizer of 0. However, in this and other
instances, the break with tradition and habit seems too great to give consis-
tency its due.
62 Normal subgroups and factor groups

shall give here is more abstract; perhaps it will give additional

insight.
We define an equivalence relation in G as follows. If a, b e G
then a ~ b means that
a—10a = b—10’b.
The verification that this is an equivalence relation is almost
trivial. The equivalence class [1] is precisely N( 0); and if
a ~ b then ax ~ bx.
The required result will then follow from a theorem which we
state separately, as follows. I]

THEOREM 4.13. Let an equivalence relation, N, be defined in

a group G, and suppose that if a ~ b then an ~ bxfor all a, b, x
in G. Then [1] is a subgroup of G, and the equivalence classes
are the right cosets of [l] in G.
Proof. We use Theorem 3.06 in showing that [l] is a subgroup.
Clearly [I] is non-empty, as 1 ~ 1 and 1 e [1]. Let x, y e [1];
thus a: N 1, g N 1.

On multiplication by 31—1, the latter gives 1 ~ g—1 and so

y_1 N 1. Then will—1 ~ 1y_1 N y_1 N 1,

that is gay—1 6 [1]. Therefore [1] is a subgroup.

Because a ~ b implies ab—1 ~ 1 the equivalence classes are
precisely the right cosets of [1], by definition of right coset.[|
The fact that N( 0) is a subgroup allows us to say that N( 0') is
the largest subgroup of G in which 0 is normal, provided 0 itself
is a subgroup. Two consequences of Theorem 4.12 are worth
recording.

COROLLARY 4.14. The number of conjugates of 0 in G equals

the index of N(0) in G. [I

COROLLARY 4.15. The number of conjugates of 0’ in the finite

group G divides the order of G.
Proof. Use Corollary 4.14 and Lagrange’s theorem 3.15.|:|
 Normal subgroups anal factor groups 63

THEOREM 4. 16. If G is a group with a proper subgroup offinite

index then G has a proper normal subgroup of finite index.
Proof. Let S be a proper subgroup of G and consider its
normalizer N(S). Since N(S) 2 S, the index [G:N(S)| divides
1G: 8' I. At all events IG:N(S)| is finite, and so 8' has only finitely
many conjugates by Corollary 4.14.
Let N denote the intersection of all these conjugates. Then N
is a subgroup of finite index in G, by Theorems 3.07 and 3.20.
We now prove normality. Let n E N and a: e G. It is clear that
n lies in every conjugate of S, and it follows that n35 has the same
property. Therefore n” e N, and so N2 g N. This suffices to
show that N <1 G. Clearly N is a proper subgroup of G, like 8. [I
This theorem tells us nothing new about finite groups, of
course; there we may always take N = 1. It is non-trivial for
infinite groups.
Associated with each class of conjugate complexes in the
group G there is a homomorphic image of G. Let {0AA e A} be
a complete set of conjugates, A being a suitable index set. The
arbitrary element as of G defines a mapping of the set {0)} into
itself, the image of 0,\ being 03. This mapping is one—one and
onto; for if 03f = 0:2 then 0A = 0‘“, and clearly the complex
0%” maps onto 0A- Thus each group element x determines a
permutation of {CA}- We now define a mapping 95 of G into the
symmetric group on the set {at} by stating that 9095 is to be the
permutation corresponding to the element x of G. Then gt is
a homomorphism, since
03?” = 314064639631 = 31403.41 = (0%)”-
The following characterization of normal complexes in terms
of conjugate elements is often useful.
THEOREM 4.17. The (non—empty) complex 0 of the group G is
normal if and only if it is the union of complete conjugacy classes
of elements of G.
Proof. Suppose C' is a non-empty complex and is also the
union of certain conjugacy classes; we mean, of course, the set—
64 Normal subgroups and factor groups

theoretic union. This implies that if x lies in 0 then so does x”

for all y in G. Thus y‘1 0y 9 0, which suffices to show that 0 is
normal.
0n the other hand, it is clear that if x lies in 0 and 0 is normal
then no” lies in 0 for all y in G. Thus 0 contains all conjugates of
each of its elements, and is the union of complete conjugacy
classes. I]

COROLLARY 4.18. The subgroup N of the group G is normal

if and only if it has a set of generators which is the union of complete
conjugacy classes of elements of G.

Proof. Suppose N is a subgroup with a set of generators of

the kind specified. An arbitrary (non—identity) element x of N
may be represented as so? 96;" where 901,..., 23,, are some of these
generators and e7; = j: 1 for l < i g n; this is possible by Theorem
3.10. Note that
(ab)x = x—labx = x‘lax.x—1ba: = axbw,
(a-1)$ = xvlal = (ac—lax)-1 = (om—1;

these facts may be extended by induction to establish that

(x? wit)” = (0010‘1 (96%)“.
But (by hypothesis) mg! lies in the given set of generators for N,
whatever element of G is chosen for y. Hence y‘lxy e N, and
Theorem 4.17 shows that N is normal in G'.
The converse statement, which is almost trivial, may be dis-
missed with the remark that if N is a normal subgroup of G then
the set of all elements of N is, by the theorem, a set of generators
of the type sought. [I
We turn from conjugate elements to conjugate subgroups.

THEOREM 4.19. If S is any subgroup and x is any element of

the group G’ then S90 is a subgroup isomorphic to 8'.

Proof. We are not excluding the case in which S and Sac are
identical; this will happen precisely when x is in N(S). In
 Normal subgroups and factor groups 65
general, let y, 2 be elements of 8', so that girl 6 S. Then general
elements of 8"” will be xrlyx, x—lzx, and we have
(x—lyx)(x-1zx)—1 = .7640;e e 8"
since 312—1 e S. The subgroup criterion of Theorem 3.06 noW
shows that 89‘ is a subgroup, as clearly 1 lies in S”.
We now set up an isomorphism from S onto 8'" by defining q
to be 3;” for all y in S. It is almost trivial to verify that
(WM = (312)” = ytzx = (y¢)(z¢),
so that 95 is a homomorphism. Further 95 is one—one as y“ = z”
implies y = z, for ally, z in 8. And 95 is onto 8“ as y in 8 maps
onto the general element ya” in 89”. Therefore, by Theorem 2. 16, 95
is an isomorphism as required.|]
If he so desires the reader may construct many illustrations
of the above theorems in S3 and in other groups.
We next consider some special normal subgroups which,
though present in every group, are sometimes trivial.

DEFINITION. The commutator of the ordered pair of elements

at, y in the group G is the element x-ly—lxy. It will be denoted
by [us]-
This concept arises from asking how near the elements as, 3]
come to commuting. We have the equation
xy = yxc
where c = [am y], and clearly [33, y] = 1 if and only if xy = yes; so
a group is abelian if and only if every commutator in it equals 1.
Example 4.20. Which elements of 6'3 are commutators ? It is
clear that every commutator, being of the form w—ly—lxy, must
be the product of an even number of transpositions, and so only
1, (123), (132) could possibly occur as commutators. In fact each
does occur:
[(12), (12)] = <12)(12)(12>(12) = 1,
[(12), (23)] = (12x23x12x23) = (123),
[(12), (13)] = (12x13x12x13) = (132).
853137 F
66 Normal subgroups and factor groups

Note that $431414; may be written in either of the forms

x—lx”, (ymrly. Therefore if either at or g lies in some normal
subgroup of G then so does [x,g].
Suppose next that the group G has normal subgroups NI, M
with N1 0 A72 = 1. Then the concept of commutator allows us
to see easily that any element a“ ofN1 commutes with any element
3/ of Nz, for [06, y] lies in both N1 and N2 and is therefore 1.
Another interesting fact may be mentioned at this point: if
x2 = l for every element a: of the group G then G is abelian. To
prove this, take the commutator [any] of elements at, y in G.
Using in succession the facts that x2 = 1, 3/2 = 1, (mg/)2 = 1, we

have [96,31] = fly—19631 = 902/9631 = 1-

Since my = get for each pair at, y of its elements, G is abelian.
DEFINITION. The derived group (or commutator subgroup) of
G is the subgroup generated by all the commutators in G. It
will be denoted by 8(G) ; the notation G’ is also current.
We emphasize that the derived group is defined to be a sub-
group of G, not the subset of commutators in G. It is true that
in S3 and in other familiar groups the subgroup coincides With
the subset. However, one must not suppose that these always
coincide. (See Problem 5 following this chapter.)
The derived group of SS is gp{(123)}. The derived group is
trivial if and only if the group in question is abelian. Now 8(G),
or rather the order of 8(G), represents a measure of how near the
group G comes to being abelian—the smaller 8(G) is, the closer G
may be regarded to being abelian. In this sense a group G is as
far removed as possible from being abelian ifand only if8( G) = G,
and we shall see that there do exist non—trivial groups with this
property (see, for example, Problem 10 following this chapter;
01‘ Example 6.22; or Theorem 11.08).
THEOREM 4.21. Let N be a normal subgroup of the group G.
Then G/N is abeltan if and only if N contains 3(G).
Proof. G/N is abelian if and only if NxNg = Nn for
arbitrary elements at, g of G (so that Nx, Ng are arbitrary
 Normal subgroups and factor groups 67

elements of G/N). This condition is equivalent to

Nxy = Ngm,
since N is a normal subgroup; and this holds if and only if
N = Nyxg—lx—l. By Corollary 3.13 this is equivalent to
yxg—lx—l e N. Now N contains the arbitrary commutator
grey—1x4 if and only if N contains all commutators; N being a
subgroup, this is the case if and only if N contains the subgroup
3(G) generated by all the commutators in G.|:l

COROLLARY 4.22. G/8(G) is abelian.

Proof. The corollary follows from the theorem as soon as we
have proved the essential fact that 3(a) is a normal subgroup
of G. It will suflice to prove that every conjugate of a commu-
tator is a commutator and to apply Corollary 4.18 about a
subgroup generated by complete classes of conjugate elements.

But 2 [£17, ylz = [xz’ 92]

as the left—hand side is 2—1x-1g—1wy2, and the right-hand side is
(z-lxz)-1(z-1yz)-1<z-1xz)(z-1yz). u
DEFINITION. The centre of the group G is the subset of those
elements which commute with every element of G'. It will be
denoted by {(G'); the notation Z(G) is also current.
The centre could equally well be defined as the intersection of
all the centralizers of the elements in G'; or as the set of elements
with no more than one conjugate each.

THEOREM 4.23. Every subgroup S of 0' contained in the centre

of G’ is a normal, abelian subgroup of G.

Proof. Since each element of G commutes with each element of

{(0’) and so with each element of S, it follows that S is normal.
The abelian property of S is obvious. |:]

COROLLARY 4.24. The centre {(0) is a, normal, abelian sub-

group of G.
68 Normal subgroups and factor groups

Proof. It is necessary to prove only that §( G) is a subgroup.

Clearly 1 e {(G). Let x, 3/ lie in {(G) so that
939 = 995: 3/9 = 93/
for all g in G. Then
(mm = my = 9W)-
Therefore my 6 §( G). The proof will be complete if we show that
(12—1 e C(G). The equation mg = gar implies
95937—1 = 9, 995-1 = “9—19:
for all g in G, as required. [I
The group G is abelian if and only if G = Z(G), clearly. The
size of G/§(G) therefore gives us another measure of the non-
abelianness of G. The antithesis of an abelian group would be a
group G for which {(G) = 1, that is a group with trivial centre.
Example 4.25. The group 83 has trivial centre. This may be
proved by elementary (and by now familiar) calculations.
Example 4.26. What is the centre of the (multiplicative)
group M2 of non-singular 2 X 2 matrices with real entries? Let
A e Cpl/2) Where b
A— a
(. a) .
Since A commutes With every element of M2 we have in par-
ticularthat
ab—10__—10ab
cal 01_Oc'
Some calculation gives b = c = 0. We have, further,
a, 0 0 1 _ 0 1 a 0
0 d 1 0 " 1 0 0 d '
It follows that a = d. Therefore A = a1 Where I denotes
the unit matrix. Since every M with a 7S 0 does in fact lie in
floflz), this centre consists of all the multiples of I by real non-
zero numbers.

THEOREM 4.27. The group G is abelian if and only if G/§(G) is

cyclic.
 Normal subgroups and factor groups 69

Proof. The statement that if G is abelian then G/C( G) is cyclic

hardly calls for proof. To prove the required converse, suppose
that a§(G) is a generator of the cyclic group G/§(G), so that an
arbitrary element of G/§(G) is a”§(G), Where n is some integer.
Thus an arbitrary element of G has the form one Where z e £(G).
Take another such element «fly, with m a suitable integer and
y e g(G); then
amgonz = among/z = anamzy = anzomg.
Since amy and one commute, G is abeh'an. I]
THEOREM 4.28. If G is a (finite) group of order p“ (where p is
prime and n is a positive integer) then £(G) 9e 1.
Proof. Let 01,...,0m be a list of the classes of conjugate
elements in G; thus each 0,; (for 1 < i g m) is a complete class
of conjugate elements. By Corollary 4.15 the set 0i has p”!
elements for a suitable integer ni with n 9 n) 2 0. Since
{0,41 g i g m} is a partition of G, we have the equation
it = pn1+".+pnm.

But one of the sets 0'; contains only the element 1, and so the
corresponding in is 0. Since by hypothesis we have n 2 1,
some n; with j 75 i must be 0 (if p > 2 then, of course, there
will be more than one such nj). The one element in the corre-
sponding 0i will have just one conjugate in G, and so lies in the
centre of G. It is not 1 as j 7E i. Therefore §(G) ¢ 1.)]
The foregoing theorem has a fundamental position in the
theory of groups ofp-power order. We shall apply it to the very
special case in which n = 2:

COROLLARY 4.29. Every group of order p2 (where p is prime) is

abelian.
Proof. Let G have order p2, so that §( G) has order 1 or p or
p2 by Lagrange’s theorem. By the preceding theorem £(G)
cannot have order 1. Suppose that C(G) has order p. Then
G/§(G) has order p and is therefore cyclic; we have seen in
Theorem 4.27 that if G/§(G) is cyclic then G is abelian, so
 70 Normal subgroups and factor groups

G = g(G) and g(G) has order 192. This contradicts the assumption
that C(G) had order 10. We are forced to the conclusion that
((G) has order 102, which implies that G is abelian.|:|
The reader may care to search for a non-abelian group of
order 23.
We now hint at some sort of ‘duality’ between 3(G) and
G/§(G), based on the question of ‘how far’ G is from being
abelian. Though 3(G) = 1 if and only if G = C(G), it is not true
that {(G) = 1 implies G = 8(G) nor is it true that G = 3(G)
implies €(G) = 1. The group 83 has {(83) = 1 but 83 > 80%),
as we have seen; an example to substantiate the other assertion
will appear in Problem 9 of Chapter 11. The next theorem
says something about the centre of a group G in which G = 8(G),
and it is a beautiful application of the homomorphism concept.

THEOREM 4.30. If G is a group in which G = 8(G) then G/€(G)

is a group with trivial centre.

Proof. We prove the required result by showing that if

G/E(G) has a non-trivial centre then 3(G) cannot coincide with G.
In other words, we are assuming the existence of an element a,
in G with the properties that aZ(G) is central in G/§(G) but
a§(G) 7E {(G), that is a ¢ C(G).
We define a mapping 49 of G into G by putting
.7c = [(1, x]
for all a: in G. Now
(will)?S = [0" my] = fly—19340595?!
= (a-Iy-lay>y-1(a-1x-1aw)y
= [a2 girl—id, 9611/-
If [am] and [(1, y] are central in G then
($3M = [mwlfiwl = WNW),
and q5 is a homomorphism. Therefore we try to show that
[(1, x] e g(G) for all x in G. We use the fact that at;(G) is central in
G/§(G), so that aC(G) commutes with x§(G). Elementary mani-
 Normal subgroups and factor groups 71

pulation shows that C(G) = (a—lx-lax)§(G), with the conse-

quences that [(1, x] lies in C(G) and 95 is a homomorphism.
The kernel K of 95 is not G for otherwise [a, at] would be 1 for
all a: in G and a would be in €(G), a contradiction to the hypo-
thesis. Therefore K < G. To complete the proof it is enough to
show that 8(G) < K. Since 45 is a homomorphism of G into £(G),
as shown above, the first isomorphism theorem shows that G/K
is isomorphic to a subgroup of €(G). This implies that G/K is
abelian. This is equivalent to 8(G) < K, by Theorem 4.21, and
the theorem is thus proved. [I
Our next topic is centred around the concept of automorphism.
That this has a place in the context of normal properties is shown
by the following result.

THEOREM 4.31 . Let G be a group and let x be afixeol element of G.

Then the mapping (,5, defined by

995 = g”
for all g in G, is an automorphism of G.

Proof. This theorem is closely related to Theorem 14.19.

Indeed, if we take 5’ = G in that theorem, then our Theorem
4.31 follows almost completely; we need show only that G” = G.
This may be done by observing that gm—1 6 G and that gym—19$ = 9,
showing that G?” 2 G and therefore that G30 = G.[|

DEFINITION. An inner automorphism of the group G is any

automorphism arising from conjugation; that is, of the form
described in Theorem 4.31. Any other automorphism of G is
called outer.
Any automorphism of an abelian group is either trivial or
outer, obviously; so the automorphism 0; described in Example
2.23 is outer, as long as the abelian group G has some element of
order neither 1 nor 2.
The next result shows that non-abelian groups have non-
trivial inner automorphisms.
72 Normal subgroups and factor groups

THEOREM 4.32. Let I( G) be the set of inner automorphisms of

the group G. Then I( G) is a group and is isomorphic to G/£(G).
Proof. Let gt“, gty be arbitrary elements of I(G), and let them
result from conjugation by x, g respectively. If g e G then
9015:: (#71) = (g¢x)¢y = (99% = (9%)” = W = g¢zy;
and so 4%: «15” = ¢w When 3/ = 90-1 we have
95a: b‘l = 751’
where zfil is the identity mapping on G, so nl = ql. Then
56295171 = ¢x¢y‘1 = 95221—1 E I(G):
and Theorem 3.06 shows that I(G) is a group.
Let a/I be the mapping G into I(G) defined by
98” = ¢a'
The previous paragraph shows that it is a homomorphism which
is onto I( G). The kernel of ‘P is the set of elements that commute
with every element of G—the centre of G. The first isomorphism
theorem then shows that G/§(G) g I(G).|:|
DEFINITION. A characteristic subgroup of the group G is a
subgroup that is mapped onto itselfby every automorphism of G.
This means that the subgroup as a whole is left fixed; in
general automorphisms will permute the individual elements
of the subgroup.
It is an immediate consequence of the definitions that a
characteristic subgroup is normal. Indeed, the normal sub-
groups of G are just those left fixed by I(G). Not every normal
subgroup is characteristic; we shall give an example to substan-
tiate this assertion below.
THEOREM 4.33. In any group G the subgroups 8(G) and ((G)
are characteristic.
Proof. Let on be an automorphism of G. Because a and “—1
are homomorphisms they each map the set of commutators into
itself:
[96:21]“ = (96‘ $15631)“ = (x‘lot)(y‘1a)(m)(?la)
= (xa)‘1(ya)‘1(m)(3l<¥) = [$062101],
 Normal subgroups and factor groups 73

and similarly for orl. It follows that on maps 3(G) onto 8(G);
that is to say, 8( G) is characteristic.
Next let 2 e {(G), so that az = za for each a in G. If follows
that (aa)(zoz) = (2a)(aa). Now aa is as arbitrary an element of G
as a was, because or is onto G. Therefore 20; e i(G). We thus have
{€(G)}oc E {(G), and the reverse inclusion follows from the fact
that 05—1 is an automorphism. Therefore ;(G) is left fixed by ac,
and is a characteristic subgroup. I]

THEOREM 4.34. In any group G, a characteristic subgroup of a

characteristic subgroup is characteristic in G, and a characteristic
subgroup of a normal subgroup is normal in G.

Proof. Let N be a subgroup of G, and suppose that N is left

fixed by some set A of automorphisms of G. Consider the action
ofA on some characteristic subgroup K of N. If a e A then a, on
restriction to N, defines an automorphism of N, which we denote
by a also. The characteristic property of K then ensures that
K06 = K.
NOW take N to be a characteristic subgroup of G, and A to be
the full automorphism group of G. Since Kon = K for all a in A,
K is characteristic in G.
Secondly take N to be a normal subgroup of G, and A to be
the group I( G) of inner automorphisms of G. Again our earlier
remarks give the required fact: Kor = K for every a e I( G),
and so K <1 G.|]
The final concept to be studied in this chapter is more closely
related to normal subgroups than may appear at first sight.

DEFINITION. The directproduct ofthe finite set {Giz 1 < i g n}

of groups is the set
{(gla'--:gn):gi 6 at for 1 g i g n}:

Where (g1,..., gn) is an ordered set of group elements. Multiplica-

tion in the direct product is defined by

(gnu-,gnxhpnnhn) = (glhl""’9nhn)
74 Normul subgroups and factor groups

where 9,, h,- lie in 0,. This direct product is denoted by

GIX ...X 0,, or by H” Gi.
1<i<n

THEOREM 4.35. The direct product 01X X on of the groups

Giforl <i<uisagroup

Proof. We verify briefly the group axioms one by one. Closure

follows immediately from the definition of direct product. The
associative law is a consequence of the same law in each factor
07;. The unit element of the direct product 1s (1,. .,1), and the
inverse of (g1,..., 91,) 1s (91—1,..., gn 1); here, of course, the symbol 1
in the ith place denotes the identity element of 9,, and gg‘l is
the inverse of g, in 0,. D

Example 4.36. Let 02 and 03 denote the groups of orders 2

and 3 respectively. Suppose am generates 0",, Where m = 2, 3.
Then the elements of 02 X 0’3 are just the set of pairs
{(ag,ag):1 g i g 2,1 <j g 3}.
Thus 0'2 x 03 is a group of order 6. Now consider powers of the
element of 02X03 in which i = j = 1:

(a2, “3)2 = ((13,003) = (1, 03),

(“2: “Q3: (“2a a3) = (“2’ 1):
(“2, “3)4 = (“2: “3) = (1, a3),
(“2’ “Q5: ((12, “3): (a2, a3),
(“2, We: (a2, a3)= (1 1)
Therefore ((12, (13) has six distinct powers. It follows that 02 X 03
is cyclic; it is, or at least it is isomorphic to, the cyclic group of
order six.1‘ [l
We remark that every group G is isomorphic to the direct
product of itself and the group of order 1; we call such a direct
product trivial, and usually interest ourselves only in non-trivial
T The learner who is now struggling may find some comfort in the story that
Cayley supposed that 06 and 02 x 03 were non-isomorphic.
 Normul subgroups and factor groups 75

direct products. Note that if the groups 0,. are finite, of order 3/,-
for 1 < i g n, then HD 0%. has order H yi; this is implied by
1<i<n 1<i<n
the very definition of direct product.

THEOREM 4.37. The direct product G’ of the set {Giz 1 g i g n}

of groups contains subgroups H,- which are isomorphic to 0,, for
1 g i g n, and which have the following properties:
(i) Hi commutes with Hj element by element if 1 g i <j g n;
(ii) 0’ = H1...Hn;
(iii) H; n H1...Hi_1Hi+1...Hn = lfor 1 g i < n.

Proof. We define H,- to be the subset of elements of G’ with the

property that g5 = 1 if 1 g j g n and j 75 i, the element gi
taking an arbitrary value in 01. The facts that this is a sub-
group and that it is isomorphic to Gi scarcely call for proof.
(i) It is clear that (1,...,l,gi,1,..., 1) and (1,..., 1, gj, 1,..., 1)
commute provided i < j; the product is
(1,..., 1, g, 1,..., 1, gj, 1,..., 1),
in whichever order the factors are taken.
’(ii) Trivially, I-I1 H” is the complex product of H1,..., H”. We
are required to show that the arbitrary element (91,..., 9%) of G’
can be expressed in the form h1 h” with h,- eHi for 1 g i g n.
We choose hi to be (1,..., 1, gi, 1,..., 1), and the rest is easy.
(iii) The elements of H1...Hi_1H¢+1...Hn are just those ele-
ments of G' with ith component equal to 1, and the only such
element in H, is (1,..., 1), which is the identity element of G.[]
Thus far we have examined a procedure which constructs a
group G’ from given groups G1,..., G”. We now suppose that we
have been given a group G instead and that we are seeking
information about whether it is, or (strictly) is isomorphic to,
some direct product.

THEOREM 4.38. Let the group Ghuve subgroups at for 1 g i g n

satisfying the conditions:
(i) Gi commutes with 09- element by element if 1 g i <j g n;
76 Normal subgroups and factor groups

(ii) G’ = 01... G”;

(iii) 0i n G1...G'i_1G'i+1... G” = 1 for 1 < i g n.
Then G g 01X ...X G”.
Proof. The arbitrary element g of G can, by (ii), be written as
91 9”, where gt 6 Gt for 1 g i g u. The hypotheses imply
that this representation is unique. Suppose that we had
9 == 91mg. = gin-9;.
where, of course, g; e Gi. We may obtain the equation
.tMgD‘1 = 9:19; 95.119%_195.1192+1 9.719;.
by means of (i). But then (iii) shows that gi(g§)-1 = 1, that is
gi = g}, and this holds for 1 < i g n. Hence the representation
of g is unique.
We now construct a mapping 4; from G to 01X...>< 0,, by
t .
Pu ting 996 = (gnu-,9.)-
Ifh = hl...hn where h e G and hi e G; for 1 g i g n, then
9h = 91 "' gnhl hn = (91h1)-"(gnhn)

by ‘1” and 5° (W = (My h.)

= (glw': gnxhlv": kn)
= (9¢)(h¢);
here we use the definition of group multiplication in G1 X X 0”.
Therefore 96 is a homomorphism.
But qS is one—one and onto. It is one—one because if q = M

then (g1,...,gn) = (hp-"5 hm),

in the above notation, 9.5 = h, for 1 g i < n, and so g = h. It
is onto because 91 gn maps onto (91,..., 9”). Therefore 96 is an
isomorphism from G’ onto 91 X X Gm and the theorem is now
proved.|:|

Example 4.39. Consider the set of all vectors in the plane 9,

or, what comes to the same thing, the set of all 1 X 2 matrices
with real entries (see Examples 1.18 and 1.32); the group
operation is addition. An easy application of the above theorem
 Narmal subgroups and factor groups 77

shows that this group is isomorphic to R1 X R2 where R1, R2 are

each isomorphic to the additive group of real numbers. Specifi—
cally, R1 is {(r, 0) :r real}, and R2 is {(0, r): 7' real}.
Example 4.40. Consider the group g of orientation—preserving
mappings of 9 into itself generated by the following two sub-
groups: the set 9 of all translations and the set .% of all
rotations about a fixed point 0. Condition (i) of the theorem
holds as g is abelian, (ii) is geometrically obvious, and (iii)
follows from the fact that the only mapping which is both a
translation and a rotation is the identity mapping. Hence
g g 9 X9.
Example 4.41. Let .1 be the set of all (n+ 1) X (72+ 1) matrices
with real entries and of the form I+ it a, Emu: where I is the
i=1
unit (n+ 1) x (91+ 1) matrix and Ei’j is the matrix with every
entry 0 except for 1 in row 75, column j. It will be found that

(I'l- iglai Emu) (I‘l' £3154 Ei,n+1) = 1+ i:1(ai+bi)-Ei,n+1'

The reader will easily be able to show that .1! is an abelian
group under matrix multiplication, and that in view of the
above equation J! N JV X X JV
= 1 n

where JV; is the set of all matrices of the form I+a¢ Emu- In
fact .4 is closely related to the group of translations in 9.
As an example of a group which is not a (non-trivial) direct
product we mention the cyclic group of order 4. It has only one
proper subgroup, and so cannot be expressed as any product of
proper subgroups.
There is an alternative form of Theorem 4.38 in which condi-
tion (i) is replaced by:
(i’) G, <1 Gfor 1 <i<m
The reader will find it an easy matter to verify that this new
version is equivalent to that originally given.
We stress again that the data in Theorems 4.37 and 4.38 are
completely different, however similar the theorems are in form.
78 Nomal subgroups and factor groups

In the former case we are given groups G1,..., 0,, and then con-
sider properties of their direct product, While in the latter we
start with a group G’ and postulate subgroups of it with certain
interesting properties. Sometimes G is called an external direct
product or an internal direct product, as the case may be.

Example 4.42. We close this chapter with an example of a

group containing a normal subgroup that is not characteristic.
Take isomorphic non-trivial groups 01, 02 and form their direct
product G. By Theorem 4.37, G’ contains subgroups H1, H2
isomorphic to G. It is clear that H1 is normal, but it is not
characteristic because an automorphism of G’ can be constructed
to map H1 onto I-12 and H2 onto H1. Detafls are left for the
reader to supply.

Problems
1. Prove that the quaternion group Q (Example 1.30) has only one
element of order 2. Determine the centre of Q, and prove that all sub-
groups of Q are normal. Does Q have any factor group isomorphic to
the cyclic group of order 4?

2. Find all classes of conjugate elements and hence all normal subgroups
in the alternating group A4 on {1, 2, 3, 4}. What is the centre of this
group ? What is its derived group?

3. Prove that, in the symmetric group S4 on the set {1, 2, 3, 4}, the subset
V = {1, (12x34), (13>(24), (14)(32)}
forms a normal subgroup of order 4. Show, by a suitable choice of trans-
versal of V in S4 or otherwise, that 84/17 is isomorphic to the symmetric
group S3 on {1, 2, 3}.

4. Which of the following multiplicative groups ofresidues are isomorphic

to (non-trivial) direct products?
(i) {[1], [3], [5], [7]} modulo 8;
(ii) {[1], [3], [9], [11]} modulo 16;
(iii) {[1]. [7], [9], [15]} modulo 16.
*5. Let G’ be the set
{(al, a2, a3, a4):a1, a2, a3, 014 are integers module 102}
 Normal subgroups and factor groups 79

Where p is any prime. The product of the elements (a1, a2, a3, a4) and
(bl, b2, b3,b4), taken in that order, is defined to be
(a1+b1+pa3 b1: “2+ b2 +10% b2, “3+ba‘l’1’a3 b2, a4+b4 +1702 b1)-
(i) Prove that G is a group under this multiplication.
(ii) Find the form of the commutator of two general elements in G,
and deduce that 3(G) g {(G').
(iii) Show that both (1;, 0, O, 0) and (0,12, 0, 0) are commutators whereas
their product is not.
6. The group G has a subgroup S with the property that every right
coset of S in G is also a left coset of S in 6‘. Must S be a normal subgroup
of G?

7. Let 8’ denote the symmetric group on the set of integers. Show that
those permutations which each move only a finite number of elements
form a normal subgroup of S.

8. (i) Let 9 = gp{A, B} Where

A=(é -3), B=(3 J)

as in Problem 5 of Chapter 3. Is gp{AB} normal in 9 Y
(ii) Let g = gp{A, B} Where
_ 1 1 _ 2 0)
A _ (0 l)’ B — (0 l
as in Problem 11 of Chapter 3. Is gp{A} normal in g?

9. The centralizer 0(X) of the complex X in the group G is defined to be

{yty e G, [x,y] = 1 for all a: in X}. Prove that
C(X) = mg: C(06)
where C(x) denotes the centralizer of {x} in G.
Prove the following statements about the centralizers of the complexes
X, Y, S in G‘:
(i) EX 9 Y then 0(X) 2 C(Y).
(ii) 8’ Q C'{O'(S)}.
(iii) 005’) = 0[0{C'(S)}]-
10. Prove that every element of A 5 is a commutator.

1 1. Prove that the multiplicative group of the non-zero complex numbers

is isomorphic to the direct product of the multiplicative group of positive
real numbers and the additive group of real numbers modulo 211- (see
Problem 10 of Chapter 2).
80 Normal subgroups and factor groups

12. (i) Let G be the subgroup of S6 generated by the set

{(12)(45), (12x46), (13>(45). (13)(46)}-
Prove that G has a normal subgroup of order 9, and find the index of this
in G.
(ii) The dihedral group of order 12 is defined in Example 1.25 as a set
of mappings leaving a 6-sided regular polygon invariant. Prove that the
group is isomorphic to a non-trivial direct product.

13. Let I( G), A(G) be the groups of inner, all automorphisms respectively
of the group G. Prove that I(G) is normal in A(G).

l4. Prove that in any group G the set

{xzx has only finitely many conjugates in G}
forms a subgroup. Prove further that this subgroup is characteristic in G.

15. Let G be the direct product of G1 and G2, and let M be a normal sub-
group of G; for i = l, 2. Which of the following statements are true?
(i) 3(G1X as g 8(01)X8(G2) and C(Gl X Ga) 2 €(G‘1)X€(G’2)-
(ii) G1 X G2 has a normal subgroup isomorphic to N1 ><Nz.
(iii) (G1 X G2)/(N1 lz) E Gi/Ni X Gz/lvz-
(iv) If GI, G2 have no proper normal subgroups then any proper normal
subgroup of GIX G3 must be isomorphic to G1 or G2.
(v) All normal subgroups of G1>< G2 have the form lNz Where
M <1 Géforli= 1,2.

*16. (i) The normal subgroupN ofthe group Gis such thatN n 8(G) = 1.
Prove that N g C(G), and deduce that {(G/N) = C(G)/N.
(ii) Prove that if N1, N2 are normal subgroups of the group G then
G/(N1 n Na) is isomorphic to a subgroup of G/N1 X G/Nz.

**17. Prove that, if {0A2A e A} is a complete set of conjugate complexes

in the group G, then the kernel K of the homomorphism which takes as
in G onto the permutation replacing CA by 0}“ is given by
K = ADA N( 0a) .

Show that N(0%) = N(0%)“6 for each A, and deduce that K also equals
J; N( 0)” where 0 is a complex arbitrarily selected from the given con-
jugate set.

18. Find a complete class 0' of conjugate elements each of order 3 in the
alternating group A‘ on {1, 2, 3, 4}. Consider the homomorphism (as
described in Problem 17) from A4 into the symmetric group on (J, and
find its kernel.
 Normal subgroups and factor groups 81
* * 19. Let S be a subgroup ofthe group G. Show that ifg is a fixed element
of G, then the mapping that takes the coset Sm to 13969 is a permutation of
the right cosets of S in G. Hence define a mapping 95 from G into the
symmetric group on these cosets. Show that 45 is a. homomorphism and
that its kernel is the intersection of all the conjugates of S in G.

*20. Let N(X), 0(X) denote respectively the normalizer, centralizer of

the (non-empty) complex X in the group G (see Problem 9 above). Prove
that 0(X) <1 N(X).
Prove further that ifX is a subgroup of G then N(X)/0(X) is isomorphic
to a group of automorphisms of X.
Deduce Theorem 4.32.

853137 G-
 5

Finitely generated abelian groups

AT the present day, to determine in any useful sense all finite

groups is a problem beyond the resources of group theory,
largely because the structure of these groups is so varied; the
classification of all infinite groups is much harder still. To
obtain structure theorems it is therefore necessary to look at
rather restricted classes of groups. In Chapter 3 (Theorem 3.21
and Problem 12) we gave a description of all finite cyclic groups;
in such groups every subgroup is cyclic, and every number
compatible with Lagrange’s theorem occurs as the order of
a certain unique subgroup. We are now going to solve corre-
sponding structure problems for all finite abelian groups. In
fact we shall do more. It will appear that it is more natural to
consider the finitely generated abelian groups, which of course
include the finite abelian groups as a subset, because much the
same tools and methods will determine their structure also. We
shall take advantage of this fact. We should perhaps caution the
reader that, in the non—abelian case, finitely generated groups
are significantly more difficult to treat than finite groups.
Unfortunately, the notation traditionally used for abelian
groups is not the multiplicative notation of the last two chapters.
An additive notation is usual—the binary operation is called
‘addition’ instead of ‘multiplication’, and one writes a—l—b for
the ‘sum’ of the elements a, b in an abelian group instead of
writing ab for their ‘product’. This new notation is logical
enough when one takes into account the existence of rings. In
these objects there are two binary operations. A ring must form
an abelian group with respect to one operation but need not have
 Finitely generated abelian groups 83

this property with respect to the other. In order to have the

ring axioms, especially the distributive laws, resemble the
familiar laws holding for the real numbers, it is necessary to use
addition for the abelian group operation and multiplication for
the other. We adhere to this practice in the study of abstract
abelian groups for the sake of uniformity.
This additive notation has certain consequences of note. The
identity element of the abelian group A is written 0 (and called
‘zero’), while the inverse of a in A is written as ~a. We write
na instead of a”, for arbitrary positive integers n. In correspon-
dence with Theorem 2.07 we have —(na) = n(——a) for such n,
and we define (—n)a as, say, —(m); of course, we define 0a to
be 0 for all a. The statement that ma-l—na = (m+'n,)a, for all
integers m, n is the new version of Theorem 2.08 according to
which 00%" = am”. Note carefully that we have
n(a+b) = na+nb
by virtue of the fact that the elements a, b of A commute—the
corresponding relation (ab)" = anb" does not hold generally in
non-abelian groups. Theorem 2.09 implies that m(m) = (mma.
The product 0102 of the complexes 01, 02 becomes their sum
014—02. In place of the direct product of groups Gym, Gn we have
their direct sum, written as f” G,- or 0163 (-9 G1,; note the need
i=1
for a special notation for direct sum to distinguish it from sum.
Sometimes the quotient group G/N is written as G—N and
called a difference group, but we shall persist with the term
‘factor group’ and the symbol G/N.
Our basic tools in the task at hand are cyclic groups and direct
sums. Since every subgroup of an abelian group is normal, it is
perhaps to be expected that direct sums should appear in view
of the criterion in Theorem 4.38 for a group to be isomorphic to
the direct sum of certain subgroups. We observe that the
direct sum of any (finite) set of cyclic groups is an abelian
group. Thus to find an abelian group of order p” (where p is
prime) we may take cyclic groups of orders p"1,..., p”r where
84 Finitely generated abelian groups

n1,..., n, are any positive integers for which n1+...+n, = n, and

take their direct sum; and a similar construction may be carried
out in more general cases. We know, of course, that cyclic
groups of all orders exist.
The fundamental result on finitely generated abelian groups
is a converse of this. It states that every such group is the
direct sum of cyclic groups (finite or infinite), and most of this
chapter is devoted to its proof. Many proofs of this result are
known. The method that we have chosen is more constructive
and may be easier for the learner to follow, as well as being more
capable of generalization, than other briefer but less lucid proofs
that have been found. (See Problems 15 and 16 following this
chapter.)
The following definition applies not only to abelian groups but
to general (even infinite) groups.

DEFINITION. A p-group (where p is any prime number) is a

group in which the order of each element is some power of p.
By Corollary 3.16 to Lagrange’s theorem any group of order
p” is a p-group, and we shall prove in the course of Chapter 7
that every finite p—group has p-power order. It will be of some
help to prove the latter statement at once in the abelian case.
Let A be a finite abelian p-group. If A has no proper subgroup
then A is cyclic, A is generated by an element ofp-power order,
and so A has p-power order. This is the first step in an inductive
proof on the order of A. If A has a proper subgroup B then both
[B] and |A/B[ are strictly less than IA I, which implies that B
and A/B each have p-power order. Since |A| = |B||A/B|, it
follows that A has p-power order. .
We now tackle finite abelian groups.

THEOREM 5.01. Let A be a finite abelian group in which

{p1,..., p} is the set of primes occurring as orders of elements.
Then the elements of A of pi-power order form a subgroup, Az- say;
and A g i” Ai.
i=1
 Finitely generated abeliom groups 85

Proof. We prove that A,- is a subgroup by applying the usual

subgroup criterion. Let x, y be elements of A with orders
101?, pg respectively, where r, s are certain non-negative integers;
we must consider x—y. If t is the larger of r and s then
20266—21) = pix—rig = pi"(r€- 96) —10€-‘S(20$ 3/) = 0
since My; = my = 0. It follows from pflw—y) = 0 that the
order of 95—3; divides p5, and so is certainly a power of 10,-. Since
A, is non-empty by hypothesis, we see that A,- is a subgroup ofA.
We prove that A is a direct sum of the required form by apply-
ing the criterion of Theorem 4.38 for a direct sum.
First, it is clear, because A is abelian, that each element of A,
commutes with each element of A, whenever 7? ;E j.
Secondly, we have to show that A = A1+...+An; that is,
that an arbitrary element a of A has an expression of the form
a1+...+a,, where a, 6A,; for each i. Let a have order m, and
suppose that the prime q divides m. Then am/q, which has order q,
still lies in A; and it follows that g e {p1,..., 10”}. Therefore
m = 123“ 10;,“ for suitable non-negative integers a1,..., an. The
Euclidean algorithm in its general formT asserts the existence
of integers B1,", [3,, for which
an m
31P11+
T ...+ .3”???
— = l

because the integers 57%,.” 1%. have 1 for their highest common
1 n

factor. It follows that

a = [31% a..+ 'ITMW

We may take a, to be 18,1)”; (1, for then

12‘

p‘é“a¢= 32-07% =13¢0 = 0

and a, e A, as (1, thus has pi-power order.
Thirdly, we must prove that, for each i, A, has trivial inter-
section with A1+...+A,-_1+A,+1+...+An. This follows from
1' See the Appendix.
86 Finitely generated abelian groups

the obvious fact that 0 is the only element of pi—power order in

this sum, and so is the only element in both A1. and the sum.
Therefore A g a Au as required. I]
i=1

COROLLARY 5.02. A finite abelian group contains non-trivial

elements ofp-power order, where p is prime, if and only if its order
is divisible by p.

Proof. Let A be a finite abelian group of order pi’l p%" where

p1,..., p” are distinct primes. The theorem, together with the
preceding remarks about p-groups, shows that A is the direct
sum of groups of orders pi’1,..., 2;". The corollary follows.|:|

COROLLARY 5.03. If the order of a, finite abeliun group A is

divisible by p? but not by p7+1 then A has a subgroup of order p7. [1
We remark that Theorem 5.01 shows that any finite abelian
group contains a unique subgroup which is maximal in the sense
that it is the set of all elements of p-power order, for each
prime p; and that it is equally true that there is a unique sub-
group containing precisely the elements Of order prime to p.
Such circumstances are generally absent, and are absent from
the symmetric group 83 in particular because the elements of
2-power order, which are also the elements with order prime
to 3, do not form a subgroup.
Example 5.04. Consider the cyclic group 0m of order m where
m = pi“ pg". The theorem above states that am is the direct
sum of cyclic subgroups of orders pi‘1,..., p21“ respectively. These
may easily be found; if c generates 0m then clearly cm: generates
the ith summand, where m,- = m/pg‘i. Incidentally, this reason-
ing shows that the direct sum of cyclic groups of orders p‘i‘l,..., p35"
is isomorphic to the cyclic group of order p‘fl pg", provided of
course that p1,..., pn are distinct primes.
Theorem 5.01 shows that we need now consider only finite
abelian p-groups in solving the problem in which we are primarily
interested, namely to ‘find’ all finite abelian groups.
 Finitely generated abelicm groups 87

THEOREM 5.05. A finite abelicm p—group is isomorphic to the

direct sum of cyclic p-groups.
Proof. Let A be a finite abelian p—group. We have seen that
A has order p" for some non—negative integer n; and we use
induction on n. When n = 0 there is, of course, nothing to prove,
as A = 0, but this does start off the induction.
When n > 0 let b be any element of maximal order in A, and
let B = gp{b}; we shall prove that B is a direct summand of A.
If B = A then A is itself cyclic and no further proof is necessary.
We may therefore take it that both B andA /B have smaller order
than A. By the inductive hypothesis, therefore, A/B is the
direct sum of cyclic p-subgroups. Let the generators of these
subgroups be a1+B,..., ar—I—B, and let their orders be 11%,”, par
where each 05,; is positive. Thus paiai E B but pat—1a,; ¢ B, for
each i. It follows that pawl. = 3,1) for certain integers Bi.
Showing that pa; divides [3,. is the key step in the present
proof. In order to derive a contradiction, therefore, suppose that
we had [9, = )3;107‘ where )3; is not divisible by 10 and 0 < y; < ai.
Thus pam- = Beam:
for 1 g i g r. This equation asserts that the order of a,- exceeds
that of b, because y, < 05,-, while of course paiai and pytb have
the same order; observe that all elements of A have p-power
order and that pat-1a, 75 0. But the order of b was to be greater
than or equal to that of any element in A. Our assumption on
[3,; has led to a contradiction, so we must conclude that p“!
divides ,8,- for each i.
Thus for each i there is an element 6,- in B for which
paid, = paibi. We put 0, = ai—bi, so that 10%, = 0. We define
0' to be gp{cl,..., on} and we intend to prove that A = B 6—) C.
If :1: is an arbitrary element of A then x+B e A /B and m—l—B
has an expression m1(a1+ B)+...+m,(a,+ B) for suitable
integers m1,..., m,. Thus
a: = m1a1+...+m,a,+y
for some 3/ e B. Since a,- = Uri-bi we have
as = m1c1+...+m,c,+z
88 Finitely generated abelian groups

where z = m1b1+...+m,b,.+y e B, and so a: 6 0+3. Therefore

A = B+ 0.
Next we suppose that x e B n 0', in the hope of showing that
a: = 0. Since 96 e 0 we have
a: = micl+...+m}c,
for suitable integers m’l,..., m,’., so that
x+B = mi(01+B)+---+m;(0r+3)-
Since an E B we have x+B = B. Recalling that oi = ai—bi, we
find that B = m;(a1+B)+...+m;(a,+ B).
This implies that m;(ai+B) = B for each i, because A/B is the
direct sum of cyclic subgroups, each generated by an a¢+B.
Therefore mint, 6 B, and so pa; divides mg. It follows (from
12%,; = 0) that mQ-cz- = 0, and the expression for :2: gives x = O
at once.
We have now shown that A = B $0, where B is cyclic,
in view of Theorem 4.38. The inductive hypothesis shows that
0 is the direct sum of cyclic subgroups, and the theorem
followslj
It should be remarked that the proof of the preceding theorem
is essentially constructive—a definite method is given for
obtaining a direct decomposition in any specific case.
Example 5.06. Consider the multiplicative matrix group g
generated by A and B where
0 1 0 —1
A=(—1 o), B=(1 0)-
This group is abelian because, as may be verified, AB = BA.
(We shall use multiplicative notation in discussing it as additive
notation would probably be even more confusing.) It will be
found that A4 = I and A2 = B2 = —I where I denotes the
unit matrix, so that f? has order 8. Ifgp{A} is taken as one direct
factor, as it may be, A being of order 4, then it can be verified
that f? g gp{A}>< gp{AB}. This is not the only decomposition
——we also have, for instance, g g gp{B} X gp{AB‘1}. (Note
 Finitely generated abelian groups 89

that g is not isomorphic to gp{A}X gp{B}, which has order 16.)

Though the decomposition is not unique, the orders of the direct
summands are uniquely determined, as may (and later will quite
generally) be proved.
We can now obtain the general result on finite abelian groups.

THEOREM 5.07. A finite abelian group is isomorphic to the direct

sum of a set of cyclic p-groups, for various primes p.
Proof. By Theorems 5.01 and 5.05, the group is isomorphic
to the direct sum ofp—subgroups, for various primes p, and these
p-subgroups are themselves isomorphic to direct sums of cyclic
p-subgroups. I]
The torsionfree abelian groups are our next consideration.
The infinite cyclic group is torsionfree, and so indeed are all
direct sums of groups each isomorphic to it.

LEMMA 5.08. Let A be an abelian group generated by the elements

a1,..., an, and suppose that there is a relation in A of the form
m1a1+...+mnan = 0

where not all the integers mum mm are 0. Then A has a set of
generators a’1,..., a; such that m’a; = O for some i and for some
non-zero integer 177/.

Proof. We use induction on the integer |m1|+...+ [mu], which

will be denoted by m. By hypothesis m > 0; and when m = 1
just one mi is 1 and the rest are 0, and we take a]. = a; for
1 g j g n.
Now suppose that m > 1. If n—l of the numbers m1,..., mm
are 0, then there is really nothing to prove. If not, then there
must be two of them, say m, and inS with r 72 s, for which
either lmr+m8| or Imr—msl is less than lmrl. We lose no
generality in taking r = 1, s = 2, for this situation occurs when
the generators a1,..., an are suitably relabelled; and in taking
lml—m2| < [m1|, for the proof with the other sign is only
trivially different from what follows.
90 Finitely generated abeliom groups

Since A is generated by a1,..., an, it is also generated by

a1, a2+a1, a3,..., an (only the second generator has been changed).
And the given relation in the given generating set is equivalent to
(ml—m2)a1+m2(a2—|—a1)+m3a3+...+mnan = O.

Wehave [ml—m21+lm21+[m31+...+1mnl<m
because [ml—m2] < lmfi. The inductive hypothesis now
applies: A has a set {a1,...,a;} of generators and a relation
m’a} = 0 (for some i) in them, with m’ # O.D

THEOREM 5.09. A torsionfree finitely generated abelian group

that is non-trivial is isomorphic to the direct sum of a set of in-
finite cyclic groups.

Proof. Let the finitely generated abelian group A be torsion-

free. It is tautologous that A has a finite generating set; but,
of course, it has many such generating sets, and we Wish to choose
a special set {a1,..., an} such that A g A163 ...@An where each
A,, is generated by oi.
Of all possible finite generating sets, we select any one with
the minimal number of members, say n. In this case there is no
relation of the form m1 (11+ ...+mn a,” = 0 in A, unless all the mi
are taken as O. For, otherwise, Lemma 5.08 shows that we can
find a set {a1,..., {1;} of generators for A with the property that
m’a; = 0 for some i and for m’ non-zero. Since A is torsionfree
we must have (1;: 0, and then {a1,..., a254, ag+l,..., cg} is a
generating set of ri—— 1 elements. This contradiction to the
choice of it shows that there is no relation in A of the form
mentioned.
Each pair of elements taken from Ai, Aj respectively commute
(A1, = gp{a,i} for 1 g i g rt); A = A1+...+An as {a1,...,an} is
a generating set; A»; n (A1+...+Ai_1+Ai+1+...+An) = 0 be-
cause of the non-existence of relations having the form described
above. The direct sum criterion of Theorem 4.38 therefore
shows that A = A1 6—) (-9 A”. Since A is torsionfree, each A1; is,
of course, infinite cycliclj
 Finitely generated abeliau groups 91

We now move on to the structure of those finitely generated

abelian groups that are neither periodic nor torsionfree, the so-
called mixed groups. Such objects exist—we have for example
the direct sum of any (non—trivial) finite cyclic group and the
infinite cyclic group. The following result applies to all abelian
groups.

LEMMA 5. 10. The elements offinite order in an arbitrary abelian

group A form a periodic subgroup P, and A /P is torsionfree.

Proof. Note that A/P may be trivial—this will be the case

if and only if A is periodic.
The fact and proof of the existence of a maximal periodic
subgroup are very similar to What was seen earlier for maximal
p—subgroups in a finite abelian group. The zero element 0 has
finite order, so the set P of elements of finite order is non—empty.
Let x, y be in P, so that rx = 0, 8y = 0 for some positive integers
r, s; then
(rs)(x-—y) = s(rx)—r(sy) = sO—rO = 0.

Hence w—y e P. The subgroup criterion of Theorem 3.06 shows

that P is a subgroup of A.
Suppose that A/P has an element a+P of finite order m.
Then m(a+P) = P, and ma 6 P. Thus ma = b Where b has
finite order. It follows that a itself has finite order, and so a e P.
Thus a+P = P. The only element in A /P With finite order,
therefore, is P. Hence A/P is torsionfree. D

Example 5.11. The elements of infinite order will not in

general form a subgroup. Let B be generated by an element b
of finite order (greater than 1), and let 0 be generated by an
element 6 of infinite order. The elements (b, c) and (0, —c) of
B G90 have infinite order, but their sum (b, 0) has neither
infinite nor trivial order.
Another illusion is the hope that the elements of finite order
in a non—abelian group form a subgroup. A counter-example will
be found in Chapter 3, Problem 5.
92 Finitely generated abelian groups

THEOREM 5.12. Let A be a finitely generated abelian group with

periodic subgroup P. Then A is the direct sum of P and a suitable
torsionfree subgroup T.

Proof. By Lemma 5.10 A /P is torsionfree, and we may suppose

that it is not trivial. Theorem 5.09 tells us that A /P, which is
finitely generated, is isomorphic to the direct sum of cyclic
groups. Let these be generated by a1+ P, . .., an—l— P, respectively.
Define T to be gp{a1,..., an}. Then there can be no (non-trivial)
relation of the form
m1a1+...+mnan = b

where b e P. For, otherwise, we would have

m1(a1+P)+-H+mn(an+P) = P
as b—l—P = P, and the fact that A /P is the direct sum Of infinite
cycles generated by the a¢+P gives m, = O for each i, and then
b = 0 follows.
The subgroup T therefore has the property that T n P = 0.
This implies that T is torsionfree. Another consequence is that
A g T (-9 P, the least trivial fact in the application of the direct
sum criterion of Theorem 4.38 now being the assertion that
A = T +P, Which follows from the definition of T. This
completes the proof of the theorem. [I

COROLLARY 5.13. The periodic subgroup of a finitely generated

abelian group is finitely generated.

Proof. In the notation of the theorem P g A/T. As a factor

group of a finitely generated group, P is also finitely generated. []
Note that Whereas P is uniquely determined in the above
theorem many choices for T are possible. Consider B 6-) 0' of
Example 5.11, in which P is precisely B. One choice for T is 0;
another is gp{(b, —c)}; in fact we may take T = gp{(nb, :lzc)}
where n is any integer.
The torsionfree subgroup T of the theorem is, however,
determined up to isomorphism, for the reason that it is isomorphic
to A/P and A itself determines the isomorphism class of A/P.
 Finitelg generated abelian groups 93

In general it is not true that the periodic subgroup of an

abelian group is a direct summand. (A suitable counter-example
will be presented after the necessary concepts have been de—
veloped; see Example 11.14.)
Our results thus far are now summarized.

THEOREM 5.14. A finitely generated abelian group is isomorphic

to the direct sum of a certain set of (finite, or infinite, or both finite
and infinite) cyclic groups.

Proof. If the abelian group A is finitely generated then

A g P (-B T where P, T denote periodic, torsionfree subgroups
respectively. Both P and T are finitely generated (recall
Corollary 5.13). By Theorems 5.07 and 5.09 both P and T are
isomorphic to direct sums of cyclic groups. The structure
theorem for finitely generated abelian groups follows. [I
Not every abelian group is isomorphic to the direct sum of
cyclic groups. According to our current definition of direct sum
such a group must be finitely generated, and even when a
generalized definition of direct sum is used (as we shall see later)
there are still abelian groups not isomorphic to the direct sum of
cyclic groups; the additive group of rationals is such a group.
Though we now know the structure of all finitely generated
abelian groups, we still do not know which of them are non-
isomorphic. For all we know at the moment we might have
Z @Z ® Z g Z (-BZ, where Z is the infinite cyclic group; or
Zps ED Zps ® Z1, g. Zps (-B s (-9 s, where Z, is the cyclic
group of order n. Our next aim, therefore, is to associate with
each finitely generated abelian group a set of integers which
determines completely its isomorphism class.
DEFINITION. An invariant of groups is a function (defined
on groups) which takes the same value on isomorphic groups.
We proceed to prove formally that such obvious integers as
the number of cyclic direct summands and their orders suflice
to determine a finitely generated abelian group, up to iso-
morphism.
94 Finitely generated abelian groups

DEFINITION. The rank of a finitely generated abelian group is

the number of infinite direct summands in any representation of
the group as the direct sum of cyclic groups.
This definition has to be justified.

THEOREM 5.15. The rank of a finitely generated abelian group

is an invariant.
Proof. Let P be the periodic subgroup of the finitely generated
abelian group A; we have seen that A determines P uniquely.
We know that A g P 6—) T where T is a torsionfree subgroup
of A, and T is determined up to isomorphism because T g A /P.
It will therefore be enough to show that rank is an invariant
of a torsionfree finitely generated abelian group.
Let S be the subset {2x : x e T} of T; that is, the set of squares
(or rather ‘doubles’, in the additive notation) of elements in T.
Since 8 is clearly a subgroup of T we may consider T/S. Suppose
now that T has a representation as the direct sum of m infinite
cyclic groups, generated by t1,..., tm respectively. We may, and
shall, identify elements of T with elements (a1t1,..., 04m tm) of
the direct sum, the oci being integers, and in that case the elements
of T/S have the form (a1t1,..., amt—S Where each oci is 0 or 1.
The 2'" elements obtained in this way are distinct, Which means
that T/S has order 27". Of course, a representation of T as the
direct sum of n infinite cyclic groups would implypin a similar
way that T/S has order 2”. Therefore 2’" = ‘m and m = n.
This proves the invariance Of the rank m Of T.|:|

DEFINITION. The type of a finite abelian p-group isomorphic

to the direct sum of a set of cyclic groups of orders p”1,..., pnr,
Where n1,..., n,. are positive integers, is the set {n1,..., m}.
Justification is again needed.

THEOREM 5.16. The type of a finite abelian p-group is an

invariant.
Proof. We suppose that the finite abelian p-group A has
 Finitely generated abeh'an groups 95

expressions both as the direct sum of cyclic groups generated by

elements (1,. of order pm: for 1 g i < r, and as the direct sum of
cyclic groups generated by elements b,- of order p": for 1 g i g s.
We may assume that
m1 2mg 2 24m,
721 2% 2 2 ns;

and then we aim to prove that r = s and that m,- = at for

1 g 73 g r.
We use induction on the order of A. The starting-point is the
case when A has order p, and then there is no doubt that the
type must be {1}.
In general, we take A of order greater than 12. Which elements
of A have order p or less? In the first direct sum representation
they have the form (a1 pm1-1a1,..., arpmr'lar), each 0% taking the
values 0,..., p—l. The elements just named are all distinct, and
there are p7 of them. The second direct sum isomorphic to A
implies that they number p8. Since the number is certamly an
invariant of A we have 10’ = 335, and r = 8.
Next we consider B = {1096293 6 A}. It is easy to see that B,
which, of course, consists of certain multiples of elements of A,
is a subgroup. In the first direct sum representation of A it
may be identified with the elements of the form (10,81 a1,..., MB, (1,)
Where 0 < B, < pmrl for 1 g i g r. The elements just named
are all distinct, becausethe subgroup they form is the direct
product of the cyclic subgroups generated by (0,..., 0, pai, 0,..., 0)
for 1 g '13 g r; the orders of these cyclic subgroups are
pm1"1,...,pmr—1 respectively. In order to investigate the type
of B we must remember that some mi may be 1. Suppose that
m,- > lfor 1 giguandmi= 1foru<i<mthenthetype
of B is {ml—1,...,mu—- 1}.
The second direct sum representation ofA gives another for B.
Supposethat Ini> 1 for 1<i<vandni= 1forv<i<r;
then a similar argument shows that the type of B is
{n1—1,..., n1,— 1}.
We may use the induction hypothesis here, B having smaller
96 Finitely generated abelian groups

order than A. We conclude firstly that u = r; and secondly

that m1—1 = nl—l, ..., mu—l = nu—l.
Thus mi=ni for 1<i<u, while mi: 1 and ni= 1 for
u < i g r. The two proposed types of A are therefore the same,
and the inductive proof that the type of A is an invariant of A
is complete. [:1
Note that a finite abelian p—group of type {n1,..., n,} has order
p” where n = n1—|—...+n,. Groups in which each n1. is 1 are
called elementary abelian p-groups; they are just isomorphic
copies of direct sums of groups of order p, or just the abelian
groups in which every element has order p or 1. We proved (in
Chapter 4) that any group with each element of order 2 or 1
must be abelian, and now we see that (ifit is finite) it is isomorphic
to the direct sum of groups of order 2.
We can now give a complete description of the groups under
study.

THEOREM 5.17. A finitely generated abelian group determines,

and is determined up to isomorphism by, the following invariants:
the rank, the (finite) set ofprimes p occurring as orders of elements,
and the type (for each p) of the subgroup formed by the elements of
p—power order.
Proof. The last two theorems, together with the facts that the
periodic subgroup and the p-subgroups are unique finite sub-
groups, show that the group determines the invariants men-
tioned. Now suppose that a set of invariants is given—rank g,
set {p1,..., 10,} of primes, and a type for each prime. If type
{n1,...,nk} corresponds to the typical prime p in the given set,
we construct the direct sum of cyclic groups of orders 12%.”, p”b
respectively; we do this for each p, take the direct sum of all the
resulting groups, and take the direct sum of this with g infinite
cyclic groups. The outcome is a finitely generated abelian group
with the given set of invariants. l]
We describe briefly another set of invariants of the finitely
generated abelian group A. The periodic subgroup P of A is
 finitely generated abeltan groups 97
the direct sum of groups of order p?“ where l g i g r and
mi. > 0, While each got—subgroup is the direct sum of cyclic
groups of orders 10’7"“, gm”, 10?“, say; here we allow some of the
”to be 0, but we still suppose that each ”u > O for at least one
value of 2'. We may without significant loss of generality
assume th t
a r1>r2>--->n,
and that m1 2 M2 > 2 ”as
for each 13. Let us write the orders of our direct summands of P
as a rectangular array the better to visualize the position.
p’fn I)?“ . . _ pill]:

20?“ I)?“ - . - 103“

pg”. purl _ . _ nrk

Now the cyclic direct summands of orders p”11”.,’.'flgenerate

a cyclic subgroup of order 10““... 11”“ ,in general, the cyclic direct
summands of orders 11”“! ., Z?" generate a cyclic subgroup of
order 101“!" .101"! for l < j"?
<k, because the primes p1," ., p, are
distinct. Put elf: 101‘”.... .Our assumptions imply that dj'1s
divisible by dj+1 for 1 g j”< Is. It is easy to see that P is iso-
morphic to the direct sum of cyclic groups of orders d1,..., dk;
and that P may be (and sometimes is) specified by these numbers,
which are invariants of A.
Although we now have a rather comprehensive description of
the finitely generated abelian groups, it does not sufl‘ice to
answer all possible questions about the structure of such groups.
The following theorem, the proofoffered for which is independent
of the theory so far developed, answers one question of this sort.

THEOREM 5.18. If an abeltan group is generated by n elements

then every subgroup of it is generated by n elements.

Proof. We use induction on n. When n = 1 the group in

question is cyclic, and Theorem 3.21 informs us that every sub-
group is also cyclic.
853137 E
98 Finitely generated abelian groups

We suppose therefore that the abelian group A is generated

by n elements, with n > 1. Note that we are not assuming that n
is in any way minimal, as we might if (for instance) we wished to
apply our structure theory. Let a1, . .., an be some set ofgenerators ;
then each element of the subgroup B has some expression as
[31 al+ +Bn an where [91,..., [in are suitable integers. If it should
happen that every element of B has such an expression with
[31 = 0, then B < gp{a2,...,an}; and in this case the inductive
hypothesis ensures that B is generated by some set of 12—1 of
its elements. It certainly follows that B is generated by n
elements.
In the general case there will be an element, b say, in B with
an expression of the above form in which [31 _-,E 0; replacing b by
—b if need be, we may take it that [31 > 0. We make a definite
choice of b; we take b as any element of B for which there is a
representation as Bl a1+...+Bn an in which [31 takes its minimal
positive value. Take next a general element x of B. It will have
some expression of the form

x = 71a1+m+7naw

The usual application of the division algorithm gives

71 = 9181+7'a 0 < 7' < I31-

The element x—qb lies in B, and we have

x—qb = (3’1“1+"'+7nan)_Q(fila1+"'+/3nan)

= 701+(72—9I32)“2+ - “ + (7n—'qpn)an'

But [31 had a certain minimal property, which tells us that since
0 g r < [31 we must have 7' = 0; that is, x—qb egp{a2,...,an}.
The significance of this is that B = gp{b}+(gp{a2,...,an} n B).
At this point the inductive hypothesis may be applied.
Therefore the elements of B which lie in gp{a2,...,an} form a
subgroup With a generating set of 12—1 elements. Since B is
the sum of this subgroup and gp{b}, it follows that B has a set ofn
generators. I]
 Finitely generated abelian groups 99

Problems
1. Express as the direct product of cyclic subgroups that subgroup of
the multiplicative group of non-zero rationals generated by {6, —-6}.
2. Show that the multiplicative group generated by the matrices A
and B is abelian if

A=(_} i), B=(:i J)-

Express this group as the direct product of cyclic subgroups in two ways.
3. Show that the permutations
(1, 2, 3, 4)(5, 6, 7,8) and (1,5, 3,7)(2, 6, 4, 8)(9, 10, ll)
generate an abelian subgroup A of the symmetric group Sn, and that A
has order 24. Express A as the direct product of cyclic subgroups.
4. Express all the abelian groups of orders 99 and 100 as direct sums of
cyclic groups.
5. How many (non-isomorphic) abelian groups of order 106 are there?
How many of them are cyclic ?
6. The infinite cyclic groups A, B are generated by the elements a, b
respectively. Find (in terms of a and b) all pairs of elements which
generate A (—9 B.
7. Find all finitely generated abelian groups A that have no characteristic
subgroups other than A and 0.
8. Let A be the abelian group formed by taking the direct sum of cyclic
groups of orders d1,..., dk. Prove that, if dj+1 divides dj for l g j < k
and if dk > 1, then no set of k— 1 elements of A can generate A.
9. The abelian group A is defined to be A1 (-13 69 A" where Ai = gp{a,-}
for l g 7} g 'n. The subgroup B is defined to be gp{b}, Where
b = “1+132‘12+---+I3nan
for certain constants [32,..., ,3". Prove that
A = B (43A2 EB... EBA”
Whenever a1 has infinite order; and that if (11 has finite order m then
A = B ®A2 (ENG—>117,
if and only if mflz a2 = 0, ..., mflnan = 0.
10. The abelian group A is generated by elements a and b for which
2a = 36. Express A as the direct sum of (not necessarily non-trivial)
cyclic groups.
1 l. Prove that, if B is a subgroup of the finitely generated abelian group
A, then the rank of A equals the sum of the ranks of B and A/B.
100 Finitely generated abeliom groups

12. Let A denote the direct sum of the cyclic groups generated by the
elements a1, a2, a3, of orders 133, pg, 322 respectively (p is a prime), and let
B = gp{(pa1, 102%, 0), (0,10%, 10%)}. Find elements b1, b2, b3 and constants
[31, [32, '83 such that
A = gP{51} 69 gp{bz}® gp{bs}.
B = gpilgl b1} Q gPiBz b2} ('9 gpifla b3}-

13. In the notation and situation of the previous question, let

0 = gp{pa1, (102+p)a2, 10003}. Can the b1; and the fl,- be so chosen that, in

add't‘
1 1°“, 0 = gpelbeegpwzbeegpwaba}
for some constants y1, ya, ya?
14. Determine which of the following statements about finitely generated
abelian groups are true and which are false.
(i) Every factor group of a torsionfree group is torsionfree.
(ii) Every infinite, finitely generated abelian group has a set of
generators, each of which has infinite order.
(iii) If A1, A2 are subgroups of the group A with A1 g A2 then
A/A1 g A /A2.
(iv) If A1, A2 are subgroups of the group A with A/A1 ; A/Az, then
A1 g A2.
(V) Every finitely generated abelian group has a subgroup of prime
order.

15. Let the abelian group A be generated by its elements a1,..., an, and
let m1,..., m" be given integers with highest common factor 1. Show, by
induction on [mlH— ...+ [mn| or otherwise, that there exists a set b1,..., b"
of elements generating A, with
61 = m1a1+ ...+mnan.

**16. Let the abelian group A be generated by a set of 1:: elements, and
choose a set {a1,..., (1),} of generators with the following properties:
(i) the order of a, is no less than the order of a,-+1 for 1 g i < k
(infinite order is taken to exceed all finite orders);
(ii) there is no generating set {b1,.. ., bk} which, When similarly arranged,
is such that for some m < k the orders of the first m pairs {a,,:, bi}
are the same, while the order of bm+1 is less than that of aml.
By using the result of the preceding problem or otherwise, prove that
A z gp{a1} 69 69 grim}-
17. An abelian group A is said to be divisible if the equation me = a has
a solution a; for all positive integers 'n and for all a e A. Prove that no
finitely generated abelian group is divisible. Show that the additive group
 Finitely generated abelirm groups 101

Q of rationals, and the group Z; (see Example 1.09), are each divisible.
Hence, or otherwise, prove that neither of these groups is finitely
generated.

18. Let Hom(A, 0*) denote the set of homomorphisms from the finite
abelian group A into the multiplicative group 0* of non-zero complex
numbers (see Problem 3 of Chapter 2).
(i) Show that Hom(A, 0*) is an abelian group. Express it as the direct
sum of prime-power cyclic groups, given such a decomposition
for A.
(ii) Prove that A is isomorphic to Hom(A, 0*) in a natural way.
(iii) Prove that, if a in A corresponds to Xa under this isomorphism,
then
b =01,
for alla,binA. X“ X
6

Normal structure

THE classical theory of finite groups (apart from the theory of

group representations) rests on two pillars—normal structure
and Sylow structure; the former is concerned with properties
defined by normal subgroups and factor groups, the latter with
the existence of certain subgroups of a group and arithmetic
properties involving the orders of these subgroups. The present
chapter and the next are therefore devoted to basic ideas of
normal structure and Sylow structure. Though finite groups are
the prime object of study we do not exclude infinite groups
except when we find it absolutely necessary.
Our first results are elementary technical theorems about
normal subgroups and factor groups, indeed they are little more
than collections of lemmas, but they are none the less essential
to our progress.

THEOREM 6.01. Let U, V, S be subgroups of the group G such

that U is a normal subgroup of V. Then S n U is a normal sub-
group of S n V.
Proof. Clearly S n U is a subgroup of S, and it is contained in
S n V since U < V; it follows that S n U is a subgroup of S n V.
Take arbitrary elements u, v in S n U, S n V respectively,
and consider 'v—luo. Since U <1 V we have o—luu e U; since
u, o e S we have o—luo E S. Hence u—luv E S n U. It follows that
o‘1(Sn U)v g Sn U
for all u E S n V. This suffices to show that S n U is normal in
SnKD
 Normal structure 103

COROLLARY 6.02. If U is a normal subgroup, and S is any

subgroup, of the group G, then S n U is a normal subgroup of 6’.
Proof. Take V = G in the theorem. [1
COROLLARY 6.03. If U is a normal subgroup of G and if U is
contained in the subgroup 5’ of G then U is a normal subgroup of 8'.
Proof. Apply Corollary 6.02.D
Having studied normal subgroups in relation to subgroups Of
a given group, we turn to the effect on subgroups of taking a
certain kind of factor group of a given group.

THEOREM 6.04. If N is a normal subgroup of the group G and if

the subgroup A of G contains N, then A /N is a subgroup of G/N.
If in addition A is normal in G then A /N is normal in G/N.
Conversely, if N is a normal subgroup of the group G, then each
subgroup of G/N has the form A /N for a suitable subgroup A of G.
In addition each normal subgroup of G/N has the form A/N for
a suitable normal subgroup A of G.
Proof. If N <1 G and N g A g G then Corollary 6.03 shows
that N <1 A. Therefore A /N exists. An arbitrary element of
A /N has the form Na where a 6A, and so a e G; therefore
Na 6 G/N, and A/N E G/N. We prove that A/N g G/N by
the usual subgroup criterion, Theorem 3.06. First, A /N is
non-empty as N E A/N. Second, if Na, Nb 6 A/N then
(Na)(Nb)-1 = Nab-1 e A/N since ab—1 E A; of course, a, b EA.
and A is a subgroup. Therefore A /N < G/N.
Now let A <1 G; thus g—lag e A for all a, g in A, G respectively.
Arbitrary elements Of A/N, G/N have the forms Na, Ng respec-

tmfly' (Na-«Nome = Ng-lag e A/N

because g— ag EA. Therefore (Ng)—1(A/N)(Ng) g A/N for all
Ng in G/N, and this suffices to establish that A/N is normal in
G/N.
Next, the converse. Although this could easily be proved by
arguments like those above, we prefer to rely on the homo-
morphism concept for the sake of simplicity and variety.
104 Normal structure

Since N is normal in G, the group G/N exists; call it H. There

is the natural homomorphism 95 from G to H, with kernel N——
it simply maps 9: onto N3:. Let K be any subgroup of H. Define
A to be the subset {a : cut 6 K} of G. We shall prove that A is the
subgroup of G which we seek.
Clearly 1 e A. If a, b e A then
(ab‘1)¢> = (a¢)(b¢)-l E K
since aq’), b¢ e K, and K is known to be a subgroup of H. By the
usual subgroup criterion A is a subgroup of G. The restriction
of 95 to A is, of course, a homomorphism of A into H and indeed
onto K; and the kernel of ¢ is, of course, N. By the first isomor-
phism theorem (4.07), A /N g K, as required.
Suppose, finally, that K is normal in H. We have to show that
A <1 G. Take arbitrary elements a, gin A, G respectively, and
consider g—lag. Since A /N is normal in G/N we have
(N9)‘1(Na)(N9) 6 A/N,
that is, Ng-lag e A/N. Thus g‘lag e A,
g-lAg S A.
This suffices to prove A normal in G. [I

COROLLARY 6.05. Let N be a normal subgroup of the group G.

Then there is a one—one correspondence between subgroups of G
containing N and subgroups of G/N; and in it normal subgroups
correspond to normal subgroups.
Proof. Let N g A g G. The procedure of the theorem was
to map A to A /N, and to associate with each subgroup K of
G/N a subgroup A of G containing N and satisfying K = A /N.
The proof makes it clear that if A /N = B/N then A = B. It is
trivial that A = B implies A,’N = B/N. The desired one—one
correspondence therefore exists. The statement about normal
subgroups should now be obviousfl
Our next task is again routine: we ask if there are operations
for normal subgroups analogous to set—theoretic intersection
and union.
 Normal structure 105

THEOREM 6.06. Let {MzA e A} be a set of normal subgroups of

the group G, where A is some index set. Then I] N,\ is a normal
EA
subgroup of G.
Proof. Theorem 3.07 implies that if N = {1N}, in these
E

circumstances then N is a subgroup of G. Take arbitrary ele-

ments 9, n in G, N respectively. Since n e N,\ for all A e A, and
since N,\ <1 G’, we have
9—1729 6 NA-
Since this holds for all A E A we deduce that g‘lng e N; that is,
g—lNg E N. This suffices to prove that N is normal in CH]
An important aspect of this theorem is that it gives a meaning
to the expression ‘the normal subgroup generated by the
complex 0 of the group G’. We may define this meaning to be
‘the intersection of all normal subgroups of G’ containing 0”,
the theorem guaranteeing that this intersection is itself normal
in 0. It may be regarded as the least normal subgroup in which
0 lies.
Example 6.07. We determine the normal subgroup N
generated by the complex {(12)} in the symmetric group Sn of
degree n 2 2; see Problem 10 of Chapter 3. Let a = (12),
b = (12...n). An easy calculation gives b—iabi = (i+1, i+2) ,
for i = 0, 1,...,n—2; therefore (i+1,i+2) e N for such i. But
every element of 8,, can be expressed as a product of such
elements. Therefore N 2 Sn, and we find that N = Sn.
The reader may be able to find a description (which we omit)
for the normal subgroup generated by a complex, along the lines
of Theorem 3.10.
We saw in Chapter 3 that unions of subgroups are not usually
subgroups, and unions of normal subgroups are no more interest-
ing to study. As the subgroups generated by a set of normal
subgroups and the complex product of the latter are closely
related, we shall analyse this situation in detail and with care.
THEOREM 6.08. Let A and B be subgroups of the group G. Then
AB is a subgroup of G if and only if AB = BA.
 106 Normal structure

Proof. Suppose AB = BA where, of course, AB is the

complex product of A and B. Since A and B are subgroups AB
is certainly non-empty; 1 GAB. Take two elements of AB,
say ai b, for t = 1, 2 (with a, e A, bi e B). We consider, as usual,
their quotient (u1 b1)(u2 b2)‘1, which is al b1 bz‘laz‘l. Since
BA = AB there are elements a, b in A, B respectively for which
(b1 b;1)u2—1 = ab.
Thus (ct1 bl)(u2 b2)'1 = a1 ab 6 AB. It follows from the subgroup
criterion of Theorem 3.06 that AB is a subgroup of G.
Conversely, suppose that AB is a subgroup of G. An arbitrary
element of BA has the form be, with a e A, b e B. Then
(ba)—1 = a—lb—1 G AB, because A, B are subgroups; hence
be GAB because AB is a subgroup. Therefore BA S AB.
We cannot claim here that AB S BA ‘similarly’, because we
do not know that BA is a subgroup. However, the proof is not
diificult. Take an element x in AB. Then 90-1 G AB (which is
a subgroup), so 96-1 has some expression as ab, with a 6A,
6 e B. But then x = b-lu—l 6 BA since A, B are subgroups.
We have thus proved that AB E BA.
It follows that AB = BA, as requiredfl
Only the former half of the above theorem is strictly relevant
to our applications.

COROLLARY 6.09. If A and B are subgroups of the group G such

that AB = BA then gp{A, B} = AB.
Proof. By the theorem AB is a subgroup, and as AB 2 A,
AB 2 B, we have AB 2 gp{A,B}. But, on the other hand,
gp{A,B} certainly contains A and B and so all elements ab
(where a GA, b e B); thus gp{A,B} 2 AB. It follows that
gp{A, B} = AB. [I

COROLLARY 6.10. If A is a subgroup of the group G and N is

a normal subgroup of G then gp{A,N} = AN.
Proof. The definition of a normal subgroup N implies that
xN = Nx for all a: in G. It certainly follows that AN = NA for
 Normul structure 107

all subgroups A of 0. Corollary 6.09 now gives the desired

result.[|

THEOREM 6.1 1. If Nl, N2 are normal subgroups of the group G’

then NlN2 is also a normal subgroup.
Proof. The previous theorem and its corollaries make it
abundantly clear that N1 N2 is a subgroup of G. Normality may
be shown directly from the definition: if x is any element of G'
then m—lMx = M for 73 = 1, 2, and so
w‘lNlNzx = x‘ll-w‘ll‘éx = lvllVZ:
as required. [I
We turn now from the study of normal subgroups to study of
factor groups. The need for the two theorems which follow may
best be felt in terms of homomorphisms, for they answer these
questions:
(i) If A is a subgroup of G and 95 is some homomorphism of G
with kernel N, what (in terms of G/N) is the image of A
under qt regarded as a homomorphism of A ?
(ii) If homomorphisms with given kernels map 0 onto G1
and G1 onto G’2 respectively, how are G, GI, 92 related?
Theorem 6.04, which showed that if N g A then A/N is a
subgroup of G/N, is related to the following answer to (i):
THEOREM 6.12 (Second isomorphism theorem). If A is a
subgroup and N 'is a, normal subgroup of the group G then
A/(AnN) g NA/N.
Proof. Note that, by Corollaries 6.02, 6.10, and 6.03,
A n N <1 A, NA is a subgroup of G, and N <1 NA.
The idea of the proof is to define a homomorphism from A
onto NA /N, to determine its kernel K, and to assert that
A /K g NA/N by the first isomorphism theorem.
We define a mapping 95 from A to NA /N by putting 00¢ = Na.
The general element of NA [N has the form NM or Na (here
u E N); so 95 is onto NA/N. Clearly 45 is a homomorphism:
(195.1156 = Na.Nb = Nab = (ab)q$.
108 Normal structure

If x e K then 9095 = N, because N is the identity element of

both G/N and NA /N. But Nx = N means precisely that 91: EN.
OfcourseqSis amapping ofA, sox EA n N. Thus K g A n N.
Conversely, if y e A n N then 31¢ is defined and y E N shows that
7145 = 1, so 3/ SK. It follows that K = A n N.
By the first isomorphism theorem, then,
A/(A n N) g NA/N.[]
Example 6.13. Take G to be the symmetric group 84, in which
N is to be gp{(12)(34), (13)(24)} and A is to be gp{(1234)}. A Venn
diagram to describe the situation may be permitted:

5+

0
It should be verified that N <1 84 and that A n N = {1, (13)(24)}.
Thus A n N has index 2 in A, and A /(A n N) consists of these two
cosets‘ A n N = {1, (13)(24)},
(A n N)(1234) = {(1234), (1432)}.
On the other hand NA consists of eight elements:
{1, (12)(34), (13)(24), (14)(23), (1234), (13), (1432), (24)}.
It follows that NA [N has just two elements:
N = {1, (12)(34), (13)(24), (14)(23)},
N(1234) = {(1234), (13), (1432), (24)}.
Just as the theorem predicted, we find that A /(A n N) g NA/N.
The homomorphism 95 of the theorem would map (1234) in A
onto N(1234), so in the final isomorphism (A nN)(1234) would
correspond to N(1234).
A lemma Will precede our answer to question (ii) above. It is
unnecessarily general for the task in hand, but significant
in numerous other contexts (such as the proof of Theorem 4.07).
 Normal structure 109

LEMMA 6.14. If 95 is a homomorphism from the group G1

into the group G2 which maps N1 into N2 (N1, N2 being normal sub-
groups of G1, G2 respectively) then there is a homomorphism gt
from G'l/N1 into G'z/Nz, defined by
(M 93M = Nz(x¢)
for all x in GI.
Proof. It is far from trivial that 1/: is a mapping—it could
conceivably happen that N2(xq$) depends essentially on the
representative at chosen from the coset l. We therefore have
to show that if ll = My then lyrics) = Nz(g¢). From
l = My we have 3/ = nx for some n eNl; so
31¢ = (“>95 = (“95)(9045) E N266?”
because g!) is a homomorphism. It follows that Nz(y¢) = N‘2(x¢),
by Theorem 3.12, as required.
To prove that ¢ is a homomorphism is now straightforward.
If g, h e 01, then

(Night-(1)71 70% = Nz(9¢)-Nz(hd>) = Nz(9<l>)(h¢)

= 2{(9h)¢}
= (M 9W1
since ¢ is a homomorphism, and the lemma is proved. D

THEOREM 6.15 (Third isomorphism theorem). If M and N

are normal subgroups of the group G such that M contains N then
(WM/(MIN) g Ci/M-
Proof. Note that we have evidence (in Corollary 6.03 and
Theorem 6.04) to assert that M[N exists and that it is a normal
subgroup of G/N.
The idea of the proof is to define a homomorphism it from
G/N to G/M, to find its kernel, and again to appeal to the first
isomorphism theorem.
We try to define 1/: by
(Nx)z/: = Mx
forallx e G, and Lemma 6.14 (With G1 = G,Nl = N, G2 = G/M,
110 Normal structure

N2 = l, and g!» the natural homomorphism from G’ to G/M)

shows that this is indeed a definition of a homomorphism.
The kernel is a normal subgroup of G/N and (by Theorem
6.04) has the form K/N where K is some normal subgroup of G.
Let Na: 6 K[N, then (Nagy/t = M. But Mac = M is equivalent
to :1; EM, and so K/N g M/N. Since it is clear that ifx EM
then (Noah/1 = M and so N90 6 K[N, it follows that K[N = M/N.
The first isomorphism theorem therefore shows that

(CHM/(MW) g G/M-D
Example 6.16. Let G be Z, the additive group of integers, and
let Ir, 8 be two positive integers. There is a homomorphism 45,8 of
Z onto the group Zn of the residue classes modulo rs defined by
ac/zrs = [at]. Its kernel is N, consisting of all integral multiples
of rs. Thus Z/N g Z”. Now e in turn has a homomorphism
95, onto the group Zr of the residue classes modulo 7‘, defined
by [ah/z, = [a]; note the two meanings of [a]. The kernel of 95,. is
{[a] :a is a multiple ofr}. This is a subgroup of e, and corresponds
to a subgroup M/N of Z[N. The theorem above confirms that M
consists of all multiples of 7'.
Example 6.17. We consider a homomorphism g6 from the
group 0 onto an abelian group, with particular reference to
Theorem 4.21. Thus the kernel 11! contains 8(G). By Theorem

“5 we have G/M g {mam/{mm}.

In terms of homomorphisms, this states that 45 = 951962 Where
¢1 is the natural homomorphism of G onto G/8(G) and 462 is
the natural homomorphism of G/8(G) onto {G/8(G)}/{M/3(G)}.
Thus every homomorphism of G onto an abelian group may be
expressed as the product of ([51 and a suitable homomorphism
of G/8(G).
We now approach the crucial result on normal structure of
finite groups, Theorem 6.23 below due to Jordan, Holder, and
others. The preparations for this include a result which is
traditionally known as Zassenhaus’s lemma but which is in fact
a fourth isomorphism theorem :-
 Normal structure 111

THEOREM 6.18. Let A, B be subgroups of the group G’ and let

X, Y be normal subgroups of A, B respectively. Then
X(A n B)/X(A n Y) g Y(A n B)/Y(X n B).
Proof. The statement of the theorem is meant to include
implicitly the fact that the required factor groups exist.
We shall prove that X(A h B)/X(A n Y) is isomorphic to
(A n B)/(A n Y)(X n B). This will sufiice, for the latter factor
group, which is unaltered when X, A are interchanged with Y,
B respectively, is then clearly isomorphic to Y(A n B)[Y(X n B)
also; the required result follows.
Because A n B is a subgroup, and X is a normal subgroup, of
A, we see (from Corollary 6.10) that X(A n B) is a subgroup of
A. But X is normal in X(A n B), and we can meaningfully con-
sider X(A n B) /X. The second isomorphism theorem gives
X(A n B)/X g. (A n B)/(A n B n X) = (A n B)/(X n B).
Now (A n B)/(A n Y)(X n B) isafactorgroup of(A n B)/(Xn B),
for quite generally if we have N1, N2 s1 G then NIN2 <1 G,
Nz <1 NlNz, and
CT'IN1N2 g. (G/N'zVWlNz/Nz)
by the third isomorphism theorem; of course both An Y and
X n B are normal in A n B, by Theorem 6.01. There is, therefore,
a homomorphism from X(A n B)/X onto (A n B) /(A n Y)(X n B).
Our tactics now are to determine the kernel, KIX say, of
this homomorphism and then to appeal to the first isomorphism
theorem. If c lies in the kernel, with lo 6 A n B, then
k e (A n Y)(Xn B), and it follows that K g X(A n Y) as we are
assuming that K 2 X. Conversely, take k EX(AnY), then
k eX(AnB), and the image of Xla is clearly trivial. Hence
K/X = X(A n Y)/X.
The first isomorphism theorem shows that
{X (A n B)/X}/{X(A n Y)/X} g (A n B)/(A n Y)(Xn B).
After simplifying the left-hand side by an application of the
third isomorphism theorem we obtain
X(A n B)/X(A n Y) g (An B)/(A n Y)(Xn B).
112 Normal structure

This, as was pointed out above, leads to the assertion made in

the theorem. L]
Preparatory definitions are also necessary. All series of sub-
groups to be considered, it should be noted, will be finite series.

DEFINITION. A normal series of a group G is a finite series

of subgroups
G=%9%9m9@flaa=l
such that Gi is a normal subgroup of G,“1 for t = 1,..., r. The
length of the series is the integer r.
Note that we do not require each G, to be normal in G, and
that a series of length r involves r+1 members (including G
and 1).
Example 6.19. The symmetric group S4 contains a normal
subgroup V of order 4, V = gp{(12)(34), (13)(24)}, as was ob-
served in Example 6.13; and V contains normal subgroups of
order 2, for instance U = gp{(12)(34)}. So we have the following
normal series for S4, in which U is not normal in S4:
&>V>U>L
We could of course insert further terms, even having repetitions
if we wanted.

DEFINITION. If the group G has two normal series then the

second is a refinement of the first if each member of the first
occurs as a member of the second also.
Thus a refinement of a given normal series may be made by
inserting repetitions. A less trivial example is the following
refinement of the above normal series for S4:
&>A>V>U>L

DEFINITION. Two normal series for a given group G, say

G: G09 G1|>
/ / / GT: 1,
l> G,_1l>
G=H012H19...I>H,_112H3=1,
 Normal structure 113

are isomorphic if there is a one—one correspondence between

the set of non—trivial factor groups Gi_1/G1- and the set of non-
trivial factor groups HJ._1/Hj such that corresponding factor
groups are isomorphic.
Example 6.20. Take G to be Z6, the cyclic group of order 6.
Two normal series for G are
Z6>Z3|>1, Z61>Z2>1
where Z2, Z3 denote the subgroups of orders 2, 3 respectively.
We make Zs/Z3 correspond to Zz/l, and Za/l to ZG/Zz; since
corresponding factor groups are isomorphic, the two series are
isomorphic.
THEOREM 6.21. Any two normal series for a given group have
isomorphic refinements.
Proof. Let the two normal series for the group G be those
named in the preceding definition. We construct a refinement
of the first by inserting between Ga;1 and 0%, for each i, the
following series:
Gi—l = Gi(Gi—1nHo) 9 Gi(G¢_1flH1) 9 9 alder—1011.9): Gi'
Observe that HP1 I; H,- implies that
Gi(Gi—1 n ‘Hj—l) 9 Gi(Gi—1 n Hi)
by arguments produced in the proof of Zassenhaus’s lemma
(Theorem 6.18). We thus obtain a refinement of length rs.
A similar refinement may be obtained for the second series by
inserting between each Hid and Hi the series
Hj—l = Hj(Gon Hj—1) 9 Hj(G1n EH) 9 9 Hj(Grn EH) = Hr
We now claim that these refinements are isomorphic. Indeed
this is a trivial consequence of Zassenhaus’s lemma, which
ensures at once that
Gi(Gi—1 n Iii—IVGilGi—l n Hy) g Luau—1 n Hj-i)/Hj(ai n Hj—l)
for 1 g i g r and 1 g j < s. The theorem is now proved.|:|
The natural thing to do next is to take a normal series for a
group G and to construct a refinement which cannot be further
853137 I
1 14 Normal structure

refined in a non—trivial manner (that is, Without the insertion of

repeated terms). This can certainly be done for a finite group G.
If it is accomplished then the factors Gil 0H1 of successive terms
will have no proper normal subgroups.

DEFINITION. A simple group is a group in which no proper

subgroup is normal.

DEFINITION. A composition series of a group G is a normal

genes G=G0|>G1|>...>G,_11>G,=1
such that G1; is a proper subgroup of GL1 and G,_1/G) is simple,
for l g i g r. The groups G’id/Gi are called the composition
factors.
The general aim of theorems about normal structure is to
relate a given group to simple groups—the composition factors
are a set of simple groups associated with G. A fundamental
task is therefore to describe the finite simple groups. For
abelian groups this is easy, for the groups in question must have
no proper subgroups and are therefore of prime order (or trivial).
From what we know so far these might be all the finite simple
groups. In fact there are many others, and a complete classifica—
tion of them is at the present time an unsolved question. Among
the well—known examples are the finite alternating groups of
degree five or more, which we shall study in Theorem 11.08
Meanwhile one specimen must suffice.
Example 6.22. Consider the alternating group A5. Its non-
trivial elements have order 2, e.g. (12)(34), or 3, e.g. (123), or 5,
e.g. (12345), so a proper normal subgroup N contains an element
of order 2 or 3 or 5. In fact there must be an element of order 3.
We have [(12345), (13)(24)] = (135),
and if N has an element of order 5 we may assume that
(12345) e N; and since a e N implies [(1, x] e N for all a: 6 A5 we
have (135) e N. Similarly if N has an element of order 2 it can
be taken to be (13)(24) and again (135) E N.
 Normal structure 115

It follows from this that N contains all the elements in A5

of order 3. For if (135) E N and. we want to show that (123) e N
then we calculate the conjugate (235)(135)(253) of (135), which
must lie in N; in fact it is (123), as required. A similar argument
shows that if (afl‘y) e N then (oz/38) E N where a, fl, 3/, 8 are
distinct. This suffices to prove that N contains all elements of
order 3 in A5.
Every element of A5 can be expressed as the product of an
even number of transpositions, and we have, with a, [3, y, 3 as

abm’ (04m = (cc/3y).

(cc/9)(r3) = (afirxa‘éfl-
Thus every element of A 5 can be expressed as the product of
elements of order 3 and thus lies in N. It follows that N = A5
and that A5 is simple.
Direct products of finite groups enable one to construct
examples of isomorphic groups with somewhat different (though
still isomorphic) composition series. The Jordan—Holder
theorem ensures that the composition series of a given group
cannot differ very widely.
THEOREM 6.23 (Jordan—Holder theorem). In a group with
a composition series every composition series is isomorphic to the
given series.
Proof. We note that a composition series of the group G
cannot be properly refined to another composition series, since
all the composition factors are already simple non-trivial groups
(as long as G 7E 1, at least). Suppose two different composition
series are given. Then Theorem 6.21 enables us to construct
isomorphic refinements, one to each series. Since each series is
isomorphic to its refinement once repeated terms are omitted, it
follows that the two original series were isomorphic. This is the
assertion of the theorem. [1
Note that in particular the set of composition factors as
well as the length of a composition series is a group invariant.
But the composition factors do not determine the group, for it
116 Normal structure

is easy to see that the two abelian groups of order 4 have iso-
morphic composition series.
COROLLARY 6.24. Any two composition series of a finite group
are isomorphic. I]
We must observe that the Jordan—Holder theorem tells us
nothing about finite abelian groups, in which all the composition
series can easily be found. Neither does it tell us anything at all
about the finite simple groups. Its use is with finite groups with
a fair number of normal subgroups which are nevertheless non-
abelian.
For infinite groups, less information on structure can be
found. There are groups with no composition series in our strict
sense, and even if we are prepared to liberalize our definition
of composition series the Jordan—Holder theorem may not be
valid.
Example 6.25. Let G be the additive group Z of integers, and
let G, = gp{p"}, where p is an arbitrary prime, for i = 0, 1, 2,... .
Because Gfil <1 Go and Gi/G141 has order p, and n G; = O, the
serles G = G0 D G1 D 02 D on

has some claim to be a composition series. But the order p of

G'i/G‘i+1 can be selected at will, and we can hardly claim that such
a composition series yields invariants of G.
Our definitions and methods do give the following result
applicable to infinite groups.
THEOREM 6.26. Let G’ be a group with a composition series of
length r, and let
G=Go> GID ...[>G,._1[>Gr> 1

be a normal series for G in which Go/Gl,..., G,_1/G', are all simple

and non-trivial. Then G, = 1.
Proof. Refine the two given series according to Theorem 6.21,
which states that isomorphic refinements exist. The first
refinement is a composition series of length r, and therefore
 Normal structure 1 17

the second is a composition series of the same length; we omit

repeated terms. It follows that G, = 1, as required. |:]
Finite group theory is the natural setting for composition
series. The idea arose from the study of polynomial equations
and the resulting Galois theory, a subject which unfortunately
cannot be described here. Composition series with abelian
factors give rise to an extensive theory of finite soluble groups,
to be touched on in Chapter 10.
The remainder of this chapter is given over to a new topic,
a tool of importance in its own right but not of immediate
application, which has a close affinity to normal subgroups. Up
to the present we have considered operations (such as the direct
product) defined on a finite set of groups, though infinite sets
were not excluded when complications did not arise; we refer to
the intersection of a set ofsubgroups, and the subgroup generated
by a set of complexes. We shall now weave a new strand into
our studies by considering products of (possibly infinite) sets of
groups.
DEFINITION. The product of the complexes {05A GA} in
a given group, where the index set A is totally ordered,1' is the
union of all complexes of the form 0A. Gig-"OM where n is a
positive integer or zero, and A1 < A2 < < An. This product
is denoted by II 0» When n = O the product is interpreted
eA
as the element 1.
There is a strong constraint in group theory that leads towards
the above definition. In the group axioms the product :1:l
of two elements x1, x2 is made meaningful, and by induction
this definition may be extended to any finite number of factors.
What we cannot do without special and elaborate precautions
is to talk sensibly of the product of infinitely many elements.
The definition of QC,“ however, avoids this difficulty.
EA

Example 6.27. Consider the abelian group Z: displayed in

Example 1.09. Take any prime for p, and use the additive
1' See the Appendix.
118 Normul structure

notation. Define On to be {zzpnz = 0} for rt = 1, 2,...; thus A is

the set of positive integers. Then 2 0” is the whole group Z5,
n>0
for each element of Z; belongs to a suitable complex 0”; if 2
has order p“ then 2 e On.

We may in particular consider products of subgroups, even

products of normal subgroups.

THEOREM 6.28. If {M : A e A} is u non-empty set of normal sub-

groups of the group G then II M is also a, normal subgroup.
AeA

Proof. To prove the subgroup property we apply the usual

subgroup criterion. Take as, g e H M. Then
AeA

x 6 M1 ....NA' (A1 < ... < Ar):

3/ ENm-"Nm (#1 < < #3).

It follows that x, 3/ lie in gp{N,\1,..., N)”, NW.” NW}, a subgroup
which we shall call N. Therefore $314 6 N. But Corollary 6.10
shows that N = IVA; NILNM NW

and NA‘NM = NMNA‘ for 1 < l g r, 1 <j g s. It follows that

N S HM, and so cog-1 lies in the product. So the product is
REA
a subgroup. The normal property is evident. [I
We now consider the direct product and the cartesian product.
The direct product of afinite set of groups was defined in Chapter
4, and both the following definitions generalize the earlier concept.

DEFINITION. The cartesian product of the set {GAzA EA} of

groups is the set of all functions from A to {0;}, so that if g is such
a function then Ag 6 GA; the product of two such functions f, g
is defined by Mfg) = (Af)()\g). This cartesian product is denoted
by HG Gk
AEA

Our first task is to verify that the cartesian product G of the

groups GA is itself a group under the given multiplication.
 Normal structure 119

Closure obtains becausefg is by the nature ofits definition a func—

tion from A to {GSl : A e A}. The associative law follows from that
law in each factor because if f, g, h e G then

A{(fg)h} = {A(fg)}(Ah) = {(Af)(A9)}(Ah),

A{f(gh)} = (Af){A(9h)} = (Af){(A9)(Ah)},
for each A e A. The identity element is the function f for which
Afis the identity element of GA. The inversef‘1 off E G' is defined
by A(f—1) = (Af)-1 for all A e A.
In the special case when A is a finite set {A1,...,An} we may
identify the element 9 of G with the following ordered set of n
elements:
0‘19”": Arug)

It follows that the cartesian product G is simply the direct

product GAIX X 0A,. as defined in Chapter 4.

DEFINITION. Ifg is a function from a set A to the set {G'yA EA}

of groups, then the support of g is {[11,n 75 1}.

DEFINITION. The direct product of the set {GAzA EA} of groups

is the set of all functions from A to {G52 A EA}, each having finite
support; the product of two such functions f, g is defined by
A(fg) = (Af)(Ag). This direct product is denoted by F” GA.
GA
It is easy to prove that the direct product is also a group. The
key point is the fact that if 01, 02 denote the supports of
f1, f2 6 H” GA then the support of f1f2 lies in 01 U 02, and the
AeA
support off1—1 is 01. Of course, the identity function (which has
empty support) is assigned finite support by courtesy.
It is also easy to see that the cartesian product no GA contains
AeA
a subgroup isomorphic to H17 GA; and that if A is finite then the
AeA
new definition of direct product coincides with the earlier.
120 Normal structure

THEOREM 6.29. The direct product G’ of the groups (0't e A}

contains subgroups HA which are isomorphic to GA for each A and
which have the following properties:
(i) HA commutes with H” element by element if A gé (1.;
(ii) G = H HA;
AeA
(iii) HA n :HAH” =1for each A E A.
[HE

Proof. We define H,\ as the set of functions h such that ‘uh = 1

for all y, except A. Obviously HA is a subgroup isomorphic to GA.
(i) Take h 6 HA, h” e H“, with A # H; and take an arbitrary
element v in A. Then

VOL/Wt) = (VkAXVhM),
VWhA) = (V Milka)-
If v is neither A nor u then both right-hand sides are trivial. If
v = A then both equal t, and if v = ,u. then both equal why.
In any case v(h,\ h”) = v(h# h,\) and hAhM = h” h)‘, as required.
(ii) Let g be an arbitrary element of G. Then
Ag = 1 ifA ¢ {A1,...,An}
for some n 2 0. We define hi for 1 g i < n by
Aiht=Aigi All/1;: IfOI‘AgéAi.

Then g = h1 ...hn. It follows that G = H HA, as claimed.

AeA
(iii) H H” contains just those functions h of G‘ such that Ah = 1;
[BEA
this follows from the definition of product. The only such
function h in HA is the identity function. [1
Note carefully that A need not be a totally ordered set for
the product in (ii) to be meaningful, as (i) implies that the order
of factors is immaterial.
The above theorem is concerned with a situation in which the
direct product has been constructed from a given set of groups.
The next is a companion theorem in that it gives a criterion for
a given group to be isomorphic to the direct product of certain
subgroups.
 Normal structure 121

THEOREM 6.30. Let the group G have subgroups {GAJ EA}

satisfying the conditions:
(i) GA commutes with G.“ element by element if A 72 M;
(ii) G = H G,\ (note again that a total ordering on A is not
AEA
needed);
(iii) GAO H at = lfor each A EA.
[5%)
Then G g Hp GA.
AeA

Proof. Take an arbitrary element g of G. By (ii) it may be

written as ghl...g)m with 9A; e G)“ for 1 < i g n and Ai EA.
Indeed g has a unique representation of this form, for if it had
two then use of (i) and (iii) would show that corresponding factors
are the same.
We may now construct a mapping 915 from G to II” GA. We
2A
define q to be that element of the direct product which has
support {A1,..., An} and which maps A; onto 9A: for 1 g i g n; this
is properly defined as g determines g;k uniquely.
Suppose that It lies in G, so that It also has a unique representa-
tion as the product of factors from a finite number of GA. It is
clear that matters regarding g and h may be so arranged that
9 = gal-"9AM, k = thing"
where g)“, h)“ 6 GM for 1 < i < m, provided that some 9)“ and
h,“ are allowed to be 1. Then the supports of 995 and M will be
subsets of {A1,...,)\,,}, and

Ai{(g¢)(h¢)} = {Ai(g¢)}{%(h¢)} = ask».

for 1 g i g m. On the other hand gh = (g,\1 hA1)---(9Amh)\m): so
that the support of (gh)¢ lies in {A1,..., Am}, and

A«;{(9'h)q5} = 9M;
for 1 < i g m. This shows that (g¢)(h¢) = (gh)q$, and so <15 is
a homomorphism from G to HD GA.
(\eA
122 Normal structure

But 45 is one-one. If q = t then A(g¢) = A0095) for all A EA,

and it follows that the product representations of g and h coincide
factor by factor, so that g = 71. Further «la is onto the direct
product; given the function in HD GA, with support {A1,...,A,,J
<AEA
such that A- g = 9A; for 1< <m, we see that it is the 1mage
under 95 of 9". 9A... 6 g.
The required isomorphism G g H” GA has now been estab-
ASA
lished. D
We remark that there is a closely related theorem in which (i)
<10 for each A.
is replaced by: GA<1
All our examples of finite direct products are also examples of
general cartesian and direct products. A less trivial illustration
follows.

Example 6.31. Let A be a periodic abelian group. There is

clearlyin A a subgroup Ap consisting of all elements of p—power
order, for each prime 10. We shall prove by means of the above
theorem that A is, or more strictly is isomorphic to, the direct
sum of all these subgroups. (As usual we speak of the direct sum
and the cartesian sum when abelian groups are involved.) There
is no trouble over the commutative property (i) or property (iii).
Regarding (ii), we have seen (in the proof of Theorem 5.01) that
if A0 is a finite abelian group then an element a of order pi“ 102",
Where ppm, 10,. are distinct primes, is the sum of elements of
orders 10‘1“," ., 11,. The same holds good in A for if a e A then
gp{a,}'1s finite and can be taken to be A. Thus a e 21A I?" and
it follows that A: 2 A10, the sum being taken over all1primes p.
HenceA = 2” A1,.

It is rather more difficult to handle the cartesian product than

the direct product, and so it is fortunate that our main use for
the latter will be to construct examples. It is certainly true that
H0 GA contains copies of each GA, for these may be found in its
AeA
subgroup HD 0". We have properties (i) and (iii) of the direct
AeA
 Normal structure 123

product given in the above theorems, for the same reason, but
(ii) is by no means valid. In other words, {gA : Ag), 6 GA, pg), = l for
y. 75 A} is not a set of generators for H0 G’,‘ when A is infinite,
AeA
for no element ofinfinite support lies in the subgroup it generates,
which is precisely Hp GA.
AeA
We shall, however, prove one elementary theorem about
cartesian products.

THEOREM 6.32. If {N,\:A e A} is a set of normal subgroups of

the group G and ifN = n N,\ then G/N is isomorphic to a subgroup
AeA
0f EA” (Ci/NA)-
Proof. We shall define a mapping 95 from G into the cartesian
product P = H0 G/M, and apply the first isomorphism
ASA
theorem.
Let x95 be the function g e P such that Ag = Mac for each A
in A. To prove that 525 is a homomorphism, suppose y e 0 maps
onto h e P so that Ah = NW. Then (wy)¢ is the function
f e P for which Af = Macy, and we have

Af = NW?! = (NMXNA?!) = (AQXM) = M9")

for all A e A. Therefore f = gh, that is

(xii/M5 = (x¢)(?/¢)-
and g!) is a homomorphism.
Let K be the kernel of 96. If x e K then 00¢ is the identity
function, so that N,\x :2 N), for all A E A. This means precisely
that x e M, and it follows that x e N. Therefore K g N.
Conversely, it is clear that if x e N then x E N,\ and 9:95 is the
identity function, so that N < K. It follows that the kernel
of ()5 is N.
By the first isomorphism theorem, then, GIN is isomorphic
to the image of G in P under 49, and this is clearly a subgroup
of P. l]
124 Normal structure

Example 6.33. Let G be the additive group Z of integers, and

let N“ = gp{n} for n = 1, 2,.... Then N = n N“ = 0, so that
n>0
G/N g Z. On the other hand, G/Nn is isomorphic to the additive
group of residues modulo n. The theorem asserts that H0 G/Nn
n>0
contains an infinite cyclic subgroup. Indeed the proof specifies
a generator for this subgroup, namely the function 9 such that
fly = Nfl+ 1.
Note, however, that [[D G/Nn does not contain an infinite
n>0
cyclic subgroup because it is a periodic group.

Problems
1. Find a composition series for the subgroup of the symmetric group
8, generated by {(1234567), (243756)}. Find a second in which the first
composition factor has a different order.
2. Find a normal series of arbitrary positive length n for Z5” (see Example
1.09) such that n— 1 of its factor groups have order 10. Deduce that Z5”
has no composition series.
3. Prove that the only finite class of conjugate elements in an infinite
simple group is {1}.
4. The group G contains subgroups A, N1, N2 such that N1 g N,, N1 <1 G,
M <1 G. Prove that (A n Nz)/(A n N1) is isomorphic to a subgroup of
ZVZ/Nl'
Illustrate this result in the case when G = S4, N1 = V, M = A4,
A = gp{(12), (123)}, in the notation of the text.
5. A group G is said to be metacyclic if it has a normal subgroup N such
that both N and G/N are cyclic. Prove that every subgroup and every
factor group of a metacyclic group are again metacyclic.
*6. (i) Prove that if the group G = AB where A g G, B g G, then
there is a one—one correspondence between the cosets of B in AB and the
cosets ofA n B in A.
(ii) Prove that if G is a. group with subgroups A, B of (finite) coprime
indices then G = AB.
7. Prove that a finite group of order 39" has a composition series oflength n.
8. The subgroup Tn of the unrestricted symmetric group G on the positive
integers is generated by the transposition (2n—1, 2n) for n = l, 2,... .
Show that gp{T.,,:n > 0} is H” Tn.
n>0
Prove that G also contains a subgroup isomorphic to H” T”.
n>0
 Normal structure 125

9. Let N be a normal subgroup of the group G and let A be an arbitrary

subgroup of G containing N. Prove that there is a one—one corre-
spondence between the cosets of A in G and the cosets of A/N in G/N.
What relation does this result bear to the third isomorphism theorem?

10. Let G1, G2 be isomorphic groups with normal subgroups N1, N2 re-
spectively. Prove that if the isomorphism between G1 and G2 induces (by
restriction) an isomorphism between N1 and M then Gl/N1 g Gz/Nz.

ll. Prove the Jordan—Holder theorem for finite groups by using induc-
tion on the least integer for which there exists a composition series of
that length and by appealing to the second isomorphism theorem.

12. Let M, M be normal subgroups of G, and let N be a subgroup of G

which is a normal subgroup of both N1 and Nz~ Prove that

(N1/N)(Nz/N) = (N1N2)/N-
Let A be a subgroup of G and let M and N be normal subgroups of G
with M 9 N. Show that the image of A, when G is mapped onto G/N
and G/N onto (G/N)/(M/N) by natural homomorphisms, g MA/M.

13. Lemma 6.14 in the text shows that if 96 is a homomorphism from G1

into G2 mapping N1 into N, (where M is a normal subgroup of G; for
73 = l, 2) then there is a homomorphism 1/1 from G1/N1 into Gz/Nz. Suppose
that ([5 is onto G2 and its restriction to N1 is onto Nz. Prove then that III is
an isomorphism if and only if the kernel of 45 lies in Nl.

14. Let A, B, 0 be subgroups of the group G. Prove that ABC is a sub-

group ifand only ifABC' = BOA and BOB E ABC’.
Let G be the multiplicative group of all matrices of the form

1 a c
0 1 b ,
O 0 1

where a, b, c are integers. Take A, B, 0 to be the subgroups generated by

these matrices, respectively:

110 100 101

010, 011, 010.
001 001 001

Prove that G = ABC. Is AB a subgroup of G?

126 Normal structure

15. Let the group G’ have, as a subgroup of index 2, the simple non-
abelian group H. Prove that H is the unique subgroup of index 2, and
deduce that it is characteristic in G.
Deduce that SS has only one proper normal subgroup.
Is a non-abelian subgroup of index two always characteristic ‘3
*16. Let C’ and D denote respectively the cartesian and the direct
sum of groups of order p, as 10 takes every priIne value. Prove that if
a is any element of O/D then the equation m: = a has a solution as for any
positive integer n. (Additive notation is used here.)

**l7. Prove that the dihedral group D” of order 2" has the property that
Dn/§(Dn) ; D,,_1 for n > 3 (Dn is described in Example 1.25, and D2 is
to be taken as 0'2 X 02 Where 0'2 is the group of order 2).
Deduce that if D = 1321’Dn then C(D) 75 1 While D/§(D) is isomorphic
71/
toD.

“‘18. The infinite dihedral group D is generated by elements a and 6,

each of order 2, with ab of infinite order (see Problem 5 of Chapter 3).
Prove that N” = gp{(ab)2"} is a normal subgroup of D for n = l, 2,... .
Let D” denote the factor group D/Nn. Show that D” is a group of order
2”+1 and is in fact the finite dihedral group of that order (see Example
1.25 and the preceding problem).
Deduce that 130” D” has a subgroup isomorphic to D.
n
7

Sylow theorems

THREE remarkable classical theorems of Sylow deal with the

existence of certain subgroups in a finite group and with their
properties in relation to the group. We shall focus attention on
questions of existence, to start with. Lagrange’s theorem 3.15
states that if G is a finite group, then |Gl is divisible by the order
of every subgroup. But this tells us nothing about the existence in
G of subgroups having for their orders prescribed factors of [G[.
In fact subgroups with such arbitrary orders may not exist, an
assertion that was made in Chapter 3 and that will now be proved.
Example 7.01. The alternating group A4 of order 12 has no
subgroup of order 6. For suppose that there were a subgroup S
with six elements. Since A4 contains three elements of order 2
and eight of order 3, there must be at least two elements of
order 3 in S. If 6’ contains three elements of order 2 and just
two of order 3, we see that 8 contains the normal subgroup of
A1 that has order 4 and so 8 = A4, a contradiction. By appro-
priately naming the set on which A4 acts, we may assume that
8 contains (123) and (124). But now a little calculation produces
seven distinct elements in S:
1, (123), (132), (124), (142), (123)(124) = (14)(23),
(124)(123) = (13)(24).
This contradiction shows that S does not exist.
Sylow’s first theorem gives a condition for existence of
subgroups. First we prove:
THEOREM 7.02. Let G be a group of order p“r where p is prime,
0; > 0, and r is not divisible by p. Then G contains a subgroup
of order p3 for each [9 satisfying 0 g )3 g a.
128 Sylo'w theorems

Proof. We use induction on the order of G. When G has no

proper subgroup, it is easy to see that G is either trivial or has
prime order; therefore its order is p, by hypothesis, and there
is no more to prove. That starts the inductive process.
In general, we assume that every group with smaller order
than G has p—subgroups of the kind required. In particular,
this will be true of all proper subgroups of G. Suppose there is
a proper subgroup S of index prime to 10. It follows that p“
is the highest power ofp to divide [S 1. By the inductive hypothe-
sis, S has subgroups of order p5 for each ,8 in 0 g ,3 < on; since
these are also subgroups of G, this case is finished.
We are now left with the case in which the index of every
proper subgroup of G is divisible by 10. If x is any element of G
then its centralizer C(x) is either a proper subgroup or is G. By
Corollary 4.14, therefore, either at has pk conjugates (for some
16 > O) or a: is central in G. We enumerate the elements of G
by conjugacy classes, much as in Theorem 4.28. If m is the order
of {(G) then G has m+ph’ elements, for some 10’ 2 O; on the
other hand G had order par, so
par = m+pk’.
Since a > O, 10 must divide m.
Thus {(G) is an abelian subgroup of G, with order m divisible
by p; and it must contain an element of order p. This is clear
from Sylow’s first theorem for abelian groups—we refer to
Corollary 5.03. Therefore I;( G) contains a subgroup N of order p.
This is normal in G as it is central in G.
We next consider G/N and apply the inductive hypothesis,
as we may because G/N has order pa—lr. We conclude that if
0 < B < at then G/N has a subgroup, M[N say, of order 125—1
for each such B. Note the use of Theorem 6.04 here. It is now
clear that M has order 105, and the proof is finished by the remark
that the case ,B = 0 is trivial. I]

COROLLARY 7.03 (Sylow’s first theorem). If the group G

has order par, where the prime 1) does not divide r, then G has at
least one subgroup of order p“. I]
 Sglow theorems 129

COROLLARY 7.04. If the prime p divides the order of the finite

group G', then G contains at least one element of order 1).

Proof. Take ,3 = 1 in the theorem; the generator ofa subgroup

Of order 12 itself has order p. [I

COROLLARY 7.05. The set ofprimes dividing the order of afinite

group is the set of primes occurring as orders of elements in it.
Proof. Use Corollary 7 .04 and Lagrange’s theorem. I]
COROLLARY 7.06. A finite group is a p—group if and only if it
has p—power order. I]
The subgroups of order p“ which appeared in Corollary 7.03
play a fundamental role in the theory of finite groups. It is on
their account that we introduce new terminology. Let 71 be
a set ofprimes (which we shall assume contains at least one prime
in order to avoid hair-splitting). We say that a positive integer
is a rr-number if all its prime factors lie in 71.

DEFINITION. A w-group is a group in which the order of every

element is a qr-number.
Thus Corollary 7.05 makes it plain that a finite group is a
w-group if and only if its order is a qr—number. We use the term
‘qr-subgroup’ in the obvious sense.
DEFINITION. A Sylow w-subgroup of a group G is a n-subgroup
which is not properly contained in any n-subgroup of G.
Thus Corollary 7.03 asserts that in the finite group 0 there is
a SlW p—subgroup of order 19“; note that there can be no sub-
group Of strictly larger p-power order, by Lagrange’s theorem.
A finite group G’— has~§ylow air-subgroups for every choice of 71.
LdAQJM/ly811.0%}tEfiiflififigfiififi
Isubgrgupf. To start with G has rr—subgroups, for 1 is a rr-sub-
group. And if S is a ar-subgroup which is not a Sylow qr-subgroup
then S is properly contained in another n-subgroup. After a
finite number Of steps this process must lead to a Sylow w-sub-
group, because G is finite.
853137 K
130 Sylow theorems

This argument does not prove Sylow’s first theorem when

71' = {p}, of course, for it does not prove that the group of order
par has a subgroup of order p“.
A natural question to ask at this point is whether Sylow’s
first theorem can be generalized in the following way. Let G be
a finite group, and let m be the largest qr—number dividing |G | , for
some set of primes 1r. When 71' = {p} we now know that G has
a subgroup of order m, and one might ask if something similar can
be proved for general 7r. That this is not so without restrictions
is shown by the next example, and what the restrictions are will
be discussed in Chapter 10. Notice, by the way, that Example
7.01 is irrelevant to the present discussion, because 6 is not the
largest w-number dividing 12 when 77 = {2, 3}.
Example 7.07. Consider the n-subgroups of A5, where
77 = {2,5}. We first show that A5 has no subgroup of order
20 (which is the maximal w-number dividing [1151). For if there
were such a subgroup of index 3, then it would have at most three
conjugates; the intersection of these would be a normal sub-
group of index at most 33 = 27, as the method of Theorem 4.16
shows. But this would be a proper normal subgroup, whereas
A5 is simple according to Example 6.22. It follows that A5 has
no subgroup of order 20. '
Nevertheless, A5 must have Sylow w-subgroups. What are
they? Any Sylow 2-subgroup, such as gp{(12)(34), (13)(24)}, is
one; if it were contained in a larger w-subgroup then the latter
would have order 20 by Lagrange’s theorem. Another Sylow
w-subgroup is gp{(12345), (25)(34)}; for it will be found that this
has order 10 and is therefore again a maximal w-subgroup. Here
is a case, then, in which Sylow n—subgroups have different
orders.
We can proceed in several ways. This state of affairs arises
because we Wish to emphasize both the elegance of the classical
Sylow theorems and the difficulties that arise when we try to
generalize them. What we shall do is to state Sylow’s second
and third theorems at once, giving some account of their
 Sylow theorems 131

applications as well, and to prove them in the course of a careful

analysis of n—subgroups of general (perhaps infinite) groups. The
reader who wishes for nothing but proof of Sylow’s theorems
will be able to select what he wants from the results to follow.
Alternative proofs to all three classical theorems will be found in
an Appendix to this chapter.
SYLow’s SECOND THEOREMT In a finite group the Sylow
p-subgroups (for a fixed prime p) are all conjugate and therefore
isomorphic.
SYLow’s THIRD THEOREM. In a finite group the number of
Sylow p-subgroups (for afixed prime p) is congruent to 1 modulo 10.
Example 7.08. Consider the Sylow 2-subgroups of A5. By
Sylow’s first theorem they each have order 4. One of them is
G2 = gp{(12)(34), (13)(24), (14)(23)}. Now A5 has a subgroup
isomorphic in a natural way to the alternating group on {1, 2, 3, 4}
and CT"2 is normal in this (compare Example 6.13). Therefore
[A5:N(Gz)| is l or 5. In fact this index is 5, because A5 is simple
(Example 6.22). Take a right transversal of N(G2) in A5; one
might choose {aizo g l < 5} where a = (12345), for instance.
Then the conjugacy class of G2 is {Gg‘z 0 < 'l < 5}, by Theorem
4.12. Sylow’s second theorem allows us to assert that this
conjugacy class is the set of all Sylow 2-subgroups of A5. Note
that there are 5 of them, and 5 =_: l modulo 2 in accordance with
the third theorem.
Example 7.09. Let G be any group of order 40; we prove that
G must contain a normal subgroup of order 5.
The first Sylow theorem implies that G has a subgroup of order
5. By the third theorem there are in fact 1 +5.70, for some integer
It. By the second theorem 1+5k divides 40, because the Sylow
5-subgroups are all conjugate and so the normalizer of any has
index 1+5k (recall Corollary 4.14). But 1+5h divides 40 if and
only if h = 1, so there is just one Sylow 5—subgroup. Since it
coincides with all its conjugates, it is a normal subgroup of G.
1' A slight liberty has been taken in the numbering of Sylow’s three theorems.
132 Sylow theorems

Example 7.10. Next we prove that an arbitrary group G' of

order 1 2 has a normal subgroup of order 3 or 4. Sylow’s theorems,
used as in the previous example, show that the number of sub-
groups of order 3 is either one or four. Suppose that there are
four. Each contains two elements of order 3, by Lagrange’s
theorem. We thus find eight elements of order 3, which are
easily seen to be distinct. There are four other elements of G.
By Sylow’s first theorem there is at least one subgroup of order
4. In the present circumstances there is precisely one, and by
the second theorem it is normal. Therefore there is a normal
subgroup, if not of order 3 then of order 4.
We now embark on an attempt to chart some of the territory
of Sylow properties, exploring such parts of it as we can reach.
Some of our theorems will be mere formal implications to start
with, demanding conditions but not specifying when these are
satisfied: this is simply part of our analysis. One difficulty which
presents itself at once is this: how do we know that Sylow w-sub-
groups exist in infinite groups? This is a tricky set-theoretical
point. But we shall make no assumption concerning it and
therefore need not even explain what the difficulties are—the
hypotheses of our theorems will always guarantee the existence
of Sylow w-subgroups, except for finite groups, since we know
that they exist in that case.
THEOREM 7.11. If 0,, is a Sylow w-subgroup of the group G,
and if a is an automorphism of G', then G‘" or is a Sylow ar-subgroup
of G.
Proof. Suppose by way of contradiction that Gfla is not a
Sylow w—subgroup of G. It is clear that G'Woz is a w-subgroup.
Therefore 07a is properly contained in some w—subgroup, G;
say. Then (G;)a"1 is a fl-SUn'OHP because oc—l is an auto-
morphism of G ; and (0;)04—1 contains 07,. But this inclusion is
proper, because if as e G; and a: ¢ Gfla then sea—1 6 (09ml and
wor1¢ G”. The fact that 0,, is properly contained in a w-sub-
group is a contradiction as 9,, was supposed to be a Sylow 71—
subgroup. I]
 Sylow theorems 133

COROLLARY 7.12. Any conjugate of a Sylow n-subgroup of G is

also a Sylow ar-subgroup of G.
Proof. Use Theorem 4.31. B
One of the general problems of Sylow theory is to find reason-
able conditions under which the obvious converse of Corollary
7.12 (the Sylow w-subgroups all lie in one conjugacy class of
subgroups) holds. A similar question emerges from the form of
the next theorem, which is usually known as Frattini’s lemma.

THEOREM 7. 13. Let the group G have a normal subgroup K with

the following properties:
(i) a Sylow w-subgroup of K exists;
(ii) if K”, K; are two Sylow w-subgroups of K such that
K; = Kfloa for some automorphism a of K, then K1, and
K; are conjugate in K.
Then G = KN(KW) where K” is any Sylow ar-snbgroup of K.
Proof. As usual N(Kw) denotes the normalizer of K“ in G.
Take any element x in G. Since K <1 G and K, < K, we have
Ki”, < K. Now conjugation by it induces an automorphism
of K, by Theorem 4.31, and K3”, corresponds to K, in this auto-
morphism. By Theorem 7 .11, therefore, K3; is a Sylow rr-
subgroup of K. By hypothesis, there exists la in K such
that Kfr=K’,$. But this implies that [01:4 EN(K7,), and so
519—1 e KN(Kw). This proves Theorem 7.13. [I

COROLLARY 7.14. Let the group G have a Sylow «subgroup

G” and a subgroup K containing N( G7,) and having property (ii)
of the above theorem. Then K = N(K).
Proof. By N(K) we mean the normalizer of K in G.
The normalizers of G,r in G and N(K) are N(G7,) and
N(K) n N(G), respectively. Since K is normal in N(K), by
definition of normalizer, we may apply Theorem 7.13 to the
group N(K). The conclusion is that N(K) = K(N(K) nN(G,, ).
But N(G7,) < K by hypothesis. It follows that N(K) = K. [J
The next theorem does not mention Sylow w—subgroups, but
134 Sylow theorems

it has no less than six corollaries which are of importance in our

work.
THEOREM 7.15. If A and B are rr-subgroups of the group G,
with A normal in G, then AB is a n-subgroup of G.
Proof. Corollary 6.10 ensures that AB is a subgroup. By
the second isomorphism theorem, AB/A g B/(A n B). Now
B/(A n B) is a rr-group because B is a w—group. Therefore, if
x G AB then x“ e A for some positive w-number n (depending on
95). But the order of x" is a ar-number since A is a w-group. It
follows that the order of x is a w-number, and so AB is a
w-group. El
COROLLARY 7.16. A normal w-subgroup of the group 0 lies in
every Sylow w-subgroup of G.
Proof. Let A be a normal qr-subgroup and B be any Sylow
w-subgroup. Then AB is a w-subgroup by the theorem. But
clearly AB contains the Sylow qr-subgroup B. It follows that
AB = B, and so A < B, as required. 1]
COROLLARY 7.17 . A group has precisely one Sylow w-subgroup
if and only if it has a normal Sylow n-subgroup.
Proof. Suppose G has precisely one Sylow ir-subgroup. By
Corollary 7 .12 this must coincide with all its conjugates; in other
words, it is normal.
Suppose G has a normal Sylow w-subgroup. By Corollary 7.16
this lies in, and therefore coincides with, every Sylow qr-subgroup .
Thus 0 has a unique Sylow w—subgroup. El

COROLLARY 7.18. A group has precisely one Sylow n-subgroup

if and only if it has a characteristic Sylow w-subgroup.
Proof. A characteristic subgroup is normal, and this remark
with the previous corollary disposes of one half of the proof.
If the group G’ has precisely one Sylow w—subgroup, 0,, say,
and if at is any automorphism of G, then Gfloc is also a Sylow
qr-subgroup of G according to Theorem 7.11, and Gnu = 07,.
That is to say, 0,, is a characteristic subgroup of G. [I
 Sylow theorems 135

COROLLARY 7.19. If G, is a Sylow w-subgroup of G then G,r is

the unique Sylow w-subgroup of N(G).
Proof. Use Corollary 7.17 in the group N(G,,). I]

COROLLARY 7.20. If G, is u Sylow w—subgroup of G then no

distinct conjugate of G7 lies wholly in N(G,).
Proof. This follows from Corollaries 7.12 and 7.19. El

COROLLARY 7.21. The normalizer of a Sylow w-subgroup of G

is its own normalizer.
Proof. Let a: e N(N(G7,)) where G, is a Sylow w-subgroup of
G. This means that N(GflY‘ < N(G7,). But since G7r g N(G,)
we also have G: g N(G7,). By Corollary 7.12 G3“, is a Sylow
w-subgroup of G, and we deduce from Corollary 7.19 that
G3”, = G". That is to say, as e N(Gfl). We have shown that
N(N(G7,)) S N(G7,), establishing the desired result. I]
Note the relation between Corollaries 7.14 and 7.21.
We now move towards incisive results. What has been said
above should make it plain that very special assumptions will
be needed. Accordingly our results deal with the case in which
77 = {p}, and we shall resume the discussion of general 7r-sub-
groups of finite groups in Chapter 10. One further lemma only
is general in nature.

LEMMA 7.22. The index of any subgroup in a w—group is

either a a-r—number or infinite.
Proof. Let S be a subgroup of finite index in the ”group G.
Theorem 4.16 guarantees the existence of a subgroup N of finite
index in G such that N g S and N <1 G. It is clear that G/N is
a finite w—group, and so by Corollary 7.05 the order of G/N is
a qr-number. Therefore Lagrange’s theorem gives the fact that
IG/N : SIN | is a w-number.
What is next required is a proof that there is a one—one
correspondence between the right cosets of S/N in G/N and the
right cosets of S in G. Now the mapping that sends the coset
136 Sylow theorems

Sc: onto the coset (S/N)Nx is clearly onto, and we claim that it
is one—one. Suppose then that (S/N)Nx = (S/N)Ny. It follows
that (Nac)(Ng/)*1 e SIN, and so avg/*1 e S; therefore Sn: = 83/.
We have proved that the mapping is one—one, establishing the
desired correspondence.
Thus the index of S in G is a n-number. I]

THEOREM 7.23. Let G be a group with a Sylow p-subgrougo

Gp which has only a finite number of conjugates in G. Then
(i) G has 1+kp Sylow p—snbgroups, for some integer 16;
(ii) every Bylaw p-snbgroup of G is conjugate to G1,.

Proof. We define certain sets of conjugates of Gp in the follow-

ing manner. Let Q be any p-subgroup of G, and put
W(Q) = {X :X is conjugate to G1,, and Q g N(X)}.
Thus every 93(Q) is a finite set; it may be empty, and if G1, = 1
it will be empty. By Corollary 7.12 the elements of fl(Q) are
Sylow p-subgroups of G.
Now take X 6.?(Q). We wish to prove that if q 6 Q then
X4 e .¢(Q). For ifX9 ¢ W(Q) then Q normalizes X4 (by definition
of fl(Q)). It is an easy step to deduce that Q normalizes X itself,
but this is impossible because X e 95(Q) and so Q cannot nor-
malize X. Therefore X4 e 9(Q).
Next we consider the number of elements in {X92q e Q}. It
is finite because G1, has only finitely many conjugates in G.
But this number equals |Q: Q n N(X)| by Corollary 4.14, and
Q n N(X) has p—power index in Q by Lemma 7.22. As Q does
not normalize X (by definition of 9(Q)) this index cannot be
1. Thus {qq E Q} has 10“ elements for some n > 0.
As we take different elements X in 9(Q) the resulting sets
{qq e Q} clearly form a partition of 9(Q). The conclusion of
the last paragraph shows that 93(Q) has k1) elements in all, for
some 16 2 0. We shall apply this fact with two special choices
of Q.
Take Q to be G1,. The only conjugate of G1, in N(G1,) is G],
itself, by Corollary 7.19. It follows that 9(Gp) contains all the
 Sylow theorems 137

conjugates of Gp except G'p itself. Since 9(911) has Icp elements,

for some 16 > 0, G1, has 1+kp conjugates.
Now let Q be an arbitrary Sylow p-subgroup, G; say, of G.
We know (from above) that W( G2) has k’p members for some 16’,
and these are all conjugates of 01,. But they cannot be all the
conjugates of G1», for these number 1+kp. (We can never have
h’p = 1 +Icp for any integers k, 10’.) Therefore G'p has a conjugate
X such that G; g N(X). By Corollaries 7.12 and 7.19, X = G},
and it follows that G; is conjugate to 010.
Parts (ii) and (i) of the theorem follow in succession. El

COROLLARY 7.24 (Sylow’s second theorem). In a finite group

the Sylow p-subgroups (for a fixed prime p) are all conjugate and
therefore isomorphic.
Proof. Sylow p-subgroups exist. Clearly they are finite in
number. Therefore Theorem 7.23 (ii) implies that they are all
conjugate. El

COROLLARY 7.25. Let G be a finite group and let G1, be any

Sylow p-subgroup of G. If P is any p-subgroup of G then P lies
in a suitable conjugate of 01,.
Proof. It is clear that P lies in some Sylow p-subgroup G; of
G. By the previous corollary G]; = 0193 for a suitable element x
of 0; so P g 0;, as desired. El

COROLLARY 7.26. Every p-subgroup of a finite group G lies

in some subgroup of order p“, if |G’| = par and p is prime to r. [I

COROLLARY 7.27 (Sylow’s third theorem). In a finiteXgroup

the number of Sylowp-subgroups (for afixedprimep) is congruent to
1 modulo p.
Proof. This follows from Theorem 7.23 (i). [I
We observe that Corollaries 7.16—7.21 may now be restated
for finite groups and for 71' = {p}, with the aid of the Sylow
theorems. We leave this task to the reader, but we shall write
down the appropriate forms of Theorem 7.13 and Corollary 7.14.
138 Sylow theorems

COROLLARY 7.28. If the group G has a finite normal subgroup

K and if K1, is a Sylow p-subgroup of K, then G = KN(KP). [I

COROLLARY 7.29. If the group G has a finite subgroup K

containing the normalizer of a Sglow p-subgroup of G, then
K = N(K). I]
The next result is important for Chapter 9.

COROLLARY 7.30. If G is a finite group in which every maximal

subgroup is normal, then any Sglow p-subgroup (p being an
arbitrary prime) of G is normal in G.
Proof. A maximal subgroup of a group G is defined to be a
subgroup S such that S < G, and S g T g G implies S = T or
T = G. It is clear that a finite group always has at least one
maximal subgroup.
Suppose that the group G, which satisfies the hypothesis of
the corollary, has a Sylow p—subgroup GI, for which N( G1,) < G.
There is then a maximal subgroup M of G, containing N(G1,).
Corollary 7.29 indicates that M = N(M), contradicting the fact
that M is normal in G. Therefore all Sylow p-subgroups of G are
normal. [I
Theorem 7 .23 imposes the strong restriction (in the context of
infinite groups) that the Sylow p-subgroups are finite in number.
It is possible to construct groups with an infinite number of
Sylow p-subgroups, non-isomorphic ones among them. Instead
of doing this we content ourselves with what is, in the circum-
stances, a rather mild example.
Example 7.31. Let G = [P Gn where G, is isomorphic to
n>0
the symmetric group 6'3 for each n. Let an, bn be the elements of
G" which correspond to ( 12), (13) respectively in 83. Put
A = n>0
HDgP{an}. B = n>0
H” gp{bn}-
An easy verification shows that if g gt A then gp{A, 9} contains
elements of order 3; this means that A is a Sylow 2-subgroup of
G. So is B, for similar reasons. Now A and B are certainly
 Sylow theorems 139

isomorphic, but they are not conjugate in G. For if A” = B for

some a; in G (that is, x is a function from {n} to {G'n}), then
me gé 1 for every 11. since an 75 bn; it follows that a; could not have
finite support, and therefore does not exist.
The final topic of this chapter is of a slightly difierent type
from what has gone before. We shall prove an existence theorem
in the case of a finite group having a normal subgroup with co-
prime order and index; the assumption of normality is made to
appear reasonable by the situation in A5, which is, of course, a
simple group. The proof falls into two cases, the first of which
is essentially due to Schur.

THEOREM 7.32. Let G be a group with an abeltan normal sub-

group K such that G/K has order n and mm = 1 for each element x
of K, m and n being coprime positive integers. Then G' has a sub-
groap O'such that G = K0 andKn 0 = 1.

Proof. We shall denote elements of G and of its abelian sub-

group K by small Roman letters, as usual, and elements of the
finite group G/K by small Greek letters. In each element a of
G/K we choose an element 9“; thus 9“ is a fixed element (depend-
ing on at) of G.
We see that gagfl and 90,5 lie in the same coset of K in G.
Therefore
(1) 91193 = gafihafi’
where hay; is some element of K. We now derive a relation
between such elements by using the associative law in G. We
obtain from (1):

ga(gflgy) = gagflyhfly = gafly kmflyhfly,

(9a 95):], = gafl has 97 = gap gyms = gapyha5,yht,p-

By hzdg we mean g;1ha,fi 97. It follows that

(2) hmflrhfia = hafiahzfi'

We now take the product on each side of (2) as or varies over
all the elements of the finite group G/K. Notice that the order
140 Sylow theorems

of the factors is irrelevant because K is abelian. If we define

63 by

(3) cB = 0513K h «,3

then we obtain from (2)

(4) 05}, kg” = 0,, 0%
where a); means 9,7105 97' It should be observed that as a runs
through the elements of G/K so does 043, provided 3 is fixed.
Since m and n are coprime, we can find an integer n’ such that
nn’ E 1 modulo m.
If we put do) = cg” then (4) gives

(5) dgdy = dfly 7‘50}

for each pair )3, y in G/K. Note that the abelian property of K
is used again, and that the order of h , divides m.
Next we define a set 0 of elements of G:
0 = {gadaza e G/K}.
We have
(gadaxgp d5) = gagpdgdp
= gafihafidgdfi
= gafl dafl

by (l) and (5). It follows that 0 is isomorphic to G/K and also

that 0' is a subgroup of G. The required facts that G = K0
and K n 0 = 1 are now clear. I]

COROLLARY 7.33. If K in Theorem 7.32 is central in G then

G g K X 0.
Proof. Since K and C’ commute element by element, the direct
product criterion of Theorem 4.38 gives this result with very
little difficulty. [I
We pass on to the case in which K need not be abelian. The
proof here is due to Zassenhaus, whence the name ‘Schur—
Zassenhaus theorem’ for the final result.
 Sylow theorems 141

THEOREM 7.34. Let G be at group with a normal subgroup K

suck that G/K and K have coprt'me orders rt and m respectively.
Titert G has a subgroup of order rt.
Proof. We proceed by induction on [G]. We take m > 1
since there is nothing to prove if m = 1. Theorem 7.32 then
starts ofi the induction.
It may be advantageous to illustrate the G
general step by a diagram showing relations
between subgroups of G. We shall not explain K N
this in detail. It is, perhaps, enough to remark
that the lines near K and N indicate that KnN H
G = gp{K,N} and that K n N is the largest
subgroup common to K and N. GP
Let p be any prime divisor of m. Since m
and rt are coprime, K contains a Sylow p-
subgroup G10 of G according to Sylow’s first
theorem. Let N denote the normalizer of G1, in
I
G. Since K <1 G, we see that K contains all the
conjugates of Gp in G according to Sylow’s second theorem.
Now N n K is the normalizer of G2, in K. It follows from
Corollary 4.14 that
lG:N| = [K:NnK|.
It is an elementary fact about indices that
|G:N|]N:NnK| = [GzKHKaKL
These two displayed equations give at once that
lNaKl = |G:K|.
Because K <1 G, Corollary 6.02 shows that N nK <1 N. We
know that G1, <1 N. It follows from the third isomorphism
theorem that
N/(NnK) g (N/GA/«Nnm/Gpt
Now IN/Gpl < [G], while [NaKl = rt as above. The induc-
tion hypothesis may be applied to N/GP, and we conclude that
G has a subgroup H such that H/G1, is of order rt.
142 Sylow theorems

If Z is the centre of G1, then Z > 1, by Theorem 4.28. Now Z

is a characteristic subgroup of G!» by Theorem 4.33, and G1, <1 N;
so Z is normal in N by Theorem 4.34. We have Z <1 H because
Z < H < N. The third isomorphism theorem gives
H/G g (H/Z)/(Gp/Z)'
Since H/G1, has order 12 while Gp/Z has order a proper factor of
[Gig], the induction hypothesis applies again. Thus H/Z has
a subgroup, L/Z say, of order n.
Schur’s Theorem 7.32 shows that L has a subgroup 0’ of order
n. This is a subgroup of G with the desired property. [I

Problems
1. Prove that there is no simple group of order 700.
2. How many subgroups of orders 2, 3, 4, 5, 6, 8, respectively, does the
symmetric group 8, contain?
3. Find all the Sylow 2-subgroups of the infinite dihedral group (described
in Problem 5 of Chapter 3). Show that they fall into two conjugacy
classes.
4. The group G has order pg, where p and q are distinct primes. Prove
that G has a proper normal subgroup. Prove further that if neither prime
is congruent to 1 modulo the other then G is abelian.

5. Prove that no pair of distinct Sylow w-subgroups of the subgroup H

of the group G can lie in the same Sylow n-subgroup of G.
6. The finite group G has an automorphism a of prime order p. Prove
that a must leave fixed at least one Sylow yaw-subgroup of G.
7. Describe a homomorphism from any group G into the symmetric
group on the right cosets of its subgroup H. Prove that if H has finite
index in G then the kernel of the homomorphism has finite index in G.
Deduce the following:
(i) If G is a p-group and H is maximal in G then H is normal in G.
(ii) If H has index p in G, and G is otherwise arbitrary, then there is
a right transversal of H in G with the form {gi : 0 < 7? < 10}.

*8. Prove that the right regular representation of a finite group G con-
tains an odd permutation if and only if a Sylow 2-subgroup of G is cyclic
and non-trivial.
 Sylow theorems 143

Deduoe that a group G with the condition mentioned cannot be non-

abelian simple.
Prove further that if n is the largest odd factor of IG] then G has a
subgroup of order n.

9. The group G has order 1029] where p, q are distinct primes, and P1, P.“
are two distinct Sylow p-subgroups in it. Prove that PI n 1'"2 is a normal
subgroup (possibly trivial) of G.
Deduce that G cannot be a simple group.
10. The group J! is generated by
0 1 —— 1 l)
A _ (1 — 1)’ B _ ( 1 0 ’
the entries of the matrices being residue classes modulo 3. Calculate A3
and B‘zABa. Prove that gp{A, B2} has order 16 and that gp{AB} has
order 3. Hence, or otherwise, find the order of .1.
11. The group G has a finite number of (non-trivial) Sylow p-subgroups
for some prime 1.0. Let g be the set of them. Show that there is a homo-
morphism from G into the symmetric group on %. Hence prove the
following statements:
(i) G has a normal p-subgroup P which has finite index in every
member of "6.
(ii) If G has precisely 1 +10 Sylow p-subgroups then P has index 22 in
every member of ‘6.
(iii) A finite group With precisely 1+{p Sylow p-subgroups is non-
simple or has a Sylow [lo-subgroup of order p.
12. Use the result of the preceding problem to prove that there is no
simple group of order less than 60.
13. Let G be a finite group with a subgroup S of index n, where n > 1,
and define a homomorphism of G into the symmetric group on the right
cosets of S in G. Hence prove the following statements.
(i) If G is simple then the order of G divides In!
(ii) Ifn = 2 or 3 or 4 then G cannot be (non-abelian) simple.
(iii) There is a finite (non-abelian) simple group with a subgroup of
index 5.
(iv) If p is the smallest prime factor of the order of G then any sub-
group of index 11 in G is normal in G.
14. The group G has a normal subgroup N, and H is minimal in the set of
those subgroups of G for which G = NH. Prove that every Sylow p-sub-
group ofN n H is normal in H.
**15. The group G has order 60. Find the possible numbers of its Sylow
p-subgroups (for p = 2, 3, 5) permitted by Sylow’s third theorem.
144 Sylow theorems

Prove that if G is simple, then there cannot be 3 Sylow 2-subgroups or

4 Sylow 3-subgroups; and that if G is not isomorphic to A 5 then there
cannot be 5 Sylow 2-subgroups.
Deduce that a simple group (other than A5) of order 60 would have
precisely 15 elements of order 2 and 15 Sylow 2-subgroups. Derive a
contradiction by considering the centralizer of an element lying in more
than one Sylow 2-subgroup, and conclude that As is the only simple
group of order 60.

*16. Prove the true statements among the following and disprove the
false.
(i) If P is a Sylow p-subgroup of the finite group G then P n H is a
Sylow p-subgroup of the subgroup H of G.
(ii) If P is a Sylow p-subgroup of the finite group G then P/(Pn N) is
isomorphic to a Sylow p-subgroup of the factor group G/N of G.
(iii) There is a finite abelian group which has non-isomorphic Sylow
w-subgroups, for a set of primes 77.
(iv) Any Sylow p-subgroup of G1 >< G2 is the direct product of Sylow
p-subgroups of G1 and G2.
(v) Ifthe group G has order 10% (where 10 is prime, a 9 l, and r is not
divisible by p) then all subgroups of order 10““ are isomorphic.
*17. Let the finite group G of order n have a subgroup H of index 7'.
Show that there are at most n—r+ 1 elements in the union of all the
conjugates of H. Deduce that if each element of G lies in some conjugate
of H then H = G.
Suppose that the prime p divides |G[ but does not divide IG: C(g)[ for
any 9 e G. Prove that a Sylow p-subgroup of G is central in G.
**18. The finite group G has an abelian normal subgroup K with index n
and order m, these being coprime integers. Let G be a subgroup of G
with order In. Prove that if G = KD for some subgroup D of G then a
suitable conjugate of 0 lies in D. (Hint: first consider the case in which
K n D = 1.)
Appendix to Chapter 7

IN recent years interesting new proofs of Sylow’s three classical theorems

have been found. They depend slightly on the theory of permutation
groups and are otherwise elementary, but they seem no easier to the
novice than the classical proofs. We give the details in this appendix,
but they will not be used in the rest of the book.
Our total formal requirement of the theory of permutation groups
consists of two definitions and a consequential theorem. Let a set A be
given, and a group G of permutations of A, and a subset I‘ of A.
DEFINITION. The stabilizer of P in the situation just described is the
subset {gzI‘g = 1"} of G; that is, the set of all elements of G which leave
I‘ fixed as a set (not necessarily element by element),
Now the stabilizer S of I‘ is a subgroup, for the following reasons.
Clearly S contains 1. And if x, y e S, so that
Ta = 1‘, Pg = I‘,
then I‘xy"1 = P, and reg—1 e S’. The usual subgroup criterion (Theorem
3.06) proves the assertion about 8.
Next suppose that I‘ is a one-element set, {y} say, so that S is {g :yg = 9/}.
Let x, y be any two elements of G for which ya: = yy. Then 3931—1 e S. By
the very definition of coset, this means that x and y lie in the same right
coset of S in G. Conversely, suppose that x and y are so situated in G; then
x = sy for some .5- e S, and ya = ysy = yy. Therefore the possible
images in A of 7, under the action of G, are in one—one correspondence
with the right cosets of S in G; and so the number of these images, being
the index of S in G, divides the order of G when the latter is finite, by the
theorem of Lagrange.
DEFINITION. The transitim‘ty class (or orbit) of y in A under G, in the
above context, is the set of images {yg:g e G} of y.
We have now proved our elementary theorem about permutation
groups.
THEOREM. If G is a finite permutation group on a set A then the number
of elements in each transitivity class under G divides the order of G. I]
We are ready to prove facts about Sylow structure.
THEOREM. Let G be a group of order par where a > 0, p is prime, and
r is not divisible by 10. Then G contains at least one sabgromo of each order
p3 where 0 g B < at.
853137 L
146 Sglow theorems

Proof. We consider all complexes of G with 105 elements, with the

intention of proving one of them to be a subgroup. Let g be the set of
all such complexes. The number of elements of if is of course (123:),

the number of combinations of 12% objects taken 103 at a time.

We define a group of permutations on (K as follows. For each 9 e G,
let 945 be the permutation of g given by 0 . q = 0g for all 0 in 9; here
09 is, of course, the complex product. To check that 5745 is a permutation
is easy; 995 is onto g because (Cg—H.945 = C', and ggb is one—one because
Cg = C’g implies O' = 0’. We have
(91¢)(9fi5) = (91 92W)
for all 91, g, in G since
(0-91¢)(92¢) = (091)(gz</>) = (09092 = 0(9192) = 0-(9192)¢-
Closure of {qzg e G} is immediate, and the associative law is easy; 14$
acts as an identity element, and g‘lgb is an inverse of 946. Hence {q : g e G}
is a group, H say. Moreover, we have shown that {6 is a homomorphism
from G onto H.
Take an arbitrary element x of G and an arbitrary transitivity class
9'of (0”. We assert that w lies in a suitable element of 9'. Take 0' e 9',
and take a e 0'; then 1 = c(;—1 6 00—1, and x e Oc—lx, and of course
06'122 e 9', as required.
This transitivity class 9' need not, of course, form a partition of G
since one element of G may lie in several elements of 9'. But we can say
that 9' contains at least pa‘flr elements of ‘6, because each 0' in 9 has only
103 elements while G itself has 27“? elements. Further, if 9' contains
precisely pa‘Br elements of 9 then the elements of 9' are all disjoint (as
complexes of G) and we do have a partition of G.
In fact, we can prove more under the assumption that 9' contains just
gamer elements. Some element of 9', say the complex 0, contains 1 (in
accordance with a fact proved above). Since 9' is a partition of G only
one complex (namely 0) of 9' contains 1. Suppose x e 0'. Then 1 6 093—1,
which also lies in the transitivity class 9'. It follows that 0 = 0313—1. If
y is another arbitrary element of C, we deduce from this that 3122—1 e 0'.
But now our subgroup criterion applies: 0 is in fact a subgroup. And
clearly 9' consists of the subgroup 0 and. its right cosets in G. Conversely,
the set of right cosets of a subgroup K in G forms a transitivity class 9'
of g with 10““31' elements, if K has order p5.
The topic of immediate interest, then, is the number t of transitivity
classes 9' of 9 containing precisely pm‘fir elements of V; the assertion of
the theorem follows from the fact t > 0, which we now try to prove.
Consider transitivity classes ‘2! of g containing more than p“—fir elements
of ‘5. How many elements does such a class 0% contain? This number
must divide the order ofH, by the previous theorem, and so the order of G,
 Sylow theorems 147
because there is a homomorphism from 0 onto H; and, being a factor of
par which exceeds god—5r, it must be divisible by pa‘fi'H. This fact suffices
for our purpose—it shows that the total number of elements of % each
lying in a transitivity class ‘7! is Irma—5+1 for some integer k.
We now enumerate the elements of g according to whether they lie
in a class like 9‘ or a class like 9%. (Of course, no transitivity class con-
tains less than pa—Br elements of g.) We obtain

(a?) =
that is, (1;:) E tpa—Br (modulo Ila—5+1).

The essential point in the present proof is to observe that this con-
gruence is largely independent of the structure of G. In fact, all we have
used is that G has order par and has 1: transitivity classes .9" with just
god—Br elements of %. Therefore the congruence will hold for any group G
of order your, provided that we regard t as a function of G.
We may, therefore, choose any group Go of order your for G; we take Go
to be the cyclic group of that order. This group exists for every order par
and has one and only one subgroup of order 103, and so one and only one
transitivity class .9" with you—Br elements; t = 1. The assertion just made
for cyclic groups follows from the theory of abelian groups developed in
Chapter 5 (for instance). Substitution of G = Go and t = 1 therefore
gives (per
1,8
) E pad-qr (modulo p“‘5+1).

It follows that, for general G and corresponding t,

tpu—Br E pm—Br (modulo paw“).
Since 10 does not divide r, which may therefore be cancelled, this con-
gruence shows that t ¢ 0. Thus t > O, and the theorem is proved.l]

COROLLARY (Sylow’s first theorem). If the group G has order par,

where the prime 39 does not divide r, then G has at least one subgroup of order
P”-

THEOREM. Let G be a group of order year where a > 0, p ispr'ime, and r is

not divisible by 10. Then the number of subgroups in G of each order p3, where
0 g )3 g a, is congruent to 1 modulo 30.

Proof. The analysis which gave the first Sylow theorem also gives this
theorem with little further development. In the course of the earlier
proof we showed that G has t subgroups of order 123, where t satisfies the
congruence tp“”3 E p“—/3 (modulo you—5+1).
This, of course, implies that t E l modulo 1). The theorem follows.l:|
148 Sylow theorems

COROLLARY (Sylow’s third theorem). The number of Sylow p-subgromzs

of the finite group G is congruent to 1 modulo 3). I]

THEOREM. If G1, is a. Sylow p-subgroup of the finite group G, and if P

is any p-subgroug; of G, then G1, contains a. suitable conjugate of P.

Proof. We suppose as usual that G1, has order p“ and that the set .9
of its right cosets in G has 1‘ elements. We define a certain group ofpermu-
tations which act on Q, as follows. If x e P then let $95 map 099 in .92
onto G1, gas in 9?. We check that $95 is a permutation of 9?. The equation
(G1) ”fix-”95) = Gr 33—117 = G11
ensures that $475 is onto .933. Since G], ya: = Gphx implies Gpg = G, h, we
see that $95 is one—one. Therefore 9645 is a permutation.
Next, we assert that (96¢)(y95) = (xy)¢ for all x, y in P because it is easy
to verify that each of (qqS) and (xy)¢ maps the coset Gpg onto the
coset G, 91:31. It follows readily that {9095 :x e P} is a group K of permuta-
tions of .933. Its order clearly divides |P[ and is therefore a power of 1),
say 109- '
Consider now the transitivity classes in .9? under K. The first theorem
of this Appendix shows that each transitivity class contains 107 cosets,
for some 3/ satisfying 0 g y g B. But the number ’1' of elements in .92 is
not divisible by p, and so we must have some 'y = 0. In other words,
there is an element Gyg of 9? Which is fixed under the action of P; we
have G, 990 = G1, 9 for all x in P. Since G9 year1 = G1, we see that gasg—1 e G”
for all x e P; that is, n‘1 g G1,. That is precisely What had to be
proved. E]

COROLLARY (Sylow’s second theorem). Any two Sylow ftp-subgroups of

a finite gram) are conjugate. [I
8

Generators and relations

THE free groups are an important and in many ways a natural

class ofgroups to consider. They arise in problems from branches
of mathematics other than algebra, and they are closely related
to several aspects of group theory.
Our approach to free groups will be through generators and
relations. Any group contains a set of generators (as we saw in
Chapter 3), for the set of all its elements will always generate
the group. There will, of course, be many relations involving
these generators; for instance, relations of the forms ab = l,
a" = l. The quaternion group Qs gives another illustration (see
Example 1.30). If we take a, b to be the matrices
0 1 0 'i
—1 0 ’ i 0
respectively, it will be found that 98 g gp{a, b} and that
a3 = b2 = (ab)2
is a pair of relations involving these generators. Finally consider
the group D = gp{a, b} of Chapter 3, problem 5. Here
1 o 1 1
“=(o —1)’ b=(0 —1)’
and we have a2 = b2 = 1 as relations involving the generators.
Some experimenting with Q8 and D may perhaps make the
reader feel that all relations in the given generators can be
derived from the stated relations and no others, and convince
him that it is natural to think of Q8 and D as being somehow
‘defined’ by the above relations. We thus have a problem
150 Generators and relations

converse to that of finding relations among generators of a

given group; this converse asks if a given set of relations can be
used to define a group in some abstract fashion. This is the sort
of problem which arises, for example, in topology and which we
shall answer once we have put the question into a rigorous
framework—the theory of free groups.
Another example may help the reader to see the difi'iculties
we face. Suppose we feel a need to construct a group with
generators at, b and with the relation a2 = b2. In what sense does
such a group exist? There certainly are such groups—the trivial
group will satisfy any relation, and this particular relation is
also satisfied in both Q8 and D. What can be said about all groups
associated with this relation? And is there a special, distinguished
group in this set?
We shall first define groups ‘with no relations’——the free
groups—and then show how to introduce such relations as we
desire into these groups. There must, of course, be equations
like min—1 = 1 in any group, but it is best to regard these as not
being proper relations.
We take any non-empty set X and proceed to define the free
group on X. Let X = {90A : A e A) where A is a suitable index set.
We take another set in one—one correspondence with X; call it
X-1 and write its elements as xfl, for A EA. Note that the
superscript has nothing to do with inversion in a group, as we
have no group in which to operate yet—though misleading at
the moment this particular symbolism will be convenient later.

DEFINITION. A word in the elements of X U X-1 is an ordered

set of n elements, each from X U X‘1 with repetitions allowed,
for some n 2 0. The length of a word is the integer n.
Notation for words will also be suggestive in a formal way.
A typical word of length n > 0 will be written as xii...xj:
where each 6‘ is 1 or —1; the (unique) word of length 0 will be
written as 1. (Of course, 901 means on.) Note carefully that
multiplication of elements in X U X“1 is not implied by the
notation. A case could be made at this stage for writing the
 Generators and relations 151

above word as (90511, ”031,-": wig), for instance. As it is an ordered

set it could be regarded as a sort of vector.
Example 8.01. If X = {95,31} then some typical words in
X U X'1 are:
$4, x‘lxyy—l, w‘lxx—l, yyyxy—l, 1.
It is a consequence of the definition that two words are equal
if and only if they have the same length and corresponding terms
are equal. The words above are a set of five distinct words, for
no two have the same length.
Next we define the ‘product’ of two words. First, let w be
an arbitrary word; then the product of 1 with w and of w with
1 is to be w. Secondly, take two words of positive lengths, say
u = xi}...x§; and v = 90311...n
,1. in the obvious notation. The
product in) is defined to be
we} 1353.222: 90%.
Note that the length of the product is n+m, and that no :73 1m
in general. The elements of X U X*1 are words of length 1, and
every word (except perhaps the empty word) is the product of
certain of these particular words. Of course, the set of all words
will not form a group under this multiplication, for no word of
positive length will have an inverse. But it is clearly true that
the associative law holds and that our set of words therefore
forms a semigroup.
Our next aim is to define a group whose elements are certain
equivalence classes of these words.
DEFINITION. Two words u, v in X U X‘1 are adjacent if there
are words 21, 22 and an element a in X U X’1 for which 11, = 2122,
'0 = 210500—122; or u = 21 (La—122, v = zl 22. (Interpret (2:4)—1
as 91:.)
Note that we cannot meaningfully talk of ‘cancellation’
because we are not working in a group. (In fact the very purpose
of defining adjacent words is to enable us to talk of cancellation
in due course.) In the examples above, 3349631314 is adjacent to
yy‘l (take 21 = 1, 22 = yy‘l, a, = x4) and also to $42: (take
152 Generators and relations

21 = $4.27, 22 = 1, a = 3/), while x—lxx—l is adjacent to 90—1. Of

course, a is adjacent to o if and only if v is adjacent to u.
DEFINITION. Two words u, o in X U X‘1 are equivalent if
there exist words 21,..., z” (with n 2 1) such that u = 21, v = z”,
and 21 is adjacent to 2H1 for i = l,...,n——1.
Two identical words are equivalent in a trivial way (with
n = 1). In our examples x—lxyy—l is equivalent to 1, for x—lxyy—l
is adjacent to gay—1 which is in turn adjacent to 1; and of course
x—lxarl is equivalent to 96—1.
It is trivial to verify that our definition of equivalence be-
tween words is an equivalence relation (in the usual sense) on
the set of all words in X U X-l; the reflexive property is a
triviality, symmetry follows from the definitions of equivalent
and adjacent words, transitivity follows from the definition of
equivalent words. We may therefore form the equivalence
classes. That class containing the word w will be denoted by [w].
DEFINITION. The product of equivalence classes [u], [v] of
words in X U X—1 is defined to be [ac].
THEOREM 8.02. The product of two equivalence classes of
words in X U X—1 is well defined, and the set of all equivalence
classes with this binary operation forms a group.
Proof. The statement about good definition is that
[u’v’] = [uc] if [u’] = [u] and [v’] = [v]. The fact that u and u’
are equivalent implies that [u’v] = [uo], because u’o and at are
adjacent if u’ and u are adjacent. Similarly [u’v’] = [u'o]
follows from the equivalence of v and '0’. Therefore [u’o’] = [uv],
as required.
We now verify the group axioms. Closure is clear from the
definition of product and its subsequent justification. The
associative law is a consequence of the associative law
u(ow) = (u'v)w for the multiplication of the words u, r), w:
[ul([v][W]) = [“1v = [u(vw)l,
([ul[v])[u’] = [WW] = [WW],
and SO [u]([01[wl) = ([u][v])[w]-
 Generators and relations 153

The class of words containing 1 is clearly an identity element,

since [w][l] = [w] for any word w. If to = alman with each
a; e X U X*1 then we may take [a;1 afl] as an inverse of the
class [w], because [wag1 ...af 1] = [1]. [I

DEFINITION. The free group on the non-empty set X is the set

of equivalence classes of words in X U X‘1 with the binary
operation described above.

DEFINITION. A word in X U X’1 is reduced if it has the form

agnaig with x :1; 7E xge‘forl = 1, 2,...,n—1. (Of course 964-1)
is to be understood as x.)
In Example 8.01 above $104 and yyyxy‘l and 1 are the reduced
words.

THEOREM 8.03. Each equivalence class of words in X U X—1

contains one and only one reduced word.
Proof. Let w be a word in X U X-1. There is no particular
trouble in showing that if w is not reduced then it is adjacent to
a word of smaller length, and so an inductive argument will
produce an equivalent reduced word. To show that [w] contains
only one reduced word requires more thought, and is an essential
step in the theory of free groups. We define a particular method
of reduction, that is of obtaining a reduced word equivalent to
a given word.
Let 7/) = alman where az- EX U X-1 for 1 <11 < n. Let
wo = l, w1 = a1. Suppose that w,- is defined, Where 1 g l < n;
we produce the definition for lab-+1, in two separate cases. It must
be remembered that 207; is a word, so that one can speak Of its
last term without ambiguity. If wi does not have last term af+11
then we put wH1 = wi ai+1; if wi does end in a531, say w; = 2a,;—11,
then 2 is uniquely determined (because 21 ai‘fil = 22 ai‘fil implies
that 2:1 = 22 by the definition of a word), and we define wqb-+1 tO be
2. This gives an inductive definition for we, w1,..., w”; of course, if
n = 0 then 10” is defined to be 1. A consequence of the definition
is that w; is reduced for 0 g l g n, and that in particular wn is
154 Generators and relations

reduced. Another easy fact is that wi is equivalent to al ...ai

for each i, and so [wn = [w]. Note also that if w is already
reduced then wn = 71;.
Next we show that two adjacent words u, '0 have reduced
forms which are identical. Let
u = a1...a,a,+1...an,
v = a1...a,xx-1a,+1...an
where x e X U X4; this choice of u and v is general enough.
Suppose the procedure described above gives the sequence
uo = 1, up ..., un for u, and v0 = 1, cl, ..., vn+2 for '0. Since wi
was determined, in the inductive definition, by the first i factors
in w we clearly have
“0 = ’00, “1 = ”I, ..., u, = ’07..

We shall show that u, = 12,”, in two separate cases.

(i) If u, does not end in 90—1 then or = u,, v,+1 = v, x, and
vr+2 = 2),. (Here we use the definition of wt.) Therefore
“1' = ’U,+2.
(ii) If ur does end in 96—1, say u, = zx-l, then 2 cannot have a
reduced form 20 x for any 20; for if it did then u, = 20 marl
would not be in reduced form. Therefore 9;, = u,,
v,+1 = z, and vr+2 = 95-1 = up as required.
Thus in either case u, = 0H2. Since the final n—r factors in the
expressions for u and 1) correspond when taken in the obvious
order, we have a,“ = vr+2+¢ for i = O, l,...,’n—r. In particular
an = on”, and it follows that the procedure for giving a reduced
form yields the same result for u as for '0.
Finally let u, v be any two reduced words in [11)]. They can be
associated by a chain of adjacent words, because they are
equivalent. We apply our procedure to all the words in the chain.
The result is the same reduced word in each case. Since the
procedure does not alter words which are already reduced it Will
not alter u and 2). Therefore u and v are identical words. There—
fore [w] contains precisely one reduced word. [I
 Generators and relations 155

We shall not always write the elements of the free group on X

as equivalence classes—we shall often merely use a representative
from each class, usually the reduced word of the class, as no real
confusion will result. In particular we write [as] as x for each
x EX UX—l, and [1] as 1.
If the sets X and Y are in one—one correspondence then the
free groups on them are (clearly) isomorphic. It is also true that
if two free groups are isomorphic then the corresponding sets X
and Y have the same cardinal (in other words, are in one—one
correspondence); we shall prove this later for finite sets X and Y.
Thus the cardinal of X is an invariant of the free group F on X,
and it is called the rank of F.
We note some properties of free groups, which are easy enough
to give us some feeling for these objects.
(i) Every free group is torsionfree. For if a is an element of
finite order in the free group F on X then a” = 1 for some n 2 1.
Let the reduced form of a be
b;1...b1‘1a1 ...a8b1... b,
where a1 72 agl; of course, r may possibly be 0. Then
a“ = bfl...bf1(a1...as)“b1...b,,
and so the reduced form of a” has length 2r+ns. But this length
is 0 because a” = 1. It follows that r = s = 0 since n > 1.
Therefore the reduced form of a is the empty word, and a = 1.
(ii) In particular, if X = {x} then the free group on X is infinite
cyclic.
(iii) If X has more than one element then the free group on X
is non—abelian, for if 90,31 6 X then my 75 gm.
It should be clear what the rigorous interpretation of the
statement that a free group ‘has no relations’ is. It is simply the
fact that if (11 an is a reduced word equal to l, with each
It, 6 X U X—l, then at = 0 and the word is empty.
Next we consider a subtler concept, the free generators of a
free group, and take as an illustration the free group F on the set
X = {95, y}. The elements 90, 3/, which certainly generate F, have
the property that any reduced word in them which equals 1 must
156 Generators and relations

be the empty word. But other generators also have this property.
Consider, for instance, {96, 9643196}. This set generates a subgroup
of F which is F itself because it contains both a: and g. A reduced
word in x, y”6 is the conjugate of a reduced word in x, y; for
example, x—l(gm)2x2(g‘”) "1 = (x—1g2x2y—1)”. Therefore the only
reduced word in {96, 3/95} equal to 1 is the empty word.

DEFINITION. Afree basis (or a set offree generators) for the free
group F is a set of generators for F with the property that the
only reduced word in them equal to 1 is the empty word. (Of
course, a free basis is said to generate F freely.)

It should be clear that a free group has many bases other than
the set on which it was defined. The concept is not applicable
to arbitrary groups (for what would be meant by a free basis for
the quaternion group?) In the above example {96, g2} freely
generates a subgroup of F that happens (and can be proved) to
be a proper subgroup of F; and {90,g, mg} is a set of generators
which is not a free basis. The length of a word depends strongly
on the free basis in use; thus x—lyx has length 3 when referred to
{96, g} in the above example, and length 1 when referred to
{50, x—lgx}.
The next few results attempt to relate free groups to arbitrary
groups.

THEOREM 8.04. Let F be a free group with free basis

X = {xkzh e A},
and let G be an arbitrary group. If {9M 6 A} is a set of arbitrary
elements of G', then there exists a unique homomorphism from F
into G' which maps at) onto 9), for all A e A.

Proof. Let ()5 be the mapping from X into 0 for which

xAgb = g), (A e A), and let [w] be any element of F. We suppose
that w = xi; rig, in the usual notation, and define [w]¢ to be
9;: gig. In order to show that qS is a mapping we have to show
 Generators and relations 157

that if [n] = [u] then [n]¢ = [v]¢; but this is clear if u and v are
adjacent, and it therefore holds when they are equivalent.
It is now easy to see that «,5 is a homomorphism. Let [u], [v]
be arbitrary elements of F, so that [no] = [n][v], as above. Then
the definition of 525 should make it clear that [nv]¢ = [u]g[>.[v]¢,
and it follows that
_ (MM)?S = [rlt- [vlt
as requ1red.
Since qt maps [my] onto 9", it is the required homomorphism.
Its uniqueness follows from the fact that the images of the
reduced words in X U X*1 are determined once the images of the
(EA are specified. I]
With the aid of this result we can discuss the invariance Of
rank.

THEOREM 8.05. Let Fm, F” be free groups of finite ranks m, n

respectively. Then Fm g F” if and only if m = n.
Proof. We mentioned earlier that if m = n then Fm g F”.
Suppose, conversely, that Fm g Fm and let G denote a group of
order 2 with generator 9. We consider homomorphisms from
Fm to G, and for this purpose we suppose that Fm has free basis
{x1,..., mm}. Such a homomorphism is completely determined by
the image of each 901., and each image may be g or 1; there are
therefore 2m—1 distinct homomorphisms from Fm onto G, in
addition to the trivial one in which every xi is mapped onto 1.
The kernel K of a homomorphism from Fm onto G is a normal
subgroup om, and Fm/K g G by the first isomorphism theorem.
Conversely, a normal subgroup of index 2 in Fm is the kernel of
a homomorphism of Fm onto an isomorphic copy of G. It follows
that Fm has precisely 2m—1 normal subgroups of index 2.
Similarly Fn has precisely 2"— 1 normal subgroups of index 2.
Because Fm g F” we have 2m—1 = 2”—-1, and so m = n, as
required. [I

COROLLARY 8.06. Every free basis for F” contains just n

elements. [I
158 Generators and relations

Theorem 8.05 can be proved for free groups of arbitrary rank,

but this requires an acquaintance with the necessary results
concerning infinite sets.
Next, a converse to the theorem stating the existence of homo-
morphisms from a free group into an arbitrary group. (For the
sake of tidiness we now define the free group on the empty set to
be the group with one element.)

THEOREM 8.07. If G = gp{g,\:A e A} is a group with the

property that, for an arbitrary group H containing a complex
{hxzh e A}, the mapping 0 : gA —> k). can be extended to a homomor-
phism of G into H, then G is a free group and {gs EA} is a free
basis for G.
Proof. We make a particular choice for H; we take H to be the
free group with {h,\:)\ e A} as a free basis. There is by Theorem
8.04 a homomorphism (b from H to G with h 45 = 9). for all A EA.
We are given, in addition, a homomorphism 6 from G to H with
9A0 = h. It is clear that
7M4”) = 7% 9A(9¢) = 9A:
and so «150, 095 are identity mappings on H, G respectively. It
follows that 0 is an isomorphism. Since H is free with free basis
{h,\:)\ e A}, it follows that G is free with free basis {gAzh e A}. I]
Sometimes a free group is defined by the property mentioned in
this theorem, but we have preferred a more constructive
definition in the present account.

THEOREM 8.08. Every group is isomorphic to a factor group of

a suitable free group. Every group with a set of n generators is iso-
morphic to a factor group of the free group of rank n.
Proof. Take a set of generators {gMA e A} for the arbitrary
group G; if need be the set of all its elements may be taken. Let
X = {xAzh e A} be some abstract set in one—one correspondence
with the generating set. Let X be a free basis for the free group
F. Then there is (by Theorem 8.04) a homomorphism <36 of F
into G such that amt = g for all A E A. The first statement of
 Generators and relations 159

the theorem follows. For the second we simply take A to be a

set of n elements. [1
Example 8.09. Let F be the free group of (finite) rank n, and
let A be the direct sum ofn infinite cyclic groups. Then F[K g A,
for some normal subgroup K of F, by Theorem 8.08. Since F/K
is abelian, 8(F) < K, by Theorem 4.21. Therefore, by the third
isomorphism theorem, F[K is a factor group of F/8(F) (compare
Example 6.17). But F/8(F) is an abelian group with a set of n
generators. It is a consequence of the structure theorem (5.14)
for finitely generated abelian groups that F/8(F) g F[K. We
have therefore proved: if F is free of rank n then F/8(F) is the
direct sum of n infinite cyclic groups.
We mention in passing the concept of a free abelian group.
This is simply the direct sum of some set of infinite cyclic groups,
and we may define free basis and rank, if we wish, in the obvious
way. The additive group of rationals is a torsionfree abelian
group which is not free abelian, as the reader may prove for
himself.
We are now ready for generators and relations. Suppose that
we try to express the quaternion group as some factor group of
a free group. We take a free group F With free basis {(16, y}, and
we take a quaternion group Q8 generated by a, b where
O 1 0 2'

Everything is arranged once we define the homomorphism 45

from F to 9,, by putting
x¢=a, y¢=b.
Now we have the relations a2 = b2 = (ab)2 in QB. They imply
that yzw—Z and (:Icy)2y‘2 are in the kernel K of ¢ because

(yaw—2M = (gazes—2 = bza—z = 1,

2b-2 = 1.
((xy)2y—2)¢ = ((x¢)(y¢))2(y¢)-2 = (ab)
Thus the kernel contains the normal closure of {312964, (xy)2y—2}—
this is the least normal subgroup containing these two elements.
160 Generators and relations

And now we have a hint as to what to do about defining relations

for groups.
Let us take another point of view. Suppose that we want to
construct a group with certain generators and relations, rather
than investigate a given one. To be specific, let us not give
ourselves a quaternion or any other group, but let us ask for
a group with generators a, b and the relation a2 = b2. We take
a free group F with free basis {(49, y}. We take a normal subgroup
K of F containing (fly—2, and we consider F/K. If a = 96K,
b = yK, then we certainly have a2 = b2 in F/K. If we take K to
be the normal closure of cozy—2 then we are justified in regarding
the resulting group G/K as being (in some sense) the most
comprehensive group with generators a, b and relation a2 = b2.
We now formalize these ideas.
DEFINITION. A group presentation consists of a free basis X
for a free group together with a set of words in X U X4.
We derive a group from a presentation in the following way.
Let F be the free group on X, and let W = {wyzA e A’} be the
given set of words in X U X*1, ie. elements of F. Define K to be
the normal closure in F of W. The group presented is then F/K.
Since each w)‘ e K we have, of course, wAK = K in F/K, and
so we may claim informally that ‘the relation w,\ = 1 is satisfied
in F/K’.
Thus in the above example X = {x,y} and W = {90231-2}.
A group presentation should ideally be denoted in some such
way as gp{X : W}. We shall, however, use a less exact notation
in which ‘relations’ replace the words of W, for example
gp{X:w,\ = l (A e A’)}, as we shall usually be more interested in
the group than in its presentation. The group discussed above
would thus be written as gp{a, bzagb—2 = 1} or even as
gp{a,b:a2 = b2}.

THEOREM 8.10 (von Dyck’s theorem). If the group G is

defined by some set of relations and the group H is defined by a
larger set of relations, then H is isomorphic to a factor group of G'.
 Generators and relations 161

Proof. It is understood that the generating sets of G, H are in

one—one correspondence so that each is isomorphic to a suitable
factor group of the same free group F. Thus F[K1 g G,
F/K2 g H, where K1 g K.3 by the condition on the relations.
By the third isomorphism theorem,
H g F/Kz a (F/K1)/(K2/K1) g GIN
where N g Kz/Kl. Thus H is isomorphic to a factor group of G,
as required. I]
Example 8.11. Consider G = gp{a, bra2 = b2}. If we add the
additional relation b2 = (ab)2 then we obtain the quaternion
group Q3, While if we add b2 = 1 (and perhaps other relations, for
all that we have proved so far) then we obtain the infinite
dihedral group D. Thus von Dyck’s theorem tells us that G has
factor groups isomorphic to Q8 and D; in particular, G is infinite.
Von Dyck’s theorem completes the solution to the general
problems about generators and relations posed earlier in this
chapter. For we have seen that there is a distinguished group
associated with given generators and relations, and every
associated group is a factor group of the distinguished one.
In view of the distinction we have made between a presenta-
tion and the group defined by a presentation, it should be no
surprise to find that one group may be presented in many ways.
The example G above could be presented as
gp{a,b,c:a2 = b2, 0 = 1}.
The less trivial instances which follow have been selected to
show the sort of ad has calculation which is often the only
efiective way of investigating a given presentation.
Example 8.12. Consider gp{a, b, c,d:bc = (1, col = a, da = b,
ab = c}. If we eliminate d between the first and third relations
in this group we find that c = a‘l. The relations then give
= a2, 6 = a3, and b = 05*; thus a5 = 1, and indeed
b5 = c5 = d5 = 1. It follows that the group has order 5 or less.
But since there does exist a group of order 5 satisfying the above
relations, namely the additive residues modulo 5 with a = [1],
853137 M
162 Generators and relations

b = [3], e = [4], d = [2], it follows that the group presented

indeed has order 5. Note that this last step, establishing the
existence of a particular group, is essential.
The point of this example is that the group presented may also
be described as gp{g:g5 = 1}. The isomorphism is by no means
evident from the presentations.

Example 8.13. Consider the following groups:

G = gp{w,b:a,4 = b2, b—lab = (14},
H = gp{c, dzcdcdc = dad, dcdcd = ado}.
In G the element a4 is central, being a power of both generators,
and 8° a4 = b—1a4b = (b—Iab)4 = (a-1)4,
from which we obtain a8 = 1. Thus gp{a} is a normal subgroup
of G, which has order at most 16. That G really does have order
16 must be proved by exhibiting a specific group satisfying the
given relations. This may be achieved in several ways, for
instance by means of the matrices
3 3 0 4
A =
(—3 3): B (4 0), =

whose entries are residue classes modulo 17; the reader should
be able to verify that
A4 = B2, B-lAB = A4,
and that gp{A, B} has order 16.
It is not obvious that G g H, but we shall proceed to prove
this. In H we have
dad = cdc.dc = dcdcd.dc,
and so cddc = 1, c2 :2 (1‘2. Therefore a2 and d2 are central
elements in H. We then find that as cdcdcd—lc'Jd—l = 1 we have
(cd)4 E Z = gp{02}, and it may be seen that gp{c} is a normal
subgroup of H/Z. Hence H/Z has order at most 8. Another
appeal to the defining relations gives
cdcd = czdcdc, cdcd = dcdcdz;
 Generators and relations 163

it follows that c2 = d2. But 62 = (Z4, and we thus have (:4 = 1;

Z has order at most 2. Therefore H has order 16 or less.
However, we know that there is a group G of order 16 with
generators a, b and relations a4 = b2, b'lab = a—1. If we
substitute 0 for b and d for ab, we find that this group satisfies the
given relations for H. Since G = gp{b,ab} we conclude that H
has order 16, and that G g H.
It is, of course, possible to consider presentations with
infinitely many generators or relations, but as an arbitrary
group has such a presentation there is some excuse to study a
restricted class only.
DEFINITION. A finite presentation of a group is a group
presentation consisting of a finite free basis and a finite set of
words.
DEFINITION. A finitely presented group is a group which has
a finite presentation.
All the examples considered explicitly above are finitely
presented. Any finite group may be finitely presented because
the set of its elements is finite and its multiplication table is
finite.
DEFINITION. The deficiency of a finite presentation gp{X : W}
of a group is the number of elements in X less the number of
elements in W.
This definition of deficiency is good enough for our purpose,
namely Theorem 8.16, but it may be enlightening to discuss how
it could possibly be improved. Given any presentation gp{X : W},
we may always decrease the deficiency by adding redundant
elements to W; perhaps of the form wf or w1 w2 Where wl, 202 are
already in W and x e X. We might wish, conversely, to rid
ourselves of such superfluous relations, but some tricky points
arlse.

Example 8.14. Consider gp{9z::ac2 = 1, x4 = 1, x“ = 1}. This

defines the group of order 2. We can change the presentation
164 Generators and relations

to gp{x:x2 = 1}, and the deficiency becomes 0. But we could

also delete one relation Without changing the underlying
group, obtaining gp{x:x4 = 1, x6 = 1}; in this case neither of
the two remaining relations is dispensable, and the deficiency
is —1.
One might compare this situation to that which arises when
sets of generators for the cyclic group of order 6 are being
examined. There is a one-element generating set, but there is
also a two-element set of which each element generates a proper
subgroup. We now consider some defining relations in these
generators.
Example 8.15. Consider gp{ar:,g/::c2 = 1, y3 = 1, [x,y] = 1}.
This clearly defines the cyclic group of order 6. None of the three
given relations is superfluous, for omission of any one gives rise
to a larger group. If x2 = 1 is omitted the resulting group has
an infinite cyclic factor group; similarly if 3/3 = 1 is omitted;
while if [x,y] = 1 is abandoned we find a non-abelian factor
group generated by the elements (12) and (123) in 83.
Nevertheless this cyclic group of order 6 can be defined by two
irredundant relations in the same irredundant generators:
gp{x,y:x = (mg/)3, y = (mg/)4}. This is quite easy to prove.
On the whole it seems futile to alter presentations by omitting
redundant relations. A better way to compare presentations
might be to consider all sets of relations, in afixed set ofgenerating
elements of G, which define the group G; and to define the
deficiency of a presentation as the maximum obtainable in this
sense. To be precise, we take a definite presentation gp{X : W}
for G. We then consider all those presentations gp{X’: W’} for
G, where X' has these properties:
(i) X and X' are in one—one correspondence.
(ii) Let F, F' be the free groups on X, X’ respectively, and
R, R’ the normal subgroups corresponding to W, W’ respec-
tively. The correspondence between X and X’ gives rise to an
isomorphism from F/R onto F’/R', in the natural way.
Then define the deficiency of the presentation gp{X : W} to be
 Generators and relations 165

the maximum deficiency (in the earlier sense) of gp{X’: W’}, as

this latter presentation varies.
We now end our digression and regard deficiency as the
concept specified in the formal definition, on page 163.
There are many examples of finite groups with negative
deficiency, and some (such as the cyclic groups and the quater-
nion group) with zero deficiency; but none exist with positive
deficiency.

THEOREM 8.16. Any group defined by a finite presentation with

positive deficiency is necessarily infinite.
Proof. Let G g F/K where F is free with free basis {x1,..., mm}
and K is the normal closure of the complex {w1,..., wn} in F; we
assume m > n. Define ax),- to be the sum of all the exponents of
x,- appearing in w (taken in its reduced form if desired, though it
makes no significant difference). Consider the n equations
”1/

glut-l3.- = 0 (1 <j < n)

for the m numbers Br Because m > n a well—known theoremT
states that there exists a non-zero solution $1,.” 3m}. Though
this solution is in the first instance a set of rational numbers
(the or,” being integers) we may clearly select it as a set of integers,
not all zero.
We shall now construct a homomorphism from G into the
infinite cyclic group 0 with generator c. First we define a
mapping (,5 from {x1,..., 96",} into 0’ by putting a,- ¢ = 05‘, for each
i (note that ,8,- is an integer). Since F is free, 45 defines a homo-
morphism from F into 0, in accordance with Theorem 8.04. The
choice of the or“ and the B,- ensures that wjd = 1 for 1 g j g n.
It follows that gp{w1,..., wn} maps onto 1, and (more relevantly)
that K maps onto 1 because K consists of products of conjugates
of the wjil. Since K95 = 1, there is a homomorphism 1/! from
F/K into 0; it is defined by
(el' = 96¢
1' See, for instance, S. Perlis, Theory of matrices, pp. 45—8 (Addison—Wesley,
1952).
166 Generators and relations

for all x e F, an assertion substantiated by Lemma 6.14. The

fact that the ,8,- are not all 0 shows that the image of F and of F[K
is an infinite subgroup of 0. Therefore F/K, and so 0', are
infinite. [I
We might now ask if a presentation with negative deficiency
necessarily defines a finite group. This is by no means the case.
Example 8.17. Consider gp{w,y:x3 = 3/3 = (mg/)3 = 1}. This
is an infinite group, for it may be verified that if
a = ...(012)(345)...,
b = ...(123)(456)...,
then a3 = b3 = (ab)3 = 1 while gp{a, b} is infinite; for ab2 has
infinite order. On the other hand the given presentation clearly
has deficiency — 1. It is possible and easy to show that none of the
givenrelations can be omitted. For instance, gp{x, gut:3 = y3 = 1}
does not have (xy)3 = 1, as we may see by taking so = a, y = b2.
Similar arguments show that the other relations are indispens-
able.
We have hinted that there are difficulties in any theory of
finitely presented groups because of a lack of general methods
and general results. This is not wholly the fault of mathe-
maticians. Many ofthe fundamental difliculties arise from logical
questions that are too technical to explain here. It will suflice
to say that general results that are obtainable tend to be negative,
and this is exemplified by the fact that the way to decide in
practice whether a given finitely presented group is finite, or
even trivial, is often by ad hoc calculation rather than by applica-
tion of a theory. On the whole the finitely presented groups are
a far less tractable class than (say) the free groups. If we
construct groups by merely writing them down we must not
be surprised to meet difficulties later on.
In conclusion we take another glance at the free groups, and
prove a result which can be reached without the introduction of
concepts too sophisticated for this book.
DEFINITION. Let 9’ be a property which groups may possess.
The group G has the property 9 residually if for each element a;
 Generators and relations 167

in G with a ¢ 1 there exists a normal subgroup Nx (depending

on x) such that x 9% Na, and G/Nw has the property .9”.

THEOREM 8.18. Every free group is residually finite.

Proof. We have to prove that if x is an element of the free

group F, with :1: ¢ 1, then there is a normal subgroup Na, of F
such that x ¢Nm and F/Nm is finite. Let F have free basis
X :2 {mph 6 A} and let the reduced form of x be xi: 93%, with
each 6,. = i 1. Thus n 2 1, and A1,..., An need not all be distinct.
Let Snfl denote, as usual, the symmetric group on {1,..., n+ 1}.
We shall define a mapping 95 from F into 8M1 such that 2195 9!: 1;
thus if K is the kernel of 45 then F/K is finite, 90 at K, and the
theorem will be proved. In fact 91:95 is to be a permutation
mapping the symbol 1 into the symbol n+1. To arrange this
we ask that semi = 0A; where oi: maps i into i+1, while xAzfi = 1
ifA ¢ {A1,...,)\,,}. Thus ”A; is to map i into i+1 if 5t = l, andi+1
into i if e,- = —1. Since F is free this mapping ¢ defined on the
free basis X would determine a homomorphism from F into
Sn“, by Theorem 8.04.
However, we have not yet completely specified the 0A,. Note
that our requirements so far are not contradictory. If 0),, maps
i into both i—l and i+1 then )‘i—l = A,“ ei_1 = —1, e,- = +1,
which is impossible because we are using the reduced form of .73.
Similarly if 0),, maps both i—1 and i+1 into i then Ai_1 = Ai,
6,4 = +1, 6,, = —1, again impossible. Thus a)“ is one—one, as
far as it is defined, and we may complete its definition as a
permutation of n+ 1 symbols in an arbitrary way.
Take Nw = K, and the theorem is proved. I]

COROLLARY 8.19. The intersection of all subgroups of finite

index in any free group is 1.

Proof. Let x lie in every subgroup having finite index in the

free group F. If x i 1 then a- ¢ NZ, by the theorem, N,c being
a certain subgroup of finite index. This contradiction shows that
x = 1. El
168 Generators and relations

The importance of residual properties is connected with the

following result.

THEOREM 8.20. If the group G has the property W residmlly

then G is isomorphic to a subgroup of the cartesian product of a
suitable set of groups each having the property 9”.
Proof. Consider the normal subgroups {Nxzx e G, 1: ¢ 1} such
that 13¢l and G/Nw has 9. Let N = n Nan. Then N S G
newest
by Theorem 6.06. Hg 6 N and y 72 1 then y 99 N” and so y ¢ N,
a contradiction (as in Corollary 8.19). Hence 3/ = 1 if y e N,
that isN = 1. By Theorem 6.32 on cartesian products G = G/N
is isomorphic to a subgroup of H0 (G/ZVac), as required. [I
laéxeG

COROLLARY 8.21. Every free group is isomorphic to a subgroup

of the cartesian product of a suitable set offinite groups.
Proof. Use Theorems 8.18 and 8.20. I]

Problems
1. Prove the following statements about the elements a, b in an arbitrary
free group:
(i) Ifa” = b”withn > Othena = b.
(ii) If amb" = bfla’" with mm at 0 then ab = ba. (Hint: use (i).)
(iii) If ab = ba then gp{a, b} is cyclic.
(iv) If the equation x" = a has a solution x for every positive integer
n then a = 1.

2. The group G has a normal subgroup N such that G/N is a free group.
Show that G contains a subgroup F such that F is a free group, G = NF,
and N n F = 1.
3. Let A, B denote the matrices

(t i), (3 i),
1: being a real number which satisfies no polynomial equation With rational
coefficients (which are understood to be not all zero). Prove that
gp{A, B} (a multiplicative group is intended) is free and that {A, B} is
a free basis.
 Generators and relations 169

4. Prove that every abelian group is isomorphic to some factor group of

a suitable free abelian group.
5. A certain free group F has {(1, b} for a free basis. Show that {ab, b} is
another free basis, and deduce that {ab", b} is a free basis for all integers n.
Which of the following are free bases for F?
(i) {“1}, bab};
(ii) {(15, b’lab};
(iii) {£16, ba};
6") {ah 5”. a2};
(V) (a, a‘lb'lab}.

6. Find an example of a free group containing elements a, b, o, d such

that [11, b] = [0, d] ¢ 1 while gp{a, b} 75 gp{c, d}.

7. Let F be the free group with free basis {90, y}. Find elements a, b in F
such that the length of [a, b] is positive but less than the length of a and
of 6 (lengths are referred to the given free basis).

8. In a certain free group w1,..., w” (where n > O) are non-trivial elements

in reduced form, for which 'w1 w" = 1. Prove that, for some i, all the
factors in w; are cancelled when w5_1wiwi+1 is written in reduced form
(wo and w”+1 being taken to be 1).

9. Prove that, for any prime p, the matrices

(5 ’i)’ (i, ‘1’)

generate (under multiplication) a free group of rank 2.

10. Let {4 = gp{A, B} where the matrices A, B are

(1 1)
0 1 ’ (2 0)
0 1
respectively (as in Problem 11 of Chapter 3). Prove (using the result of
that problem, or otherwise) that
g ; gp{a,b:bab‘1 = a2}.
11. Prove that gp{a, b} is infinite if
a = ...(l,..., n)('n+ 1,..., 2n)...(i’n+ l,..., (i+ l)n)...,

b = ...(2,...,n+1)(n+2,...,2n+1)...(in+2,....(1Z+1)n+l)...,
and n. 9 2.
Using this result prove that
gp{a,b:a—1b°‘a = b5, b‘la“b = afi}
is infinite if a, ,8 have a common factor greater than 1.
170 Generators and relations

12. The elements a, b of the group G are such that

b—lmb = (1)9, (1711)"?! = b8
where or, y are positive. By finding two expressions for a5”, or otherwise,
prove that I)”’6 commutes with a” and with at”. Deduce that if a and B
are coprime then bV—a commutes with a, and that 6(7‘5)” = 1.
Prove that gp{a, b :b—la,“b = aw”, ar‘lbm = b7+1} has order 1; and that
gp{a, bzb—laab = a5, a’1b7a = b9} is the quaternion group.

13. What is the order of the group presented by generators a, b and

defining relations
nab-1a“zba—1b2ab‘3a‘1b“ab—ab"1053ba—3b3a‘1b‘2ab3a—1b—2ab3a—1b—3a, = 1,
b3a—1b‘3ab'1a2ba‘3b“lazba'sa“1b”ab—”aab‘la’zbaab—la—2ba3b—1a—2b = l?

*14. (i) Prove that if a, b are elements of some group for which there
exist integers 'y, 3 giving aVbasb = 1 then ay'a commutes with b.
(ii) Let G = gp{a,b:ban°‘b2 = a2, baflb = a}. Prove (by considering
G/8( G) or otherwise) that G is infinite if 205—— 33 = 1 and, in particular, if
a = = — 1.
(iii) Assume now that the integers oz, [3 associated with G are arbitrary.
Prove (by means of (i) or otherwise) that as“ and an” both commute
with b.
(iv) Prove that if c = aa—B—lb then acor1 = c”.
(v) Deduce that if 2a—3B ¢ 1, then G is finite.
(vi) Show that (iii) implies that G is abelian if [3+1 and 2a+2 are
coprime.

*15. Which of the following statements are correct? Prove each answer
you give.
(i) If every subgroup generated by three elements of G is free then
G itself is free.
(ii) No free group is a (non-trivial) direct product.
(iii) Every subgroup of a residually finite group is residually finite.
(iv) Every factor group of a residually finite group is residually finite.
(v) If the free group Fm of rank m is generated by a set of n elements,
where n > m, then a subset of m of these elements will provide
a free basis for Fm.

***16. Let F be a free group with free basis X = {xAzh e A}. Prove
that F is residually a finite p-group by means of the following steps:
(i) Take an arbitrary non-trivial reduced word 963;"; 963’)" in F with each
m; non-zero and hi 7E Ai+1 for 1 g i < M. Show that there is a power 10“
of p for which m1 ...mn sé 0 modulo 1)“.
(ii) Let ”I! be the set of all (n+ 1) x (n+ 1) matrices with integers
modulo p“ for entries, each element on the diagonal being 1 and each
 Generators and relations 171

element below the diagonal being 0. Prove that ”71, and all its subgroups,
form (under multiplication) finite p-groups.
(iii) Let I be the unit matrix in ‘ZI, and E575 the matrix Whose only non-
zero entry is 1 in row j, column Ia. Prove that EjlklEg-‘ki = 0 if 161 75 jg,
and that Ejlk1 EM. = EN“ if 101 = jg, Deduce that (I +Ejk)m = I +mEjk,
ifj aé k.
(iv) Define X“ =AH (I +1?”+1) with X” = I if no A,- equals [1,, and
show that the entry in Saw 1, column n+ 1 of X321 ...Xfif‘?‘ is m1 ...mn.
(v) Construct a. homomorphism from F into 01 under which the image
of xi”? ...xgy is non-trivial.
9

Nilpotent groups

THE theorem that states that a finite group with p“ dividing

its order has a subgroup of order 3)“ indicates that the study of
the finite p-groups is basic to any theory of finite groups. Many
of the properties of finite p-groups are shared by a wider class of
groups, which might be expected to be the general (finite or
infinite) p-groups, but which are in fact the nilpotent groups
defined below. These are much more agreeable to work with
than the infinite p-groups, and seem to be the ‘correct’
generalization of the finite p-groups.
In connection with the series about to be introduced, we again
state that all series considered in this book are finite.

DEFINITION. A central series in the group G is a normal series

64:63; 019...? GM; Gr=1
such that
(i) 0,; is a normal subgroup of G, for 0 g i g r; and
(ii) aid/G..- lies in the centre of G/G’i, for l g i g 1'.
(Note that (i) follows from (ii) and the fact that the central
subgroup G,_1 is normal in G.)

DEFINITION. A group is nilpotent if it has a central series.

A group with a central series of length r is said to be nilpotent of
class 1'.
Note that, as we have defined it, the class is not an invariant
of a nilpotent group—the length of a central series may always be
increased by inserting repeated terms. The least value of 1*
associated with the group is, of course, an invariant.
 Nilpotent groups 173

Abelian groups are nilpotent, being of class 1. Other examples

of nilpotent groups are the groups G in which 8(G) g {(G), for
we may take r = 2, G1 = ((G) and then G/§(G) is abelian, in
accordance with Theorem 4.21. Further examples will appear
in the natural course of our discussion.
It will be found, on the other hand, that the symmetric group
8’3 has no central series. The term Gr_1 in the above standard
series must be central in G, and S3 has trivial centre.
We need some minor work in order to apply the above defini-
tions with greater ease. The first item is essentially a matter of
notation. If A, B are subgroups of the group G then [A, B] will
denote gp{[a,b]:a e A, b e B}. We make two observations: if
A <1 G then [A,B] g A, and if A g A1, B < B1 then
[A B] < [141,31].
LEMMA 9.01. Let
G=G0>G1>u. >GT_I>G,=1

be a series of subgroups each normal in G. This is a central series

if and only if [Gi_1, G] g Gi for 1 < i < r.

Proof. The central property for the series just mentioned is

equivalent to G¢_1/Gi g C(G/Gz.) for each i. This in turn is
equivalent to [yGfixGi] = G;- or [y,x]Gi = Gt or [y,x] e Gi for
each i and for all at in G, g in Gi_1. But this is equivalent to
[Gr—1: G] < at [I
THEOREM 9.02. All subgroups and factor groups of the nilpotent
group G are nilpotent. If G is of class r then every subgroup and
factor group has class r.

Proof. Let G have the standard central series Which we

described in Lemma 9.01 and let S be a subgroup of G. Consider

these“ S=SO>SI>N2SLI>ST=I
in S, Where 8;- is defined to be S n G; for 0 g i g r. We have
8; <1 8 by Corollary 6.02, because G1, <1 G. By Lemma 9.01,
174 Nilpotent groups

the series in S will be central if [524, S] g Si for 1 < 13 g r.

But IS;_1 < (AH, S < G, and so [SQ-4, S] g [03-4, 0]. The series
for G’ was central, so by the lemma [Gi_1, G] g 02., and we have
[8254, S] < G’i. Clearly [:3;_1, S] < S, and we thus have
[Si-fl, S] g S n 05 = Si, as desired.
Next let N be any normal subgroup of G', and consider the
following series in G/N:
‘G/N = GoN/N 2 GIN/N 2 2 G,_1N/N 2 o,N/N = N.
That this is a series, and that GiN/N <1 G/N, require only
elementary verification. Typical elements of G/N, Gi_1N/N
are wN, yN respectively where w e G’, g 6 G14; and
[yN,xN] = [y,x]N e G’iN/N
because [31,06] e 63,. Hence [Gi_1N/N, G/N] < G4; N/N, and the
theorem is proved. I]
Note that 8,, cannot be nilpotent for n 2 3, by this theorem.
For if 8,, were nilpotent then its subgroup 6’3 would also be
nilpotent.

THEOREM 9.03. The direct product of a finite set of nilpotent

groups is nilpotent.

Proof. It will sufl‘ice to prove this theorem for the direct

product of two nilpotent groups; once that is done, an easy
inductive argument (which is omitted) would complete the
proof. We lose no further generality in supposing that we are
given nilpotent groups H, K with central series of the same
length r: H = Ho 2 H1 > 2 ILL—1 9 Hr=1a

K=K0>K1>m >Kr-12K7'
for ifnot we may lengthen the shorter series by inserting repeated
members. We shall prove that G = H X K has a central series
of length r, given by G'i = Hix K1, for 0 g 't' g r.
It is not hard to see that G‘i <1 G, so we omit formal proof of
this fact. We have yet to show that [02%, G] g G" for 1 g t g r,
 Nélpotent groups 175

if we are to use Lemma 9.01 in this proof. But an easy computa-

tion gives
[Hi—IXKi—i: HXK] = [Hi-1’ H] X [Ki—19K] < HiXKt
for 1 < 93 < 7‘; that is, [Gi_1, G] < Gt. The proof is complete. I]
Notice that in the proof just given we are identifying direct
factors with subgroups of the direct product in the usual way,
which was explained in Theorem 4.37.
It is not true that the direct product of an arbitrary set of
nilpotent groups is nilpotent. To see this one forms H” Gn Where
n>0
G1, G2,", Gm... are groups such that G” is nilpotent but of no
class less than n. Such groups do exist (see Problems 1 and 3 of
this chapter). If the direct product were nilpotent then it would
have class a, say, and so each G” as a subgroup of G would also
have class 0 (by Theorem 9.02), a contradiction. Of course
no G” certainly cannot be nilpotent, either.
n>0

THEOREM 9.04. Any fim’te p-group is nihzotent.

Proof. Let G be a finite p-group with order 10". (We use

Corollary 7.06 Without explicit reference from now on.) The
proof goes by induction on n. The cases 12. = 0, 1 are trivially
easy, so we take 71. > 1. Theorem 4.28 ensures that ((G) > 1,
and this is the key to the present proof because the inductive
hypothesis may be applied to G/§(G). Since this group has
order strictly less than |G|, it has a central series:
GMG) = Gama) > G3/464) > > salsa) 2 Gr/€(G) = HG),
the normal subgroups G1- of G each containing £(G). We have
[G,-_1/§(G), G/§(G)] g Gi/L'(G) for 1 g 'i g r, by Lemma 9.01.
Therefore we have [y§(G), x€(G)] e Gi/§(G) for all x e G, :1; 6 $54.
Thus [31, x]§(G) e Gi/é(G), and [g], x] 6 G1. Now consider the series
a:GO>GI>"'>GT—1>GT>GT+1= 1

for G. We have Gi <1 G and [G¢_1, G] g Gi for 1 g 13 g r, as

above; further, [G,., G] g G,+1 since G, = £(G) and G,+1 = 1.
Thus we have a central series for G, and G is nilpotent. El
176 Nilpotent groups

In fact the proof just given can be made to yield a little more:
if the group G is such that G/§(G) has class r then G itself has
class r+ 1. The reader should prove this for himself.
The next result is of fundamental importance.
THEOREM 9.05. If the m'lpotent group G has a proper subgroup
H then H is a proper subgroup of its normalizer.
Proof. Take a central series for G:
a: GO>G1>...>G,_1>G,= 1.
We have [G¢_1, G] g G, for 1 < '13 g r, by Lemma 9.01. Suppose
that Gk < H while G,c_1 g H. Such a value of k exists because
1 = G, g H and G = G0 g H, H being aproper subgroup of G.
Th” [0].-.. G] < a. < H.
and so [Gk_1, H] g H. In other terms, G,c_1 normalizes H. By
the choice of lo there is an element of G,c_1 which does not lie in H,
and it follows that N(H) > H. [I

COROLLARY 9.06. If the nilpotent group G has a maximal

subgroup M then M Q G and G/M has prime order.
Proof. We have M < G and so N(M) = G, by the theorem,
since M g H < G implies M = H or H = G. Therefore
M <1 G. By Corollary 6.05 G/M has no proper subgroup, and
this easily implies that G/M has prime order. CI
Note carefully that we do not assert that a nilpotent group
must have a maximal subgroup. In fact this is not the case even
for abelian groups. The infinite abelian group Z; of Example
1.09 has every proper subgroup finite cyclic, as the reader may
care to prove, and this ensures that no maximal subgroup
exists.

COROLLARY 9.07. If the m'lpotent group G has a Sglow w-sub-

group G”, for some set 71' ofprimes, then G7r is normal in G.
Proof. If G7, is a Sylow w-subgroup then N(G) = N(N(G"))
by Corollary 7.21. On applying Theorem 9.05 to the subgroup
N(Gfi) of G, we conclude that G = N(G7). That is, G, <1 G. I]
 Nilpotent groups 177

We have not asserted that any Sylow n—subgroup exists. But

we shall soon prove that G77 does exist, by a method that avoids
set-theoretic difficulties.
At this point it is possible to give much information about
finite nilpotent groups and, in particular, to characterize them
in terms of finite p—groups. The following theorem generalizes
some familiar facts about abelian groups.

THEOREM 9.08. The following conditions on the finite group G

are equivalent.
(i) G is nilpotent.
(ii) All maximal subgroups of G are normal.
(iii) Any Sylow p-subgroup G1, is normal (p is an arbitrary
prime).
(iv) Elements of coprime orders in G commute.
(v) G is the direct product of its Sylow p-subgroups.

Proof. (i) implies (ii) by Corollary 9.06.

Since G is finite, maximal subgroups exist; and so (ii) implies
(iii) by Corollary 7.30.
To prove that (iii) implies (iv), we first observe that elements
of orders p“, q3 (Where p, q are distinct primes) lie in normal
Sylow subgroups G1,, Ga respectively. Since their commutator
lies in G1, n G, = 1, they commute.
Next suppose that a, b are elements of orders p‘fl... p21“,
fling.) respectively, where the primes p1,..., p“, q1,...,_q,, are
distinct. A routine application of the Euclidean algorithm
shows that a = almau, b = bl...b1, where ai, bj have orders
pg“, (1% respectively (compare Theorem 5.01). Since each ai
commutes with each bj, as above, we find that a commutes with b.
Therefore (iii) implies (iv).
We deduce (v) from (iv) as follows. Take one Sylow p-sub-
group GP for each prime p dividing IGl. Then (iv) asserts that
G1, commutes with Ga element by element, provided p 75 q.
Suppose that [GI = p‘i‘l pg“, so that |GmI = pg“; it follows on
comparison of orders that G = G1,,1 G1,“ (note the use of
853137 N
178 Nilpotent groups

Theorem 3.20 here). It should now be clear that

G = G‘mx ...>< G1,”
by virtue of the direct product criterion of Theorem 4.38.
Therefore (iv) implies (v).
To finish off the proof, we show that (v) implies (i). This,
however, is an easy consequence of Theorems 9.04 and 9.03. [I
Some results below are best reached by means of special
central series. We therefore define inductively subgroups
MG) and ”(G) of the arbitrary group G. Let §0(G) = l, and
let (AG) be that subgroup of G for which
“GHQ—1(a) = ({G/Z¢_1(G)}
for each'l 2 1. Let y1(G) = G, and let ”(G) = [y¢_1(G), G] for
each i > 1. Note that §1(G) = C(G) and y2(G) = 3(G).

DEFINITION. If {8(G) = G for some integer s then G is said to

have the upper central series
1 = €o(0)<C1(G)< "<Es—1(G) S §,(G) = 0
DEFINITION. If yt+1(G) = 1 for some integer t then G is said to
have the lower central series
= r1(G)> y2(G)> >y.(G) > mm) = 1.
We ought to show that these are indeed central series as this
term was defined earlier. The normal property of each (AG)
should be clear, and our earlier definition ensures that the {1(G)
give a central series. We prove that ‘Vzl G) <1 G by induction on 75,
the case 13 = 1 being trivial. If a: e G, g e yi_1(G) then
[31,96]” = [31”, a3"] 6 [Vi—1(a): G]
for all g e G; it follows that gab-(GP g yi(G), which suffices. That
the ”(G) give a central series follows from Lemma 9.01.
It is not hard to show that MG) and ”(G) are characteristic
subgroups of G.
Example 9.09. Let G—
-- QSX 0 where
=,gp{a b: (12—
= b2: (ab)2},
0 = gp{c:c2 = 1}.
 Nrhmtent groups 179

It is easy to calculate that

72(0) = gPW}: MG) = 1,

{1(9) = gP{a226}: 5(0) = G,
in a lax but clear notation. Here is a case in which the upper
and lower central series exist and have the same length, but
possess differing terms.

THEOREM 9.10. If G has a central series

G=GO2GI>...>G,_1>G,=1
then GP,- g MG) and G’i > y¢+1(G) for 0 g ’5 g r.

Proof. When r3 = 0 both results are trivial. We use induction

oni,takingl <i<r.
In the former case we suppose inductively that

Gr—'t+1 g {rt—1(a).

By the third isomorphism theorem

Gila—1(0) E (G/Gr—i+1)/(€r'—1(GVGr—JHI)‘
There is therefore a homomorphism 95 from G'/G’,_H1 onto
G/§¢_1(G) with kernel §i_1(G)/G'r_i+1. Because G,_i/Gr.di+1 is
central in G'/G,_H1 we have (G'Ht/ G,_i+1)qS central in G/€i_1(G).
But

(GM-lGr—i+1)¢ % ((Gr—i/Gr—i+1)(€i—1(G)/Gr-i+1))/(§i—1(G)/Gr—i+1)
% «Gr—4 :i—1(G))/Gr—i+1)/(€i-1(G)/Gr—i+1)
% (Gr—r {1—1(G))/€1—1(G):
by the third isomorphism theorem. It follows that

(Gr—12 €H( G))/Ci-1(G)

is central in G/§i_1(G). The definition of the upper central series

gives arm-MG) < are).

This implies, since {14(0) g are), that GM g MG), which is
the result we require.
180 Nilpotent groups

For the second case we make the inductive assumption that

Gi‘l 2 311(G). Then

7i+1(G) = [71(G): 0'] < [Gr-1: G] g as:

the last inclusion following from Lemma 9.01. Thus we have
the desired result Gi 2 yi+1(G’). [I

COROLLARY 9.11. If a group is nilpotent then its upper and

lower central series have the same length, and this is the least
length for any central series.
Proof. The theorem clearly implies that the given series of
length r is at least as long as the upper and lower central series.
Since the given series was an arbitrary central series we may thus
compare the lengths of the upper and lower central series. They
are equal. [I
We turn to a closer study of the lower central series.

LEMMA 9.12. If x, y, z are any elements of an arbitrary group

then
(i) [ml/’2] = [23, z]”[y, 2];
(ii) [96,212] = [96, 2196, 2/]?
Proof. We recall that u” means v—luv. We have

(mg/)5 = 9023/”,
[m 2l = (my)‘1(wy)z = y—lx'lw’y” = (ac-196”)”(y‘1yz)
= [$3 2]"[a Z]:
and this gives (i). Because [u,v]‘1 = [tau] inversion of both
sides of (i) gives [2, xy] = [z, 3/12, 96]”,

which is (ii) with suitable relabelling. El

COROLLARY 9.13. In the same notation

(iii) [96—1, y] = [96, 21W";
(iv) [stay—1] = [x, y]—1"1.
 Nllpotent groups 181

Proof. Here u‘” means (7171)”. Put 3/ = x4 in (i) and obtain

1 = [93, z]“'1[w—1, 2]
Then (iii) follows, and the proof of (iv) is similar. [1

DEFINITION. The simple commutator [x1,...,xn] where n 2 2

and x1, . .., 90,, are elements of a fixed group G is defined inductively
by [971,952] = 931— 1x; 1371962 and by [ml’m’xn] = [[x1""’xn—1]’xn]
for all n > 2.

THEOREM 9.14. If G = gp{aA:A EA} then, for each i> 1,

31,-(G')[yi+1( G') is generated by the set of all elements [b1,..., bi]yi+1(G),
where each b]. is chosen arbitrarily from {a},:)\ e A}.
Proof. Note that the theorem would be valid for 'l = 1 if we
defined [b1,...,b,.] to be b1 when l: 1. We shall adopt this
convention and use the case '5 = 1 as the basis of an inductive
proof. (It is, of course, feasible to prove first the case i = 2
and then the inductive step, but they are essentially the same
proof and our device avoids such repetition.) Let 13 > 1 and
suppose that yi_1(G)/yi(G) has a generating set of the required
form.
The definition of 321(0) implies that
MG) = gp{[ya 96M e 7t—1(G): x e G}
for all 13 > 1. Therefore the elements [31, nob/“1(a) generate
”(Gm/“1(0). Now the arbitrary element 3/ of y,_1(G) has the
form cz where c = c1 0,, each cj being a commutator [61,..., baa—1]
or the inverse of such a commutator, and z E ”(0); this is the
inductive assumption. But
[cz, x] = [6, x]z[z,x]
= [c,x][c,x,z][z,x],
by formula (i) of Lemma 9.12, and the fact that a” = aft», 2)],
respectively. Because [0, x, z] and [2, a] lie in yi+1(G) we have
[02’ li+1(G) = [0: x]7i+1(G);
and it follows that the elements [c,x]y,-+1(G) generate
”(Gym-”(6%).
182 Nilpotent groups

Consider [cab/”1(a). A similar application of Lemma 9.12

gives [ext/MG) = [w] [win-Hm).
If it happens that it is 6,71 which has the form [b1,..., bi_1], then
we apply formula (iii) of Corollary 9.13:
[01:90] = [0?1: w]“”
= [6f1,Wl‘1[[6f1,%l‘laca-l-
(Again we also use the fact that u” = u[u, 12].) Now
[[6?1, ml“: 0;] 6 mm), and so
[02" xJYt+1( G) = [Oi—1: wl—17i+1(al-
We may conclude that the elements [b1, . .., bi_1, x]7t+1( G) generate
mam-M).
Here a: is still an arbitrary element of G. We put a: = x1 x8
where each xi equals some a,\ or ail, and expand the commu-
tator [b1,...,bi_1,x] using (ii) and (iv). The details follow, with
b0 = [b1""’b’i—1]:

[60, x1 $s]7t+1(G) = [bo’ x1] [bm xs]7i+1(G):

[1)!» “X Ill/“1(0) = [bo,afl‘1y¢+1(G).
Therefore yi(G)/yi+1(G) is generated by the elements
[bu-n, 51—1, “Ala/“1(0),
and this completes the inductive proof of the theorem. 1]

COROLLARY 9.15. If the group G is finitely generated then

y¢(G)/yi+1(G) is finitely generated for each 13 2 1. El
Next we apply the above theorem to nilpotent groups.

THEOREM 9. 16. Every subgroup of a finitely generated nilpotent

group is finitely generated.
Proof. Let H be a subgroup of the finitely generated group G
and let G have lower central series of length 8. Theorem 9.14:
shows that 7i(G)/Yi+1(a) is finitely generated for l g 1? g s. It is
easy to verify again that the series
H=H12H2>...>IL>H.+1=I,
 Nilpotent groups 183

where 19,; = H n ”(G) for 1 < i g 8—}— 1, is a central series

(perhaps not the lower central series) for H. Further, we have
Ht/Hrn = {H n 7¢(G)}/{H n Yi(G) n 7241(0)}
E 7¢+1(G){H n 7i(G)}/7’i+1(G)
by the second isomorphism theorem. Therefore Iii/Hi+1 is
isomorphic to a subgroup of yi(G)/yi+1(G). But any subgroup of
a finitely generated abelian group is finitely generated, according
to Theorem 5. 18. It follows that He/I1Q+1 is finitely generated for
1<i<a
It is easy to see that if N <1 G and both N and G/N are finitely
generated then G is finitely generated. An extension of this
remark, which can be proved rigorously by induction, shows that
each Iii/LEI”1 being finitely generated implies that H is finitely
generated. [I

THEOREM 9.17. A nilpotent group is finite if it is generated by

afinite number of elements each havingfinite order.
Proof. Let G be nilpotent and let G be generated by the subset
{a1,..., an} of elements, ai having finite order for 1 g 1; g n. The
fact that G has a lower central series makes it sufficient to prove
that Yi(G)/Y¢+1(G) is finite for each 13 2 1. This factor group is
abelian and, as Corollary 9.15 indicates, finitely generated.
Therefore it will suffice to produce for yi(G) /-yi+1(G) a generating
set each element of which has finite order.
If we refer to Theorem 9.14 we will find that the key step was
this fact: if
7i—1(G)/7i(G) = gp{b'1r,-(G),m, baa/1(9)}
(where '5 > 1) then
71(G)/'}’i+1(a) = gp{[bi’wkll’i+1(a)31 <3 < m, 1 < k < 77'}
If (1%" = 1 for some positive integer ark then it is certainly true
that [6}, (1%] e yi+1(G). The commutator [6}, (25%] may be
expanded by means of formula (ii) in Lemma 9.12, and (because
b} e yi_1(G) with the result that [b;-, wk] 6 yi(G) while
[14" “k, x] E Yi+1(G)) we find that [b;-,ak]°‘byi+1(G) = 'Yz'+1(G)-
184 Nilpotent groups

Therefore the element [bJ’-,ak]y,-+1(G) of yi(G)/'y,-+1(G) has order

dividing 0%, and certainly has finite order.
Since we now have a finite set of elements generating
yi(G)/yi+1(G), each element having finite order, the theorem
follows by the remarks above. [I
COROLLARY 9.18. In any nilpotent group G the elements of
finite orderform a normal subgroup N such that G/N is torsionfree.
Proof. Let x, y be elements offinite order in G. By the theorem
gp{:z:, g} is finite, and it follows that avg—1 has finite order. Since 1
has finite order, the subgroup criterion (Theorem 3.06) now
shows that the set of elements of finite order forms a subgroup,
N say. It is clear that N <1 G. If Na is an element of G/N with
finite order a then (Na)°‘ e G/N and so a“ e N. Since N is
periodic a“, and so a, have finite order. Therefore a e N, and
Na = N. This shows that G/N is torsionfree. I]
COROLLARY 9.19. In any nilpotent group elements of coprime
orders commute.
Proof. We are referring, of course, only to elements of finite
order. If a, b are elements of a nilpotent group with orders
a, )3 respectively then gp{a, b} is finite, by Theorem 9.17. Let
a, ,3 be coprime. An earlier theorem (9.08) about finite nilpotent
groups then shows that a and b commute. El
COROLLARY 9.20. For each prime p the nilpotent group G
contains one and only one Sglow p-subgroup, which is therefore
normal in G.
Proof. Given p and G, we consider the set S of all elements of
p-power order. A proof very like the one given in Corollary 9.18
(using Theorem 9.08 as well) shows that S is a subgroup. It is
then clear (from the definition of S) that there is precisely one
Sylow p-subgroup, namely 8. Corollary 7.17 implies that
S <1 G. I]

COROLLARY 9.21. The periodic subgroup of the nilpotent group

G is the directproduct ofthe Sglowp-subgroups of Gfor allprimes p.
 Nilpotent grooms 185

Proof. By the ‘periodic subgroup’ of G we mean that subgroup

N defined in Corollary 9.18. That 810 g N, where 8;, is the Sylow
p-subgroup described in Corollary 9.20, is a trivial fact. A simple
application of the direct product criterion (6.30), with the aid of
Corollary 9.19, gives the result—the details are no harder than
in an earlier case in which G was finite. [I
Example 9.22. We construct a finitely generated nilpotent
group G Whose periodic subgroup is not a direct factor of G. Let
Q8 = gp{a, b : a2 = b2 :2 (ab)2} be the quaternion group of order 8,
and let Z = gp{z} be infinite cyclic. Our example G is the sub—
group of s Z generated by {95, y} where x = (a, z), y = (b, 1).
It is easy to see that the periodic subgroup P is gp{y,[x,y]},
which is abelian, and that G/P is infinite cyclic. This implies that
either P is a direct factor and G is abelian, which is not the case,
or P is not a direct factor.
In Chapter 11 we shall construct an abelian group whose
periodic subgroup is not a direct summand (see Example 1 1.14).
The theory ofnilpotent groups developed here can be extended
greatly, but, of course, our account is restricted by the fact that
only a small elementary portion is accessible to us. One general
remark that we might make is that its conclusions do not apply
to all p-groups, for there exist infinite p-groups which are not
nilpotent. An example was indicated on page 175, and another
example that even has trivial centre will be found among the
problems at the end of this chapter. (See Problem 17.)
The less general theorems that follovsr may be found to have
some appeal on the grounds of interest, elegance, or importance.
The theory of nilpotent groups has an application to finite
p-groups and their sets of generating elements. Our context Will
now be finite groups (not necessarily p—groups), so that we may
avoid difficulties associated with infinite set theory.
DEFINITION. The Frattimi subgroup of the group G is the inter—
section of all the maximal subgroups of G. We adopt the notation
(I>(G) for this subgroup, and the convention that (I)(G) = G if G
has no maximal subgroup.
186 Nilpotent groups

THEOREM 9.23. The Fruttint subgroup of afim’te group consists

of all elements which may be omitted from any generating set (for
the group) in which they occur.

Proof. Let x be an element of the finite group G and suppose

that a: ¢(I>(G). Then :1: $1M where M is a certain maximal
subgroup of G. It is then clear that G = gp{M, :0} while
G > gp{M} = M. Thus {M, x} is a set of generators for G While
{M} is not. It follows that if the element x of G can be omitted
from every generating set in which it occurs then x e (I)(G).
Conversely, take any element 3/ in @(G), and suppose that
G = gp{0,g} where C' is some complex; thus {0, y} is a set of
generators for G. To derive a contradiction, let us suppose that
if S = gp{0} then S < G. Clearly we have 3/ ¢ 6’. We now
consider the set .5” of subgroups of G:
y={H:H<o,s<H,y¢H}.
Since G is finite .5” has maximal elements; in other words there is
at least one subgroup H0 in 5” not contained properly in any
element of y. We prove that such a subgroup H0 is maximal in
G. Suppose that H0 < H1 g G; by choice of [-10 we have y EH1,
and so 1511 2 gp{H0, y} = G. Therefore H0 is maximal in G, and
g ¢ H0. This contradicts the fact that y lies in (I)(G) and so in
all maximal subgroups of G. Therefore G = gp{0}, and y may
be omitted from any generating set {0, g} for G. I]

THEOREM 9.24. If G is a finite group then <I>(G) is nilpotent.

Proof. We see that (13(G) is normal, even characteristic, in G.

Apply Corollary 7.28 with K = (I)(G); thus if Kp is a Sylow
p-subgroup of K we have G = KN(K1,). But the previous
theorem shows that if G = gp{K, N(1%)} then G = gp{N(Kp)}.
That is, Kp <1 G. It certainly follows that each Sylowp—subgroup
of(I)(G) is normal in (l)(G). By Theorem 9.08, @(G) is nilpotent. [I

THEOREM 9.25. The finite group G is nilpotent if and only if

<I>(G) 2 8(G).
 NMmtent groups 187

Proof. Suppose G is finite and nilpotent. Maximal subgroups

of G exist (unless G is trivial); if M is one of them thenM <1 G by
Theorem 9.08. In fact it is clear that G/M is of prime order since
it has no proper subgroup. Thus 8(G) < M by Theorem 4.21,
and it follows that 8(G) < ©(G).
Suppose next that <I>(G) 2 8(G); we have to prove that G is
nilpotent. Let Gp be a Sylow p—subgroup of G. If N(G1,) yé G
then N(G1,) g M where M is a maximal subgroup of G. Thus
M 2 ©(G). The hypothesis that (D(G) 2 8(G) shows that
M 2 8(G); and it follows that M <1 G. But Corollary 7.29 shows
that any subgroup containing N(G1,) is self-normalizing, so M
cannot be normal in G. This contradiction forces us to accept
that N(G1,) = G, and so by Theorem 9.08 G is nilpotent. [I
We return to finite p-groups.
THEOREM 9.26. If G is a finite p-group then
(MG) = sway], 2pm, 31.2 e 67}-
Proof. Put N = gp{[x,g], zpzx, g, z e G}. From the proof of
the previous theorem it is easy to see that N lies in every maximal
subgroup of G, and so N < CD(G).
We have yet to show that if g e <I>(G) then g e N. Suppose this
is not the case for some particular g. Thus g e (I)(G) and g at N.
Then Ng 75 N. The group G/N is clearly elementary abelian and
if its order is p’ then it is the direct product of r groups of order 17.
We may choose r—l elements generating a subgroup of order
pr—l not containing Ng (this follows from our constructive proof
of Theorem 5.05 describing finite abelian groups). If these
elements are a1,..., ar_1 then gp{N, a1,..., a,_1} is a subgroup of G
having index 10 and not containing g. Therefore g ¢(I)(G). It
follows that <I>(G) = N, as required. I]
We observe that G/(I)(G) is elementary abelian, that G/(I>(G)
is trivial if and only if G is trivial, and that (D(G) = 1 if and only
if G is elementary abelian; these remarks apply in the case when
G is a finite p-group.
THEOREM 9.27. If G is a finite p—groap with G/(I)( G) of order p'
then any generating setfor G contains a generating set of r elements.
188 Nthrotent groups

Proof. Let G = gp{gl,...,gs}. Then

G/®(G) = gP{¢(G)91,---:¢(G)9s}-
Because G/<I>(G) is the direct product Of r groups of order p, we
can select h1,..., h, from {gl,..., gs} so that
G/CI>(G) = gp{Cl)(G)h1,...,®(G)h, .
Therefore G = gp{(D(G),h1,...,h,}. But the elements of (I)(G)
may be omitted from any generating set of G. We find that
G = gp{h1, ..., h,}, as desired. [1

COROLLARY 9.28. Let 0 be any complex of the finite p-gronp G

such that 0, but no proper subset of 0, generates G. Then the number
of elements in 0’ is an invariant of G. [I
We repeat our caution that results such as the preceding two
are peculiar to finite p-groups. Perhaps we could support it by
mentioning that Z; is a nilpotent group with no maximal
subgroup, so that Z; = @(f), and by asserting that if
G = gp{g:g6 = 1} then the generating sets {9} and {92,93} for
G contain different numbers of elements.
We turn now to a quite different topic, connected with the
fact that a nilpotent group generated by finitely many elements
each of finite order must itself be finite. The so-called Burnside
problem seeks information on a finitely generated group in
which each element has finite order, and in the case of p-groups
we might ask Whether a finitely generated p-group must be finite
and so nilpotent. It is beyond the scope of this work to describe
the counter-examples that provide a negative answer. Very
special conditions, however, do give positive results.

DEFINITION. A group G has exponent n if x" = 1 for each

element x of G.
Even a condition on the exponent of a p—group does not give
good general results. Groups of exponent 2 are trivially easy, as
we have seen (in remarks after Example 4.20), and groups of
exponents 3 and 4 will now be treated in an elementary fashion.
 Nilpotent groups 189

LEMMA 9.29. If x, y, z are arbitrary elements in any group then

(V) [90, 2’ 21$][y’ 90’ zyllz, a 96”] = 1-
Proof. We simply expand each factor in (v):
[3% z, 21”] = [90: fifty—”[96, Zly”
= (2496—1290) (ac—ly—lx) (x‘lz‘lxz) (x—lyx)
= (zyz—lxz)-1(xzx”1yw).
It should now be clear that if the two other expansions are
similarly carried out then the left-hand side of (v) will cancel
completely. I]
LEMMA 9.30. If a, b are elements of a group for which
a3 = (ab)3 = (ab—U3 = 1 then [a, b, b] = 1.
Proof. Since (ab)3 = 1 we have bab = a—lb'la—l. Substitution
of this in a—lbab gives
a-lbab = a‘zb—la—l.
But a3 = 1. We therefore have
a—Zb'la—1 = ab*1a2.
Finally, (ab-1)3 = 1 gives ab'la = Ina-1b. Therefore
ab—la2 = ba—lba.
When combined these three equations give a—lbab = ba—lba.
Thus b commutes with a—lba and so with [b, a] = b-la-lba and
with [b,a *1 = [a, b]. We conclude that [a, b,b] = 1. I]
COROLLARY 9.31. If x, y are arbitrary elements in a group of
exponent 3 then [x,y, y] = 1. I]
We now consider the class of groups for which [x, y, y] = l for
every pair at, y of elements. We note that this condition means
precisely that each element y commutes with every conjugate
y” of y and y'” of girl; or alternatively that the normal closure
of {y} is abelian.
THEOREM 9.32. Ifthe group G satisfies the condition [90, y, y] = 1
for every pair of its elements a, y then G is nilpotent of class 3;
8(0) is abelian; and y3(G') has exponent 3.
190 Nilpotent groups

Proof. Ifx, y, z are arbitrary elements of G’ then [x, yz, 3/2] = 1.

We expand the commutator by means of the relations (ii) and
(i) above (p. 180); thus by (ii)
(1) [06, 3/2, .112] = [96, 3/2, lx, W: W-
Application of (ii) and (i) gives
[96, 21%: Z] = [[90, 2196, gilt Z]
= [[96, yl": Z]
because [95, z, z] = 1. Since 2 commutes with the conjugates of z
and 2'1 we have
[[22, yllz’ 2] = [95: y: 2F = [‘3’ 3/, 2}
Hence
(2) [x,yz,z] = [x,y,z].
Similar treatment is given to [90, yz, y]:
[90, W, .71] = [[90, z][x, 31M]
=W%MWW%l
But [96,11]]2 can be expressed in terms of two conjugates of 3;,
namely as (yxzrlyz; so [[x,y]z,y] = 1. Similarly [x,z,y] and
[90, y]z are products of conjugates of y and 3/4, and so commute.
Thus
(3) [£16, 3/23 3/] = [27, z: 2/]-
Combination of (l), (2), (3) gives
(4) [96, 3/, zllw, 2,31] = 1-
But by (iii) we have
[2), 2/, z] = [[yax]_15 2]

= [ya 3’: z]—[y’w]—1:

and the argument about 3/ commuting with all its conjugates

and their inverses gives
(5) [x,y,z] = [y,x,z]-1
at this point. We deduce from (4) and (5) that
(6) [90, z, y] = [31, x, z],
 Nilpotent growps 191

and a. renaming of x, y, 2, which are arbitrary elements, gives

(6’) [y,x,z] = [z,y,x].
Next we apply (ii) to the terms of (v). Thus
[£17, z: 3/90] = [58, z: y[y: ad]

= [96: z, 3/]
since [90, z] commutes with [3], x] as the usual argument shows.
Hence and similarly (v) becomes
[23, 2: y][?/: x: 2:":2’ y, "19] = 1'
Now we use (6) and (6’):
(7) [50: z: y 3 = -
Take a further arbitrary element w. Then
[95, 3/, z: w] = [(12, y, w? z]_1 by (4)
= [w,x, y, z]-1 by (6)
= [g,z,[w,x]]'1 by (6)
= [[w, x], [2h zl]
= [2], z, w, w] by (6)

= [3% z: x: w]_1 by (4)

= [w,y,z,w "1 by (6).
We conclude that [06, y, z, w] has order dividing 2. But this
commutator has order dividing 3 by (7). Hence
(8) [x9yaznw] = 1'

The previous calculation therefore gives, further,

(9) [[11, 2], [10:96]] = 1-
Now (9) proves that 3(a) is abelian. Since commutators of the
form [1», y] generate y2(G) an easy application of (i) shows that
commutators of the form [x,y,z] generate 713(0). By (8), 'y3(G')
is a subgroup of {(G), and so 0 has class 3. It is now clear from
(7) that y3(G) has exponent 3. I]

COROLLARY 9.33. A group G of exponent 3138 nihootent of class

3, and 3(0) is abdiam. l]
192 Nilpotent groups

We observe that G need not have class 2, as Problem 14

of Chapter 11 indicates.
Our results on groups of exponent 4 are less precise.
THEOREM 9.34. A finitely generated group of exponent 4 is
finite.
Proof. Let G' = gp{a1,...,an} be such that x4 = 1 for all x in
G. If G,- = gp{a1,...,a¢-} for 1 g t g n and Hi = gp{0i, (134.1}
for 1 g i < n, then Gm = gp{I-Ii,a¢+1} and a?+1 6 151,. These
remarks show that an inductive proof of the theorem may easily
be constructed once we have this key fact: if A is a group of
exponent 4 with an element a and a finite subgroup B such that
A = gp{a, B} and a2 e B, then A is finite. We proceed to prove
this.
A typical element g of A may be written in the form
b0 able abr+1
where bi e B for 0 g i g r+ 1. It suffices to prove that, for any
element of A, such a form may be given with r bounded from
above, for then A will have only finitely many elements. We
therefore seek such a bound on r, and we suppose that the ex-
pression chosen for g has r minimal. Thus bf, 05a obj,+1 7e 9
for s < r, whatever elements 6; are chosen in B, and this implies
thatbi 7E 1for1 gtgrasazeB.
Let t be fixed. We have (abi)4 = 1, which implies
abia = bile—1bi‘1a'1bg‘1 = ciaciabfl
Where oi = b51072 e B. Thus substitution for abia (1 <1? g r)
may be made in g, and there results an expression bf, ab’1 a ab;+1
for g in which b§+1 = bflbifl. Note that 12;“ 75 bi“ because
b1; 75 1. By carrying out such substitutions repeatedly, a number
of expressions for g will be found. We shall describe inductively
a method for obtaining 7; distinct expressions for g, differing one
from another in the value assigned to 6,, for 1 g i S r.
When 1? = 1 there is nothing to prove (and we may ignore the
trivial case when r = 0). Let 1 g i < r and suppose that we
have i expressions for 9 corresponding to 2' distinct values of bi,
While bi+1,..., b, are as given in the original expression for g.
 Nilpotent groups 193

Make the recommended substitution for ab, (1, in an arbitrary

one of these expressions, so that the fixed element b,-+1 is replaced
by bg+1 = b; 1bfil. The inductive hypothesis ensures that the
resulting i values of b; H are all distinct. Further, each is distinct
from the fixed element bH1 because b, ;E 1, the integer r associa-
ted with 9 having the minimal property explained above. We
therefore have i+1 expressions for g which difier from one
another in the value of b,+1, and which agree in the values of
bi+2,..., 1),. This completes the inductive step.
The result is r expressions for g, each with its particular value
of b,. But each 6, lies in the finite group B. Hence 1' cannot
exceed the order of B, and this is the required upper bound on r. I]

COROLLARY 9.35. A finitely generated group of exponent 4 is

nilpotent.

Proof. By the theorem such a group is a finite 2-group, which

must of course be nilpotent. I]
Groups of exponent 3 could be treated by the method used for
the case of exponent 4, provided we are content with the finitely
generated case. In fact all groups of exponent 3 were nilpotent.
This cannot be proved when the exponent is 4, because it is not
in fact true.

Problems
1. Prove that the dihedral group (see Example 1.25) of order 2"+1 is
nilpotent of class 'n, but not of class 71,— 1, for each n > 1. Do the terms
of the upper and lower central series coincide?

2. Prove that every proper subgroup of the group Z§° (see Example 1.09)
is cyclic While Z1? itself is not cyclic. Deduce that Z}? has no maximal
subgroup.

3. The group G" is defined as gp{or,b:a“”‘+1 = 1, b?" = 1, bvlab = a1+1’}

for n > 1 and p an odd prime. Prove, by considering the set of pairs
{(a,l3):oz,B residues modulo 1W“, 1)” respectively}, or otherwise, that G"
has order 102"“ and is nilpotent of class n+ 1 but not of class n.
853137 0
194 Nihmtent groups

*4. Prove that if G is any Sylow 2-subgroup of the symmetric group

S8 then G has no set of fewer than 3 generators and is nilpotent of class 4
(but not 3), while all proper subgroups and proper factor groups of G
have class 3.
5. Prove that if G is nilpotent of class 3 then 8(G) is abelian.
6. Prove that the finite group G can be generated by n elements if and
only if G/<I>(G) can be generated by n elements.
7. A group in which every finitely generated subgroup is nilpotent is
called locally nilpotent. Prove the following properties of a locally
nilpotent group G:
(i) there is a unique Sylow p-subgroup of G for each prime 1);
(ii) any maximal subgroup of G is normal in G;
(iii) G > 8(G) unless G = 1.
8. Show that the set of all commutators of the form [x1,...,xn] in an
arbitrary group G generates 91,,(G).
9. Establish the true statements and discredit the false statements
among the following:
(i) G is nilpotent if and only if G/§(G) is nilpotent.
(ii) G is nilpotent if and only if 8( G) is nilpotent.
(iii) If G has order p4 (p is prime) then 8( G) is abelian.
(iv) If G is a finite p-group and 3(G) is cyclic then a suitable commu-
tator generates 8(G).
(v) If each group in {GA:/\ 6 A} is nilpotent of class 71. then {1" GA is
6A
nilpotent of class n.
10. Prove that N n {1(G) > 1 where N is a normal subgroup of the nil-
potent group G and N 72 1. Show that the normal abelian subgroup A
of G is properly contained in no normal abelian subgroup if and only if
it is properly contained in no abelian subgroup.
11. Show that if a nilpotent group G has G/8(G) cyclic then G is abelian.
Deduce that no nilpotent group can be generated by one class of con-
jugate elements unless it is cyclic.
12. Show that a finitely generated nilpotent group with finite centre is
finite.
13. Prove that the group G has exponent 3 if and only if G is generated
by elements of order 3 and satisfies the relation [90, y, y] = l for an arbit-
rary pair x, y of its elements.
14. The group G is such that G/y2(G) is finitely generated. Prove that
Yi(G)/7¢+1(G) is finitely generated for all n 2 1.
Deduce that if G/y2( G) is finite then so is each y.,;(G)/yH1(G); and that
the nilpotent group H is finite if and only if H/y2(H) is finite.
 Nilpotent groups 195

15. Use the identity (v) of Lemma 9.29 to show that ifA, B, G are normal
subgroups of the group G then
[[A, B], 0] < [[B, 0],A][[0.A], B]-
Deduce that if G has a central series with terms denoted (as in the text) by
G,- for 0 g t g r then [G13 G5] < Gi+j+1

where0<i<r,0<jgnandGHk:1forlc>0.

16. Use the ‘three-subgroup theorem’, enunciated in Problem 15, to

prove that [315(0), {5(G)] = 1 for 12 = 1,2,... in any group G.
**17. Let G be that subgroup of the symmetric group on the positive
integers generated by all the permutations p“, for 'o' 2 1 and j > 0, Where
i—l

Pia' 31:11(jpi+k,jp"+p¢“‘+km(j+1)p‘—pi‘1+k)
and p is a fixed prime. We define subgroups Gi and FL; of G as follows:
0.1-. = gP{Pu3j > 0}.
, H.- = 8P{G1:---: Gr}:
With G0 = Ho = l.
Prove the following statements:
(i) PEIGiPrj = 94,330” > i > 1 andj > 0;
(ii) G‘s! GandI-Q <1 Gfori 9 1;
(iii) G4 and Hi/H;_1 are abelian p-groups for 11 9 1;
(iv) every finitely generated subgroup of G is a finite p-group;
(v) no non-trivial element of Hi/ H134 is central in G/Hi_1 for 73 > 1;
(vi) G is not nilpotent.

***18. Prove that a group of order 1)” has a central series

G=Go>Gi>--->Gn—1>Gn=l
in which G4_1/Gi has order p for l g i g n. Deduce that if elements g;
are chosen so that gi e GL1, gi gt G1- for 1 g i g n then each element of G
has a unique expression 9;“ 9;?" where 0 g a; < p. Deduce again that
there are integers )3“, yiJ-k such that
91-” = gist-3+1 913‘",
[9239i] = a???“ 3“,." (7’ < .7):
and 0 < Bu < P, 0 < Vim < P-
Show that the multiplication table of G is completely determined by
theIS’M-with 1<i<j<nandtheyijkwith 1<i<j<k<nh
Establish that there are no more than p‘""")/“ groups of order p".
10

Soluble groups

IT is now time to examine the remarkable interplay between

normal structure and Sylow structure in the class of groups
called finite soluble groups. We must repeat that all series
considered in this book are finite.

DEFINITION. A group G is soluble if it has a normal series

0:00? 01? n. B 07—1? GT=1

in which 01:1]Gi is abelian for 1 g i g r. A group with such a

series of length r is said to be soluble of length (or rank) 1‘.
The soluble groups of length 1 are precisely the abelian groups,
while G is soluble of length 2 (or metabelian) if and only if 8(G)
is abelian (by Theorem 4.21). Comparison of definitions shows
that every nilpotent group is soluble. On the other hand, there
are metabelian groups such as S3 and A4 that are not nilpotent
of any class.
The concept of invariant series is specially suitable for treating
the properties of soluble groups. It is elementary, and could
have been introduced and discussed long ago, without the
context of soluble groups, if it had seemed worth while.

DEFINITION. An invariant series

0:002012...2G’,_1>G,=1
of a group G’ is a normal series in which Gt. <1 G for 1 g i g r.
Thus every central series is an invariant series; of course 83
and A4 have invariant series but not central series.
 Soluble groups 197

THEOREM 10.01. Any two invariant series for a given group

have isomorphic refinements.

Proof. Though we have not formally defined a refinement of

an invariant series and isomorphism of two such series, the
meanings of these terms should have been made evident by the
discussion of normal series in Chapter 6. Indeed the proof of
this theorem is very like that of Theorem 6.21 and will only be
sketched. Let the group G have the invariant series
G=GO>G1>...>G,_1>G,=1,
G=H0>H1>...>Iag_l>fl$=1.
Between each GL1 and Gz- in the first series we insert terms
GAGE;1 n H,-) where j = 1,...,8—1;’ between each Hj_1 and [ii
in the second we insert [fa-(61¢ fl Hj_1) Where i = 1,..., r—l. The
terms inserted are of course normal subgroups of G, and so the
resulting series are invariant. An application of Zassenhaus’s
lemma (Theorem 6.18) proves the isomorphism of the two
refinements. [I

DEFINITION. A chief series of a group G is an invariant series

G=G0>Gl>u->Gr—1>Gr=1
such that Gi_1 > 01- and if GL1 2 N 2 0%- with N Q G then
G'i_1 = N or N = 6171-, for 1 g i < r. The groups G¢_1/G‘i are
called the chief factors.
Thus a chief series is an invariant series that cannot be refined
in a non—trivial manner. Note that chief factors need not be
simple groups; a glance at a chief series for A4 will confirm this.

THEOREM 10.02. In a group with a chief series every chief series

is isomorphic to the given series.
Proof. This result is analogous to the Jordan—Holder theorem
(6.23). Take two non-isomorphic chief series and construct iso-
morphic refinements as in the preceding theorem, omitting
repeated members. Each chief series is isomorphic to its
198 Soluble groups

refinement (by the definition of chief series), so they are them-

selves isomorphic. |:|

COROLLARY 10.03. Any two chief series of a finite group are

isomorphic. I]
We note that the chief factors are thus group invariants. It is
false that, conversely, a group is determined by its chief factors
(even if they exist), because there are non-isomorphic groups of
order 4, for example, with isomorphic chief series.
We return to soluble groups. If the group G has a normal
genes G = 00 I; G112 9 G',_1 9 a, = 1
and if Gig/G). is abelian, then 8(G'i_1) g G) (by Theorem 4.21).
This suggests the following inductive definition: put 30(G') = G',
and let 34G) = 8(3i_1(G)) for each i 2 1. Thus 81(0) = 8(0).

DEFINITION. If 88(9) = 1 for some integer s then G is said to

have the derived series
0 = 80(0) 2 81(0) > > 3.40) 2 38(0) = 1.
THEOREM 10.04. The group G has an invariant series with
abelian factors if and only if G has a normal series with abelian
factors.

Proof. One assertion of the theorem is trivial because any

invariant series is also a normal series. So we concentrate on the
less trivial part, assuming that G has a normal series
G’: G012 G11; ".12 0,412 0,: 1.
We have already noted that 8(Gi_1) g Gt for 1 g i g r. If
8i_1( G) < Gi_1 for some particular i then
BAG) = 8(8i—1(G)) < Nat—1) < Gi-
But since 80(0) = G0 we have found an inductive proof that
81(0) g at for 0 g i g r. In particular SAG) = 1.
The series

G = 30(0) 2 31(0) 2 2 3749) > 3A0) = 1

 Soluble groups 199

is clearly normal with abelian factors. The proof will be complete

as soon as we show that it is an invariant series. We again use
induction on i to prove that 8,-(G) <1 G for 0 < i g r; in fact,
it is just as easy to show that each 3i(G) is a characteristic sub-
group of G. When i = 0 the assertion is trivial. Suppose then
that i 2 1 and that 8i_1(G) is characteristic. We know from
Theorem 4.33 that, in any group H, 8(H) is a characteristic
subgroup. Take H = 8i_1(G) and apply Theorem 4.34. The
conclusion is that 84G) = 3(8i_1(G)) is characteristic in G. It
certainly follows that 84G) $1 G for 0 g i < r. Thus the derived
series is invariant, and the theorem is proved. [1
Two statements established in the course of the above theorem
are worth making explicitly.
COROLLARY 10.05. If G has a normal series with abelianfactors
then BAG) g G; for 0 g i g r, in the notation above. [I
COROLLARY 10.06. The derived series is an invariant series. [I
The derived series is thus distinguished in much the same way
as the lower central series, in their respective contexts. Note that
While every nilpotent group is soluble, the converse is far from
true.
Three definitions (in terms of normal or invariant or derived
series) are now available for soluble groups. Pride of place was
given to the one based on normal series, and this is historically
justified because of the link with Galois theory (which we
cannot touch on here). The alternative definitions make it clear
that soluble groups have many normal subgroups, and together
they provide many possible ways of proving the routine but none
the less essential theorem which follows.
THEOREM 10.07. Let G be a group with a subgroup S and a
normal subgroup N. Then
(i) if G is soluble of length r then S is soluble of length r;
(ii) if G is soluble of length r then G/N is soluble of length r;
(iii) if G/N, N are soluble of lengths s, t respectively then G is
soluble of length s—l—t.
200 Soluble groups

Proof. (i) We use the fact that BAG) = 1. Since 8 g G we

(clearly) have 3,;(6') g SAG) for i = O, 1,..., and it follows that,
as required, 8,08) = 1.
(ii) We have a normal series
G=GOI> 0'1 12!; 0H? 0,: 1
for G' in which the factors are abelian. We also have the normal
senes G B N 9 1.

Theorem 6.21 enables us to construct isomorphic refinements.

The first refinement has abelian factor groups while the second
contains N as a term. Truncation of the second therefore gives
“ems G=H09H19 ".9114; H,=N
with abelian factors; note that the length has not increased
from 1‘ because of the method of construction of the refinements.
The existence of the series

GW=MWBEWBWBE4NKWN=N
shows that G/N is soluble of length r because the third iso-
morphism theorem gives
_ (Hi-l/NVUJi/N) % ‘Hi—l/Hi
for 1 g 1/ g 7'.
(iii) Take normal series with abelian factor groups for G/N
and N :
G/N = Go/N 9 Gl/N x; I; GH/N I; Gs/N = N,
N=MBMB"BM&M=L
Here Gym, G's_1 are certain subgroups of G containing as = N.
We have Gigi Gab;1 for 1 < i g 8 since Gi/Nsl Gig/N, and
01-1]G'i is abelian because
(Gt—1/N)/(G¢/N) E Gi—l/Gi
by the third isomorphism theorem. The normal series
0:672); G’1 1212 GHBNO 12M 9...2M_191\T,= 1
proves that G is soluble of length 8-H. [I
 Soluble groups 201

COROLLARY 10.08. The direct product of any finite set of soluble

groups is soluble.
Proof. It is sufficient to show that if G = G1 >< G2 with GI, G2
soluble then G' is soluble, for only an elementary induction is
lacking. But G76?1 g G2 and part (iii) of the theorem at once
shows that G is soluble. I]
The direct or cartesian product of an arbitrary set of soluble
groups need not be soluble. The existence of soluble groups of
length 12 but not n—l for each positive integer n would sufl‘ice
to prove this, but we shall not make the necessary digression.
An essential difference between soluble and nilpotent groups
has appeared in part (iii) of the preceding theorem (note that
83 is not nilpotent but has a normal subgroup N such that both
S3/N and N are nilpotent). Indeed (iii) makes it often convenient
to prove theorems giving sufficient conditions for finite groups
to be soluble by induction on the group order. Let us suppose
that y is a property which may be possessed by finite groups
and which is possessed by every subgroup and every factor
group of G if it is possessed by G. For instancej' ? might be
(a) having odd order, or (b) having a w-number as order where
71' = {10, q}. Suppose further that we want to prove that a finite
group with Q is soluble. Then the required inductive proof can
be constructed once the following fact is proved: the only finite
simple groups with 9” are abelian.
We note in the interests of completeness that a finite group
need not be soluble if all its proper subgroups are soluble. Take a
finite non-abelian simple group of least order (this is the alternat-
ing group A5 according to Problem 12, Chapter 7). Every proper
subgroup is non-simple or abelian, and therefore soluble by an
argument like that sketched in the last paragraph; but the
original group is certainly insoluble.
The infinite cyclic group is soluble but lacks a composition
series or (what is the same thing in this case) a chief series.
Therefore it is reasonable to confine our attention to finite
1' That either (a) or (b) really does imply solubility is a fact beyond the
scope of this book to establish.
202 Soluble groups

soluble groups when discussing composition and chief series.

Since the only simple abelian groups are those of prime order,
the composition factors of a finite soluble group have prime
order. The next result concerns the chief factors.
THEOREM 10.09. Any chieffactor of a finite soluble group is an
elementary abelian p-group, for some prime 12.
Proof. Let the finite soluble group 0 have the chief series
G=Go> 01> ...> G,_1>G,= 1.
Each factor group Gi_1/Gi is a finite abelian group. If the Sylow
p-subgroup S/Gi were a proper subgroup of 6124/91. for some
prime p, then S/G’i would be a characteristic subgroup of
Gi-l/Gi and so a normal subgroup of G/Gi, by Theorem 4.34.
Thus the given chief series could be properly refined, which is
a contradiction. So Gi_1/ G1. is a finite abelian p-group for some
prime p. Next we consider the subgroup T/ Gt- Of Gi—l/Gz'
generated by all the 10th powers in Gi_1/Gi. Again, T/Gi is
characteristic in Gi_1/G, and so normal in G/Gi. The chief series
property shows that T = 0,, for if T = 691,»;1 then GL1 = G,-
by the theory of abelian groups. Thus Gig/G4 has exponent p.
This is the assertion of the theorem. D
COROLLARY 10.10. A minimal normal subgroup of a finite
soluble group is an elementary abelian p-group, for some prime p.
Proof. By a minimal normal subgroup we mean, of course,
a normal subgroup that contains properly no normal subgroup
except 1. Such a normal subgroup N is a term of some chief
series, Which may be obtained by refining the invariant series
G I; N I; 1, according to Theorem 10.01. Thus N/ 1 is a chief
factor and Theorem 10.09 describes its structure. I]
The last few results have completed our preparations for the
SlW—like theorems that are valid in finite soluble groups (and
which in fact characterize them, though we shall not prove this).
Our method is induction on the group order, and we shall, of
course, use in an essential way the fact that there exist proper
normal subgroups.
 Soluble groups 203

DEFINITION. A Hall w-subgroup of a finite group G is a 77-

subgroup with coprime order and index. (7r will denote a non-
empty set of primes throughout this chapter.)
Thus a Hall n—subgroup is certainly a Sylow w-subgroup.
We saw in Example 7.07, however, that the insoluble group A5
has Sylow n-subgroups but not Hall w-subgroups, if 71 = {2, 5}.
The following theorem is of the greatest importance.
THEOREM 10.11. Let G be a finite soluble group and 77 a set of
primes. Then
(i) G contains Hall w-subgroups;
(ii) any two Hall w-subgroups of G are conjugate;
(iii) every w-subgroup of G lies in a suitable Hall n-subgroup.
Proof. An induction on [G] is started by the remark that the
theorem is true, even trivial, when |G| is a ar-number. Suppose
then that |G| = mn where m is the largest a-r-number dividing
[G1 and n > 1. In the general step of the induction we distinguish
two cases.
The hypotheses imply that G has proper normal subgroups
(compare Corollary 10.10). In the first case we suppose that K
is such a subgroup of order m1 n1 with n1 < n; m1 and n1 are
to be such that they divide m, n respectively. The inductive
hypothesis indicates that G/K has a subgroup, S/K say, of
order m/ml. Then S has order mnl. On applying the inductive
hypothesis to S, we find a subgroup M of order m. This is the
required subgroup of G, and (i) is proved.
To establish (ii), let M1 and M2 be a pair of subgroups of order
m. Since ALE/K g JIL/(K n M) for i = 1, 2 by the second iso-
morphism theorem (6.12), and since |K n Mi| divides m1, we see
that MI, K/K I = (m/m1)m2, where m2 divides m1. On the other
hand MI, K/K [ divides |G/K l = mn/(m1 n1), by Lagrange’s
theorem (3.15), and so m2 divides n/nl. Therefore we have m2 = 1,
MI,- K[KI = m/ml, and M,- K/K is a Hall w—subgroup of G/K. It
follows from the inductive hypothesis that M1 K/K and M2 K/K
are conjugate in G/K. Thus, for some a in G,
MIK/K = (Kw)‘1(M2K/K)(Kw),
204 Soluble groups

from which it follows that

1111K = x‘1(M2K)x, and x‘lx g MIK.
Now 1111 and 90—l x are Hall w-subgroups of G and therefore of
M1 K, and a further appeal to the inductive hypothesis shows
that they are conjugate in MIK for
[MlKI = [MlK/K] [K] = (m/m1)(m1n1)< mn = [G I.
Hence M1 and M2 are conjugate in G.
We turn to (iii). Let P be an arbitrary w-subgroup of G. Then
PK/K is a n-subgroup of G/K. The inductive hypothesis shows
that PK/K g S/K, the latter being a Hall qr-subgroup of G/K.
Therefore P < S. Again, induction shows that P lies in some
Hall w-subgroup M of S and therefore of G.
The first case is now complete, and we have to consider the
alternative, in which every proper normal subgroup K of G has
order m1 n for some m1; that is, n1 = n for all K. Now G certainly
has minimal normal subgroups, and by Corollary 10.10 these
are p—groups for some prime 10. We conclude that n = p“
where p is a fixed prime and a > 0; and that all the minimal
normal subgroups have this order. Such a subgroup K is a
Sylow p-subgroup of G because m and n are coprime. It follows
(from Corollary 7.17) that K is the unique minimal normal
subgroup of G. The Schur—Zassenhaus theorem (7 .34) shows
that there is a subgroup of order m, at this point. We shall,
however, prove this in another way, for this will make the
present proof more coherent.
By Corollary 10.10 G/K has a minimal normal subgroup L/K,
say. Its order will be QB where q is a different prime from p,
and [3 > 0. Now L will have a subgroup Q of order qB, by Sylow’s
first theorem, and clearly L = KQ. By Corollary 7.28
G = LN(Q) where N(Q) is the normalizer of Q in G. Therefore
G = LN(Q) = KQMQ) = KN(Q)-
Our aim now is to show that K n N(Q) = 1, for it will then
follow that [N(Q)[ = |G/K| = m, and N(Q) will be the desired
subgroup of order m.
Put D = K n N(Q). We have (a) D <1 K because K is abelian
 Soluble groups 205

(by Corollary 10.10); and (b) D Q N(Q) because K Q G implies

K n N(Q) Q N(Q) (by Corollary 6.02). It follows that
D Q KN(Q) = G. But D Q K, a minimal normal subgroup
of G. Therefore D = 1 or D = K. If we had D = K then
K Q N(Q), and so G = KN(Q) = N(Q); this would imply that
Q Q G, but K was the unique minimal normal subgroup of G
and K Q Q because 10 # q. We therefore conclude that D = 1.
As remarked above, this shows that N(Q) has order m.
Next we show that all subgroups of order m are conjugate in
G. If M is any such subgroup then G = LM since |LM | is
divisible by both [L[ = pug/3 and [M | = m. The second iso-
morphism theorem shows that G/L and M/(M n L) have equal
orders. Since {GI = mp“ we find that IM n L] = qfi.
Now any subgroup of L which has order we is conjugate to Q,
by Sylow’s second theorem. Because lN(Q)| = m, and because
N(Q95) = N(Q)”, its normalizer has order m and is conjugate to
N(Q). Therefore, if M is any subgroup of order m, M n L is
conjugate to Q, M = N(M n L), and M is conjugate to N(Q).
It follows that all such subgroups M are conjugate.
Finally we prove that any w-subgroup S of G lies in a Hall
w-subgroup. Let the order of S be m’ where m’ < m. It is easy to
see that [SK] = m’p“ < [G|. BecauseM and SK have coprime
indices in G, it follows from Theorem 3.20 (ii) that [M n SK [ is the
greatest common divisor of |M | and |SK I, which is m’. Thus
S and M n SK are subgroups of SK with equal orders. The
inductive hypothesis shows that they are conjugate in SK.
Therefore S lies in a suitable conjugate of M, a Hall w—subgroup. [J
This basic result on the structure of soluble groups makes
possible a considerable theory, of which we give only a specimen.
DEFINITION. A set V of Sylow p-subgroups of the finite group
G (one for each prime p dividing [G l) is a Sylow basis if PQ = QP
for all P, Q E 5”. '
DEFINITION. Two Sylow bases V, .7" of the finite group G
are conjugate if there is an element x in G for which P e .7’
implies P70 E 3'.
206 Soluble groups

DEFINITION. The normalizer of a Sylow basis 9 is the set of

elements of G normalizing each element of 57.
THEOREM 10.12. If G is a finite soluble group, then
(i) G has at least one Sylow basis;
(ii) any two Sylow bases are conjugate;
(iii) the normalizer of a, Sylow basis is ntlpotent.
Proof. (i) Let 71 = {p1,..., 10,} be the set of primes dividing [G I,
and let 711 be 1r less pi for 1 g t g r. Theorem 10.11 ensures that
G contains a Hall wi-subgroup Hi for each i. We put I} = 9m-
7: .7
Since the indices in G of the H1. are coprime, an obvious
extension of Theorem 3.20 (ii) shows that E is a Sylow Pf
subgroup of G.
To prove that Pi = PkPj for j 75 Is, it is sufiicient to show
that I}P,c g G, by Theorem 6.08. It is clear that P] and Pk are
contained in n Hi if the intersection is taken over all 1} except
j, k. Therefore Pi is contained in this intersection. The usual
argument about orders gives equality, so that P, Pk is a subgroup
of G.
(ii) Take two Sylow bases 9 = {E},.7 = {nor G. Construct
Hall wi—subgroups Hi, Ki from 5”, .7 respectively, by defining
Hi = II I; and Ki = LI Qj. Let .5"6 = {Pf} be a conjugate of
net i
.7, with the greatest possible number of Hg” in common with {K1-}.
If 5’90 ¢ 9', then some H1? does not coincide with the correspond-
ing Ki. But it is clear that Hg” and Ki are Hall art—subgroups of G
and are therefore conjugate in G by Theorem 10.11: H?” = Ki,
say. Since G = H? Pf, we may even assume that y e Pf.
But then H?” = H? since P? g H? ifj 9’5 t. We now have {H197}
containing one more element in common with 9’ than 5?",
contradiction to the choice of as. We therefore have 5”“? r: .7,
as required.
(iii) Let N be the normalizer of the Sylow basis 5” of G.
Clearly N is a subgroup of G. Let E be the pi—subgroup in 5’,
and let g EN have pi-power order. Since g eN(Pi) we have
E Pi, by Corollary 7.19. Therefore N n Pi is the Sylow pr
 Soluble groups 207

subgroup of N. The fact that the Sylow pi-subgroup of N is

unique shows thatN is nilpotent (by Theorem 9.08 and Corollary
7.17). D
We pass on to what is clearly a special class of soluble groups.
DEFINITION. A group is supersoluble if it has an invariant
series with cyclic factors.
Not every soluble group is supersoluble; this is established by
the example A4. Neither is every nilpotent group supersoluble;
consider Zz‘f. But finite p-groups and finite nilpotent groups are
supersoluble :
THEOREM 10.13. A finitely generated nilpotent group is super-
soluble.
Proof. If G is nilpotent then G has a (finite) central series.
The factors are finitely generated because all subgroups of G are
finitely generated (Theorem 9.16). These factors are therefore
direct products of cyclic groups, by Theorem 5.14. Because of
the central property of the series for G, it can be refined to a
central series with cyclic factors without losing finiteness of
length. This is an invariant series, so G is supersoluble. D
We note that supersoluble groups need not be nilpotent, for
83 is not nilpotent. For finite groups, supersolubility occupies an
interesting intermediate position between nilpotence and
solubility.
Next we shall consider cyclic factors of invariant series in
detail. A lemma is required.
LEMMA 10.14. The automorphism group of a cyclic group is
abelian.
Proof. Let a, ,3 be automorphisms of the cyclic group G with
generator g. Then go; = g”, gfi = g” for some integers u, 1;; it
follows that get/3 = 9“” = gBa. Thus {ca/3 = xfia for all x in G,
so 043 = Boa. Therefore the full automorphism group of G is
abelian. D
THEOREM 10.15. Let the group G have the invariant series
G>8(G)=K0>KI> ...>KH>K,= 1
208 Soluble groups

where Kill/K5 is cyclic for 1 g i g r. Then

5(a) =K0 >K1 > >K,_1 >K,= 1
is a central series for 8(G’).
Proof. We have to show that Ki_1/K; lies in the centre of
8(G)/Ki for 1 g i < r. Now Ki_1/K,- is cyclic, and the auto—
morphisms arising from conjugation by elements xKi, yKi of
G/Ki commute, by the lemma. Therefore [x,y]Kz- induces the
trivial automorphism on Ki—l/Ifi' Therefore 3(G)/K, and Ki_1/Ki
commute element by element, and this is the desired result. [I
COROLLARY 10.16. If the group H has 8,.(H)/8,-+1(H) cyclic but
not trivialfor some fixed i 2 1 , then there is no normal subgroup N
of H such that N < 8i+1(H) and 8¢+1(H)/N is cyclic.
Proof. Suppose that there is in fact such a subgroup N, and
apply the theorem with G = 3¢_1(H)/N. We find that 3i+1(H)/N
is central in 8i(H)/N, since
{8i(H)/N}/{8i+1(H)/N} g 5,-(H)/3¢+1(H)
(by the third isomorphism theorem), and this group is cyclic by
hypothesis. By Theorem 4.27, 3,-(H)/N is abelian. By Theorem
4.21, N 2 8i+1(H). We therefore find that N = 8i+1(H), a
contradiction. There is, therefore, no such subgroup as N. I]
COROLLARY 10.17. If 8i(H)/8i+1(H) and 8i+1(H)/8i+2(H) are
both cyclic for some group H and for some i 2 1, then
3i+1(H) = 8i'+2(H)- D
This has a number of interesting implications; for example,
8(0) g 83 is impossible for any group G.
COROLLARY 10.18. If G is supersoluble then 8(G’) is nilpotent.
Proof. Let
G=GO>G1>...>G,_1>G',=1
with each G,_1/G,. cyclic be an invariant series for G. Put
K; = 3(a) n am for 0 g i < r; thus K0 = 8(0) since Clo/Cl1 is
abelian and so 8(0) g G1 by Theorem 4.21. NOW
3(G) 2K0 2K1 9 219—2 >Kr—1 = 1
 Soluble groups 209

is an invariant series for (8 G) because Gi+1 <1 G and so K,- sl 8(G);

and IQ_1/IQ is cyclic for 1 g i < r because
Ki—l/Ki = (8(G) n Gi)/(3(G) n Gi+1>
E (8(G) n Gi)Gi+1/Gi+1
(by the second isomorphism theorem) and this is a subgroup of
the cyclic group Gi/GHI. Omit repetitions in this invariant
series for 3(G) and apply Theorem 10.15, concluding that 8(G)
is nilpotent. [I
Some alteration of invariant series of a supersoluble group is
possible.

THEOREM 10.19. A supersolnble group has an invariant series

in which every factor is cyclic of infinite or prime order.

Proof. Let G have the invariant series

G=G0>G1>...>G,_1>Gr=1
in which Gi_1/Gi is cyclic for 1 g i g r. If a factor is infinite, we
disregard it; if it is trivial, we omit it. Otherwise we examine its
subgroups with a view to refining the invariant series which we
started with. If |G,L-_1/G¢| = p1...pt Where p1,..., p, are (not
necessarily distinct) primes, then Gi_1/Gi has precisely one sub-
group Hj_1/Gi of order pimp) for 1 g j g t; remember that
Gi_1/ G4; is cyclic. Such a subgroup 119-1/ Gi must be characteristic
in Gi_1/Gi, and therefore normal in G/Gi, by Theorem 4.34.
Therefore H];1 <1 G and Hj_1/Hj has prime order for 1 g j g t
(with H, = Gi). It follows that
a:00>...>Gi_1>H1>...>Is_1>ai>...>ar=1
is an invariant series for G with H];1/Hy of prime order. Sufficient
repetition of this process will produce an invariant series of the
required kind for G. I]

COROLLARY 10.20. A chief series for a finite supersoluble group

is also a composition series.

Proof. This is now obvious from the definitions. [I

853137 1’
210 Soluble groups

LEMMA 10.21. Let G be a group of order pg where the primes

p, g are such that p > g. Then G has a characteristic subgroup
of order p.

Proof. We use the Sylow theorems. A subgroup P of order p

has 1 +kp conjugates for some 16; as usual 1 +kp divides pg, and
since p > q we see that l+kp divides 1, that is h = 0. There is,
therefore, only one subgroup of order p, and it is characteristic
by Corollary 7.18. D
(Note that a similar argument applies to the Sylow q-subgroup,
showing that G is abelian unless p E 1 modulo q.)

THEOREM 10.22. Let the group G have an invariant series

a:002G12'">Gt—l>at>a‘+l>u->Gr-1>Gr=l
7;

in which Gi_1/G¢, G,./G,.+1 have prime orders p, g respectively. Then

G has an invariant series
G=G0>G12m>G,_1>H>G,+1>...>G,_1>G,=1
in which Gi_1/H, H/ Gt+1 have orders q, p respectively provided
P > 9'

Proof. The group G,-_1/Gi+1 has order pq and therefore con-

tains a characteristic subgroup H/ G,-“ of order p, by the lemma.
The second invariant series is now constructed in the obvious
fashion. I]

COROLLARY 10.23. Let G be a finite supersoluble group and let

|G| = pi“... p31" where p1 > 102 > > pu, these pi being prime.
Then G has normal subgroups of orders pi“... pg“ for 1 g i g u.

Proof. Let
G=GO>G1>...>G,_1>G,=1
be an invariant series for G. By Theorem 10.19 we may suppose
that each Gi_1/ G1. has prime order. By repeated application of
Theorem 10.22 we may assume that the last a1 factors of the
 Soluble groups 21 1

series have order 121. This proves that G has a normal (even
characteristic) subgroup of order pi“. An obvious induction
(on u) would complete a rigorous proof ofthe theorem, but details
of this are naturally omitted. [3

Problems
1. Prove that two finite abelian groups have the same order if and only if
their chief series are isomorphic.
2. Let G be a finite group. Prove that the orders of the chief factors of
a subgroup divide the orders of the chief factors of G.
Prove also that the orders of the chief factors of a factor group are a
subset of the orders of the chief factors of G.

3. Prove that a finite group is soluble if and only if every factor group
other than 1 has a normal abelian subgroup other than 1.

4. Prove that 8(GIX G2) ; 8(G1) x3(G2). Deduce that if G1 and G2 are
soluble of length r then so is G1 >< G2.

5. Let G be a group of order 19q where p, q are primes and p > q. Prove
that G has a normal Sylow p-subgroup if p > 3. Is this the case when
p = 3? Show that G is always soluble.

*6. The group G has the form AB for certain abelian subgroups A, B.
Prove (by considering commutators or otherwise) that G is metabelian.
Show also that if G 72 1, G ¢ A, G 7’: B then one of A, B contains a
proper normal subgroup of G; and that one of A, B is contained in a
proper normal subgroup of G.

7. Find all the Hall w—subgroups of A 5. Does there exist a Sylow basis ?
Prove that all the subgroups of order 6 are conjugate, and investigate
conjugacy in the set of subgroups of order 10.

8. Show that the symmetric group S4 is soluble, and find all its Sylow
bases.

9. Prove that the subgroup G of S7 generated by {(1234567), (243756)}

has order 42 and is soluble.
Find a Sylow 2-subgroup and a Sylow 3-subgroup that do not lie in
the same Sylow basis; and find two Sylow bases, each containing one of
these subgroups.

10. Prove that a finite soluble group is nilpotent if and only if its Hall
Tr-subgroups of prime-power index are all normal.
21 2 Soluble groups

1 1. Let h be the number of Hall w-subgroups of a finite soluble group G.

Prove that h is the product of a set of numbers with both the following
properties:
(i) each is congruent to 1 modulo some prime in 1r;
(ii) each divides the order of some chief factor of G.

12. Is it true that a group with a composition series also has a chief
series? (As usual, all series are finite.)

13. Prove the following statements about the supersoluble groups G

and H:
(i) Every subgroup of G is supersoluble.
(ii) Every factor group of G is supersoluble.
(iii) G XH is supersoluble.
(iv) G is finitely generated.
(v) If G ; K/N, where N is a cyclic normal subgroup of K, then K is
supersoluble.
14. The group G has order 2})" where p is an odd prime. Prove that G is
soluble, and that
(i) a minimal normal subgroup M of p-power order lies in the centre
of the Sylow p-subgroup P;
(ii) ifM = P then M is cyclic;
(iii) G is supersoluble.
*15. The group G is the subgroup of the symmetric group 88 generated
by {(12)(35)(47)(68), (2345687), (346)(578)}. Prove that the normalizers
of the Sylow systems have order 3 and are contained in cyclic subgroups
of order 6. Is G supersoluble?
16. Prove the true and disprove the false statements among the following.
(i) gp{a,b:a-1ba = b2} is soluble.
(ii) If G = AB Where A, B are soluble subgroups then G is soluble.
(iii) If all the Sylow subgroups of the finite group G are abelian then G
is soluble.
(iv) A soluble group is metabelian if and only if it has a chief series.
(v) If all proper subgroups of G are supersoluble then G is supersoluble.
17. The soluble group G has a normal series in which every factor is
finitely generated abelian. Prove that G has a normal series in which
every factor is either the direct product of a finite number of cyclic
groups of prime order or a similar product of cyclic groups of infinite
order. Show that the number of infinite factors in such a series is a
group invariant.
*18. The finite group G has a normal Hall w-subgroup K, and so by the
Schur—Zassenhaus theorem G has a subgroup of order IG/K I. Prove that
if either K or G/K is soluble then any two such subgroups are conjugate.
11

Survey of examples

IN this final chapter we return to the subject of Chapter 1:

examples of groups. We shall not only discuss our earlier
examples in the light of the theory we have developed, but also
produce further examples that have been promised or that are
relevant to various facts in this book. We should perhaps warn
the reader that the examples of Chapter 1 are well-known groups
and have reasonable properties; further experience may perhaps
lead him to the conclusion that most groups tend to have un-
expected and even discouraging aspects, and indeed some in-
transigent counter-examples have already appeared among the
problems.
Perhaps our first duty is to find all groups of small order.
There are no surprises, for we have become familiar with most of
them in one context or another.
Groups of order p or p2 are abelian (see Corollary 4.29), and
abelian groups are treated as in Theorem 5.14. There is one
isomorphism class of order p, and two of order 192. The possible
types in the latter case are {2} and {1, 1}; that is to say, the group
is cyclic or the direct product of two groups of order 1). These
remarks are all that is necessary for the groups of order less than
16, except those of orders 6, 8, 10, 12, 14, 15.
We proceed to groups of order 210, where p is an odd prime;
this will deal with 6, 10, and 14 as one case. As for abelian groups
of order 21), Theorem 5.01 shows that such a group is the direct
product of a group of order 2 and a group of order p, and this
is of course cyclic. Suppose then that G— is a non—abelian group
of order 21). By Sylow’s first theorem G’ contains elements a, b
214 Survey of examples

of orders 2, 1) respectively. By Sylow’s other theorems gp{b} has

only one conjugate in G. Therefore b“ = b“, and since a2 = 1
we have b“2 = b. It follows that or E —1 modulo 1). We have
the following relations in G:
a2 = 1, b1" = 1, b“ = ()4.
Conversely, these relations define a group of order 210, namely
the dihedral group of Example 1.25. There are, therefore, two
isomorphism classes of groups of order 210.
Theory shows that there are three distinct abelian groups of
order 8; the types are {3}, {2, 1}, {1, 1, 1}. A non—abelian group G of
order 8 must contain an element of order 4, for otherwise every
element would have order 2 or 1 which implies that G’ is abelian
(see remarks following Example 4.20). This element b of order 4
generates a subgroup of index 2 in G, which is normal (see
Example 4.01). If a ¢ gp{b} then b“ = b“. Since b has order 4,
a = i1. Further, on2 e gp{b}, and of course a, has order 4 or 2
(if a has order 8 then G’ is cyclic); so a2 = 1 or a2 = b”. Therefore
two possible sets of relations emerge:
b4 = 1, a2 = 1, b“ = (2‘1;
64 = 1, a2 = b2, b“ = b—l.
These correspond to non-isomorphic groups, namely the di-
hedral and quaternion groups of order 8 (see Examples 1.25 and
1.30). Non-isomorphism follows from the fact that the former
has three elements of order 2 and the latter only one.
Next we discuss groups of order 12. Here each Sylow 3-sub-
group has order 3, while each Sylow 2-subgr0up may be either
of the two groups of order 4. In the abelian case, therefore,
there are just two possibilities: Z3>< Z4 or Z3>< ZZX Z2, where
Zn denotes the cyclic group of order 12. Note that the former of
these groups is simply Z12. In the non—abelian case, Example
7.10 shows that a group G’ of order 12 still has one normal
Sylow subgroup. If this is the Sylow 2—subgroup the reader will
be able to verify that G = A4 is the only solution to our problem,
because the cyclic group of order 4 has no automorphism of
order 3. Suppose finally that the Sylow 3-subgroup of G is
 Survey of examples 215

normal. There are then two possibilities for 0’, depending on the
structure of its Sylow 2-subgroup, and the following presenta-
tions for them can be found:
gp{a,b:a3 = 1, b4 =1, ab = a-l},
gp{a,b,cza3 =1, b2 = c2 = [b,c] = 1, ab = a4, a.” = a}.
We leave the details to the reader, noting that the latter group
is just 8'3 X Z2.
Finally we glance at the groups of order 15. Lemma 10.21
and a remark following show that such a group must be abelian,
and we soon conclude that it is in fact cyclic.
We now reconsider the examples of Chapter 1, starting with
abelian groups. Examples 1.01, 1.02, 1.03, 1.09, 1.10, whichwere
Z, Q, R, 2,3,", Zn respectively, are of fundamental importance.
Their structure can hardly be given in simpler terms, rather
they are themselves used to elucidate the structure of more
complicated groups.
Example 1.04 was the additive group 0 of complex numbers.
This is not of such great significance, because it is now obvious
that it is isomorphic to R ® R (in additive notation).
The multiplicative group (2* of positive rationals, in Example
1.05, can also be decomposed. It is generated by the set
{121, p2,..., pn,...} Where p" is the nth prime, because every
positive rational (except perhaps 1) can be written in the form
palm? 13%; Where the a,- are non—zero integers. In fact Theorem
6.30 shows that 0* is the direct product of the subgroups genera—
ted by 121, p2,..., pm... respectively. The group of non-zero
rationals is the direct product of Q* and the group generated
by —-1 (which has order 2).
The multiplicative group of positive reals R* (Example 1.06)
is isomorphic to the additive group R; this follows from well-
known properties of the logarithmic function, for we may define
an isomorphism (i) from R* onto R by 96¢ = lnx (logarithm to
any positive base except 1). The multiplicative group of non-
zero reals is the direct product of R* and the group of order 2.
Next we examine 0*, the multiplicative group of non-zero
216 Survey of examples

complex numbers, in Example 1.07. Of course, such a number 2

can be expressed in the form r6277“ where r > 0 and 0 < 0 < 1;
some thought along with the use of de Moivre’s theorem shows
that 0* is isomorphic to R*X R/Z, and as we have seen this is
isomorphic to R X R/Z. We shall not discuss R/Z further;
we merely note that this is the abstract group of Example
1.08.
We turn to a brief discussion of the various groups of con-
gruence mappings. It will be seen that Example 1.26 is the
algebraic description of Example 1.16, the group of translations
of .9; this group is isomorphic in an obvious manner to R.
Similarly Example 1.27 arises from Example 1.17. Now it is
easy to verify that ¢_a zfio a = BL.» and 5&3 = 1; so Example 1.17
is generated by R and the further element 5110 of order 2. Note
that 1,00— 1¢a 1,00 = 95%; that is, do induces the automorphism of R
according to which every element is inverted.
It should now be clear that the groups of Examples 1.18 and
1.21 are isomorphic to RXR and to RXRXR respectively.
Example 1.19 is another representation of R/Z, since each
rotation of 9 about a fixed point can be specified by some 6
in 0 g 0 < 271-, and. addition is modulo 277. The group (given
in Example 1.22) of all rotations of 5’ about a fixed line is iso-
morphic to this. These groups may also be presented as groups
of matrices. We leave this task, and further investigation of
Examples 1.20, 1.23, and 1.24, to the reader.
Example 1.25 is a finite group to which frequent reference has
been made, and we now write down generators and relations
for it. Let b denote any rotation through 277/17, and a, any reflec-
tion; then it will be found that

a2 = 1, b” = 1, a—lba = ()4.

Conversely, the group defined abstractly by these generators

and relations coincides with the dihedral group of order 217.. It
will be found that if n is odd a has n conjugates and these are all
reflections; and that the powers of b are all rotations. The most
importantZspecial cases arise Where n is aprime or is a power of 2.
 Survey of examples 217

The quaternion group (Example 1.30) is another example of

importance. In Chapter 8 we saw that it may be presented
abstractly as gm“, bza2 : b2 = (ab)2}.
Example 1.31 is of course isomorphic to the additive group R of
reals, while Example 1.32 is isomorphic to the direct sum of
mn copies of R, provided the elements of the matrices are real
numbers.
Example 1.11 will now be generalized and discussed at some
length. Let m be an integer greater than 1. The set {[n]} of
residue classes modulo m, where 1 g n < m andnis coprime to m,
is the object of interest; we denote it by Gm. A routine proof, using
the Euclidean algorithm, shows that Gm is an abelian group with
identity element [1] under the usual multiplication of residue
classes—for details compare Example 1.11. It is now reasonable
to ask for the structure of Gm in terms of the theory of finite
abelian groups given in Chapter 5. For the present we use
multiplicative notation for all groups.
At this point we indicate another way of viewing Gm.

THEOREM 11.01. The automorphism group A(Zm) of the cyclic

group Zm of order m is isomorphic to Gm.

Proof. Let 2 be a generator of Zm. It follows from the equation

(zk)oc = (2007‘ that any automorphism 0c of Zm is completely
specified by 20:. Another easy fact is that 20c” = 2” determines
an automorphism an of Zm if n is non—zero and prime to m; and
all the automorphisms of Zm are determined by such values of n
in 1 g n < m. Consider the correspondence in which can is
paired with [n]. That this is an isomorphism of A(Zm) with Gm
is evident. CI
Note that Lemma 10.14 has now been superseded.

LEMMA 11.02. Let H = lx ...><Hpn where p1,..., p” are

distinct primes and Hp denotes the Sylow p—subgroup of H. Then
A(H) g Amp!) >< ><A(Hp").
218 Survey of examples

Proof. We note that each Hp is a characteristic subgroup of H.

If therefore an automorphism or of H is given, then a determines
by restriction an automorphism a,- of Hz”, for each i. Conversely,
suppose that an automorphism 061; of H1,, is given, for each i. It
is then easy to see that an automorphism or of H is determined
by (h1,...,hn)oc = (kl 051,...,hnozn),
where h,- 6 HM; one merely verifies that a is well defined, is a
homomorphism, is one—one, and is onto H.
We now turn to the direct product criterion of Theorem 4.38.
It is clear that the AMI“) generate A(H), from the remarks
above. The requirements (i) and (iii) are evidently satisfied, and
therefore A(H) g 11(l) >< >< A(Hpn). [I

COROLLARY 11.03. If m = p531 pE' where p1,..., p, are distinct

primes, then A(Zm) g A(Zp§1) X >< A(Zp§,).
Proof. Use Theorem 5.01 and Lemma 11.02. [I

COROLLARY 11.04. If m = p1l pE' where p1,..., p, are distinct

primes, then G g Gpglx >< G'pgr.
Proof. Use Theorem 11.01 and Corollary 11.03. [I

THEOREM 11.05. Ifp is an odd prime then 0105 is cyclic of order

(p—1)p3-1. If p = 2 then 02;; is the direct product of a cyclic
group of order 25-2 and a cyclic group of order 2 unless B = 1
and G2 is trivial.
Proof. It is easy to see that pig-1 of the numbers {n: 1 g n < p5}
are divisible by p, so [Gppl = (p—1)pI3—1, whether p is even
or odd.
Take p odd, to start with. It will clearly suffice to produce
elements of app with orders p— 1 and p54 respectively, for their
product will be a generator of 0135.
Consider [1+p]. We have, by the binomial theorem,
(1+p)1’5"1 E 1 modulo p3,
(1-}—p)1!”3_2 E 1+p3‘1 modulo [)5
ifB > 1 andp > 2; so in this case [1 +p] has order pig—1.
 Survey of examples 219

To produce an element of order p— 1 is rather more difficult.

Let d be a factor of p—l and consider the congruence
oath—1 E 0 modulop.
We shall need the well—known fact’r that this has at most d
solutions modulo p. There are, therefore, at most d elements of
order dividing d in the multiplicative group of residues modulo p,
for each divisor d of p— 1. Application of the fundamental
theorem (5.07) on finite abelian groups, therefore, shows that
as is cyclic. If [a] generates G1, then a10‘1 E 1 modulo p, and so
the element [am—1] of G105 has p—power order, by binomial calcula-
tions similar to those above. A suitable power of [a] will have
order p— 1 as element of Gpp, and this is what we want.
Next we work with p = 2, neglecting the trivial case ,8 = 1.
Indeed we take [3 > 2, for it is easy to see that A(Z4) has order 2.
A binomial calculation shows that [5] has order 25—2 as element
0f 923‘ 52“” = (1+4)2H E 1 modulo 213,
525‘“ = (1+4)2"“ E 1—|—2I9—1 modulo 219.
Clearly [— 1] has order 2. Further, [— 1] is not a power of [5] ; if
it were its order would force us to
[—1] = [5125—1
—1 E 1+23-1 modulo 2/3,
which is impossible as B > 2. It is now evident that 025 is the
direct product of gp{[5]} and gp{[— 1]} [l

COROLLARY 11.06. Each of the isomorphic groups Gm and

A(Zm), where m = p11...pfr, p1 < < p, are primes, and 31,...,
B, arepositive, is the direct product of abeliau groups of orders
(25— 1)p/131‘1,..., (12,— 1)pf"1 respectively. These are cyclic when pi
is odd, and have type {fill—2, 1} when p1 = 2.

Proof. Use Theorem 11.01, Corollary 11.04, and Theorem

1 1.05. [I

1' See G. Birkhofi and S. MacLane, A survey of modern algebra (third edition,
Macmillan, 1965), p. 58.
220 Survey of examples

Example 11.07. Gm need not be cyclic; for instance 015 is the

direct product of cyclic groups of orders 2 and 4.
We pass on to an examination of Examples 1.13, 1.14, and
1.15, which were permutation groups. We note that any group
G is a subgroup of a suitable symmetric group SX, for Theorem
2.18 shows that G’ < SX where X is the set of elements in G'.
In particular, if G has finite order n then G < Sn.
We now generalize Example 6.22, according to which A5 is
simple.

THEOREM 11.08. The alternating group An is simple if n 2 5.

Proof. We proceed by induction on n, starting off with the
known fact that/A5 is simple. Letrz, 2 6 and suppose that AW1 is
simple. Let N be a normal subgroup of An other than 1; we
want to prove that N = An.
We first show that N contains a non-trivial permutation which
leaves fixed an element of the set {1,...,n} on which An acts.
Suppose not. Then 95 e N where
(195 = b, 045 = d
and a, b, c, d are distinct elements of {1,..., n}. Since 72 2 6,
An contains (/1 such that
a:/x=b, bz/I=a, c¢r=d, d¢r=e
Where e is different from a, b, o, (1. Now git—1955!: e N, and

I’M—19W!) = 0», deb—1M) = e.

We also have (fist—195.}: e N, and
a(¢¢‘1¢¢) = a, Owl/"19M = 6-
Thus ¢¢—1¢¢ is a non-trivial element of N leaving the symbol a
fixed. This is a contradiction. We conclude that N contains a
non—trivial element fixing some symbol in {1,..., n}.
Suppose (Without loss of generality) that qS E N and 195 = 1.
It is easy to see that the set of all elements in An which fix 1
forms a subgroup 8(1) which is isomorphic to An_1. Thus we
haveN n 8(1) 7e 1,and, of course,N n S(1)isanormal subgroup
 Survey of examples 22]

of 8(1). But 8(1) is simple because A,”1 is simple by the induc-

tive hypothesis. Therefore N n 8(1) = 8(1); that is, N 2 8(1).
Next N 2 8(2') for each i in {l,..., n}; for if M; = 1 for some
it 6A,, then 8(i) = ¢8(1)¢-1, and the normal property of N
gives N 2 8(92). We observe that 8(2) is isomorphic to A,“1 in
the natural way, and that 8(1) n 8(2) g An_2.
We now consider gp{8(1), 8(2)}, which is contained in N.
The order of this group is at least [8(1)]|8(2)H8(1) n 8(2)]‘1,
as we see by counting the cosets of 8(1) n 8(2). Thus

IN! > sm—1):}s<n—1>!}/s<n—2>r} = ”T‘lm—m

But A” has order %n! Since n— 1 does not divide n, we find that
|N| = |An|. Thus N = An, and An is simple. I]

COROLLARY 11.09. AX is simple if the set X is in one—one

correspondence with the positive integers.

Proof. We recall that AX is a subgroup of RX, each element of

which moves only a finite number of symbols. Now AX contains
a subgroup isomorphic to An for each n 2 1, namely the
permutations affecting the subset {1,...,n} of X only; we have
AM1 2A”, forn 2 1, and AX = U An.
n21
Suppose that 1 < N <1 A x- Then N contains an element
p 75 1 of some An. Further, p 6 AM,- for eachi 2 0. But AM)
is a simple group for n+t' 2 5, by the theorem, and N n A n+1:
is normal in Amt; so A n+7: <N. It follows that AX <N.
Therefore AX is simple. I]
We have made a simple choice of X in the above corollary,
for the benefit of the reader unfamiliar with infinite set theory.
However, this suffices to produce a group with a composition
series and a subgroup with no composition series; namely AX
and its subgroup generated by
{(123), (456), (789),...}.
The details are left for the reader to complete.
222 Survey of examples

Our final remarks on the examples of Chapter 1 concern

Examples 1.28 and 1.29, multiplicative groups of non-singular
matrices. One aspect of their importance is that a finite group
of order n can always be exhibited as a suitable matrix group,
according to Corollary 2.21. Our present remarks go no further
than definitions.
A field K is a set in which two operations, called addition and
multiplication, are defined; K is to be an abelian group with
identity element 0 under addition, the non-zero elements of K
are to form an abelian group under multiplication, and the
familiar distributive law a(b+c) = ab+ac is to be valid for all
a, b, c in K. Familiar examples of fields are the real numbers,
the rational numbers, and the residue classes modulo a prime
(by Example 1.11).
We consider matrices with entries from K. A non-singular
matrix has an inverse with entries from K, as may easily be
shown, and it is natural to generalize Examples 1.28 and 1.29 by
taking multiplicative groups of non—singular n X n matrices with
entries from K.
Example 11.10. The set of all such matrices forms a group,
the general linear group GL(n, K). Detailed proof is easy and is
omitted, in this and the following examples.
Example 11.11. The centre Z of G'L(n, K) consists of all
matrices 161 where k is a non-zero element of K. The proof of this
statement is indicated in Example 4.26. The projective general
linear group PGL(n, K) is defined as GL(n, K) /Z.
Example 11.12. The special linear group SL(n, K) is the sub-
group of GL(n, K) Whose elements have determinant 1.
Example 11.13. The projective special linear group or linear
fractional group LF(n, K) is the central factor group of SL(n, K).
We state without proof that the groups LF(n, K), with a few
exceptions, are simple.
We now examine further examples, first fulfilling a promise
made in Chapter 9 (just after Example 9.22).
 Survey of examples 223

Example 11.14. There is an abelian group A which is not the

direct sum of its periodic subgroup and a torsionfree subgroup.
Notice that A cannot be finitely generated, by Theorem 5.12.
Let p be any prime and let An be the cyclic group of order p",
for each n > 0. The example is A = 2: A". Suppose that P is
10>

the periodic subgroup of A. We observe that P contains more

than just 21’ An; if An = gp{an} then f e P where
n>0
nf = gun—1a”.
It will suffice for our purpose to show that A /P is not isomorphic
to any subgroup of A.
NOW A /P has the property that it contains certain non-trivial
elements a+P such that the equation
pM<x+P) = a+P
hasasolutionx—l— P, for allm > 0. For instance, takea e Awhere
(2%)“ = pnaZn:
(2n—l)a = 1,
for n = 1, 2,..., and fix m. Ifx in A is defined by
(2n)x = Ian-mam for n 9 m, and 1 otherwise;
(2n—1)x = 1,
for n = 1, 2,..., then pmx—a, e P, so that pm(x+P) = a+P, as
required.
On the other hand, if the element b of A is such that 107% = b
has a solution for all m > 0, then b = 1. For if b 72 1 then
nb 7’.- l for some n > 0 while ib = 1 for 1 g i < n. It follows
that pn+1x = 71.6 has a solution x in An, which contradicts one
of two facts, namely nb 7E l and that [An] = 1)”.
Therefore A /P is not isomorphic to any subgroup of A, and P
cannot be a direct summand of A.
The remaining examples of this chapter require the concept of
‘split extension’. Let G’ be a group with a normal subgroup N,
and suppose that a transversal F of N in G’ can be chosen so
that F is in fact a subgroup of G. This is a very special situation;
one instance arises when lG/N | and [N [ are finite and coprime,
224 Survey of examples

by Theorem 7.34. Each element of G can be expressed as gun

where gm 6 F, n e N; and this expression is unique. Multiplica—
tion in G goes as follows:

(90‘ n1)(g,e 7r2) = (gagxn‘lfinz),

since gm g3 = gag. There is a homomorphism from F into the
automorphism group A(N) of N, the images being defined by
conjugation. Since G/N g F, we have a homomorphism from
G/N into A(N).
The object is to reverse the above procedure, to construct a
group from given abstract groups F and N.

THEOREM 11.15. Let groups F and N be given, together with

a homomorphism 95 from F into A(N). Then there exists a group
G containing a normal subgroup N’ and a subgroup F’ with the
following properties:
(i) F'g F, N’gN;
(ii) G/N’ g F;
(iii) the automorphism of N’ induced by conjugation with
a: e F’ is the automorphism of N defined by and, elements of
F and F’ and of N and N’ being identified in the natural way.

Proof. First we define a set G as follows:

G = {(ga,n):oc e F, n EN}.
Here {get} is merely a set of symbols in one—one correspondence
with {a}, and G is a set of ordered pairs. Next we define multi—
plication in G by

(9a,%1)(gpan2) = (gafianiflnz)
where n‘l’B denotes nfifigb); thus B e F and [375 e A(N).
We now prove that G is a group. Closure is obvious. The
associative law may be verified by a little calculation, left to the
reader. The identity element is (g6, 1) where e, 1 are the identity
elements of F, N respectively. Finally, the inverse of (gel, n)
is (g5, (rt-1W) Where ,8 = oc—l.
 Survey of examples 225

The following isomorphisms should be clear:

F g {(ga, 1):oc E F},
N g {(gE,n):n EN}.
We therefore define F’, N’ in the obvious way, and (i) is complete.
By now (ii) is obvious. As for (iii), we observe that
(gap 1)—1(ge’n)(gw 1) = (gs, n9“)
follows from the definitions. But n94 was defined as n(a¢). We
therefore have (iii). D
The group G constructed in the theorem is called a split
extension ofN by F, or a semidirect product of N by F; it is perhaps
better not to call it the split extension or semidirect product
because it depends on 45 as well as N and F. Its use, which we
illustrate, is in the construction of particular groups.
Example 11.16. Consider again Example 1.27. This group G
has a subgroup of index 2 isomorphic to R. It also contains an
element z/Io which induces the inverting automorphism on R, and
zlfi = 1 ; we discussed this earlier in Chapter 1 1. In fact it happens
that G’ is a split extension of R by the group of order 2, and 36 maps
the generator of the latter onto the inverting automorphism.
The dihedral group of Example 1.25 may be similarly
regarded as a split extension.
Example 11.17. We now use split extensions to construct a
pair of important non—abelian groups of order p3, where p is an
arbitrary odd prime.
Let N1 be the direct product of two groups of order p, with -
respective generators b, 0. Let F1 be of order p, and let (151 map
a generator of F1 onto or, the automorphism of N1 specified by
ba = bc, soc = c.
It is easy to see that this mapping does specify an automorphism
a of order p. The resulting split extension GI may be presented as
gp{a,b,c: a)" = 1, b1" 2: 1, c? = 1, [11,0]: 1, b“ 2 be, c“: c}
or, equivalently, as
gp{a,b: up: 1, b1": 1, [a,b]19 = 1, [a,b,a.] = 1, [a,b,b] = 1}.
853137 Q,
226 Survey of examples
The second construction starts by taking for N2 the cyclic
group of order p3; let it be generated by b. Again F2 is to have
order p, and ¢2 is to map a generator onto the automorphism
a of N2 specified by bu = b1+1’.

After the details are tidied up, we obtain a group G2 with

presentation
gp{a,b: a” = 1, b1"2 =1,b“ = b1+10}.
Example 1 1.18. Let N be cyclic of order 2"“, for any positive
integer n, and let F be cyclic of order 4 with generator a. Further,
qS is to be specified by wt being the inverting automorphism of N.
The split extension G which results has order 2””.
Now the element a2b2” is central in G, if b generates N; put
Z = gp{a2b2"}. The factor group G/Z, of order 2””, is of some
interest. Since it coincides with the quaternion group of order
8 when n = 1, it is called the generalized quaternion group. It
has the following presentation:
gp{a,b: b“ = 6—1, a2 = b2"}.

Next we discuss the holomorph of an arbitrary group G—

this is a split extension of G by A(G). Theorem 2.18 shows that
the right regular representation of G is a subgroup of 80 iso-
morphic to G, where 80 is the symmetric group on the elements
of G. Now A(G) is also a subgroup of 86,, because any auto—
morphism of G is a permutation of its elements. We have
p(G) n A(G) = 1, where p(G) denotes the copy of G in 86,, as
may be seen from the actions of p(G) and A(G) on the element 1
of G. Before describing a split extension we need the following
result.

THEOREM 11.19. If g6 e A(G) then ¢-1pg¢ = pug, where pg is

the image of g in p(G).

Proof. Take any element x in G. Then

90(95‘1q5) = ((96<I5“1)Pg)<lS = ((x¢‘1)9)¢ = x(9¢) = n¢~

The desired result follows. [J
 Survey of examples 227

COROLLARY 11.20. A(G) normalizes p(G) in Sa. [I

It follows that gp{A(G), p(G)} = A(G)p(G) and that p(G’) is
normal in this. Indeed A(G)p(G) is the split extension of p(G) by
A(G), the automorphisms of p(G) being specified as in Theorem
11.19. This group A(G)p(G) is called the holomorph of G'.

Example 11.21. Let G be cyclic of order 10, an odd prime. Then

A(G) is cyclic of order p—~1, and the holomorph of G’ has the
following presentation:
gp{a,b: I)” = 1, (110-1 = 1, b“ = b°‘},
where [or] has order p—l modulo 1) (compare Theorem 11.05).
Note that the holomorph contains the dihedral group of order
21) as a subgroup.

Example 11.22. Consider the automorphism group A(Z) of

the infinite cyclic group Z. If Z = gp{z} then there are only two
elements which individually generate Z, namely .2 and 2—1.
Since Z clearly has an automorphism interchanging these two,
A(Z) has order 2. It is now easy to give a presentation for the
holomorph of Z:
gp{z,a: z“ = z—l, a2 = 1}.
The reader may be able to show that this group is isomorphic to
that of Problem 5 of Chapter 3; it is the so-called ‘infinite
dihedral group’.

Problems
1. Let .A’ be the set of all m x n matrices with entries from an abelian
group A. Show that J? forms a group under matrix addition (as in
Example 1.32) and that “I is isomorphic to the direct sum of W copies
of A.

2. Show that the two groups of order p3 in Example 11.17 are non-iso-
morphic. Prove that there are no other non-abelian groups of that order,
1) being odd.

3. Prove that Q/Z is isomorphic to the direct sum of the groups

Z$,..., 2,33,... where p” is the nth prime.
228 Survey of examples

*4. Find all groups of order 16.

5. (i) Prove that every finite group occurs as a subgroup of a suitable
finite simple group.
(ii) Prove that every finite group can be represented as a (multiplica-
tive) group of matrices of determinant 1.

6. Prove that the group of Example 1.23 is isomorphic to the group of

3 X 3 matrices With real entries and determinant l.

7. An abelian group A is said to be divisible if the equation mt: = a has

a solution x for each n > 0 and each a E A. Which of the groups Z, Z",
Z5, Q, R, Q/Z, R/Z are divisible?
8. Which of the following statements are true? Justify your choice.
(i) The symmetric group 8,, has a unique subgroup of index 2.
(ii) Let p and g be primes with p > q. There is precisely one non-
abelian group of order pg if and only if p E 1 modulo q.
(iii) Let 10,, denote the nth prime and Zn the group of that order. Then

41102..) g H0 M.»
n>0 n>0
(iv) If G = H” G" where each G), is a characteristic subgroup of G,
n>0
then A(G') ; flD Aw").
n>0
(v) All groups of (finite) order m are abelian if and only if m is divisible
by the square of no prime.

"*9. Let K denote the field of residue classes With 5 elements.

(i) Prove that the centre Z of SL(2,K) has order 2.
(ii) Prove that Z is the only proper normal subgroup of SL(2,K).
(iii) Deduce that G = 3(G) when G = SL(2, K).
(iv) What is the Sylow 2-subgr0up of SL(2,K), as abstract group ‘I
(v) Give an alternative description of LF’(2,K).
10. Let G be a finite group. Show that, in the regular representation of G,
any two isomorphic subgroups of G map onto conjugate subgroups of 62;.

11. Show that gp{w,b:a2 = b2 = (ab)4 = 1}

is a presentation of the dihedral group of order 8, and that this group G‘
has an automorphism oz of order 2 which interchanges a and I).
Let H be the split extension of G by a group of order 2, with or playing
the obvious role; thus H has order 1 6. Is H a dihedral group, or a general-
ized quaternion group, or neither?

*12. Prove that

gp{a,b:ab = “31, a1“ = bio, aloo = 1}

has order 103, and identify its Sylow 2-subgroup.

 Survey of examples 229

Prove further that none of the three given relations is redundant, but
that the same group may be presented as follows:
gp{c,d:c10 = dl0 = [c,d]}.

*13. A hamiltonian group is a non-abelian group in which every sub-

group is normal. Let G’ be a hamiltonian group.
(i) Prove that [39, y, y] = 1 for all x, y in G.
(ii) Prove that G is periodic.
(iii) Deduce that G’ is the direct product of nilpotent p-groups for
various primes 1).
(iv) Prove that the Sylow p-subgroups, for odd primes p, are abelian.
(v) Prove that the Sylow 2-subgroup G2 contains a quaternion group
(28; that the centralizer O of Q28 in 92 has exponent 2; and that
62 = 0623.
Deduee that a finite group is hamiltonian if and only if it is the direct
product of a quaternion group, an abelian group of exponent 2, and an
abelian group of odd order.

***l4. Let G be a finite group of exponent 3, and generated by {(1, b, 6}.

Assuming for the moment that G is nilpotent of class 3 but not of class 2,
prove the following facts about G:
(i) 3(6‘) has order 34 and is elementary abelian;
(ii) G/3(G) has order 33 and is elementary abelian;
(iii) y3(G‘) has order 3.
Now try to construct a group H of order 37, exponent 3, and class 3
but not 2, as follows. Take an elementary abelian group of order 34 and
form a split extension H1 of this by a group of order 3; form a similar
split extension H2 of H1, and a third split extension H of H2. At each
stage use information gained earlier about the structure of G.
Revision problems

1. Distinguish the true statements among the following from the false,
giving appropriate proofs.
(i) a and b are conjugate in G if and only if (17% 6 3(6‘).
(ii) The non-empty complex 0 in the group G is a subgroup if and
only if Car: = 01/ implies x = 3/.
(iii) There is a group G such that G/§(G’) is isomorphic to the additive
group Q of rationals.
(iv) IfX= Li __ 3) then the distinct positive powers of X form
a group.
(v) A group generated by an arbitrary distinct pair of its non-trivial
elements is cyclic.
(vi) Every group of order n2 is abelian (for each positive integer M).
(vii) Any subgroup of order 2 in a group is central.
(viii) Ifa, b are elements of the group G then the orders of the elements
a'lb, ab—l, ba—l, b‘la are all equal.
(ix) All cyclic groups of infinite order are isomorphic, and therefore
each is generated by any one of its elements.
(x) If a, b are any elements of any group then ab 6 gp{a}.

2. The central product of groups G, H with isomorphic centres £(G), £(H)

is defined as follows: it is (G XH)/N where N consists of all pairs (9:, y)
with x e {(0), y e {(H), and x, y corresponding under the given iso-
morphism.
(i) Prove that the central product is well defined.
(ii) Prove that the central product of two quaternion groups is iso-
morphic to that of two dihedral groups of order 8.
(iii) Is this group isomorphic to the central product of one quaternion
group and one dihedral group?

3. Are the following groups isomorphic?

gp{a,b:a8 = 1, a4 = b“, ab = ‘1},
gp{a,b:a3 = l, a4 = b4, (1" = a3}.

4. The subgroup B of the abelian group A is said to be pure if the equation

me = b (7» positive integer, b e B) has a solution x in B whenever it has
a solution in A.
 Revision problems 231

Prove that a subgroup of a finitely generated torsionfree abelian group

is a direct summand if and only if it is pure.
Is the corresponding result true for subgroups of finitely generated
abelian groups?

5. Prove that it is impossible for G/Z(G) to have a normal subgroup

isomorphic to the quaternion group, whatever group G is.

6. Prove that if G is any group and N Q G then yn(G/N) = yn(G)N/N

for n 9 1.
Deduce that if F is a free group then F/'}’n+1(F ) is nilpotent of class n.
Hence (or otherwise) show that a free basis for F cannot contain two
distinct conjugate elements.

7. Let G be any group. Prove that C(G) is cyclic if and only if G contains
an element x such that gp{a:} = O[N{0(w)}]. (As usual 0, N indicate
centralizer, normalizer respectively.)

8. Let a, b be fixed elements of the group G, and n be a positive integer.

Prove that [(1, b]"‘ = (a‘lb—l)”(ab)"c1 cn_l

for suitable commutators 01,..., cn_1. Prove further that if |G/{(G)| = n

then (ab)" = (ba)”, and that [a, 6]” may be expressed as the product of
some n~1 commutators. Deduce that if G/§(G) is finite then so is 3(G).

9. The group G contains a finite normal subgroup H in which lies a Sylow

p-subgroup G’p of G. Prove that G = HNg{NH(Gp)} where Ng, NH denote
normalizers in G, H respectively.

10. The normal subgroups A, B of the group G are nilpotent of classes

cl, 02 respectively. Prove that AB is nilpotent of class 01-1—02.

11. Let G be a group and y a mapping of G into G. For each a; in G define

[90, y] to be (filmy) and [71, w] to be [my y]—1. Let a, b be two fixed elements
of G.

Prove that [(1, bly = [00” by]

if and only if

[[bfil’ [7» b]][[b, uln'l‘y'b‘b’, 1’: (171W, )6 bl] = 1-

12. The group G is a finite 2-group in which G/3(G) has order 4.
(i) Prove (by induction, for instance) that G contains a cyclic sub-
group of index 2.
(ii) Show that (i) is false if 8( G) is replaced by @(G).
(iii) Show that (i) is false if the prime 2 is replaced by any other prime.
232 Revision problems

13. Let G be a finite p-group. Prove that if @(G) has centre of order p
and is non-abelian then €(G) n 11>( G) = {{CIJ(G)}; and deduce that Such
a group G does not exist. Hence prove:
(i) All groups of order p5 are metabelian.
(ii) All groups of order 26 are metabelian (use the previous problem).

14. The element 9 of the group G has finite order and lies in a finite
conjugacy class. Prove that g lies in a finite normal subgroup of G.
Deduce that no group contains an even number of elements of order 2.

15. A subgroup H of the group G is said to be quasinormal if H commutes

with every subgroup of G.
Let G = A X B, where
A = gp{a,b:a” = a1”, 6” = 1},
B = gp{c:c1’2 = 1},
and p is an odd prime. Find two quasinormal subgroups of G Whose
intersection is not quasinormal.

16. The subgroups A, B of the group G are such that AB = BA. Prove
that A and B are not conjugate in AB unless A = B.
Deduce that if a group G has a maximal subgroup M which is quasi-
normal in G (see the previous problem) then M <1 G.
Show that any subgroup of index 10 in the finite group G is normal ifp
is the smallest prime dividing [G].
17. Prove that a p-group of finite exponent is nilpotent if it contains a
nilpotent subgroup of finite index.

18. The group G has G/81(G) a cyclic p-group (p prime). Prove that no
normal subgroup of G has index a positive power of p. Deduce from this
that if 32(G)/33(G) is a finite cyclic p-group then 32(0) = 33(G).
19. Let G be a group in which, for each n, 8(G) contains an element not
in any way expressible as the product of n commutators. Let G" g G
for n = 1, 2,... and put H = H0 G“. Show that there exists an element
n>0
of H0 3(Gn) Which does not lie in 3(H).
n>0
Is it true that {(H) = 1‘10 C(Gn)?
n>0

20. In the arbitrary group G, the subgroup H3 is generated by all ele-

ments in G which do not have order 3. Let h 6 H3, :4: ¢ H3, and prove:
(i) th‘”2 = 1;
(ii) UL”, h7’] = 1 if 901173 7E s;
(iii) H3 = 1, unless H3 has index 3 or 1 in G.
Deduce that a finite (non-trivial) group G has exponent 3 if and only
if every element outside (13(G) has order 3.
 Revision problems 233

21. Let the group G have a fixed finite set {gl,..., gn} of generators, and
consider subgroups of fixed finite index k in G. Such a subgroup K
determines an equivalence relation in G in the usual way: 91: ~ 3; means
any—1 6 K. Show that if {h1,..., hk} g {g1i1,..., gfil} then two of the following
elements of G are equivalent:
1, hl, hlhz, ..., hlh2 hk.
Deduce that G’ has only a finite number of subgroups of index 10.

22. Use the general method of the previous problem to show that a sub-
group offinite index in a finitely generated group is itselffinitely generated.

23. Prove that a finitely generated periodic soluble group is finite. (Hint:
use induction on the solubility length and apply the result of the previous
problem.)

24. Prove that if the group G is nilpotent of class 2"— 1, where n is a

positive integer, then G is soluble of length n.

25. Let H1, H2 be given groups and let K be a subgroup of G = H1 >< H2.
Define subsets F1, F2 of H1, H2 as follows:

F1 = {h1:(h1, kg) 6 K for some h2 6 H2},

F2 = {h2:(h1,h2) eK for some kl 6 H1}.

(i) Prove that F, g H1 for 13 = 1, 2.

(ii) Prove that ifKi = K n Hi then Ki £1 E for i = 1,2.
(iii) Prove that K/K1 g F2, K/K2 ; F1.
(iv) Prove that I’ll/K1 ; Fz/KZ; and that (kl, ha) 6 K if and only if the
images of hi in Fi/Ki (i = 1, 2) correspond in this isomorphism.

26. Let G’ be a finite p-group on two generators, having exponent p2 and

containing an element of order 102. Show that the following statements
are equivalent:
(i) Every element of order p2 lies in @(G).
(ii) [G/H(G)] > p, where H(G) is the subgroup generated by the
elements of order 102.

27. The abelian group A has subgroups A1,..., An and homomorphism

(ti from A to Ai, and «[1,; fromAi, to A, satisfying the following conditions for
1 g 7} g n.
(i) {/15951- is the identity mapping on At;
(ii) (piq- maps A1; onto 0 ifj ;E i.
Prove that A ; A1 69 ...(JaAn if and only if i aqSix/Ji = a for each
i=1
element a of A.
234 Revision problems

28. The finite group G has a. subgroup H for which [GzHI is a power of
the prime 12. Show that if G1, is a Sylow p-subgroup of G then H n G1,,
is a Sylow Ira—subgroup of H, and G = H0,.
Prove that the following conditions on the element x of p-power order
of the finite group G are equivalent:
(i) The least normal subgroup of G containing 90 is a p-group.
(ii) There is a normal subgroup N of G containing ac, and :70 has n
conjugates as an element of N where n is some power ofp.

29. The finite group G has the property that no maximal subgroup is
normal. Prove that either G is abelian or two distinct maximal subgroups
have non-trivial intersection, by assuming that the latter alternative is
false and counting the elements of G in each conjugacy class of maximal
subgroups.
30. Suppose that G is a finite group in which every proper subgroup is
nilpotent. Prove that if G is simple then any two distinct maximal sub-
groups have trivial intersection. Deduce from this and the result of the
last problem that G must in fact be soluble.
31. Prove the theorem that a finite non-abelian group G contains a pair
of distinct conjugate elements which commute, by means of the following
steps.
(i) Reduce the problem to the case in which every maximal subgroup
of G is abelian but not normal.
(ii) Prove that M1 0 M2 = {(G), where M1, M2 are distinct maximal
subgroups.
(iii) Apply Problem 29 to investigate G/{( G).
*32. Let H be a finitely generated subgroup of G, which is defined to be
Ana FA, each FA being isomorphic to the finite group F. Prove that, for
EA
fixed A, the mapping from H to F,‘ which sends h to Ah is a homomorphism
and that its kernel K,‘ has finite index in H. Deduce that H is finite.
Prove further that if each FA is finite and [FA I is bounded independently
of /\ then every finitely generated subgroup ofA110 FA is finite.
GA

*33. Show that the following statements are equivalent:

(i) Every residually finite group having (1 generators and (finite)
exponent e is finite.
(ii) The finite groups with d generators and exponent e have their
orders bounded by a function of d and e.

34. Let 'n be the nuInber of distinct primes dividing the positive integer m.
Find a set of n elements which generates the cyclic group Zm of order m,
such that no proper subset generates Zm. Find a set of n defining relations
in such a set of generators, with the property that no relation is redundant.
 Revision problems 235

35. Prove that gp{a,b,c:b" = ()3, c“ = 03, ab = a3} is a finite 2-group

which cannot be generated by any two of its elements. What is the order
of the group?

36. The group G is defined by generators at, b, c and relations

be 2 b3, Ga = 6—1, ab = (1—1,
a4=b4=c4, (15:1.
(i) Prove that G has order 27.
(ii) Prove that if x, y #5 <I>(G) then gp{w} n gp{y} 75 l.

37. (i) Prove that a finite group G is nilpotent if and only if G/<I)(G) is
nilpotent.
(ii) Prove that, if G is a finite group with H g G, N <1 G, and
N g @(H), then N g @(G).
(iii) Prove that, if G is a finite group with a normal subgroup H such
that both H and G/3(H) are nilpotent, then G is nilpotent.

*38. The upper p-se'm'es

1=PO<NO<P1<...<P,<NT=G

of the finite group G is defined as follows: 1) is any prime, Ni/Pi is the largest
normal subgroup of G/fi with order prime to p for 0 g 1} g r, and B+1/M
is the largest normal p-subgroup of G/M- for 0 g i < r. The group G is
said to have p-length r.
(i) Prove that the upper p-series is a well-defined invariant series
for G.
(ii) Prove that, if G is soluble, then P1 contains the centralizer 0 of
Pl/No. (C is {mm s G, [9,96] 6 N0 for all g 6 P1.) (Hint: the Schur—
Zassenhaus theorem may be useful.)
(iii) Prove that a finite soluble group with abelian Sylow p-subgroups
has p-length l.

*39. The scale 0(G) of a finite group G is defined to be the subgroup

generated by all the minimal normal subgroups of G. Prove that a(G) is
in fact the direct product of a selection of these normal subgroups.
Let N g o-(G), and N Q G. Prove that N is also the direct product of
a suitable set of minimal normal subgroups of G. (Hint: Problem 25
may be helpful.)
Deduce that a finite group with no proper characteristic subgroup is
the direct product of isomorphic simple groups.

“‘40. An important property of a finite soluble group G Whose Sylow

subgroups are all abelian is that 8(G) n C( G) = 1. Prove this in the steps
indicated below, or otherwise.
236 Revision problems

(i) Reduce the problem to the ease in which 8(G) n {( G) is non-trivial

and is the unique minimal normal subgroup of G. Consider this case only
in what follows.
(ii) Show that G is metabelian, and that 8( G) is a p-group. Deduce that
G 2 G1, G7, where G1" is a Sylow p-subgroup and G,r is a suitable abelian
Hall n-subgroup.
(iii) The elements of order p in 3( G) form a subgroup P of the form
PIXPZ, where P1 = 8(G) n {(G) and P2 is a suitable subgroup. Each
element of G,r induces an automorphism 0 of P which fixes P1. If :1: e P
then x = 2:11:2(905 E Pi for i = 1, 2) and we may form 933 = I; {(9100)2 0‘1}.

Prove that the set of all such x3 is a normal subgroup of G, and deduce
thatP = 8(G) n ((6‘).
(iv) Show that G cannot exist.

*41. We define a subcartesian product H of the groups A1,..., Aw... as

follows. Let G = H0 A” and let H g G; then we require that
Tl>0
{n961q5 e H}: A”
for each n > 0.
N0W let A be a fixed group and B be a subgroup of A. Take An g A
for each n. Show that the constant functions on {1,...,n,...} generate a
subgroup A’ of G isomorphic to A, and that A’ contains a subgroup B’
isomorphic to B.
Prove that B’ and H: A" generate in G a subcartesian product of
11>
A1,..., Am... .
Deduce that B may be obtained from A by means of these operations:
(i) taking a subcartesian product S of isomorphic copies of A;
(ii) taking a factor group of S.

*42. Let F be a free group with free basis {(1, b}, and suppose that F
contains elements u, v for which
[(1,1)]2 = [MW]-
Show that even if u, v are reduced words then some cancellation occurs
when u—Lv—lu'u is reduced.
Let u, v be chosen so that the sum of their lengths is minimal. Prove
that
(i) none of u—l, 72—1, M, v cancels completely in an adjoining word; and
(ii) u and v are not of the form x—lwx unless a: = 1.
Deduce that the cancellation in [um] occurs either between v‘1 and
u only; or between u—1 and 0'1, or u and v, or both.
Hence show that [(1, b]2 is not a cormnutator in F.

*43. Prove that gp{a,b:a“ = b3 = (ab)3 = 1} is the alternating group

on 4 symbols.
 Rem'sio’n, problems 237

The finitely generated group G contains an element of each order 1, 2, 3

but no elements of other orders. Prove the following statements about G:
(i)Two elements of order 2 generate S3 or Z2 (of order 2) or Z2 >< Z2.
(ii)No element of order 2 lies in both 83 and Z2 >< Z2.
(iii)G does not contain both 6'3 and Z2 >< Z2.
(iv) G does not contain both Z2>< Z2 and Z3><Z3 (where Z3 is the
group of order 3).
(v) G is finite, and contains either a normal subgroup ofexponent 2 with
index 3 or a normal abelian subgroup of exponent 3 with index 2.
**44. Let H be a subgroup of the group G. Prove that G has a subset
which is both a left and a right transversal of H if and only if any two
conjugate subgroups (in G) lying in H have the same index in H.
Is this condition satisfied when
G = gp{a,b: ab = a2},
and H = gp{a}?

**45. Prove that if G is the set-theoretic union of its subgroups G1,..., G",
but not of any proper subset of them, then G1 n n Gn has finite index
in G.
46. Let G denote the group defined by generators a, b, c and relations
a2”=b4=c2= 1,
ab = a1+2"—1, bc = 6—1, [0,a] = 1,
where n 2 3. Prove the following facts about the full group of auto-
morphisms of G:
(i) it is a 2-group;
(ii) it is abelian.
***47. In any group G the set of elements {a : [(1, ac, x] = l for all x in G}
is denoted by E2(G). Prove the following facts about 102(61):
(i) If a e E2(G) then (1 lies in an abelian normal subgroup of G, and
(1‘1 6 102(0).
(ii) Expand the relation [(1, anyway] = l to prove
[0», x, y][a, y, x][a, 96, y, x] = 1
for all x, y in G.
(iii) By replacing an by 90—1 in (ii), show that

[0" m: 9““) y: x] = 1!

[a,x,y,x] = 1.
(iv) Deduce that E2(G) g G.
(v) By making use of Lemma 9.29, prove that

[0: x, Ell—2 = [$10, 3!, a]-

238 Revision problems

(vi) Prove that [(1, x, y,z 4 = l for all x, y, z in G; and deduce that
E2(G) g {3(0) provided no element of G has order 2.

M48. Let G be the (multiplicative) group generated by a, b where

_ (12 3) b _ (7 6)
a _ 14 5 ’ _ 7 11 ’
and the entries are residue classes modulo 17. Prove that
on2 = b3 = (ab)4.
Hence (or otherwise) show that G is a soluble group of order 48, with
{(G) of order 2 and G/§(G) ; S4. Is G isomorphic to GL(2, K) where K
is a field with 3 elements?

**49. In the group G the derived group 3(G) is infinite cyclic and
8(G) g {(G). Prove that 8(G) is generated by one commutator.
50. Give a critical discussion of the following statements.
(1) The definition of a group is when you have a set and it also has a.
multiplication table.
(ii) The group of all non-singular 2 X 2 matrices is an isomorphic group.
(iii) All groups of prime order are cyclic, and since 13 and 17 are prime
all groups of those orders are cyclic and so abelian. But clearly 15 is the
mean of 13 and 17, and it follows that any group of order 15 is abelian.
(iv) Let the group G have a subgroup H of index 2. The set of elements
not in H is a coset but not a subgroup, so H is the only subgroup of G
having index 2. Therefore H is a characteristic subgroup of G.
(v) Let S be a subgroup of the free group F. If w = 1 Where w is a
reduced word in S, then the freeness of F shows that w is the empty word.
Therefore 8 is free.
Appendix

Elementary set and number theory

THE purpose of this survey of basic concepts does not comprise formal
definition of sets, mappings, numbers, and the like. We are more con-
cerned with settling genuine queries and doubts that might appear in the
mind of the absolute beginner, as well as with describing notation and
conventions.
Sets of numbers are the most familiar and elementary examples of sets.
The elements of the set Z of integers are 0, j; 1, i2,..., and we write

2 = {0, i1, i2,...}.

The set of positive integers is {1, 2,...}. The set Q of rational numbers is
described in our set notation as

Q={x:x=%, peZ, q, 9720}.

The symbols following the colon give the defining property of Q, and
‘p e Z’ means thatp is an element of the set Z, that is p is an integer. Other
fundamental number systems are the set of real numbers and the set
of complex numbers.
If A is a set and if each element of the set B belongs to A, then we say
that B is a subset of A and write B Q A. If there is an element x such that
x e A and x ¢ B, then we say that B is a proper subset of A and write
B c A. Thus we have Z C Q, for example. Note that A = B if and only
if A g B and B 9 A are both valid statements.
The best attitude towards the empty set 0 is, perhaps, to regard it
as an interesting curiosity, a convenient fiction. To say that .70 E 6 simply
means that a; does not exist. Note that it is conventionally agreed that 0
is a subset of every set, for elements of Q} are supposed to possess every
property.
Union and intersection of sets A, B are defined in the standard way

AUB={w:meraceB},

AnB={x:xeAandxeB},
240 Appendix

respectively. Since 0 is a set, A n B is always a set. If A is the set of

three-sided polygons and B is the set of four-sided polygons then it is
clear that A n B = 0.
The reader should be at ease with the idea of a set of sets, that is a set
Whose elements are themselves sets. Suppose A is some fixed set. Then
{some is a subset of A}
is a perfectly respectable set of sets. (Factor groups are sets of sets.)
A partition of a set A is another important example of a set of sets.
It is a set P whose elements are non-empty subsets of A such that each
element of A lies in one and only one element of P. A possible partition of
Z is {Z1, Z2} where Z1, Z2 are the subsets of even, odd integers respectively.
We must also mention sets of ordered pairs. If A, B are sets and if
a e A, b e A, we can form the pair {a, b}; it is simply a set with two
elements. We may think of a, b as occurring in the order ‘first a, then b’,
and in so doing form the concept of an ordered pair, conventionally
written as (a, b) in this case. We can clearly form sets Whose elements are
ordered pairs; for instance
AXB = {(a,b):aeA, b 6B}.
This can readily be generalized. Given n sets A1,..., A" we can consider
n-tuples, or ordered sets of n elements; an n-tuple is written as (a1,..., (1,)
Where a.) 6 Ai for l g 73 g n, and the set of all n-tuples is denoted by
A1 X XAn in these circumstances.
Next we consider mappings from a set A into a set B. A mapping 95
associates with each element x of A precisely one element 3/ of B, a fact
described in the equation y = 0695. Note that we write the mapping
symbol </> ‘on the right’, which is the usual convention in algebra. Two
mappings 95 and it: from A into B are equal if and only if $55 = 901/; for all
x e A.
We say that 45 is onto B, or simply onto, if b = 93¢ has at least one solu-
tion :0 in A for each element b in B; thus each element b in B appears as
the image of a suitable element of A . We say that 95 is one—(me if $95 = y¢
implies :1) = 3/; thus distinct elements of A have distinct inlages in B. The
case in which (75 is a one—one mapping from A onto B is of special interest ;
¢ is called a one—one correspondence. A set is called countable if it is
finite or is in one—one correspondence with the set Z of integers.
We turn to mappings of rather special sets. Suppose that A is some set
and ()5 is a mapping of A X A into another set B; then 96 is called a binary
operation onA. All that 95 does is to map each ordered pair (a1, (1.2), Where
a1 6 A and a2 6 A, onto an element of B. If 95 maps A XA into A then
we say that ‘ A is closed under the binary operation qS ’. (Group ‘ multiplica-
tion’ is of course a. binary operation on the group.)
As far as this book is concerned there is no essential difference in the
meanings of the terms mapping, function, and operation.
 Appendix 241

Suppose next that 75 is a mapping from A into B, and 1/1 is a mapping

from B into 0', Where A, B, C’ are certain sets. Then we can define a
mapping from A into 0 by applying first gt and then 1/1, obtaining (x4990
from the element x of A. This mapping is called the product of ([5 and ill,
and is written (fin/r. The definition and notation yield the equation
70(951/1) = (WW1-
The reader will probably find it easy to persuade himself that (fit/r is onto
only if 1/: is onto, and. that 961/: is one—one only if 95 is one—one; the product
of two one—one correspondences is also a one—one correspondence.
Another interesting exercise arises when we have mappings ([51, 952
from A into B, and mappings 5&1, 1/12 from B into 0'. The reader is invited
to prove that 915 is onto if and only if gin/:1 = 961,02 implies 1/11 = 51/2, and that
5!: is one—one if and only if 9512,11 = (#251: implies (t1 = 932; the mappings
(151, 452, {[11, 1,02 are to be regarded as arbitrary.
A most important property of the multiplication of mappings that has
just been described is its so-called associative property. Let A, B, C, D
be sets and let 95, 1/1, a) be mappings from A to B, from B to 0, from 0' to D
respectively. The property in question is expressed by the equation

¢(¢'w) = («Mm-
In terms of an arbitrary element x of A, this is just
(x¢)(¢'w) = {w(¢¢)}w;
the result of acting on x first with 9S and then with {/10} coincides With the
result of acting first with 451/; and then with w; the products 50w, 95¢]; are
defined as indicated above.
The associative law is intuitively very obvious. Its truth should be
manifest once its meaning is grasped. We give a formal account of its
proof for the sake of completeness, rather than with an intention of
thereby increasing understanding.
The proof is just a calculation based on the product definition. The
meaning of x/Jw shows that
(x¢)(z/1w) = {(x¢)¢}w,
While the meaning of ¢lp shows that
MM) = (MW!
and hence {$(¢¢I)}w = {(q):/J}w.
We therefore have (x¢)(z/Iw) = {x(q$1/x)}w,
both sides being equal to {(x¢)x/I}w. Since at was an arbitrary element of A
we have the desired equation (fix/1(4)) = (ch/Aw.
Note the special case of this result in which A = B = 0 = D, so that
(I), «[1, w are mappings from A into A. In that case products like l/la),
¢(1/;w), etc., are of course also mappings from A into A.
853137 R
 242 Append'ix

The identity mapping 1. of a set A onto itself is defined by xi 2 a: for

all x in A. It may perhaps not seem natural to regard this as a mapping
until one is convinced of its utility (this sort of feeling may arise from the
set 0, in another context).
Next, let 45 be a mapping from A to B and (/1 be a mapping from B to A,
and suppose that 96¢; is the identity mapping (on A); then we say that 1/1
is a right inverse of 95, and 95 is a left inverse of (/3. If, on considering git/I
and 1,095, we find that 90 is both a right inverse and a left inverse of (I) then
we call 1/: an inverse of ¢ and write 1/: = 43—1. The reader may like to find
examples of right or left inverses that are not inverses.
In fact gt has a right inverse if and only if (I) is one—one, and 95 has a left
inverse if and only if 9!; is onto; no proof of these statements will be
attempted. They imply that 4i has an inverse if and only if ¢ is a one—one
correspondence; this is almost trivial to verify independently.
The next concept we consider is that of a one—one mapping of a set A
onto itself; this sort of mapping is called a permutation of A. It is nothing
but a one—one correspondence of A with itself. Clearly a permutation has
an inverse, and clearly this inverse is also a permutation. We note that,
in the special case when A is a finite set, a mapping ¢ from A into A is
onto if and only if it is one—one; so either of these conditions is equivalent
to «)5 being a permutation of A.
Suppose A is a finite set. If A has only one element then the only
permutation of A is the identity mapping. How many permutations
does A have when it contains N elements? Suppose that there are
precisely pm) one—one mappings from one set An with n elements onto
another set B." also containing n elements, for each In 2 1; clearly p( 1) = 1.
Let n > 1. A definite element x of An may be mapped onto any one of
the elements of B”; therefore the image of a: may be chosen in 11. ways.
A one—one mapping of A” onto 3,, may then be completed by mapping
the other n— 1 elements of A onto the remaining n— 1 elements of B", in
a one—one manner, and this may be effected in precisely p(n— 1) ways.
Therefore pm) = np(n— 1) for n > 1. An easy induction shows that
p(n) = m, the well-known factorial function. Now take B" = A”.
The conclusion is that there are N! permutations on a set of N elements.
Special notation is needed to obtain such properties of permutations
as we require. It is traditional, but by no means essential, to regard
permutations as acting on a set whose elements are labelled with distinct
positive integers, if the set is finite; the positive integers or all the integers
may be used for countable sets. We note that these numbers appear only
as markers or labels, their additive and multiplicative properties are not
required.
Suppose then that p is a permutation of the set {1, 2,. . ., n}. The standard
notation for p is 1 2 n
( lp 2p i.I np > ' (1)
 Appemlix 243
An abbreviated notation is common and useful. Consider the integers
1, 1,0, 1,03,... . We must come to an integer m, where l g m g n, such
that 1pm = 1; for we must have a repetition of integers some time (they
number only n) and if lpml = 1pm with 0 g m1 < m2 then 1pm8_ml = 1.
The action of p on {1, lp,..., lpm—l} is officially given as
1 1p 1pm—1
(1p 1p2 1 )’ (2)
but we often abbreviate this to (1, lp,..., lpm—l) for convenience. Ifm < n
then the above process is repeated with 1 replaced by some integer not in
{1, lp,..., ham—1}. In this way p is associated with a number of cycles, as
permutations of the form (2) are called; and it is easy to see that these
cycles involve distinct symbols and that each pair of them commute as
permutations. The product of the cycles of p is in fact p. For example,
lete 123456
(421365)'
Then we write p in the form (1, 4, 3)(2)(5, 6). Often cycles like (2) involving
only one symbol are omitted; thus p = (l, 4, 3)(5, 6). The decomposition
just explained of a permutation into the product of cycles that contain
no symbol in common is unique, apart from the order in which the cycles
appear.
There is another important way of decomposing permutations. For
simplicity consider the special permutation p0 = (1, 2,...,m). It will be

““01 the“ P0 = <1,2>(1,3)...(1.m>.

To verify this statement is an easy exercise in multiplying permutations.
Permutations of the form (a, b), where a aé b, are called transpositions.
Every cycle is the product of suitable transpositions, as we have just
seen; but this product is not unique, symbols may be common to different
transpositions, and the transpositions in it may not commute. By way
of proof we note that
po = (m, 1)(m, 2)...(m,m— 1),
(12)(13) = (123) 75 (132) = (13)(12).
It is now clear, however, that any permutation, not only a cycle, is the
product of suitable transpositions.
There is a useful algorithm for calculating [3—1043 when a and 3 are given
permutations. We illustrate its use when a is Written in the form
(1 2 n)
1a 2a not '
The reader will find it easy to prove that ,8‘1043 is

(35,; 2255 :3: :55)-

 244 Appendix

In the case when a = (1, 2,...,m) we have B—lafi = (13,2/3,...,mfi), and

products of cycles can be treated on an obvious basis.
All the above remarks apply to quite general permutations; we have
not essentially assumed that any permutation affects only a finite number
of symbols.
Suppose now that we have a permutation p which does have the
property just mentioned, so that it leaves fixed all symbols except
a finite number, which we take to be {1,..., n}. Then p can be expressed
as the product of suitable transpositions. We define the parity of p to
be 1 if it is the product of an even number of transpositions, and — 1 if it
is the product of an odd number; the identity permutation is assigned
parity 1. It is clear that if parity is well defined then it has a multiplica-
tive property—namely, if p1, p2 have parities 101, 102 respectively then
P1P2 has parity 201202-
To prove that parity is well defined is a notoriously tricky businessxf
What has to be established is essentially this: no permutation can be
expressed both as the product of an even number of transpositions and
as the product of an odd number.
We first investigate the outcome of multiplying a permutation p by
a transposition 7-, with p expressed as the product of cycles each con-
taining more than one symbol and no two cycles containing a common
symbol. Let r = (a, b) with a :;E b. There are three cases.
(i) Neither a nor b appears in the expression for p. Then pr has an
obvious expression like that for p, but with the additional cycle (a, b).
(ii) Just one of {a, b}, say a, occurs in the expression for p. We may
write the cycle pa in which a occurs as (a, b1,..., bm). Then
p07 = (a, b1,...,bm)(a,b) = (a, b1,...,bm, b).
Thus the effect of right-multiplying p by -r is to place an extra symbol in
one cycle and to leave the others as they were.
(iii) Both a and b occur in the expression for p. Ifa and b occur in the
same cycle multiplication goes like this:
p01- = (a, b1,...,bm, b, 61,...,ck)(a,b)
= (a, b1,...,bm)(b,cl,...,ck).
However, a and b may also occur in different cycles:
p07- = ((1, b1,..., bm)(b, chm, ck)(a, b)
= (a,b1,...,bm,b, 01,...,ck).
Thus the number of cycles increases or decreases by one, and the number
of symbols appearing is unchanged.
1' For a fallacious demonstration on which a proof of invariance of parity
is based, see F. M. Hall, An introduction to abstract algebra, vol. i, Theorem
12.6.2 (Cambridge University Press, 1966).

[Not literally true; there [corrected in the second

are other cases such as edition]
r0 = (a, b, c1, ..., ck).]
 Appendix 245

This suggests defining a function P(p) as follows. Take the standard

expression for p, involving say n cycles containing m1,..., m" symbols
respectively, with each mi 2 2. Define
P(p) = (— 1)”"1‘1(— 1)“2—1 (— 1)”‘"‘1,
with P(L) = 1. Our remarks above show that if 1- is any transposition
then P(p7) = —P(p).
The way ahead is now clear. Suppose that a permutation p can be
expressed as the product of r transpositions. An obvious induction based
on the last paragraph shows that P(p) = (— D". But P(p) is independent
of the expression of p in terms of transpositions. It follows that P(p) = l
for one such expression if and only if P(p) = 1 for all such expressions.
In other words, p cannot have expressions as the product of an even
number of transpositions and of an odd number.
It follows that parity is well defined. In fact the function P(p) described
above is the parity ofp. A permutation is said to be even or odd according
as its parity is l or — 1.
Consider next the set R of all permutations that move only a finite
number of symbols in some set. Let Re, R0 denote the subset of even, odd
permutations respectively. We wish to prove that these subsets are in
one—one correspondence, assuming that the trivial case when R0 is
empty is avoided. Let 1- be any transposition in R0. Then any element
of ItD may be expressed as [27, for some p 6 Re ; the desired correspondence
is that in which p and p7- are matched. Note that in particular there are
13%! even and «En! odd permutations on a finite set with n elements,
providedn > 1.
The concept of a mapping of a set into itself admits a fruitful generaliza-
tion in the concept of a relation on a set. We shall define only the notation,
not the concept. If x, y are elements of a set S and a: is related to y then
we write x ~ y. We note that the conditions x N y and :0 ~ 2, for some
2 e 8, do not together imply y = 2, as in the case of a mapping; but that
a relation does have this in common with a mapping, that it determines
a subset of S X S.
Equivalence relations are of special interest in mathematics. The basic
reason for this is that it is often much easier to say when two objects have
the same property than to describe the property; consider the property
of being a right-handed (not left-handed) set of axes, or of having a
certain number of elements. Equivalence relations are the tool used in
comparing objects in respect of their properties.
To be specific, the relation ~ is an equivalence relation on S if and only
if it has the following three features.
(i) The reflexive property: 36 ~ 96.
(ii) The symmetric property: if 2: ~ y then 3/ N :16.
(iii) The transitive property: if 90 ~ g and 3/ ~ z then 91: ~ 2.
Here x, y, z are arbitrary elements of S. Note that (i) may be written
246 Appendix

as an implication, like (ii) and (iii), for it is equivalent to: if at: = y then
.71: ~ g.

A most important equivalence relation is that of congruence between

integers. The set concerned is the set Z of integers. Let n be a fixed
positive integer. We define the relation :1: a y, where m, y eZ, to mean
that x—y is divisible by n. The reader will find that verification of (i),
(ii), and (iii) requires only the most elementary properties of integers.
Suppose we have a partition of the set S. It should be clear that we
obtain an equivalence relation on S by defining x N y to mean that x
and y belong to the same element of the partition. The basic theorem
about equivalence relations is a converse of this: every equivalence
relation on S determines a partition of S.
To prove this we introduce the so-called equivalence classes. We
define [a] to be {xzm ~ a} for each a in S. Now each element of S lies in
some class, because a e [a] by (i). We therefore require to prove that no
element of S lies in two distinct classes. Suppose then that a e [b] and
a e [o]. By definition of equivalence class we have (1. ~ b and (1 ~ 6.
By (ii) we have b ~ (1 and a ~ 0. By (iii) we have b ~ 0. Now ifx e [b]
then :1: ~ b, and so :1: ~ 0 by (iii), that is x e [6]. Therefore [b] g [c],
similarly [c] g [b], and finally [b] = [c]. We have shown this: either
[b] n [c] = 9 or [b] = [c]. The theorem stated in the last paragraph is
therefore established.
Consider the equivalence classes defined in Z by congruence modulo n.
We see that [a] = {xzx E amodulo n}.
Therefore [a] may be described as the set
{..., a—n, a, a+n, a+2n,...}.
The distinct equivalence classes are [0],..., [n— 1]; note that [n] = [0].
We often refer to these classes as the residue classes modulo n, and we
call [0] the zero residue class. (A factor group is a set of equivalence
classes with a certain structure, and more generally a coset is an equi-
valence class.)
We mention a relation of quite a different sort from an equivalence
relation. A total ordering on a set S will be written thus: a: g y if x, y are
elements of S which the total ordering relates. We assume that if x, y E S
then either a: g y or y g :20 must hold, in addition to:
(i) 96 < 96;
(ii) ifxgyandy g xthenx2y;
(iii) ifx g yandy g zthenac g z.
The natural ordering of the real numbers leads to the most familiar
examples of total orderings.
Our final consideration in this Appendix concerns the beginnings of
number theory. We take it for granted that every integer greater than
1 can be expressed as the product of primes, uniquely apart from the order
 Appendix 247

of the primes. It thus makes sense to talk of the greatest common divisor
of two positive integers; if we suppose it to be positive then it is uniquely
defined. Two such integers are said to be coprime (or relatively prime)
when their greatest common divisor is 1. We shall also assume the
division algorithm: if d is a, positive integer and c is any integer then there
are integers 'y, 8 for which 6 = yd-f—S and 0 g 3 < d.
The so-called Euclidean algorithm is often referred to in elementary
group theory. Its basic form asserts that if the non-zero integers a, b are
coprime then there are integers a, [3 (not necessarily positive) for which
aa—l—l = 1.
We prove this. Our proof consists in a deduction of the Euclidean al-
gorithm from the statement that every non-empty set of positive
integers contains a least integer (which also implies the division algorithm,
by the way). Consider the set S of numbers,
S = {Au—I—‘ub: A eZ, p. eZ}.
Clearly S contains a non-empty subset of positive integers, and we
suppose that d is the least of these. Let a, ,8 be elements of Z such that
aa+Bb = d. Suppose further that c is any element of S, then A0 a+[1.0 b = c
for some A0, #0 in Z. By the division algorithm there are integers y, 8 such
that c = yd+8 and 0 g 3 < d. NOW 3 e S as the following calculation
shows: 8 _ 6—),d

= (A0 a‘i’llvo b)—y(aa+/3b)

= (Au—Ya)a+(Po—713)b-
The facts that d is the minimal positive element in S and that 0 < 8 < (1
now show that 3 = 0. That is, d divides c, and so the special element (1
divides every element of S. It is clear that a, and b are in S (take /\ = l,
p. = 0 and A = 0, y. = 1 respectively). Therefore d divides the two co-
prime integers a and b, and so d = 1. This completes the desired proof.
A generalization is at once possible: if h is the greatest common divisor
of the non-zero integers a and b, then there are integers a, )3 such that

curl—[36 = h.
The idea of the proof of this is to apply the preceding result to the integers
a/h, b/h; the reader will be able to fill in the details.
A further generalization concerns the case of n non-zero integers
a1,..., a," Whose greatest common divisor is 1. It is then possible to find
integers a1,..., can such that
a1a1+...+oc,,an = l.
The proof is by induction on In and uses the now known result for n = 2.
Suppose that n > 2 and that the result has been established for n— 1. If
the greatest common divisor of a1...., an_1 is k then h and a,” are coprime,
248 Appendix

in accordance with the hypothesis. Further, the greatest common divisor

ofa1/h, ..., c2L,,_1/h is 1, and the inductive hypothesis gives integers Oti: . .., 04,4
Suchthat a1a1+...+aq'1_1a,,,_1 = h.

It also gives integers BI, [32 such that

[31 71+]?2 a1. = 1.
We take a1 = 0431,.” 017,4 = 04,431, an = B2, and the inductive step
is complete.
Little more than the definition is needed as regards the use made of
complex numbers in this book. We remark that any such number may
be expressed in the form a+1§b where a, b are real and i is a particular
complex number for which i2 = —— 1. Alternatively, a non-zero complex
number may be expressed as r(cos 0—H sin 0) Where r > O and 0 g 0 < 211,
1' being called its modulus. We note de Moivre’s theorem:
{r(cos 0+i sin 0)}” = r"(cos n0+i sin n6),
for any integer n. There is an easy proof, using induction on n combined
with a special argument for the case n = — 1.
Note on Further Reading

THE present book on group theory is essentially introductory. After

mastering it the young mathematician should be ready for the more
specialized works listed below; however, of the more advanced
general works still in circulation we mention the books of Kurosh and
Hall, emphasizing infinite and finite group theory respectively.
The student of finite groups will have to learn a great deal about
group representations, on which the book of Curtis and Reiner is an
excellent treatise. Huppert is writing a very comprehensive work on
finite groups, including lengthy accounts of representations, coho-
mology, permutation groups, etc. ; the first volume on finite soluble
groups has just become available, and a second volume on finite
simple groups is promised. '
A thorough grounding in ordinals and cardinals, as well as use of
equivalents of Zorn’s lemma, is necessary before more about infinite
groups can be tackled; this material may be learnt from the many
good available books on infinite set theory. The book of Magnus,
Karrass, and Solitar will then be found first-rate for topics which are
generalizations of freeness in groups and of properties defined by
series. More special, but none the less an attractively written book,
is Hanna Neumann’s monograph on varieties of groups.
Wielandt’s work affords an introduction to permutation groups,
Fuchs’ book is a standard work on abelian groups, and that of
Dickson, though old, is still a good account of linear groups.
No heed has been taken in the present book of applications of
groups to physics, and we may remark that it is representations of
topological groups that are usually required; good references here are
the books of Boerner and Hamermesh, the former being on the
mathematical side and the latter more physical.
H. BOERNER, Representations of groups with special consideration for
the needs of modern physics (North-Holland, 1963).
C. W. CURTIS and I. REINER, Representation theory of finite groups
and associative algebras (Interscience, 1962).
L. E. DICKSON, Linear groups with an exposition of the Galois field
theory (Dover, 1958).
250 Note on Further Reading

L. FUCHS, Abelian groups (Publishing house of the Hungarian

Academy of Sciences, 1958).
M. HALL, The theory of groups (Macmillan, 1959).
M. HAMERMESH, Group theory and its application to physical problems
(Addison-Wesley, 1962).
B. HUPPERT, Endliche Gruppen I (Springer-Verlag, 1967).
A. G. KUROSH, The theory of groups, vols. i and ii (Chelsea, 1955,
1956).
W. MAGNUS, A. KARRASS, and D. SOLITAR, Combinatorial group
theory (Interscience, 1966).
H. NEUMANN, Varieties of groups (Springer-Verlag, 1967).
H. WIELANDT, Finite permutation groups (Academic Press, 1964).
Index

Algorithm, division, 247. Degree of alternating group, 28.

Euclidean, 247. of symmetric group, 28.
Anti-automorphism, 33. Divisible, 100, 228.
Automorphism, 30, 71, 72. Divisor, greatest common, 247.
inner, 71, 72.
outer, 71.
Automorphisms, groups of, 31, 72, Element, identity, 1, 2, 16 et seq.
80, 81, 207, 217—20, 224—7, 237. inverse, l, 2, 16 et seq.
Axioms, group, l—... order of, 44.
unit, 34.
Bases, conjugate Sylow, 205 et seq. zero, 83.
Basis, free, 156 et seq. Elementary abelian, 96.
Basis, Sylow, 205 et seq. Elements, commuting, 7.
Exponent, 188.
Cancellation, 17, 19, 151. Extension, split, 223 et seq.
Cayley, 26, 74 n.
Centralizer, 61, 79, 81.
Centre, 67 et seq. Factor, chief, 197.
Class, conjugacy, 61 et seq. composition, 114.
equivalence, 246. Factor group, 56 et seq.
nilpotency, 172. proper, 57.
residue, 4, 246. Field, 222, 228, 238.
transitivity, 145. First isomorphism theorem, 58.
Closure, 1, 36, 240. Fourth isomorphism theorem, 110.
Commutator, 65 et seq. Frattini, 133, 185.
simple, 181. Function, 240.
Complex, 34 et seq. Functions, groups of, 118—19.
conjugate, 60 et seq.
inverse, 38, 41, 47.
normal, 60, 63, 64. Generators, 40 et seq., 149 et seq.,
Complex number, 248. 186—8.
Complex numbers, additive group of, free, 156 et seq.
3, 215. Group, abelian, 7, 82 et seq.
multiplicative groups of, 4, 215. abstract, 26, 28.
Complexes, commuting, 35, 53, 61. alternating, 9, 28, 54, 114—15, 220-
groups of, 36—7, 52, 56. 1.
Congruence, 246. axioms for, 1—2.
Conjugate, 60 et seq., 205. commutative, 7.
Coprime, 247. cyclic, 49 et seq., 217—20.
Correspondence, one—one, 155, 240. derived, 66 et seq.
Coset, left, 43 et seq. dihedral, 11—12, 216.
right, 42 et seq. divisible abelian, 100, 228.
Cycle, 243. factor, 56 et seq.
finite, 27.
Deficiency, 163 et seq. finitely generated, 40, 233.
252 Index

Group (cont.) Law, associative, 1, 18—19, 241.

finitely presented, 163. distributive, 83, 222.
free, 153 et seq. reflexive, 245.
free abelian, 159. symmetric, 245.
general linear, 222, 238. transitive, 245.
generalized quaternion, 226. Lemma, Frattini’s, 133.
hamiltonian, 229. Zassenhaus’s, 110.
infinite dihedral, 126, 227. Length, 112, 150, 156, 196.
linear fractional, 222, 228.
locally nilpotent, 194. Mapping, 240.
metabolism, 196. congruence, 9 et seq., 216.
metacyclic, 124. identity, 242.
nilpotent, 172 et seq. inverse, 8, 23 et seq., 242.
order of, 44. one—one, 8, 240.
periodic, 45. onto, 8, 240.
projective general linear, 222. Mappings, groups of, 7 et seq., 216.
projective special linear, 222. product of, 241.
quaternion, 13, 217. Matrix, 12.
quotient, 56 et seq. permutation, 29.
residually finite, 167, 234. unitriangular, 15.
restricted symmetric, 9, 79, 221. Matrices, groups of, 13—14, 28-30,
simple, 114—15, 143—4, 220—2. 217, 222.
soluble, 196 et seq.
n-tuple, 240.
special linear, 222, 228.
Non-zero complex numbers, multi-
supersoluble, 207 et seq. plicative groups of, 4, 215—16.
symmetric, 8, 28, 51, 54.
Non-zero rationals, multiplicative
torsionfree, 45.
group of, 3, 215.
Non-zero reals, multiplicative group
Holomorph, 226—7. of, 3, 215.
Homomorphism, 21 et seq. Non-zero residue classes, multiplica-
natural, 58. tive groups of, 6, 217 et seq.
Homomorphisms, groups of, 32, 101. Normalizer, 61, 81, 205.
Notation, additive, 82—3.
Identities, commutator, 180, 189. Number, complex, 248.
Inclusion, set, 239.
Index, 47 et seq. Operation, 240.
Integer, 239, 246—8. binary, 1, 2, 240.
positive, 239, 246—8. Orbit, 145.
Integers, additive group of, 2, 215. Order, 44.
Invariant, 93. Ordering, total, 117, 246.
Inverse, right, 242.
p-group, 84, 129, 175, 185, 187—95.
left, 242.
elementary abelian, 96.
Isomorphism, 23 et seq. p-length, 235.
Isomorphism theorem, first, 58.
p-series, upper, 235.
fourth, 110.
p-subgroup, Sylow, 129.
second, 107.
pair, ordered, 240.
third, 109.
parity, 9, 244, 245.
partition, 240, 246.
Kernel, 55 et seq. permutation, 8, 242.
even, 244, 245.
Lagrange, 46. odd, 244, 245.
 Index 253

permutations, groups of, 8~9, 26—31, Series, central, 172 et seq.

145—8, 220—1. chief, 197.
w-group, 129. composition, 114.
air-number, 129. derived, 198.
w-subgroup, Hall, 203. invariant, 196.
Sylow, 129. isomorphic invariant, 197.
Positive rationals, multiplicative isomorphic normal, 113.
group of, 3, 215. lower central, 178.
Positive reals, multiplicative group normal, 112.
of, 3, 215. upper central, 178.
Power, 20—1. Set, 239.
Presentation, finite, 163. countable, 240.
group, 160 et seq. empty, 239.
Prime, relatively, 247. totally ordered, 117, 246.
Problem, Burnside, 188. Sets, intersection of, 239.
Product, cartesian, 118. set of, 57, 240.
central, 230. union of, 239.
complex, 35 et seq., 117. Socle, 235.
direct, 73 et seq., 119. Stabilizer, 145.
external direct, 78. Subgroup, 34 et seq.
internal direct, 78. characteristic, 72 et seq.
semidirect, 225. commutator, 66.
subcartesian, 236. Frattini, 185.
Property, residual, 166—8. index of, 47.
maximal, 138.
Quotient group, 56 et seq. minimal normal, 202.
proper, 57. normal, 53 et seq.
proper, 34.
pure, 230.
Rank, 94, 155, 159, 196. quasinormal, 232.
Rational, 239. Subset, 239.
Rationals, additive group of, 2, proper, 239.
215. Sum, cartesian, 126, 223.
Reals, additive group of, 3, 215. direct, 83 et seq., 233.
Refinement, 112, 197. Support, 119.
Reflection, 9—12, 216.
Relation, 149 et seq., 245—6.
Relation, equivalence, 245—6. Theorem, Cayley’s, 26.
Representation, right regular, 27, de Moivre’s, 23, 40, 216, 248.
142, 228. first isomorphism, 58.
Representatives, coset, 44. fourth isomorphism, 110.
Residue class, 4, 246. Jordan—Hfilder, 115, 125.
Residue classes, additive group of, 5, Lagrange’s, 46.
215. Schur—Zassenhaus, 140.
multiplicative groups of, 6—7, 217— second isomorphism, 107.
20. Sylow’s first, 128, 147.
Ring, 82—3. Sylow’s second, 131, 137, 148.
Rotation, 10—12, 77, 216. Sylow’l third, 131, 137, 148.
third isomorphism, 109.
Schur, 139. three-subgroup, 195.
Second isomorphism theorem, 107. von Dyck’s, 160.
Semigroup, 7, 13, 15, 27, 33. Third isomorphism theorem, 109.
254 I ndex

Translation, 9—11, 77, 216. Word, 150 et seq.

Transposition, 9, 51, 243. reduced, 153 et seq.
Transversal, left, 44, 47, 237. Words, adjacent, 151.
right, 44, 47, 61, 237. equivalent, 152.
Type, 94. product of, 151.

Vector, row, 12, 28-30. Zassenhaus, 110, 140.

 PRINTED IN GREAT BRITAIN
AT THE UNIVERSITY PRESS, OXFORD
BY VIVIAN RIDLER
PRINTER TO THE UNIVERSITY