Notes on Group Theory

Mark Reeder
March 7, 2014
Contents
1 Notation for sets and functions 4
2 Basic group theory 4
2.1 The denition of a group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Group homomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Cosets and quotient spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.5 Normal subgroups and quotient groups . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.6 The rst isomorphism theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.6.1 Exact sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.7 The second isomorphism theorem (correspondence theorem) . . . . . . . . . . . . . . 10
2.8 The third isomorphism theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.9 Direct products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.10 Semidirect products (internal view) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.11 Conjugacy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Basic Examples of Groups. 14
3.1 Cyclic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Finite abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1
3.2.1 Unit Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 The symmetric group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Linear groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.4.1 Conjugacy classes in GL
2
(F) . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.4.2 S
n
as a subgroup of GL
n
(F) . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.4.3 The Bruhat decomposition for GL
n
(F). . . . . . . . . . . . . . . . . . . . . . 27
3.5 The dihedral groups D
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.6 The quaternion and generalized quaternion groups Q
4n
. . . . . . . . . . . . . . . . . 30
3.7 p-groups, a rst look . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.8 Simple groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.8.1 Simplicity of alternating groups . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8.2 Simplicity of PSL
2
(F) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.9 Exceptional isomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4 Group actions 38
4.1 The left regular action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.2 Group actions on coset spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 Applications to simple groups . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Actions by S
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.4 Actions by GL
n
(F) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.5 Double cosets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.6 Conjugation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5 Sylow Theorems and Applications 47
5.1 Sylow p-subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.1.1 Small examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.1.2 Groups of order pq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Sylow subgroups in GL
n
and ag varieties . . . . . . . . . . . . . . . . . . . . . . . . 53
2
5.3 The Burnside Transfer Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Simple groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.4.1 The simple group of order 60 . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.4.2 The simple group of order 168 . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.4.3 Simple groups of order 720 . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.4.4 Almost-simple groups of order 720 . . . . . . . . . . . . . . . . . . . . . . . 70
6 Solvable and nilpotent groups 73
7 p-groups, a second look 77
7.1 Groups of order p
3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.1.1 Automorphisms of the Heisenberg group . . . . . . . . . . . . . . . . . . . . 80
7.2 Higher powers of p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.3 Projective limits and pro-p groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.4 Toward the classication of p-groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8 Presentations of Groups 86
8.1 Free Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.2 Generators and Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.3 A presentation of the symmetric group . . . . . . . . . . . . . . . . . . . . . . . . . . 89
8.4 Coxeter groups and reection groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
8.5 Presentations of alternating groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
8.5.1 A presentation of A
5
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
8.5.2 The exceptional isomorphism PSL
2
(9) A
6
. . . . . . . . . . . . . . . . . . 96
8.6 The Platonic Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9 Building new groups from old 98
9.1 Automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
9.1.1 Automorphisms of S
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3
9.2 Semidirect Products (external view) . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.2.1 Groups of order p
2
q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
9.3 Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.4 Metacyclic groups and extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
1 Notation for sets and functions
For any set S, we write [S[ for the number of elements in S if S is nite, and put [S[ = if S is
innite. The empty set is denoted S = .
If f : S T is a function, we write f(s) or f
s
for the value of f at an element s S. The set
imf = f(S) = f(s) : s S of these values is the image of f and f
1
(t) = s S : f(s) = t
for the ber of f over an element t T. Thus, imf = t T : f
1
(t) ,= . For any subset T

T,
the set f
1
(T

) = s S : f(s) T is the union of the bers f

1
(t) for t T.
A function f : S T is injective if [f
1
(t)[ 1 for all t T, in this case we sometimes write
f : S T to emphasize injectivity, and to indicate that we may identify S with its image in T.
A function f : S T is surjective if [f
1
(t)[ 1 for all t T, in this case we sometimes write
f : S T to emphasize surjectivity.
Finally, f is bijective if it is both injective and surjective, that is, if [f
1
(t)[ = 1 for all t T. When
this holds, we have [S[ = [T[. We sometimes write f : S

T to indicate a bijection, or S T to
mean that there exists a bijection between S and T.
2 Basic group theory
2.1 The denition of a group
A group is a set G together with a function : G G G, assigning to each pair (a, b) of elements
of G another element a b G, satisfying the following three axioms:
G1 (associativity) We have a (b c) = (a b) c, for all a, b, c G.
G2 (existence of identity) There exists an element e G such that e a = a e = a for all a G.
G3 (existence of inverses) For all a G there exists an element a

G such that a a

= a

a = e,
where e is an identity element as in axiom G2.
There is no requirement that a b = b a for all a, b G. If this property does hold, we say that G is
abelian.
4
The element e of axiom G2 is unique: for if e

is another identity element, then e

= e

e = e by
applying axiom G2 rst to e, then to e

. For each a G, the inverse element a

in axiom G3 is unique:
for if a

is another inverse element, we have

a

= a

e = a

(a a

) = (a

a) a

= e a

= a

,
by applying successively axioms G2, G3 (for a

), G1, G3 (for a

) and nally G2 again.

We usually use multiplicative notation and abbreviate a b as ab or a b, and write 1 or 1
G
instead of
e for the identity element of G, and a
1
instead of a

for the inverse element of a. For any positive

integer n, we write a
n
for the product of a with itself n times, and a
n
for the product of a
1
with itself
n times. Finally, we put a
0
= 1.
If Gis abelian, and only in this case, we sometimes use additive notation, writing ab = a+b, denoting
the identity element by 0, and the inverse element of a by a.
The order of Gis the cardinality [G[, either a positive integer or . Any group Gof order one consists
of the identity element only and is called the trivial group.
We can extend the product of elements in G to subsets: If S and T are subsets of a group G, we dene
ST = st : s S, t T.
2.2 Group homomorphisms
The structure of a group G is revealed by its subgroups and homomorphisms into other groups.
A homomorphism of groups G, G

is a function f : G G

satisfying
f(ab) = f(a)f(b) a, b G.
This implies that f(1
G
) = 1
G
and that f(g
1
) = f(g)
1
for all g G.
The kernel of a homomorphism f : G G

is the subset of G dened by

ker f := g G : f(g) = 1
G
.
For all a, b Gwe have f(a) = f(b) if and only if ab
1
ker f. Hence f is injective iff ker f = 1
G
.
The image of f is the set theoretic image, dened as above by
imf := f(G) = g

G : f
1
(g

) ,= .
There may be many homomorphisms between two given groups. We set
Hom(G, G

) = homomorphisms f : G G

.
An isomorphism f : G G

is a bijective group homomorphism. Thus f is an isomorphism if and

only if ker f = 1
G
and imf = G

. We sometimes write f : G

G

to indicate that f is an
5
isomorphism. Two groups G, G

are isomorphic if there exists an isomorphism f : G

G

. We
write G G

to indicate that G and G

are isomorphic, without specifying any paticular isomorphism

between them.
We sometimes abuse terminology and say that G is or is a copy of G

, when we really mean only that

G G

. For example, any two trivial groups are isomorphic, so we allow ourselves to say that the
trivial group is the unique group with one element.
Continuing in this vein, we say that a homomorphism f : G G

is trivial if imf = 1
G
. This is
equivalent to having ker f = G, so being a trivial homomorphism is the opposite of being an isomor-
phism. Hence trivial homomorphisms are just as important as isomorphisms.
An isomorphism from a group to itself is called an automorphism. We set
Aut(G) = automorphisms of G.
The set Aut(G) forms a group under composition, whose identity element is the identity automor-
phism, which sends g g for all g G.
Many automorphisms arise from within the group itself as follows. For each element g in a group G,
the map
c
g
: G G given by c
g
(x) = gxg
1
x G
is an automorphism of G, called conjugation by g. Automorphisms of this form are called inner
automorphisms. The function c : G Aut(G) sending g c
g
is a homomorphism.
2.3 Subgroups
A subgroup of G is a subset H G with the following three properties:
SG1 (closure) ab H for all a, b H.
SG2 (identity) The identity element of G is contained in H.
SG3 (inverses) For all a H we have a
1
H.
The subsets 1 and G are subgroups of G. All other subgroups of G, if any, are called proper
subgroups. We write H G to indicate that H is a subgroup of G which is possibly equal to G itself.
We write H < G for a subgroup which is not equal to G.
Lemma 2.1 Let G be a group and let H be a nonempty nite subset of G. Then H G if and only if
SG1 holds.
Proof: Let h be an element of the nonempty set H. Since H is nite, the powers h, h
2
, h
3
, . . . must
eventually repeat, so we have h
i
= h
j
for some positive integers i < j. It follows that h
ji
= 1, so
SG2 holds, and h h
ji1
, so SG3 holds. Hence H is a subgroup of G.
6
This proof moved from H to a particular kind of subgroup of G. A group C is cyclic if there exists an
element c C such that C = c
n
: n Z. In this case we write C = c.
In an arbitrary group G, any element g G is contained in the cyclic subgroup
g = g
n
: n Z.
The order of g is the order of the group g. The order of g is the smallest positive power m such that
g
m
= 1, if such an m exists. In this case, the order can be characterized by the useful property that for
any integer d, we have g
d
= 1 iff m [ d. If g
d
,= 1 for any nonzero integer d, we say the order of g is
innite.
More generally, if S is any subset of G, the subgroup generated by S is the smallest subgroup S of
G containing S. More precisely,
S =

SHG
H
is the intersection of all subgroups H of G which contain the subset S.
2.4 Cosets and quotient spaces
A left coset of a subgroup H < G is a subset of G of the form gH = gh : h H. Two left cosets
are either equal or disjoint; we have
gH = g

H g
1
g

H.
In particular, we have gH = H if and only if g H. The set of left cosets of H in G is denoted G/H,
and is called the quotient of G by H.
A right coset of H in G is a subset of the form Hg = hg : h H. Two right cosets are either equal
or disjoint; we have
Hg = Hg

g
1
g

H.
In particular, we have Hg = H if and only if g H. The set of right cosets of H in G is denoted
G/H.
A coset is a left or right coset. Any element of a coset is called a representative of that coset. We have
canonical bijections H gH and H Hg, sending h gh and h hg, respectively. Hence if H
is nite, all cosets have cardinality [H[.
There are an equal number (including innity) of left and right cosets in G. We denote this number by
[G : H] = [G/H[ = [HG[,
and call it the index of H in G. If G is nite, then G is partitioned into [G : H] cosets, each of
cardinality [H[. It follows that [G : H] = [G[/[H[. In particular we have
Lagranges Theorem: If H is a subgroup of a nite group G, then [H[ divides [G[.
7
Example 1: If [G[ = p a prime, then G has no proper subgroups. Hence for any nonidentity element
g G we have G = g, so G is cyclic.
Example 2: The order of any element in a nite group G divides [G[. in particular, we have g
|G|
= 1
for all g G. The smallest positive integer e such that g
e
= 1 is called the exponent of G. If g has
order m, then m divides e, which in turn divides [G[.
The converse of Lagranges theorem is false. The smallest counterexample is the group A
4
of order 12,
which has no subgroup of order 6. However, the converse of Lagranges theorem is true for subgroups
H of prime power order. This is part of the Sylow theorems, which we will prove later. However, one
special case is easy:
Proposition 2.2 Any group of even order contains an element of order two.
Proof: Suppose G has even order [G[ = 2m. Pair the nonidentity elements with their inverses. Since
there are 2m1 such elements, at least one of them is paired with itself. This is a nonidentity element
g G such that g = g
1
. Thus, g is an element of G of order two.
2.5 Normal subgroups and quotient groups
Let G be a group and let H G be a subgroup. One attempts to dene a group structure on the set
G/H by the rule:
gH g

H = gg

H, g, g

G. (1)
However, this rule is only well-dened when every left coset of H in G is also a right coset of H in G.
The subgroup H is said to be normal in G if gH = Hg for all g H. On the level of elements, this
means that ghg
1
H for all g G and h H. If G is abelian, then ghg
1
= h, so every subgroup
is normal in G. Thus, being normal in G is a weakening, with respect to H, of the abelian condition.
We write H G or H G to indicate that H is a normal subgroup of G.
When, and only when H G, the set G/H = HG becomes a group under the operation given by
(1). We call G/H the quotient of G by H. It is a group of order equal to the index of H in G:
[G/H[ = [G : H].
Example: The center of a group G is the subgroup
Z(G) = z G : zg = gz g G.
This is clearly a normal subgroup of G. We will see the quotient group G/Z(G) appearing in several
contexts. One useful fact is
Proposition 2.3 If G/Z(G) is cyclic then G is abelian.
Proof: Exercise.
8
2.6 The rst isomorphism theorem
Any group homomorphism f : G G

induces an isomorphism from a quotient of G to a subgroup

of G

. More precisely, we have the following.

Theorem 2.4 (First isomorphism theorem) Let f : G G

be a group homomorphism with kernel

K = ker f. Then the kernel K = ker f is a normal subgroup of G, and there is an isomorphism

f : G/K

imf, given by

f(gK) = f(g). (2)
Proof: It is a good exercise to check that

f is well-dened and bijective.
It is useful to have this result in slightly more general form:
Theorem 2.5 Let f : G G

be a group homomorphism with kernel K = ker f. Let H be a normal

subgroup of G contained in K. Then there is a surjective homomorphism

f : G/H

imf, given by

f(gH) = f(g), (3)
with ker

f = K/H.
Proof: Again, this is a good exercise.
We often say that

f is induced by f or that f factors through G/H.
Conversely, every normal subgroup H G is the kernel of a surjective homomorphism from G into
another group. Namely, the canonical homomorphism

H
: G G/H, given by
H
(g) = gH
is surjective with ker
H
= H.
Example: If x, y are two elements of G, the commutator
[x, y] = xyx
1
y
1
is an element of G that measures the failure of x, y to commute. The Commutator Subgroup
[G, G] = [x, y] : x, y G
is the subgroup generated by all commutators. For g G we have g[x, y]g
1
= [gxg
1
, gxg
1
]. It
follows that [G, G] G. The quotient G/[G, G] is called the abelianization of G, and is often denoted
G
ab
, because of the following result.
Proposition 2.6 The quotient G/[G, G] of a group G by its commutator subgroup is abelian. More-
over, G/[G, G] is the largest abelian quotient of G, in the following sense: If f : G A is a homo-
morphism from G to an abelian group A, then [G, G] < ker f, so f factors through a homomorphism

f : G/[G, G] A.
Proof: Exercise.
9
2.6.1 Exact sequences
A composition of group homomorphisms
G
1
f
1
G
2
f
2
G
3
is exact at G
2
if imf
1
= ker f
2
. A sequence of group homomorphisms
. . . G
i1
f
i1
G
i
f
i
G
i+1
. . .
is an exact sequence if it is exact at G
i
for all i. A short exact sequence is a sequence
1 G
1
f
1
G
2
f
2
G
3
1
with
ker f
1
= 1, imf
1
= ker f
2
G
1
, imf
2
= G
3
G
2
/G
1
.
2.7 The second isomorphism theorem (correspondence theorem)
The subgroups of a quotient group G/H are related to subgroups of G as follows.
Theorem 2.7 Let G be a group with normal subgroup H G, and let
H
: G G/H be the
canonical homomorphism.
1. If K is any subgroup of G containing H, then H K and K/H =
H
(K) is a subgroup of
G/H.
2. Conversely, if J is any subgroup of G/H, then
1
H
(J) is a subgroup of G containing H as a
normal subgroup and J =
1
H
(J)/H.
3. K/H G/H if and only if K G, in which case (G/H)/(K/H) G/K.
Thus, we have a one-to-one correspondence
subgroups of G containing H subgroups of G/H
K K/H

1
H
(J) J,
and this correspondence preserves normal subgroups.
There is also a correspondence theorem for homomorphisms, the rst part of which is just Thm. 2.5
above.
Theorem 2.8 Let G be a group with normal subgroup H G.
10
1. If f : G G

is a group homomorphism with H ker f then f induces a well-dened homo-

morphism

f : G/H G

, given by

f(gH) = f(g).
2. If : G/H G

is any homomorphism, then

H
: G G

is a homomorphism whose
kernel contains H.
Thus, we have a one-to-one correspondence
Hom(G/H, G

) f Hom(G, G

) : H ker f

H

f f
2.8 The third isomorphism theorem
The nal isomorphism theorem concerns products of subgroups. If H and K are subgroups of a group
G, the product HK = hk : h H, k K contains the identity, but it need not be closed under
the operation in G, hence HK need not be a subgroup of G. However, it will be so under an additional
condition.
The normalizer of H in G is the subgroup
N
G
(H) = g G : gHg
1
= H.
We have N
G
(H) = G if and only if H G. In general, N
G
(H) is the smallest subgroup of G
containing H as a normal subgroup.
Returning to our two subgroups H, K in G, let us assume that
K N
G
(H).
(This assumption holds automatically if H G.) Then, for h, h

H and k, k

K, we have
(hk)(h

) = h(kh

k
1
) kk

(where we insert () and to help parse the product), and kh

k
1
belongs to H since k N
G
(H), so
(hk)(h

) HK. Similarly, (hk)

1
= k
1
h
1
= (k
1
h
1
k) k HK. Hence HK is a subgroup of
G. Since both H and K are contained in N
G
(H), it follows that HK is also contained in N
G
(H). In
other words, H is normal in HK.
This proves the rst part of the following
Theorem 2.9 (Third Isomorphism Theorem) Let H and K be subgroups of a group G and assume
that K N
G
(H). Then
1. HK is a subgroup of G and H is a normal subgroup of HK.
11
2. H K is a normal subgroup of K.
3. We have a group isomorphism K/(K H) HK/H, induced by the map f : K HK/H
given by k kH.
Proof: We have already proved the rst part, and the second part is easy. As for the third part, it is
clear that f is surjective, so it remains to determine ker f. Let k K. Then f(k) = 1 in HK/H iff
(k) H, which means that k H. But k K, so we have f(k) = 1 iff k H K, as claimed.
2.9 Direct products
Let H and K be groups with identity elements 1
H
and 1
K
. Then the direct product
H K = (h, k) : h H, k K
is a group under the operation
(h, k)(h

, k

) = (hh

, kk

)
and identity element (1
H
, 1
K
). The direct product of nitely many groups G
1
, . . . , G
n
is dened simi-
larly; we conne our discussion to the case n = 2.
Let G = H K, and write 1 = (1
H
, 1
K
). The maps : H G and : K G given by
(h) = (h, 1
K
) and (k) = (1
H
, k) are injective homomorphisms. Their images H

= (H) H
and K

= (K) K are normal subgroups of G such that

H

= 1 and H

= G.
Conversely, this can be used to recognize direct products as follows.
Proposition 2.10 Let G be a group with subgroups H and K. Assume that
1. H and K are both normal in G;
2. H K = 1;
3. G = HK.
Then G H K, via the map f : H K G given by f(h, k) = hk .
Proof: Let h H, k K and parse the commutator [h, k] = hkh
1
k
1
in two ways. On the one
hand, [h, k] = (hkh
1
)k
1
K since K G. On the other hand, [h, k] = h(kh
1
k
1
) H, since
H G. But H K is trivial, so [h, k] = 1. Hence h and k commute for all h H and k K. It is
now immediate that f is a homomorphism, which is surjective by assumption 3. Finally, if f(h, k) = 1,
we have h = k
1
H K = 1, so h = k = 1. Therefore f is an isomorphism, as claimed.
If H and K are abelian groups then we often write H K instead of H K, in accordance our use of
additive notation for abelian groups.
12
2.10 Semidirect products (internal view)
Recall that a group G with two normal subgroups H, K G is the direct product G = H K iff
HK = G and H K = 1. This situation often occurs with the variation that only one of the
subgroups is normal.
Denition 2.11 A group G is a semidirect product of two subgroups H, K G if the following
conditions hold.
1. One of the subgroups H and/or K is normal in G.
2. H K = 1.
3. HK = G.
Suppose G is the semidirect product of H and K and (say) H is normal in G. On the set HK, dene
a group law as follows:
(h, k)(h

, k

) = (hkh

k
1
, kk

).
Let H K denote the set H K with this group law.
Proposition 2.12 If G is a semidirect product of two subgroups H and K with H G, then the map
(h, k) hk is a group isomorphism
H K

G.
Proof: exercise.
2.11 Conjugacy
For g, x G, let us set
g
x = gxg
1
.
Two elements x, y G are conjugate in G if y =
g
x for some g G. Conjugacy is an equivalence
relation on G, whose equivalence classes are the conjugacy classes of G. Thus any group is partitioned
into conjugacy classes. We write
G
x =
g
x : g G
for the conjugacy class of x in G.
Some of our earlier notions can be expressed in terms of conjugacy. For example, we have x Z(G)
if and only if
G
x = x. And H G if and only if H is a union of conjugacy classes in G.
The centralizer of a given element x G is the subgroup C
G
(x) = h G :
h
x = x consisting of
all elements in Gwhich commute with x. This is generally not a normal subgroup of G, so the quotient
space G/C
G
(x) is not a group. However, we have
13
Proposition 2.13 For every x G, the map g
g
x induces a well-dened bijection
G/C
G
(x)

g
x.
In particular, if G is nite, we have
[
G
x[ = [G : C
G
(x)] =
[G[
[C
G
(x)[
.
Proof: Exercise
The last formula in Prop. 2.13 is very useful for computing the sizes of conjugacy classes and central-
izers in nite groups. It implies for example that [
G
x[ divides [G[. Since G is the disjoint union of its
conjugacy classes, we have
Corollary 2.14 (The class equation) Let G be a nite group, let X
1
, . . . , X
k
be its conjugacy classes,
choose x
i
X
i
for 1 i k, and let G
i
= C
G
(x
i
). Then we have
k

i=1
[G : G
i
] = [G[.
Alternatively,
1
[G
1
[
+
1
[G
2
[
+ +
1
[G
k
[
= 1.
Corollary 2.15 If [G[ is a power of a prime p then G has nontrivial center.
Proof: We have G partitioned as G = Z(G) noncentral elements. Every conjugacy class has
size a power of p. This power is zero precisely for those classes consisting of a single element
in Z(G) and every conjugacy class of noncentral elements has size a positive power of p. Hence
[noncentral elements[ is divisible by p. As [G[ is also divisible by p, it follows that p divides [Z(G)[.
Since [Z(G)[ 1, it follows that a positive power of p divides [Z(G)[.
3 Basic Examples of Groups.
We give brief introductions to the most fundamentally important groups here.
3.1 Cyclic groups
Let Z be the group of integers under addition, with identity element 0. Since addition of integers is
commutative, the group Z is abelian. Using the division algorithm, one proves that for any subgroup
14
H Z, there is an integer n 0 such that H = nZ, the set of multiples of n. If n = 0 we have
H = 0. Assume now that n 1. Then the quotient Z/nZ is nite of order n, and consists of
the cosets nZ, 1 + nZ, . . . , (n 1) + nZ. The subgroups of Z/nZ correspond to the subgroups of Z
containing nZ. These are the subgroups dZ/nZ, for positive integers d [ n. Note that multiplication
by d
1
induces an isomorphism
dZ/nZ Z/d
1
nZ.
Every subgroup is normal in Z, and we have
(Z/nZ)/(dZ/nZ) Z/dZ.
Cyclic groups play an important role in any group G. For each element g G determines a homomor-
phism
e
g
: Z G, given by e
g
(n) = g
n
.
The image of e
g
is the subgroup generated by g
ime
g
= g = g
n
: n Z.
The kernel of e
g
is a subgroup of Z, hence is of the form mZ, for some integer m 0. If m = 0 then
g has and g Z. If m > 0 then m is the order of g.
A group G is cyclic if G = g for some g G. That is, G is cyclic iff there exists g G such that
every element of G is a power of g. We can also say that G is cyclic iff there exists g G such that the
homomorphism e
g
: Z G is surjective.
We have seen that every innite cyclic group G is isomorphic to Z, and every nite cyclic group of
order n is isomorphic to Z/nZ. Usually we will encounter cyclic groups while we are working with
multiplicative notation, where we will let C
n
denote a generic cyclic group of order n. Thus, C
n
= g,
for any element g G of order n.
Since C
n
Z/nZ, it is immediate from our discussion of Z and its subgroups that C
n
has a unique
subgroup of every order d dividing n, namely g
d
C
d
. Every subgroup of C
n
is of this form and we
have
C
n
/C
d
C
n/d
.
This is a complete description of all of the subgroups and quotients of C
n
.
Example: Let m, n be positive integers. The subgroup C
n
[m] = x C
n
: x
m
= 1 of C
n
is the
kernel of the homomorphism C
n
m
C
n
sending x x
m
, so it ts into the exact sequence
1 C
n
[m] C
n
C
m
n
1.
If g is a generator of C
n
, one can check that C
n
[m] = g
n/d
, where d = gcd(m, n). Thus, C
n
[m] C
d
and C
m
n
C
n/d
.
A nal remark on cyclic groups: The word isomorphic does not mean equal. We have Z/nZ C
n
,
but there is a subtle distinction between these groups: Note that Z/nZ has a canonical generator,
15
namely 1 + nZ. But C
n
has no canonical generator. For if g generates C
n
then so does g
k
for any
integer k with gcd(k, n) = 1 (exercise...). For example, if
G =
__
1 0
0 1
_
,
_
0 1
1 1
_
,
_
1 1
1 0
__
then G C
3
and either nonidentity matrix generates G, but here is no natural preference for either
generator. The root of this issue is that the isomorphism Z/nZ

C
n
induced by e
g
depends on the
choice of generator g of C
n
; a different choice would give a different isomorphism. An isomorphism
of this sort, which depends on one or more arbitrary choices, is called noncanonical.
3.2 Finite abelian groups
Every nite abelian group is a direct product of cyclic groups. The rst basic result in this direction is
as follows.
Proposition 3.1 Let A, B, C be nite abelian groups tting into the exact sequence
1 A C

B 1.
Assume that the orders of A and B are relatively prime. Then C A B.
Proof: Set m = [A[ and n = [B[, so that [C[ = mn. Let D be the set of elements in C whose
order is relatively prime to m. Then D A = 1. I claim that (D) = B. Let b B and choose
c C such that (c) = b. Since [C[ = mn we have (c
m
)
n
= 1, so the order of c
m
divides n, which is
relatively prime to m. Hence c
m
D, and (c
m
) = b
m
. But the map x x
m
is an automorphism of
B, again since gcd(m, n) = 1. Hence (D) = B
m
= B. It follows that C = AD, and that maps D
isomorphically onto B. By Prop. ?? we have
C = A D A B.

However, a product decomposition of an abelian group need not be unique.

Proposition 3.2 Suppose n
1
, n
2
, . . . , n
k
are relatively prime integers with product n = n
1
n
2
n
k
.
Then
C
n
1
C
n
2
C
n
k
C
n
.
Proof: Let g
i
be a generator of C
n
i
for each i. I claimthat the element (g
1
, g
2
, . . . , g
k
) C
n
1
C
n
k
has order n. We have (g
1
, g
2
, . . . , g
k
)
n
= (g
n
1
, g
n
2
, . . . , g
n
k
) = (1, 1, . . . , 1), since n
i
[ n for all i. And if
g
m
i
= 1 for all i then n
i
[ m for all i, so n [ m, since the n
i
are relatively prime. Hence n is the order
of (g
1
, g
2
, . . . , g
k
).
We can get a unique decomposition of a nite abelian group G as follows. For each prime p let G(p)
be the set of elements of G whose order is a power of p.
16
Theorem 3.3 Let G be a nite abelian group of order n. Then
1. G

p|n
G(p) is the direct product of its nontrivial subgroups G(p).
2. For each prime p there exist unique positive integers e
1
e
2
e
k
> 0 such that
e
1
+ e
2
+ + e
k
is the power of p dividing n and
G(p) C
p
e
1 C
p
e
2 C
p
e
k .
Proof: Part 1 follows from Prop. 3.2, using induction on the number of primes dividing [G[. We will
prove part 2 later, using modules over principal ideal domains. See Milne for an elementary proof.
Example 1: If n = p
1
p
2
p
k
is a product of distinct primes, then there is only one abelian group of
order n, up to isomorphism, namely C
n
. For the unique decomposition of Thm. 3.3 would be
G C
p
1
C
p
2
C
p
k
,
which is isomorphic to C
n
, by Prop. 3.2.
Corollary 3.4 A nite subgroup of the multiplicative group of a eld is cyclic.
Proof: Let F be a eld and let G be a nite subgroup of the multiplicative group F

. If n is a positive
integer and g G has order dividing n, then g is a root of the polynomial x
n
1, which has at most n
roots in F. Hence G has at most n elements of order dividing n, for any n 1. Write G =

p
G(p),
according to part 1 of Thm. 3.3. Each G(p) has at most p elements of order dividing p. Hence G(p) is
cyclic, by part 2 of Thm. 3.3. Now G is cyclic, by Prop. 3.2.
Example 2: If F is a nite eld with [F[ = q, then F

C
q1
. Consequently, for any positive integer
m the subgroup
m
(F) = x F : x
m
= 1 is also cyclic, of order gcd(m, q 1):

n
(F) C
gcd(m,q1)
(4)
3.2.1 Unit Groups
Let m be a positive integer and consider Z/mZ under multiplication (that is, as a ring):
(k + mZ)(k

+ mZ) = kk

+ mZ,
which is a well-dened operation. It is not a group operation, however, since the element 0+mZ has no
multiplicative inverse. In fact, from the Euclidean algorithm it follows that k +mZ has a multiplicative
inverse in Z/mZ iff gcd(k, m) = 1. Hence the set
(Z/mZ)

= k + mZ : gcd(k, n) = 1
forms a group under multiplication in Z/mZ, with identity element 1 +mZ. Since multiplication in Z
is commutative, the group (Z/mZ)

is abelian, of order
(m) = k Z : 1 k < m, gcd(k, m) = 1.
17
Proposition 3.5 Let m = p
r
1
1
p
r
k
k
be the factorization of m into a product of powers of distinct
primes p
i
. Then
1.
(Z/mZ)

i=1
(Z/p
r
i
i
Z)

.
2. For any prime p and integer r 1 we have
(Z/p
r
Z)

_
C
(p1)p
r1 C
p1
C
p
r1 if p 3
C
2
C
2
r2 if p = 2 and r 2
1 if p = 2 and r = 1.
Proof: Part 1 follows from the Chinese Remainder Theorem. We prove part 2 for p 3 and leave
p = 2 as an exercise. Reduction modulo p gives a surjective map
: (Z/p
r
Z)

(Z/pZ)

.
By Prop. 3.2 and Example 2 above, it sufces to show that ker is cyclic. Clearly 1 + p ker . We
will show that 1 + p has order p
r1
modulo p
r
.
Recall that p [
_
p
k
_
for all positive integers k < p. Now if a and b are congruent integers modulo some
power p

, it follows that a
p
b
p
mod p
+1
. Now using induction on 2 we have
(1 + p)
p
2
1 + p
1
mod p

,
which implies that 1 + p has order p
1
modulo for any 2.
3.3 The symmetric group
For any set X, the set S
X
of bijections : X X from X to itself forms a group under composition
whose identity element is the identity function e(x) x, called the symmetric group of X. The
elements of S
X
are usually called permutations, and the identity is the trivial permutation.
If f : X Y is a bijection between two sets X and Y , then we get an isomorphism S
f
: S
X

S
Y
,
dened by
S
f
() = f f
1
.
If X = 1, 2, . . . , n, we write S
n
instead of S
X
. In the literature, S
n
is sometimes called the sym-
metric group on n letters. This is a slight abuse of terminology, for the following reason. Sup-
pose X is any nite set, with [X[ = n. By labelling the elements of X we obtain a bijection
f : X 1, 2, . . . , n, whence an isomorphism S
f
: S
X

S
n
. However, this isomorphism is
noncanonical, because it depends on the chosen labelling f.
18
By counting permutations of 1, 2, . . . , n, we nd that
[S
n
[ = n! = 1 2 3 n.
For 1 k n, a k-cycle is an element of S
n
obtained as follows. Choose distinct numbers
i
1
, i
2
, . . . , i
k
1, . . . , n. Then (i
1
i
2
. . . i
k
) is the element of S
n
which sends
i
1
i
2
i
3
i
k
i
1
and xes all numbers in 1, . . . , n i
1
, . . . , i
k
. Note that the cycles
(i
1
i
2
. . . i
k
), (i
r
i
1
i
2
. . . i
k1
), (i
k1
i
k
i
1
i
2
. . . i
k2
), , (i
2
i
3
. . . i
k
i
1
)
all give the same permutation of 1, . . . , n. Hence we regard these cycles as equal.
We will see how to express each element of S
n
as a product of cycles. Given S
n
, dene an
equivalence relation

on 1, . . . , n as follows. For a, b 1, . . . , n, we declare that

a

b a =
i
b
for some i Z. Let A
1
, . . . A
q
be the equivalence classes (here q depends on ), numbered so that
[A
1
[ [A
2
[ [A
q
[, and set
i
= [A
i
[ for each i. Thus, we have a set partition
1, . . . , n =
q

i=1
A
i
and a corresponding numerical partition
n =
q

i=1

i
,
where the
i
are positive integers such that
1

2

q
1. If
1
= 1 then all
i
= 1 and = e
is the identity element of S
n
. Assume that
1
> 1 and let p q be the largest index such that
p
> 1.
Now choose one element a
i
A
i
arbitrarily, for each i = 1, 2, . . . , p. The cycle

i
= (a
i
a
i

2
a
i
. . .

i
1
a
i
)
preserves A
i
and acts there just as does, while ignoring the numbers outside A
i
. It follows that
=
1

2

p
(5)
is a product of disjoint cycles. Disjoint cycles commute, as we will soon see. Hence the product (5) is
independent of the order of the terms
1
, . . . ,
p
. The cycle type of is the partition
() = [
1
,
2
, . . . ,
q
]
recording the lengths of the cycles
i
, as well as the number of xed-points of , which is q p.
For example, the element S
6
sending
1 4, 2 6, 3 1, 4 3, 5 5, 6 2
19
may be written as
= (143)(26),
and has cycle type () = [3, 2, 1]. Note that 5, which is xed, is omitted in the cycle decomposition
of , but is counted in the cycle type of .
We multiply using the cycle decomposition by following the path of each number, starting with the
right-most cycle. For example, we have
(143)(26)(465) = (142653).
Note that the cycle type of = (143)(26)(465) is not [3, 3, 2], because the cycles are not disjoint. We
rst have to make them disjoint, as we have done above. We obtained a single 6-cycle (142653), so the
cycle type of in S
6
is () = [6].
We illustrate the cycle types for all elements of S
3
and S
4
, as follows.

[3] (123), (321)
[21] (12), (23), (13)
[111] e

[4] (1234), (1324), (1243), (1423), (1342), (1432)
[31] (123), (321), (124), (142), (134), (143), (234), (432)
[22] (12)(34), (13)(24), (14)(23)
[211] (12), (13), (14), (23), (24), (34)
[1111] e
The cycle types determines the conjugacy classes in S
n
.
Lemma 3.6 Let S
n
be any element and let = (i
1
i
2
i
k
) be a cycle in S
n
. Then

1
= (i
1
i
2
i
k
).
Proof: Let j 1, 2, . . . , n = 1, 2, . . . , n. If j = i
p
for some 1 p k then

1
j = i
p
= i
p+1
,
where subscripts are read modulo k. And if j / i
1
, . . . , i
k
, then
1
j =
1
j, so
1
j = j.

Proposition 3.7 Elements , S

n
are conjugate if and only if () = ().
Proof: Suppose =
1
for some S
n
. Let () = [
1

2
. . .
q
] be the cycle type of . Then
the cycle decomposition of is =
1

2

p
, where the
i
are disjoint cycles of length
i
, and

p
> 1 =
p+1
if q > p. Now
=
1
=
1

1

2

1

p

1
.
By Lemma 3.6, each
i
:=
i

1
is a
i
-cycle and the
i
s are disjoint cycles since is injective.
Therefore the disjoint cycle decomposition of is =
1

2

p
, and we have () = ().
20
Conversely, suppose () = () = [
1

2
. . .
q
]. Then the disjoint cycle decompositions of and
have the form =
1

2

p
and =
1

2

p
, where both
i
and
i
are
i
-cycles for 1 i q.
There are a
i
, b
i
1, . . . , n such that

i
= (a
i
a
i
. . .

i
1
a
i
), and
i
= (b
i
b
i
. . .

i
1
b
i
),
and

j
a
i
: 1 i q, 1 j
i
= 1, 2, . . . , n =
j
b
i
: 1 i q, 1 j
i
.
Hence there is a permutation S
n
such that
j
a
i
=
j
b
i
for all i, j. In particular, a
i
= b
i
for all
i, and we have

j
b
i
=
j
a
i
= (
1
)
j
b
i
so that =
1
, as claimed.
A permutation S
n
is even if its cycle type () contains an even number of even numbers;
otherwise is odd. Thus, (12)(34) and (123) are even, while (12) and (12)(345) are odd. We put
sgn() =
_
+1 if is even
1 if is odd
Note that 1 is a group under multiplication.
Proposition 3.8 The function sgn : S
n
1 is a surjective group homomorphism.
We will prove this in the next section, using determinants.
The Alternating Group A
n
is dened as
A
n
= ker sgn = S
n
: is even.
Thus, A
n
S
n
and
[A
n
[ =
1
2
n! = 3 4 n.
For example, A
4
has order 12; it consists of the eight 3-cycles, the three 22-cycles and the identity
element of S
4
.
The conjugacy classes in A
n
are determined as follows. First, A
n
consists of the even classes in S
n
.
However, two elements of A
n
which are conjugate in S
n
need not be conjugate in A
n
. Hence an even
S
n
-class could break up into several A
n
classes. To see when this happens, let A
n
and restriction
of the character sgn to the centralizer C
Sn
() of in S
n
. We have
C
An
() = ker[C
Sn
()
sgn
1],
so that the index
h

:= [C
Sn
() : C
An
()] =
_
2 if C
Sn
() , A
n
1 if C
Sn
() A
n
.
21
Now the size of the conjugacy class
An
in A
n
is given by
[
An
[ =
[A
n
[
[C
An
()[
=
1
2
[S
n
[
1
h
[C
Sn
()[
=
h

2
[
Sn
[.
It follows that if C
Sn
() , A
n
then
Sn
is a single conjugacy class in A
n
and if C
Sn
() A
n
then
Sn
breaks up into two A
n
-conjugacy classes:

Sn
=
An

An
,
where is conjugate to in S
n
but not in A
n
.
For example, in S
4
elements in the [22]-class contain 2-cycles in their centralizer (which is D
4
), so
the [22] is a single class in A
4
. But elements in the [31]-class generate their own centralizer, which is
therefore contained in A
4
. So the [31] class breaks up into two classes with four elements each. These
classes are mutually inverse. If A
4
is viewed as rotations of the tetrahedron, then one class consists of
clockwise face rotations and the other class consists of counterclockwise face rotations.
3.4 Linear groups
Let V be a vector space over a eld F. The set GL(V ) of invertible linear transformations T : V V
forms a group under composition, called the general linear group of V , whose identity element is the
transformation I
V
(v) v.
If V has nite dimension n over F and we choose an ordered basis v
1
, . . . , v
n
of V then each
T GL(V ) corresponds to an invertible n n matrix A
T
whose j
th
column is
_

_
a
1j
a
2j
.
.
.
a
nj
_

_
,
where T(v
j
) = a
1j
v
1
+ a
2j
v
j
+ + a
nj
v
n
. Matrix multiplication is dened so that
A
S
A
T
= A
ST
, for all S, T GL(V ).
It follows that sending T A
T
is a group isomorphism
GL(V )

GL
n
(F), (6)
where GL
n
(F) is the group of n n invertible matrices under matrix multiplication, whose identity
element is the n n identity matrix
I
n
=
_

_
1 0 . . . 0
0 1 . . . 0
.
.
.
.
.
. 0
0 0 . . . 1
_

_
.
22
As was the case for S
n
, the isomorphism (6) is noncanonical, because it depends on a choice of ordered
basis v
1
, . . . , v
n
of V . A 1 1 matrix is just a number, so
GL
1
(F) = F

is the group of nonzero elements of the eld F under multiplication. This group F

appears in GL
n
(F)
as both a normal subgroup and a quotient.
First, we have an injective homomorphism
F

GL
n
(F), given by a a I
n
=
_

_
a 0 . . . 0
0 a . . . 0
.
.
.
.
.
. 0
0 0 . . . a
_

_
,
whose image Z F

is the center of GL
n
(F). The Projective Linear Group PGL
n
(F) is dened
as
PGL
n
(F) = GL
n
(F)/Z.
Elements of PGL
n
(F) are no longer linear transformations, but they permute the lines in the vector
space F
n
. Thus, the elements of PGL
n
(F) are transformations of the Projective Space P
n1
(F)
which is the set of lines in F
n
.
Next, recall that an n n matrix A is invertible if and only if its determinant det(A) is nonzero.
Moreover, if A and B are two n n matrices, we have det(AB) = det(A) det(B). Thus, det is a
homomorphism
det : GL
n
(F) F

.
The Special Linear Group SL
n
(F) is dened as
SL
n
(F) = ker det = A GL
n
(F) : det(A) = 1.
When F = R, elements of SL
n
(F) are the linear transformations of R
n
which preserve volume.
Returning to general F, consider the restriction of det to Z. Since
det(aI
n
) = a
n
, (7)
it follows that
det(Z) = F
n
is the subgroup of n
th
powers in F

. Hence the composition

GL
n
(F)
det
F

/F
n
induces a surjective homomorphism
PGL
n
(F)
det
F

/F
n
.
The latter group depends on the eld F. For example, we have [C

/C
n
[ = 1, while [R

/R
n
[ = 1
or 2 according as n is odd or even. For some elds, F

/F
n
can be innite.
23
It also follows from (7) that the intersection
Z
1
= SL
n
(F) Z = aI
n
: a
n
= 1
n
(F),
where
n
(F) = a F

: a
n
= 1 is the subgroup of n
th
roots of unity in F

. The group Z
1
is the
center of SL
n
(F) and the quotient
PSL
n
(F) = SL
n
(F)/Z
1
= ker det
is the image of SL
n
(F) in PGL
n
(F). The center Z
1

n
(F) of SL
n
(F) also depends on the eld F.
For example, we have [
n
(C)[ = n, while [
n
(R)[ = 1 or 2 according as n is odd or even. For general
elds F,
n
(F) is always nite of order dividing n.
In summary, we have four closely related groups GL
n
(F), SL
n
(F), PGL
n
(F), PSL
n
(F), related to
each other via the following commutative diagram with exact rows.
1
n
(F) F

n
F
n
1

_
1 SL
n
(F) GL
n
(F)
det
F

_
1 PSL
n
(F) PGL
n
(F)
det
F

/F
n
1
If F is a nite eld with [F[ = q, we write GL
n
(q), SL
n
(q), PGL
n
(q), PSL
n
(q) instead of GL
n
(F), SL
n
(F), PGL
n
(F), PSL
n
(F).
Often in the literature one nds the abbreviation L
n
(q) for PSL
n
(q). The orders of these groups are
given as follows.
[ GL
n
(q)[ = q
n(n1)/2
(q 1)(q
2
1) (q
n
1)
[ PGL
n
(q)[ = [ SL
n
(q)[ = q
n(n1)/2
(q
2
1)(q
3
1) (q
n
1)
[ PSL
n
(q)[ = [ SL
n
(q)[/ gcd(n, q 1)
(8)
In the last line, recall from (4) that gcd(n, q 1) = [
n
(F)[. All of these orders then follow from the
calculation of [GL
n
(q)[, which can be done as follows. To have a matrix A GL
n
(q), we can take any
of the q
n
1 nonzero vectors in F
n
for the rst column, then any of the q
n
q vectors not in the line
spanned by the rst column, then any of the q
n
q
2
vectors not in the plane spanned by the rst two
columns, etc. This gives
[ GL
n
(q)[ = (q
n
1)(q
n
q)(q
n
q
2
) (q
n
q
n1
)
= q
1+2++(n1)
(q
n
1)(q
n1
1)(q
n2
1) (q 1)
= q
n(n1)/2
(q
n
1)(q
n1
1)(q
n2
1) (q 1),
as claimed above.
24
3.4.1 Conjugacy classes in GL
2
(F)
We assume a bit more familiarity with elds in this section. Let F be a eld. Assume either that F is
nite or that 2 ,= 0 in F. Let

F be a xed algebraic closure of F. We identify F

with the center of

GL
2
(F), as in ().
A quadratic extension of F is a eld K such that F K

F and K is a two-dimensional F-vector
space. This means K = F F for some (any) K F. The elements 1, ,
2
are then linearly
dependent over F so we have a dependence relation
2
T+N = 0, where T = T() and N = N()
(the norm and trace) are elements of F. The roots of the polynomial x
2
Tx +N are distinct; let

be
the root other than .
Using the basis 1, the action of K

on K by multiplication gives an embedding

: K

GL
2
(F).
For an arbitrary element a + b K

we have (a + b) 1 = a + b and
(a + b) = a + b
2
= a + b(T N) = bN + (a + bT).
Hence

is given explicitly by

(a + b) =
_
a bN
b a + bT
_
.
The eigenvalues of this matrix are a + b and a + b

; in particular these eigenvalues lie in K

.
A different choice

of K F gives a new basis of K over F, and the subgroups

(K

) and

(K

) are conjugate in GL
2
(F). We denote any of these subgroups by K

, with the understanding

that they are only determined up to conjugacy. If L and K are distinct quadratic extensions of F the
groups L

and K

are not conjugate in GL

2
(F) and any two members in the class L

or K

will
intersect in F

. This is because the eigenvalues of elements of L

lie in L and similarly for K, and

K L = F.
For g, h GL
2
(F) we write g h to mean that g and h are conjugate in GL
2
(F).
Proposition 3.9 Let g GL
2
(F) have eigenvalues , in

F.
1. If = F

, then either g =
_
0
0
_
or g
_
1
0
_
.
2. If ,= F

, then g
_
0
0
_

_
0
0
_
.
3. If ,= / F

then and =

belong to a unique quadratic extension K and g

()

) K

.
25
Proof: In case 1, the matrix g I
2
has nonzero kernel. If g I
2
, choose any vector v F
2
such that the vector u := (g I
2
)v is nonzero. Since (g I
2
)
2
= 0, the vectors u, v are linearly
independent. Using the basis u, v, we have g
_
1
0
_
.
In case 2, we use the basis of eigenvectors of g to see that g
_
0
0
_
.
In case 3, and

are the roots of the characteristic polynomial of g so they generate a quadratic
extension K of F. The element

()

(K

) has the same eigenvalues. Hence g and

() are two
elements of GL
2
(F) which are conjugate to
_
0
0

_
in GL
2
(K). From the theory of rational canonical
form, or Hilberts theorem 90, it follows that g and

() are conjugate in GL
2
(F).

If F has no quadratic extensions, for example if F = C or if F is the eld of constructible numbers,

then case 3 does not arise. If F is nite, say [F[ = q, then F has only one quadratic extension K, and
[K[ = q
2
. In this case there are (q
2
q)/2 conjugacy classes of type 3.
3.4.2 S
n
as a subgroup of GL
n
(F)
The symmetric group S
n
is (isomorphic to) a subgroup of GL
n
(F), as follows. Let e
i
F
n
be the
vector with 1 in the i
th
component and zero in the other components. For each permutation S
n
, let
A

be the n n matrix such that A

(e
i
) = e
(i)
. Then for , S
n
we have
A

(e
i
) = A

(e
(i)
) = e
(i)
= A

e
i
.
Hence the map
f : S
n
GL
n
(F) given by f() = A

is a homomorphism. It is easy to check that the map f is injective and the image of f consists of the
matrices with a single entry = 1 in each row and column and all other entries = 0. These are called
permutation matrices.
Suppose = (i
1
i
2
i
r
) is an r-cycle in S
n
. The matrix A

sends e
i
1
e
i
2
e
ir
e
i
1
and
xes the remaining e
j
s. Since the determinant of a matrix is unchanged by simultaneous interchanges
of rows and columns, we have
det(A

) = det
_
A

0
0 I
nr
_
= det(A

)
where A

is the r r matrix
A

=
_

_
0 0 0 . . . 1
1 0 0 . . . 0
0 1 0 . . . 0
.
.
.
.
.
.
.
.
.
.
.
. 0
0 0 . . . 1 0
_

_
.
26
Expanding along the top row, we compute
det(A

) = (1)
r1
.
Hence det(A

) = +1 or 1 according as r is odd or even.

Now take a general S
n
and write it as a product of disjoint cycles. We have A

= +1 or 1
according as has an even or odd number of even cycles. This agrees with our denition of sgn above,
so we have shown that
det(A

) = sgn().
Hence sgn = det f is a group homomorphism, as claimed in Prop. 3.8.
3.4.3 The Bruhat decomposition for GL
n
(F).
The invertible matrices are characterized as those whose determinant is nonzero. However, it is not ob-
vious how to write down a general invertible matrix. That is, we have the locus, but not the parametriza-
tion of invertible matrices. The Bruhat decomposition remedies this by expressing a general invertible
matrix in terms of simpler ones that can be parametrized.
We have seen that the group G = GL
n
(F) has the subgroup W S
n
consisting of permutation
matrices. We also have the subgroup B < G consisting of upper triangular matrices with all diagonal
entries nonzero.
Theorem 3.10 (Bruhat Decomposition) The group G = GL
n
(F) is a disjoint union
G =

wW
BwB,
where BwB = b
1
wb
2
: b
i
B.
Proof: Let e
1
, . . . , e
n
be the standard basis of F
n
. Then B is the subgroup of elements b G for
which be
j
is contained in the span of e
i
: i < j, for all 1 j n.
Take g G and write
ge
1
=
n

i=1
g
i
e
i
,
and let i
1
= maxi : g
i
,= 0. Dene b B by
be
j
=
_
g
1
i
1
_
e
i
1

i
1
1
i=1
g
i
e
i
_
if j = i
1
e
j
if j ,= i
1
.
Then
bge
1
= b
i
1

i=1
g
i
e
i
=
i
1
1

i=1
g
i
e
i
+ g
i
1
g
1
i
1
_
e
i
1

i
1
1

i=1
g
i
e
i
_
= e
i
1
.
27
For j > 1 let g

j
be the coefcient of e
i
1
in bge
j
, and dene b

B by
b

e
j
=
_
e
j
g

j
e
1
if j > 1
e
1
if j = 1.
We then have
bgb

e
1
= bge
1
= e
i
1
and for j > 1 the coefcient of e
i
1
in bgb

e
j
is g

j
g

j
= 0. Thus, the matrix bgb

has column 1 equal

to e
i
1
and row i
1
equal to e
1
.
We repeat the above procedure with column 2 of the matrix h := bgb

. Write
he
2
=
n

i=1
h
i
e
i
,
and let i
2
= maxi : h
i
,= 0. Note that h
i
1
= 0, so i
1
,= i
2
. Dene b

B by
b

e
j
=
_
h
1
i
2
_
e
i
2

i
2
1
i=1
h
i
e
i
_
if j = i
2
e
j
if j ,= i
2
.
Then
b

he
1
= b

e
i
1
= e
i
1
, since i
1
,= i
2
,
and
b

he
2
= b

i
2

i=1
h
i
e
i
=
i
2
1

i=1
h
i
e
i
+ h
1
i
2
h
i
2
_
e
i
2

i
2
1

i=1
h
i
e
i
_
= e
i
2
.
For j > 2, let h

j
and h

j
be the coefcients of e
i
1
and e
i
2
in b

he
j
, and dene b

B by
b

e
j
=
_
e
j
h

j
e
1
h

j
e
2
if j > 2
e
j
if j = 1, 2.
We have then
b

hb

e
1
= b

he
1
= e
i
1
, b

hb

e
2
= b

he
2
= e
i
2
and for j > 2 the vectors e
i
1
and e
i
2
each have coefcient = 0 in the vector
b

hb

e
j
= b

he
j
h

j
e
i
1
h

j
e
i
2
.
Thus, for j = 1, 2, the matrix b

hb

= b

bgb

has column j equal to e

i
j
and and i
th
j
row equal to e
j
.
Repeating this process up to j = n shows that there are elements b
1
, b
2
B such that b
1
gb
2
is the
permutation w W sending j i
j
for all j. Hence g BwB, as claimed.
28
3.5 The dihedral groups D
n
A reection is an isometry of a Euclidean space E whose set of xed-points in E is a hyperplane. A
reection group is a group of isometries of E which is generated by reections. The dihedral groups
are the reection groups in two dimensions.
Suppose E is a plane. Each line in E determines a reection r with xed-point set . If P is a point
not on , then r(P) is the mirror image of P with respect to . We say that r is the reection about .
Note that r is a nontrivial involution of the isometry group of the plane.
Suppose and

are two lines in the plane, with reections r and r

. If and

are not parallel, then

they meet in a point P, and the product rr

is rotation about twice the angle at P from

to . In
particular, r and r

commute precisely when and

are perpendicular. If and

are parallel, the

product rr

is translation by twice the perpendicular vector from

to .
For any integer n 1 consider n lines
1
, . . . ,
n
in the plane meeting in a common point, with equal
angles (= /n) between adjacent lines. Let r
i
be the reection about
i
. The dihedral group D
n
is the
group generated by the reections r
1
, . . . , r
n
. We dene D

similarly, by taking a countable number

of parallel lines equally spaced apart (all meeting at innity, with equal angle zero).
For n = 1 we have just one line, with reection r and
D
1
= 1, r C
2
.
For n = 2 we have two perpendicular lines
1
,
2
, whose reections r
1
, r
2
commute. Hence
D
2
= 1, r
1
, r
2
, r
1
r
2
C
2
C
2
.
For n = 3 we have three lines
1
,
2
,
3
intersecting at the angle /3. Let r, s be reections about
adjacent lines. Then rs is a rotation of order 2/3, hence has order three. The equation (rs)
3
= 1 can
be written as
rsr = srs.
This element rsr = srs is the third reection. It follows that D
3
is generated by r and s only, and its
elements are
D
3
= 1, r, s, rs, sr, rsr.
The product of any two elements in D
3
is completely determined by the three rules:
r
2
= 1, s
2
= 1, rsr = srs.
For example, we have rs rsr = rs srs = r rs = s.
It is a similar story for an arbitrary nite n 2. We again let r, s be reections about adjacent lines

r
,
s
, so that rs is a rotation by 2/n and hence has order n. The equation (rs)
n
= 1 can be written as
rsrs . . .
. .
n terms
= srsr . . .
. .
n terms
29
The element srs is reection about the line s(
r
), which is the other line adjacent to s. It follows that
all reections can be written in terms of r and s, as s(rs)
i
for some 1 i n and that the elements of
D
n
are
D
n
= (rs)
i
, s(rs)
i
: 1 i n
and [D
n
[ = 2n. The element t = rs generates a cyclic subgroup t C
n
, consisting of all rotations
in D
n
, which has index two in D
n
. The reections in D
n
are precisely the elements outside of t, and
for any reection r

, we have r

tr

= t
1
.
The subgroup lattice of D
n
can be described as follows. For each divisor mof n, we have rst of all the
unique subgroup C
m
C
n
= t, as well as n/m copies of D
m
obtained as follows. Index the lines
by Z/nZ, and partition them according to the bers of the natural map
m
: Z/nZ Z/(n/m)Z.
Let D
(i)
m
be the subgroup of D
n
generated by the reections about lines in the ber
1
m
(i). Since this
ber has m equiangular lines, we indeed have D
(i)
m
D
m
. As i ranges over Z/(n/m)Z, we obtain
n/m subgroups D
(i)
m
, all containing the same cyclic subgroup C
m
. In particular, we have n/1 = n
subgroups D
(i)
1
D
1
, each generated by the reections about one of the lines. Finally, if [ m [ n, we
have D
(j)

< D
(i)
m
iff
1

(j)
1
m
(i), iff j i mod
n
m
.
The situation for D

is similar but simpler. The element t = rs now has innite order, hence generates
a copy of Z in D

, and for any reection r

we again have r

tr

= t
1
. We leave the subgroups D

to
the exercises.
Returning to nite n, we can viewD
n
as subgroup of GL
2
(R). Assume the lines intersect at (0, 0) R
2
and that the reecting line
1
is the x-axis. The reection r
1
has matrix
r
1
=
_
1 0
0 1
_
.
For 1 k < n let
k+1
be the line rotated from counterclockwise by k/n. Then
2
is adjacent to
1
and the rotation (r
1
r
2
)
k
has matrix
(r
1
r
2
)
k
=
_
cos(2k/n) sin(2k/n)
sin(2k/n) cos(2k/n)
_
.
3.6 The quaternion and generalized quaternion groups Q
4n
The generalized quaternion groups are best understood as subgroups of the group SL
2
(C) of 2 2
complex matrices with determinant =1. Let T be the subgroup of diagonal matrices in SL
2
(C). Its
normalizer N(T) in SL
2
(C) consists of two cosets of T:
N(T) = T wT, where w =
_
0 1
1 0
_
.
The nite subgroups of N(T) are of two types: Those contained in T are cyclic. The generalized
quaternion groups are the nite subgroups of N(T) which are not contained in T. Let Q < N(T) be
such a subgroup. Since Q ,< T, it contains an element of the form wt for some t T. Replacing Q by
30
s
1
Qs, where s
2
= t, we may and shall assume that t = 1, so that w Q. Since w has order four, the
order of Q is divisible by four.
The generalized quaternion group Q
4n
is the unique subgroup of N(T) containing w and having order
4n.
To see Q
4n
explicitly, let
k
= e
i/n
, which has order 2n as an element of C

. Then Q
4n
is generated
by the two matrices
t
n
=
_

n
0
0
1
n
_
, and w =
_
0 1
1 0
_
.
Note that t
n
has order 2n, that w has order four, and we have
wt
n
w
1
= t
1
, w
2
= t
n
= I.
Thus, we have
Q
4n
= t
i
n
w
j
: 0 i 2n 1, j = 0 or 1,
so that [Q
4n
[ = 4n, as claimed. Note that
t
n
w = t
n
n
= I,
where I =
_
1 0
0 1
_
. This is the unique element of order two in Q
4n
, and the subgroup I is
normal in Q
4n
, with quotient
Q
4n
/I D
n
.
Hence the subgroups of Q
4n
containing I are in bijection with the subgroups of D
n
. Recall the
subgroups of D
n
are cyclic rotations or dihedral. The subgroups of Q
4n
corresponding to cyclic rota-
tions are cyclic. These are precisely the subgroups of even order in Q
4n
. The odd-order subgroups are
cyclic and contained in t
n
.
For k = 1 we have Q
4
= w C
4
. This is the only generalized quaternion group which is abelian.
Indeed, we have Z(Q
4n
) = I when n 2.
For k = 2 the group Q
8
is commonly known as the quaternion group (of order eight), and has different
notation. It is common to write
Q
8
= 1, i, j, k,
where
1 =
_
1 0
0 1
_
, i =
_
1 0
0

1
_
, j =
_
0 1
1 0
_
, k =
_
0

1 0
_
,
which have relations i
2
= j
2
= k
2
= 1 and
ij = k, jk = i, ki = j, ji = k, kj = i, ik = j.
The proper subgroups of Q
8
are i, j, k, 1. We have
1 = i j k
and this subgroup is both the center Z(Q
8
) and the commutator subgroup [Q
8
, Q
8
]. The group Q
8
is
the simplest non-abelian group in which every subgroup is normal. It can be shown that any nite
non-abelian group with all subgroups normal is isomorphic to Q
8
A, where A is abelian.
31
3.7 p-groups, a rst look
A nite group G is a p-group if the order of G a power of a prime p.
Each abelian p-group is a direct product G = C
p
n
1 C
p
n
2 C
p
n
k of cyclic p-groups, there being
one isomorphism class of such groups for every set of positive integers n
1
, . . . , n
k
. When all n
i
= 1,
the group C
k
p
= C
p
C
p
C
p
is called elementary abelian of rank k. The dihedral groups D
2
n
and generalized quaternion groups Q
2
n are examples of nonabelian 2-groups.
One cannot hope to classify all p-groups, except those whose orders are small powers of p.
Proposition 3.11 Let p be a prime and let G be a p-group.
1. If [G[ = p then G C
p
.
2. If [G[ = p
2
then G is abelian. We have G C
p
2 if G is cyclic and G C
p
C
p
if G is not
cyclic.
3. If [G[ = p
3
then either G is one of two nonabelian groups or G is one of C
p
3, C
p
C
p
2 or
C
p
C
p
C
p
.
Proof: We already noted that part 1 is a consequence of Lagranges theorem. We will prove part 2
here, and postpone the proof and a more detailed statement of part 3.
Assume [G[ = p
2
. We have seen in Prop. 2.15 that every p-group has a nontrivial center Z(G). By
Lagranges theorem, we have [Z(G)[ = p or p
2
. If [Z(G)[ = p then G/Z(G) has order p, hence is
cyclic, so G is abelian, contradicting Z(G) ,= G. Hence G is abelian.
The order of every element of Galso divides p
2
. If Ghas an element of order p
2
then G C
p
2. Assume
G has no element of order p
2
. Then every nonidentity element of G has order p. Choose h, k G with
h ,= 1 and k / h. The subgroups H = h and K = k haver order p and are both normal in the
abelian group G. Now HK is a subgroup of G properly containing H. Since [G : H] = p, it follows
that HK = G. Likewise, HK is a proper subgroup of K, which has order p, so HK = 1. Now
by Prop. 2.10 we have G H K C
p
C
p
.
Prop. 2.15 can be extended to prove the converse of Lagranges theorem for p-groups. First we need a
lemma.
Lemma 3.12 If A is a nite abelian group whose order is divisible by a prime p then A contains an
element of order p.
Proof: By induction, we may assume the result is true for groups of smaller order. Let b A have
order m > 1, and let B = b. If p [ mthen b
m/p
has order p. Assume p m. Then p divides [A/B[ and
[A/B[ < [A[, so A/B has an element of order p, by the induction hypothesis. This element is aB for
some a A such that a / B, but a
p
B. Therefore a
p
= b
r
for some integer r. Since gcd(p, m) = 1,
32
we can write r = kp + m for integers r, . The element c = ab
k
does not belong to B since a / B,
but since A is abelian we have
c
p
= a
p
b
kp
= b
rkp
= b
m
= 1.
Hence c A has order p.
As we will see in the next result, the lemma is true without the assumption that A is abelian, but the
proof is not as constructive.
Proposition 3.13 Let G be a nite group of order p
r
, where p is a prime. Then G has a chain of
subgroups
1 = G
0
< G
1
< G
2
< < G
r1
< G
r
= G
such that for all 0 i < r we have
1. [G
i
[ = p
i
;
2. G
i
is a normal subgroup of G and G
i+1
/G
i
C
p
;
3. G
i+1
/G
i
is contained in the center of G/G
i
.
Proof: We argue by induction on r. By Prop. 2.15, the center Z(G) is a nontrivial abelian p-group.
By Lemma 3.12, there exists a subgroup G
1
< Z(G) of order p. Since G
1
is central in G we have
G
1
G. The group G = G/G
1
has order p
r1
. Applying the induction hypothesis to G, there is a chain
of subgroups
1 = G
0
< G
1
< G
2
< < G
r2
< G
r1
= G
such that for all 0 i < r 1 we have [G
i
[ = p
i
and G
i
G and G
i+1
/G
i
is contained in the center of
G/G
i
.
By the Correspondence Theorem applied to G/G
1
there are normal subgroups G
i
G such that
G
i
= G
i
/G
1
.
Moreover, the canonical projection G G induces isomorphisms
G/G
i

G/G
i
which restrict to isomorphisms
G
i+1
/G
i

G
i+1
/G
i
for each 0 i < r. It follows that G
i+1
/G
i
is contained in the center of G/G
i
, as claimed.
Thus, every p-group has a tower of normal subgroups whose quotients are cyclic of order p. Despite
this apparent simplicity, the number of isomorphism classes of groups of order p
r
grows rapidly with
r, especially for the prime p = 2. Below is a table of the number of 2-groups for exponent r 10.
33
[G[ number of groups
2 1
2
2
2
2
3
5
2
4
15
2
5
51
2
6
267
2
7
2 328
2
8
56 092
2
9
10 494 213
2
10
49 487 365 422
It has been determined
1
that the total number of all groups of order 2000 is 49 910 529 484, so over
99% of these groups have order 2
10
.
3.8 Simple groups
A group G is simple if G has no normal subgroups other than 1 and G itself. Such groups have
remarkable properties. For example,
Every homomorphism from a simple group to another group is either injective or trivial.
For if f : G G

is a nontrivial homomorphism from a simple group G into some other group G

then
f is automatically injective, since ker f is a normal subgroup of G.
Likewise,
If G is nonabelian simple, then the center Z(G) = 1 and the commutator [G, G] = G.
For both Z(G) and [G, G] are normal subgroups of G. As G is nonabelian, we have Z(G) ,= G and
[G, G] ,= 1, so we must have Z(G) = 1 and [G, G] = G.
By Lagranges Theorem, any group of prime order is simple. All other simple groups are nonabelian;
they are extremely rare and interesting. Of the 49 910 529 484 groups of order at most 2000, exactly
six are nonabelian simple groups, namely
simple group G [G[
A
5
PSL
2
(5) 60
PSL
2
(7) GL
3
(2) 168
A
6
PSL
2
(9) 360
PSL
2
(8) 504
PSL
2
(11) 660
PSL
2
(13) 1092
1
The groups of order at most 2000, Besche et al., AMS Elec.Res.Ann. 2001.
34
These small simple groups belong to two families A
n
and PSL
2
(q) for n 5 and q 5 a prime
power. We prove that the groups in these families are simple.
3.8.1 Simplicity of alternating groups
Theorem 3.14 For n 5 the alternating group A
n
is simple.
Proof: Every element of A
n
is a product of an even number of transpositions. Hence A
n
is generated
by elements of the form (a b)(c d) and (a b)(b c). Now
(a b)(c d) = (a c b)(a c d), and (a b)(b c) = (a b c),
so A
n
is also generated by 3-cycles.
Since n 5, the centralizer of a 3-cycle in S
n
contains a transposition, so this centralizer is not
contained in A
n
. It follows that the 3-cycles form a single conjugacy class in A
n
.
Let N A
n
be a nontrivial normal subgroup of A
n
. We must show that N = A
n
. For all N and
A
n
the commutator
1

1
belongs to N, since N A
n
. We use this procedure to show that
N contains a 3-cycle. Since N is a union of conjugacy classes of A
n
, it will follow that N contain all
3-cycles and therefore N = A
n
by our previous remarks.
We write elements of N as products of disjoint cycles.
Case 1: Suppose N contains a disjoint product of the form
= (a
1
a
2
. . . a
r
), r 4.
Let = (a
1
a
2
a
3
). Then we compute

1
= (a
1
a
3
a
r
) N,
so that N contains a 3-cycle in this case.
Case 2: Suppose N contains a disjoint product of the form
= (a b c)(d e f).
Let = (a b d). Then we compute

1
= (a d b f c) N.
Then case 1 applies, and shows that N contains a 3-cycle.
Case 3: Suppose N contains a disjoint product of the form
= (a b)(c d).
35
We have already observed that (a b)(c d) = (a c b)(a c d), so that
= (a c b)(a c d).
Then case 2 applies, and shows that N contains a 3-cycle.
Every nonidentity element of A
n
can be written in one of these three forms. Hence N must contain a
3-cycle.
3.8.2 Simplicity of PSL
2
(F)
We next prove that the group PSL
2
(F) is simple for any eld with at least four elements. The proof
depends a series of lemmas, each interesting in its own right. We work in the group G = SL
2
(F), with
the following subgroups and element:
B =
__
a b
0 a
1
_
: a F

, b F
_
, U =
__
1 b
0 1
_
: b F
_
, U =
__
1 0
c 1
_
: c F
_
,
T =
__
a 0
0 a
1
_
: a F

_
, Z =
__
1 0
0 1
_
,
_
1 0
0 1
__
= Z(G), w =
_
0 1
1 0
_
.
Lemma 3.15 (Bruhat Decomposition) We have G = B BwB, a disjoint union.
Proof: A matrix
_
a b
c d
_
G lies outside of B exactly if c ,= 0. In this case,
_
a b
c d
_
=
_
c
1
a
0 c
_

_
0 1
1 0
_

_
1 dc
1
0 1
_
BwB.

Lemma 3.16 The subgroup B is a maximal proper subgroup of G.

Proof: Suppose H is a subgroup of G properly containing B. Then there exists h H with h / B.
By Lemma 3.15, we can write h = b
1
wb
2
, with b
i
B. It follows that w H, hence BwB H, so
H = G.
Lemma 3.17 The group G is generated by U and U.
Proof: Let H be the subgroup of G generated by U and U. If a F

, we have
_
a 0
0 a
1
_
=
_
1 1
0 1
_

_
1 0
a 1 1
_

_
1 a
1
0 1
_

_
1 0
a a
2
1
_
.
36
Since B = TU, we have B H. And
w =
_
1 1
0 1
_

_
1 0
1 1
_

_
1 1
0 1
_
,
so w H. From Lemma 3.15, it follows that H = G.
Lemma 3.18 If [F[ 4 then SL
2
(F) is its own commutator subgroup.
Proof: By Lemma 3.17 it sufces to show that the elements of U and

U are commutators. This
depends on the fact that T normalizes U and U. Indeed, we have
_
a 0
0 a
1
_ _
1 b
0 1
_ _
a 0
0 a
1
_
1
=
_
1 ba
2
0 1
_
, and
_
a 0
0 a
1
_ _
1 0
c 1
_ _
a 0
0 a
1
_
1
=
_
1 0
ca
2
1
_
.
We get the following commutators
__
a 0
0 a
1
_
,
_
1 b
0 1
__
=
_
1 b(a
2
1)
0 1
_
, and
__
a 0
0 a
1
_
,
_
1 0
c 1
__
=
_
1 0
c(a
2
1) 1
_
.
Now if F has at least four elements, we can nd a F such that a / 0, +1, 1. For any x F we
take b = x/(a
2
1) and c = x/(a
2
1), and nd that
_
1 x
0 1
_
and
_
1 0
x 1
_
are commutators, as claimed.
Lemma 3.19 The intersection

gG
g
B = Z.
Proof: Since Z < B, and Z is central, it is clear that Z <
g
B for all g G. Conversely, we have
B
w
B = T. Letting v =
_
1 0
1 1
_
U, one checks that T
v
B = Z. Hence

gG
g
B Z, proving
equality.
Lemma 3.20 Assume that [F[ 4. Let H be a normal subgroup of G = SL
2
(F). Then either H Z
or H = G.
Proof: Since H G, the product HB is a subgroup of G containing B. By Lemma 3.16 we have
either HB = B or HB = G. If HB = B then H B. But since H G, we have H

gG
g
B =
Z.
Suppose HB = G. Then we can write w = hb for some h H, and b B. One checks that B
normalizes U and
w
U = U. It follows that U =
w
U =
hb
U =
hU
. As U and U generate G, we have
then G = HU. By the second isomorphism theorem, we have
G/H = HU/H U/U H.
37
The latter group is abelian since U is abelian. Hence H [G, G]. But [G, G] = G, as we proved in
Lemma 7.1. Hence H = G, as claimed.
Now we can prove our result.
Theorem 3.21 If F is a eld with at least four elements then the group PSL
2
(F) is simple.
Proof: By the Correspondence Theorem, the normal subgroups of PSL
2
(F) are the projections of
the normal subgroups of SL
2
(F) which contain Z. From Lemma 3.20 it follows that every normal
subgroup of PSL
2
(F) is either trivial or all of PSL
2
(F). Hence PSL
2
(F) is simple.
Remark: If [F[ 3 then PSL
2
(F) is not simple. Indeed, we have
PSL
2
(2) S
3
, PSL
2
(3) A
4
.
These are the rst two of the exceptional isomorphisms discussed in the next section. For n 3, the
group PSL
n
(F) is simple for every eld F (see [Lang, XIII.9]).
3.9 Exceptional isomorphisms
The previous two sections exhibit two families of nite simple groups, namely the alternating groups
A
n
for n 5 and the groups PSL
2
(q) = PSL
2
(F) where F is a nite eld with [F[ = q 4. A
small number of groups are common to both families via isomorphisms whose subtly ranges from the
non-obvious to the miraculous. We list these exceptional isomorphisms, and a few others.
S
3
GL
2
(2) = PSL
2
(2)
A
4
PSL
2
(3)
A
5
PSL
2
(4) PSL
2
(5)
PSL
2
(7) GL
3
(2) = PSL
3
(2)
A
6
PSL
2
(9)
A
8
PSL
4
(2).
(9)
The rst two, as well as the isomorphism A
5
PSL
2
(4), arise easily from the theory of group actions
in the next section. For the remaining exceptional isomorphisms, see sections 5.4.1, 5.4.2, 8.5.2 and
??.
4 Group actions
We say that group G acts on the set X if there is a homomorphism : G S
X
from G into the group
S
X
of permutations of X. The pair (X, ) is called a G-set or a G-action. Thus, each g G gives a
38
permutation
g
of X, which sends any x X to an element (g)x. If is understood or is completely
general, we usually omit it from the notation, writing gx or g x instead of (g)x.
If (X, ) and (Y, ) are two G-sets, a function f : X Y is called G-equivariant if f((g)x) =
(g)f(x) for all g G and x X. We say that (X, ) and (Y, ) are equivalent G-sets if there exists
a G-equivariant bijection f : X Y .
The notion of a G-set generalizes the notion of a group. For we can regard the action as a map
GX X, given by (g, x) g x such that g (g

x) = (gg

) x for all g, g

G and x X.
Some standard terminology associated with group actions is as follows.
The stabilizer or xer of a point x X is the subgroup of G given by
G
x
= g G : g x = x G.
The orbit of an element x X is the subset of X given by
G x = g x : g G X.
Orbits are equivalence classes under the equivalence relation x y if y = g x for some g G. Hence
two orbits are either equal or disjoint; the orbits form a partition of X.
The kernel of a group action : G S
X
is which is the normal subgroup of G consisting of the
element acting trivially on X. We have
ker =

xX
G
x
,
AG-action on X is faithful if ker is trivial. Equivalently, the action is faithful if no nontrivial element
of G acts trivially on X. In this case, G is isomorphic to a subgroup of S
X
.
Finally, a group action is free if g x = x for some x X implies g = 1. That is, a group action is
free iff all stabilizers are trivial. Clearly free actions are faithful. An example of a free action is where
a subgroup H of a group G acts on G by left multiplication: h x = hx. Here, G is the set and H is
the group which is acting. The orbits are the right cosets Hx.
Proposition 4.1 For all g G and x X, we have
G
gx
= gG
x
g
1
.
In particular, the stabilizers of all elements of the same orbit are conjugate.
A G-action on X is transitive if for all x, y X there exists g G such that g x = y. Equivalently,
the action is transitive iff X consists of a single G-orbit. For a general group action, each orbit is a
transitive G-set. Thus, transitive group actions are the essential ones. An example of a transitive group
action is where X = G/H, for some subgroup H G, and the action is g xH = gxH. We will see
39
that all transitive G-actions are of this form. More generally, a G-action on X is k-transitive if G is
transitive on k-element subsets of X. This is a measure of the strength of transitivity.
The Main Theorem of Group Actions
If a group G acts on a set X, then for each x X we have a G-equivariant bijection
f : G/G
x

G x, given by f(gG
x
) = g x.
In particular, any transitive group action is equivalent to an action on cosets.
Proof: The map f is well-dened because for all h G
x
we have (gh) x = g (h x) = g x.
The map f is injective because if g x = g

x then g
1
g

x = x, so g
1
g

G
x
, which means that
gG
x
= g

G
x
. The map f is surjective, by the denition of the orbit G x. Finally, for all g, g

G and
x X we have
f(g g

G
x
) = f(gg

G
x
) = (gg

) x = g (g

x) = g f(g

G
x
),
which shows that f is G-equivariant.
As a corollary, we have one of the most useful formulas in group theory.
The Counting Formula. Let G be a nite group acting on a set X. Then the cardinality of an orbit
equals the index of the stabilizer of any point in the orbit. That is, for any x X we have
[G[
[G
x
[
= [G x[. (10)
Note that the right hand side of this equation depends only on the orbit, while the left side appears to
depend on the stabilizer of a particular point in the orbit. In fact, the left side depends only on the orbit
of the stabilizer. By Prop. 4.1 all stabilizers in a given orbit are conjugate, hence have the same order.
Let O
1
, . . . , O
k
be the orbits of G in X, and choose x
i
O
i
. Applying the counting formula to each
orbit, we have the weaker but still useful formula
[X[ =
k

i=1
[O
i
[ =
k

i=1
[G : G
x
i
]. (11)
4.1 The left regular action
Any group G acts on itself by left multiplication. More precisely, this action is given by the homo-
morphism L : G S
G
such that L
g
x = gx for all g, x G. This is called the left regular action;
one easily checks that it is free and transitive. Moreover, any free transitive G-action on a set X is
isomorphic to the left regular action (Exercise ...). If G is nite, say [G[ = n, then S
G
S
n
via a
labelling of the elements of G. Since the left regular action is faithful, this proves:
Proposition 4.2 A nite group G of order n is isomorphic to a subgroup of S
n
, via the left regular
action L : G S
n
.
40
In other words, S
n
contains every group of order n as a subgroup.
What are the cycle types of the elements of L(G)? For any permutation S
n
, the numbers in
the cycle type are the sizes of the orbits of on 1, 2, . . . , n. Since G acts freely on itself, the
subgroup g also acts freely on G and the orbits of g on G are just the cosets of g. The order d of
g divides n and there are n/d cosets of g in G, all of size d. This proves
Proposition 4.3 Let G be a nite group of order n, let g G, and let d be the order of g. Then under
the left regular action L : G S
n
the cycle type of L
g
is a product of n/d cycles of length d.
This has the following surprising corollary.
Corollary 4.4 Suppose G is a group of order n containing an element of order d where d is even and
n/d is odd. Then G has a normal subgroup of index two. In particular, G cannot be simple.
Proof: If g G has even order d with n/d odd, then L
g
is a product of an odd number of even cycles,
so sgn(L
g
) = 1. Hence the homomorphism sgn L : G 1 is nontrivial, so ker(sgn L) is a
normal subgroup of G of index two.
From this, we can prove another partial converse to Lagranges theorem.
Corollary 4.5 Suppose G is a group of order 2m where m is odd. Then G has a normal subgroup of
order m.
Proof: By Prop. 2.2 there exists g G of order two, satisfying the conditions of Cor. 4.4.
Since a group G also acts on itself by right multiplication, we also have the right regular action given
by the homomorphism R : G S
G
given by
g
x = xg
1
. [Check that this is a group action- in so
doing, youll see why we need the inverse.] Analogues of the above results all hold for R as well.
4.2 Group actions on coset spaces
Let G be a group with identity element e, let H G be a subgroup and let G/H be the set of left
cosets of H in G. Instead of H acting on G, we can consider G acting on G/H, via left multiplication:
g xH = gxH. When H = 1 we recover the left regular action. For nontrivial H, this action is
transitive (because any xH G/H is x eH) but no longer free, because the stabilizer of eH is the
nontrivial subgroup H. More generally, the stabilizer of xH is xHx
1
.
Every transitive group action is isomorphic to one of this form. Indeed, if Gacts transitively on a set X,
and we pick any x X, then the Main Theorem of Group Actions shows that this action is isomorphic
to the action of G on G/H, via left multiplication as just dened, where H = G
x
. However, describing
a transitive G- set as G/H contains the additional data of a basepoint, namely eH.
If G is nite, this gives a useful way to study subgroups of G, as follows.
41
Proposition 4.6 Let G be a nite group having a subgroup H G of index [G : H] = m. Then there
is a homomorphism
H
: G S
m
whose image is a transitive subgroup of S
m
and whose kernel is
given by
ker
H
=

gG
gHg
1
.
4.2.1 Applications to simple groups
Proposition 4.7 Let G be a simple group of order [G[ > 2 and let H be a subgroup of G of index m.
Then G is isomorphic to a subgroup of the alternating group A
m
. In particular, the order of G divides
1
2
m!.
Proof: The action of G on G/H gives a homomorphism
H
: G S
m
which is automatically
injective so G is isomorphic to the image G

=
H
(G). The composition
G

H
S
m
sgn
1
cannot be injective since [G[ > 2, so it must be trivial. This means that G

< A
m
.
Corollary 4.8 Let G be a nonabelian simple group and let H be a proper subgroup of G. Then
[G : H] 5.
Proof: We know that [G : H] ,= 2. If [G : H] = 3 then [G[ divides 3, so G is abelian, a contradiction.
If [G : H] = 4 then G is a subgroup of A
4
. The subgroup K < A
4
generated by the 22-cycles is
abelian and normal in A
4
, so GK is normal in G. If GK = G then G < K so G is abelian, which
it is not. So G K = 1. Then the composition G A
4
A
4
/K is injective. As [A
4
/K[ = 3, we
get same contradiction as before. Thus, we cannot have [G : H] = 4 either.
The inequality in Cor. 4.8 is sharp. For we will later see that the alternating group A
5
is simple. And
A
5
contains A
4
with index [A
5
: A
4
] = 5.
4.3 Actions by S
n
By denition, the symmetric group S
n
acts on the set 1, 2, . . . , n. But also for any positive integer
k n the group S
n
acts on on the set X
k
= x 1, 2, . . . , n : [T[ = k of k-element subsets of
1, 2, . . . n. One such subset is x
0
= 1, 2, . . . , k, whose stabilizer is isomorphic to S
k
S
nk
. The
counting formula says that
[X
k
[ =
[S
n
[
[S
k
[ [S
nk
[
=
n!
k!(n k)!
,
as it should.
42
4.4 Actions by GL
n
(F)
Let F be a eld. By denition the general linear group GL
n
(F) acts on the set of vectors in F
n
. But
also for any positive integer k n the group GL
n
(F) acts on the set X
k
of k-dimensional subspaces
of F
n
. We observed this already when k = 1, where X
1
= P
n1
is the projective space. For general k,
the set X
k
is called the Grassmannian of k-planes in F
n
.
One such subspace is x
0
= the span of e
1
, . . . , e
k

2
whose stabilizer P
k
consists of matrices mapping
x
0
to itself. We have
P
k
=
__
A C
0 B
_
: A GL
k
(F), B GL
nk
(F), C an arbitrary k (n k) matrix over F
_
.
3
By the Main Theorem of group actions, we have a GL
n
(F)-equivariant bijection
X
k
GL
n
(F)/P
k
.
Suppose now that F is a nite eld, say [F[ = q. Then GL
n
(F) = GL
n
(q) is a nite group. We
can use the Counting Formula to re-compute [GL
n
(F)[ and [X
k
[ (cf. (8)). First, for n = 1 we have
GL
1
(F) = F

, so
[GL
1
(F)[ = [F

[ = q 1.
For n > 1 the group GL
n
(F) acts transitively on the set Y
n
of nonzero vectors in F
n
. The stabilizer of
the vector e
1
is the subgroup
H
n
=
__
1 C
0 B
_
: B GL
n1
(F), C an arbitrary 1 (n 1) matrix over F
_
.
Counting the possible choices for B and C and observing that [F
n
[ = q
n
, we nd that
[Y
n
[ = q
n
1, [H
n
[ = [GL
n1
(F)[ q
n1
.
The Counting Formula says that
q
n
1 = [Y
n
[ =
[GL
n
(F)[
[H
n
[
=
[GL
n
(F)[
[GL
n1
(F)[ q
n1
.
This gives the recursive formula
[GL
n
(F)[ = q
n1
(q
n
1) [GL
n1
(F)[.
and by induction we recover the formula from(8):
[GL
n
(F)[ = q
n(n1)/2
(q
n
1)(q
n1
1) (q
2
1)(q 1). (12)
2
Recall that e
1
, . . . , e
n
is the standard basis of F
n
.
3
Clearly P
k
depends also on n and F. We use the letter P because P
k
is a special case of a parabolic subgroup in the
theory of algebraic groups.
43
To simplify this formula, we dene the q-factorial
[n!]
q
=
(q
n
1)(q
n1
1) (q
2
1)(q 1)
(q 1)
n
.
If we pretend that q is a variable, we have (n!)
q
n!, the usual factorial, as q 1. With this notation,
we have
[GL
n
(F)[ = q
n(n1)/2
(q 1)
n
[n!]
q
.
Now to nd [X
k
[ we need only divide by [H
k
[. Considering the possible choices for A, B, C in the
element
_
A C
0 B
_
H
k
, we nd
[P
k
[ = [GL
k
(F)[ [GL
nk
(F)[ q
k(nk)
= q
k(k1)/2
(q 1)
k
[k!]
q
q
(nk)(nk1)/2
(q 1)
nk
[(n k)!]
q
q
k(nk)
= q
n(n1)/2
(q 1)
n
[k!]
q
[(n k)!]
q
.
It follows that the number of k-dimensional subspaces of F
n
is given by
[X
k
[ =
[n!]
q
[k!]
q
[(n k)!]
q
,
the q-binomial coefcient. If we pretend that q is a variable, this reduces to the ordinary binomial co-
efecient
_
n
k
_
as q 1. In this vague sense, k-element subsets of 1, 2, . . . , n are like k-dimensional
subspaces of F
n
over the (nonexistent) eld of one element, and S
n
is like GL
n
(F) over this eld.
All of these formulas came from the Counting Formula. Later we will see other relations between S
n
and GL
n
(F), where the eld F is arbitrary.
4.5 Double cosets
Suppose a group G acts transitively on a set X and we wish to study the action of a subgroup K < G
on X. From the Main Theorem on Group Actions, we may assume that X = G/H for some subgroup
H < G, possibly equal to H.
So let us begin with a group G and arbitrary subgroups K, H. Let the group K H act on the set G
by the rule
(k, h) g = kgh
1
.
The orbits of this action are called (K, H)-double cosets. Each double coset is of the form
KxH = kxh : k K, h H
for some x G. As with all group actions, the set G is partitioned as
G =

x
KxH,
44
where x runs over representatives for the (K, H)-double cosets.
Each double coset KxH is a union of left cosets of H and KxH/H G/H is the set of the left cosets
of H contained in KxH. In fact KxH/H is exactly the K-orbit of xH in G/H. By the Main Theorem
of Group actions, this orbit is a K-set which must be equivalent to K/J for some subgroup J K.
Proposition 4.9 The stabilizer of xH in K is K xHx
1
. Hence we have a K-equivariant bijection
K/(K xHx
1
)

KxH/H,
sending k(K xHx
1
) kxH/H.
Proof: If k K, then
kxH = xH x
1
kx H k xHx
1
K.
The rest of the proposition follows from the Main Theorem of Group Actions.
If G is nite, we can use Prop. 4.9 to compute the number of elements in a given double coset KxH,
namely
[KxH[ =
[K[ [H[
[K xHx
1
[
.
We can also view KxH as the H-orbit of Kx in KG, and the stabilizer of Kx in H is x
1
Kx H,
so that KKxH (x
1
Kx H)H, as H-sets.
The set of (K, H)-double cosets in G is denoted by KG/H. The number [KG/H[ of double cosets
may be thought of in three ways:
[KG/H[ is the number of K H orbits on G.
[KG/H[ is the number of K-orbits on G/H.
[KG/H[ is the number of H-orbits on KG.
In contrast to ordinary cosets, there is no simple formula for [KG/H[ in general, because the action
of K on G/H (or the action of H on KG) need not be free.
It is especially interesting to know when [KG/H[ is small. For example, it follows from the deni-
tions that
[KG/H[ = 1 G = KH K is transitive on G/H.
Here is a deeper situation. Let G be a group acting transitively on a set X. To avoid trivialities we
assume X has more than one element. Then G also acts on X X, via g (x, y) = (g x, g y). This
action cannot be transitive, because the diagonal X = (x, x) : x X is an orbit. Hence there are
at least two orbits on X X. We say that G acts doubly transitively on X if G has exactly two orbits
on X X. This means G acts transitively on (x, y) X X : x ,= y. In other words, G is doubly
transitive on X if for any two pairs of distinct elements (x, y) and (x

, y

) in X X, there is a single
element g G such that g x = x

and g y = y

.
45
Proposition 4.10 Let G be a group acting transitively on a set X, and assume [X[ > 1. The following
are equivalent.
1. [G
x
G/G
x
[ = 2 for any x X.
2. For any x X, the stabilizer G
x
is transitive on X x.
3. G acts doubly transitively on X.
Proof:
(1 2): We have [G
x
G/G
x
[ = 2 exactly when G
x
has exactly two orbits in X. One of these orbits
is x so [G
x
G/G
x
[ = 2 exactly when G
x
is transitive on X x.
(2 3): Assume G
x
is transitive on X x for any x X. Let (x, y) and (x

, y

) be two pairs
of distinct elements of X. Since G is transitive on X, there is g G such that g x

= x. Note that
g y

,= x since x

,= y

. By our assumption, there is h G

x
such that h (g y

) = y. For the element

hg G we have hg x

= h x = x and hg y

= y, so hg (x

, y

) = (x, y). Hence G is doubly

transitive on X.
(3 2): Assume G is doubly transitive on X and let x X. Let y, y

be two elements of X x.
Since G is doubly transitive, there exists g G such that g x = x and g y = y

. This shows that G

x
is transitive on X x.
In other words, if H is any subgroup of G then G acts doubly transitively on G/H if and only if
G = H HgH, (disjoint)
For any element g outside of H.
Example 1: S
n
Let G = S
n
and let H S
n1
be the subgroup stabilizing the number 1. The
permutations in H are precisely those whose disjoint cycle expression does not involve the number 1,
and we have S
n
/H 1, 2, . . . , n as S
n
-sets. Given 1 i < j n there exists S
n
sending 1, 2
to i, j respectively. It follows that S
n
is doubly-transitive on 1, 2, . . . , n and we have
S
n
= H HH.
The elements of HH are precisely those permutations whose disjoint cycle expression involves the
number 1.
Example 2: GL
n
(F) Let G = GL
n
(F) and let P be the subgroup stabilizing the line through e
1
. The
matrices in P are precisely those whose rst column is of the form
_

0
.
.
.
0
_

_
,
46
and we have G/P P
n1
(F) as GL
n
(F)-sets. Any two linearly independent vectors v
1
, v
2
F
n
are
contained in a basis v
1
, v
2
, . . . , v
n
of F
n
. The matrix g GL
n
(F) whose j
th
column is v
j
sends
the lines through e
1
, e
2
to the lines through v
1
, v
2
respectively. It follows that GL
n
(F) acts doubly
transitively on P
n1
(F) and we have
GL
n
(F) = P PgP
where g is any matrix in GL
n
(F) not in P.
4.6 Conjugation
Conjugation can be viewed as an action of G on itself. This action is the homomorphism c : G S
G
given by c
g
(x) = gxg
1
. Understanding this action is another important way to know a given group.
In this context, the stabilizer of an element x G is the centralizer C
G
(x)g G : gxg
1
= x and
the orbit of x G is its conjugacy class x
G
= x
g
: g G. Finally, the kernel of the conjugation
action is the center Z(G) of G.
If G is nite, the formula
[C
G
(x)[ [cc(x)[ = [G[, for all x G (13)
is a the Counting Formula for the conjugation action.
5 Sylow Theorems and Applications
We have seen that the converse of Lagranges Theorem is false, in general: If G is a group and d is
a divisor of [G[ then G need not have a subgroup of order d. The simplest example is G = A
4
with
[G[ = 12, which has no subgroup of order d = 6. Note that 6 is a product of the two smallest distinct
primes. On the other hand, A
4
does have proper subgroups of orders 2, 3, 4, which are the other proper
divisors of 12.
5.1 Sylow p-subgroups
A p-subgroup of a nite group G is a subgroup of G whose order is a power of a prime p. By
Lagranges theorem, G can have nontrivial p-subgroups only if p divides the order of G. Write the
order as [G[ = m p
r
, where p m. A Sylow p-subgroup of G is a p-subgroup P G having the
maximal order [P[ = p
r
allowed by Lagranges theorem. Equivalently, a p-subgroup P Gis a Sylow
p-subgroup exactly when p does not divide the index [G : P].
Example: Let F be a eld with [F[ = q, a power of a prime p. We have seen that the group
GL
n
(F) = GL
n
(q) has order
[ GL
n
(q)[ = q
n(n1)/2
(q 1)(q
2
1) (q
n
1).
47
Let U
n
(q) GL
n
(q) be the subgroup of upper triangular matrices with 1s on the diagonal. Thus U
n
(q)
consists of all matrices of the form
_

_
1 . . .
0 1 . . .
0 0 1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 . . . 1
_

_
,
where the entries above the diagonal of a matrix in U
n
(q) can be arbitrary elements of F, so we
have
[U
n
(q)[ = q
0+1+2++(n1)
= q
n(n1)/2
,
which is a power of p. Since p does not divide (q 1)(q
2
1) (q
n
1), it follows that U
n
(q) is a
Sylow p-subgroup of GL
n
(q).
For general groups it is not obvious that Sylow p-subgroups exist.
Lemma 5.1 Let G be a nite group and let H be a subgroup of G. Assume that G has a Sylow
p-subgroup. Then H has a Sylow p-subgroup.
Proof: Let P be a Sylowp-subgroup of Gand consider the H-orbits on G/P under the action h gP =
hgP. Since p does not divide the index [G : P], there exists an H-orbit O G/P such that p does not
divide [O[. Let gP O and let Q be the stabilizer in H of gP.
For all q Q we have qgP = gP, which implies that q gPg
1
. Hence Q is contained in the Sylow
p-subgroup gPg
1
. By Lagranges theorem, Q is a p-group. We chose O so that p does not divide
[O[ = [H : Q]. It follows that Q is a Sylow p-subgroup of H.
Remark: The proof actually shows that if G has Sylow p-subgroups, then one of them meets H in a
Sylow p-subgroup of H.
Theorem 5.2 (Sylow I) Let G be a nite group whose order is divisible by a power p
i
of a prime p.
Then G has a subgroup of order p
i
.
Proof: By Prop. 3.13 it sufces to prove that G has a Sylow p-subgroup. Let n = [G[. Then G is
isomorphic to a subgroup of S
n
by Cayleys theorem. And S
n
is isomorphic to a subgroup of GL
n
(p).
Hence G is isomorphic to a subgroup of GL
n
(p). We have seen in the example above that GL
n
(p) has
a Sylow p- subgroup. By Lemma 5.1, the group G has a Sylow p-subgroup P.
Alternate proof: The proof just given is very simple, and is my favorite, but one can object that such
a fundamental fact as Thm. 5.2 should not depend on auxilliary groups like S
n
and GL
n
(p). Here is
another, more intrinsic proof that produces p-subgroups for all 0 i r at once. We rst need a
lemma about nite sets. For any nonzero integer n, let v
p
(n) be the highest power of p dividing n.
Lemma 5.3 Let S be a nite set of cardinality mp
r
, where p is a prime. Let X be the set of subsets of
S of cardinality p
r
: X = A S : [x[ = p
r
. Then v
p
([X[) = v
p
(m).
48
Proof of the lemma: We have
[X[ =
_
mp
r
p
r
_
= m
p
r
1

k=1
_
mp
r
k
p
r
k
_
.
Let 1 k p
r
1 and write k = np
s
for some integer n not divisible by p, with s = v
p
(k) < r. Then
mp
r
k = mp
r
np
s
= p
s
(mp
rs
n),
so v
p
(mp
r
k) = v
p
(k) does not depend on m. Hence v
p
(mp
r
k) = v
p
(p
r
k) and the lemma is
proved.
Now we can prove Thm. 5.2. Let X be the set of subsets of G of cardinality p
r
and let s = v
p
(m), so
that v
p
([G[) = r + s.
The group Gacts on X by left multiplication: gA = gA, for g Gand A X. Thus, X is partitioned
into G-orbits. By the lemma, v
p
([X[) = s. Hence there exists a G-orbit O X such that [O[ is not
divisible by p
s+1
. Choose A O and let G
A
be the stabilizer of A. That is,
G
A
= g G : gA = A.
We will prove the theorem by showing that [G
A
[ = p
r
.
By the Orbit Counting Theorem, we have
[O[ = [G : G
A
].
Since p
s+1
does not divide [O[, it follows that p
r
must divide [G
A
[. On the other hand, choose a A
and let f : G
A
A be the function f(g) = ga, which makes sense for g G
A
. If f(g) = f(g

) for
g, g

G
A
then ga = g

a so g = g

. Hence f is injective. Since [A[ = p

r
, this shows that [G
A
[ p
r
.
It now follows that [G
A
[ = p
r
, as claimed.
The next Sylow Theorem asserts that Sylow p-subgroups are unique up to conjugation. The proof of
this also reveals information about the number of Sylow p-subgroups, which we put in the statement.
First we need two lemmas.
Lemma 5.4 Let P be a Sylow p-subgroup of a group G. Then every p-subgroup of the normalizer
N
G
(P) is contained in P.
Proof: Let H N
G
(P) be a p-subgroup. Since H normalizes P, the product HP is a subgroup of
N
G
(P) and P HP. Since
HP/P H/H P
is a quotient of the p-group H, and P is a p-group, it follows that HP is a p-group. And P HP. But
P is a maximal p-subgroup of G, by the denition of Sylow p-subgroup. Hence HP = P, meaning
that H P.
49
Lemma 5.5 Let H be a p-group acting on a nite set X, and let X
H
= x X : h x = x be the
xed-point set of H in X. Then
[X
H
[ [X[ mod p.
Proof: The size [X[ of X is the sum of the sizes of the orbits of H in X. If O is an H-orbit in X
and [O[ 1 then [O[ is a power of p, since H is a p-group. In this case [O[ 0 mod p. Hence [X[
is congruent modulo p to the number of orbits consisting of one element only. That is, [X[ [X
H
[
mod p. .
Now we can state and prove the second Sylow theorem.
Theorem 5.6 (Sylow II) Let G be a nite group of order mp
r
, where p is a prime not dividing m.
Then the following hold.
1. Any two Sylow p-subgroups of G are conjugate. That is, if P and Q are two subgroups of G of
order p
r
then there exists g G such that gPg
1
= Q.
2. The number n
p
of p-Sylow subgroups of G divides m and is of the form 1 + kp where k is an
integer.
3. Every p-subgroup of G is contained in a Sylow p-subgroup of G.
Proof: Let X = P G : [P[ = p
r
be the set of Sylow p-subgroups of G. Then G acts on X by
conjugation. For this action we have g P = gPg
1
for g G and P X, and the stabilizer of P is
the normalizer N
G
(P).
Consider the xed points of a Sylow p-subgroup P acting on X by conjugation. I claim that
X
P
= P. (14)
That is, I claim that P normalizes no other Sylow p-subgroup but itself. For suppose Q X
P
is any
xed point of P in X. This means P N
G
(Q), so that P is p-subgroup of the normalizer of the Sylow
p-subgroup Q. By Lemma 5.4, we have P Q, hence P = Q since both P and Q have order p
r
. This
proves (14).
Now let P X and let O = gPg
1
: g G be the G-orbit of P in X. By (14), we have O
P
= P.
From Lemma 5.5 we have [O[ [O
P
[ mod p. Since [O
P
[ = 1, this means that
[O[ 1 mod p. (15)
I claim that that X = O. Suppose not. Then there exists Q X with Q / O. Applying (14) to
the Sylow p-subgroup Q, we have X
Q
= Q. Since Q / O, it follows that O
Q
is empty. Applying
Lemma 5.5 to the action of the p-group Q on the set O, we have
[O[ [O
Q
[ = 0 mod p. (16)
50
This contradicts equation (15). Hence Q cannot exist and we have X = O.
Now equation (15) says that [X[ 1 mod p. And since X is a single orbit, the Orbit Counting
Formula says that
n
p
= [X[ = [G : N
G
(P)] = [G : P]/[N
G
(P) : P] = m/[N
G
(P) : P],
or
m = [N
G
(P) : P] n
p
,
so n
p
[ m. Thus, items 1 and 2 are proved.
For item 3, let H be any p-subgroup of G. Then [X
H
[ [X[ 1 mod p, by Lemma 5.5 again and
item 2 which has been proved. It follows that X
H
is nonempty. This means H N
G
(P) for some
P X. By Lemma 5.4 again, we have H P.
From item 1 of Thm. 5.6 we get the following condition for a Sylow p-subgroup to be normal.
Corollary 5.7 A Sylow p-subgroup of G is normal in G if and only if it is the unique Sylow p-subgroup
of G.
It follows from Thm. 5.6 that the order of G may be written as
[G[ = p
r
(1 + kp), (17)
where
p
r
= [P[, = [N
G
(P)/P[, 1 + kp = [G : N
G
(P)].
Of course r, , k all depend on p, and p . Let us call (17) the p-factorization of [G[. Besides giving
[G[, it displays [P[ = p
r
and [N
G
(P)[ = p
r
, as well as n
p
= 1 + kp. We have k = 0 if and only if
P G.
5.1.1 Small examples.
For G = A
4
, S
4
, A
5
, S
5
, SL
2
(7), PSL
2
(7), the p-factorizations for each prime p dividing [G[ are shown.
A
4
:
p
r
(1 + pk) N
G
(P)
3 1 (1 + 3) C
3
2
2
3 (1 + 0) A
4
S
4
:
[p
r
(1 + pk) N
G
(P)
3 2 (1 + 3) S
3
2
3
1 (1 + 2) D
4
A
5
:
p
r
(1 + pk) N
G
(P)
5 2 (1 + 5) D
5
3 2 (1 + 3) S
3
2
2
3 (1 + 2 2) A
4
S
5
:
p
r
(1 + pk) N
G
(P)
5 4 (1 + 5) F
20
3 4 (1 + 3 3) S
3
S
2
2
3
1 (1 + 2 7) D
4
The normalizer N
S
5
(P) of the Sylow 5-subgroup P = (1 2 3 4 5) in S
5
is generated by P and the 4-
cycle which squares (1 2 3 4 5) via conjugation, namely (2 3 5 4). It is called F
20
in honor of Frobenius,
51
although Galois discussed it at length much earlier. One can think of F
20
as the ax + b-group over the
eld of ve elements, isomorphic to
__
a b
0 1
_
: a (Z/5Z)

, b Z/5Z
_
,
which corresponds to a Borel subgroup of PGL
2
(5) under the isomorphism S
5
PGL
2
(5).
SL
2
(7) :
p
r
(1 + pk) N
G
(P)
7 6 (1 + 7) G
21
C
2
3 4 (1 + 3 3) C
3
C
4
2
4
1 (1 + 2 10) Q
16
PSL
2
(7) :
p
r
(1 + pk) N
G
(P)
7 3 (1 + 7) G
21
3 2 (1 + 3 3) S
3
2
3
1 (1 + 2 10) D
4
Here G
21
is the unique nonabelian group of order 21 (see Prop. 9.6) realized here as the subgroup of
the upper-triangular matrices in SL
2
(7) whose diagonal entries have odd order. The group C
3
C
4
is
the nonabelian group of order 12 other than A
4
and Q
16
is the generalized quaternion group of order
16 (see 3.6).
5.1.2 Groups of order pq
We can use Sylows theorems to classify groups of order pq, where p and q are distinct primes. Assume
p < q and let P, Q be Sylow p- and q- subgroups of G, respectively. Consider the q-factorization
qp = q (1 + qk). Since p is prime and p < q we must have k = 0 and = p. Hence Q G
and QP is a subgroup of G, with Q P = 1 because the orders of P, Q are relatively prime. Since
QP/P Q/QP = Q, it follows that [QP[ = pq, so QP = G. Therefore G is a semidirect product:
G = QP.
The structure of G is now completely determined by the homomorphism
: P (Z/qZ)

given by xyx
1
= y
(x)
,
where x P and y Q. For if y generates Q then every element g G can be uniquely expressed as
g = y
b
x, for b Z/qZ and x P, and the product of two such elements is
(y
b
x)(y
b

) = y
b
(xy
b

x
1
) xx

= y
b
y
b

(x)
xx

= y
b+b

(x)
xx

.
If is trivial then G is abelian and G = QP C
pq
.
Suppose is nontrivial. Then is injective, since p is prime, and we must have q 1 mod p. The
group (Z/qZ)

is cyclic of order q 1, hence has a unique subgroup A

p
of order p, and we have
(P) = A
p
.
Let G
pq
be the following subgroup of the ax + b-group over Z/qZ:
G
pq
=
__
a b
0 1
_
: a A
p
, b Z/qZ
_
.
52
The function : G G
pq
dened by
(y
b
x) =
_
(x) b
0 1
_
is clearly bijective. In fact it is a group homomorphism, for we have
(y
b
xy
b

) = (y
b+b

(x)
xx

) =
_
(xx

) b + b

(x)
0 1
_
=
_
(x) b
0 1
_

_
(x

) b

0 1
_
= (y
b
x)(y
b

x).
Hence is an isomorphism G G
pq
. To summarize, we have shown the following.
Proposition 5.8 Let G be a group of order pq where p, q are primes with p < q.
1. If G is abelian then G C
pq
is cyclic of order pq.
2. If G is nonabelian then q 1 mod p and G G
pq
.
For example the two groups of order 2q, where q is an odd prime, are C
2q
and the dihedral group
D
q
G
2q
.
5.2 Sylow subgroups in GL
n
and ag varieties
Let F be a nite eld with [F[ = q, a power of a prime p and let G = GL
n
(q). We have seen that the
subgroup
U = U
n
(q) =
_

_
1 . . .
0 1 . . .
0 0 1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 . . . 1
_

_
G (18)
is a Sylow p-subgroup of G. The normalizer of U is the Borel subgroup B appearing in the Bruhat
decomposition:
N
G
(U) = B =
_

_
. . .
0 . . .
0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 . . .
_

_
,
where the entries are arbitrary in F as before, and the diagonal entries are nonzero elements of F.
We have B = U T, where T is the diagonal subgroup of G, and [B[ = q
n(n1)/2
(q 1)
n
. The
p-factorization of [G[ is therefore
[G[ = q
n(n1)/2
(q 1)
n

(q
n
1) (q
2
1)(q 1)
(q 1)
n
.
53
We observe that
(q
n
1) (q
2
1)(q 1)
(q 1)
n
=
n

k=1
(1 + q + + q
k1
) 1 mod p,
as guaranteed by Sylows theorem. This is the number of Sylow p-subgroups of G, of which U is only
one. Let X be the set of all Sylow p-subgroups of G. By the Main Theorem of Group Actions, the
mapping
G/B X, sending gB gUg
1
is a G-equivariant bijection. The set X involves the complete projective geometry of the vector space
V = F
n
, as I will explain.
For n = 2, we have seen that B is the stabilizer of the line
o
= Fe
1
in V , so in this case G/B is also
identied with the set P(V ) of all lines V . Thus we have a G-equivariant bijective correspondence
P(V ) = g
o
gB gUg
1
= U

X
between lines in V and Sylow p-subgroups of G = GL
2
(q). Given a line , the subgroup U

is the set
of elements in G which act trivially on both and V/. All elements of U

have the form

_
1
0 1
_
with
respect to any basis v
1
, v
2
of V with v
1
, but not all elements of U

have this form with respect to

the original basis e
1
, e
2
, unless =
o
.
For n = 3 we have both lines and planes in V = F
3
, and a given plane may or may not contain a given
line. A ag in V is a pair (, ), where is a line contained in a plane V . Such congurations
comprise the complete projective geometry of V . Let T(V ) be the set of all ags in V . The group
G = GL
3
(F) acts on T(V ) via g (, ) = (g , g ). This G-action is transitive and B is the
stabilizer of the ag (
o
,
o
), where
o
= Fe
1
and
o
= Fe
1
Fe
2
. Thus we have a G-equivariant
bijective correspondence
T(V ) (, ) = g (
o
,
o
) gB gUg
1
= U
(,)
X
between ags in V and Sylow p-subgroups of G = GL
3
(q). Given a ag (, ), the subgroup U
(,)
is the set of elements in G which preserve and and act trivially on , / and V/. All elements
of U
(,)
have the form
_
_
1
0 1
0 0 1
_
_
with respect to any basis v
1
, v
2
, v
3
of V for which v
1
and
v
1
, v
2
, but not all elements of U
,
will have this form with respect to the original basis e
1
, e
2
, e
3
,
unless =
o
and =
o
.
For general n 2, a ag in V = F
n
is a sequence of subspaces
f = (V
0
, V
1
, . . . , V
n1
, V
n
), where 0 = V
0
V
1
V
2
V
n1
V
n
= V,
and dimV
i
= i for all 0 i n. The ag variety of G is the set T(V ) of all ags f V . The action
of G = GL
n
(q) on T(V ) permutes the ags in V . That is, we have
g (V
0
, V
1
, . . . , V
n
) = (V
0
, gV
1
, . . . , gV
n1
, V
n
).
54
This G-action is transitive and B is the stabilizer of the ag
f
o
= (0, Fe
1
, Fe
1
Fe
2
, , Fe
1
Fe
2
Fe
n1
, V ).
Thus we have a G-equivariant bijective correspondence
T(V ) f = g f
o
gB gUg
1
= U
f
X
between ags f V and Sylow p-subgroups U
f
G = GL
n
(q). Given a ag f = (V
0
, V
1
, . . . , V
n
),
the subgroup U
f
is given by
U
f
= g G : gV
i
= V
i
and g acts trivially on V
i
/V
i1
for all i n.
All elements of U
f
have the form (18) with respect to any basis v
1
, v
2
, . . . , v
n
of V for which V
i
=
Fv
1
Fv
i
, but not all elements of U
f
have this form for the original basis e
1
, . . . , e
n
unless
f = f
o
.
The total number of ags in V is
[T(V )[ = [G/B[ =
(q
n
1) (q
2
1)(q 1)
(q 1)
n
,
which we have seen to be a polynomial P
n
(q) in q of degree n(n 1)/2.
For any eld F there is still a complete geometry of ags in V , dened in the same way. Let us take
F = C, with G = GL
n
(C) and U, B dened as above for the eld F = C. The polyomial P
n
(q) still
counts something about G/B, but not something so elementary as points, because now the ag variety
G/B is innite.
In fact, G/B is a complex projective variety of dimension n(n 1)/2.
4
Any smooth complex pro-
jective variety X of dimension d has cohomology groups H
i
(X) for i = 0, 1, . . . , 2d which are nite
dimensional complex vector spaces whose dimensions dimH
i
(X) are called the Betti numbers of
X. For example if X = P
d
(C) then dimX = d and the Betti numbers are dimH
i
(X) = 1 for
i = 0, 2, 4, . . . , 2d and dimH
i
(X) = 0 for all other i.
It turns out that the polynomial P
n
(q) above encodes the Betti numbers of G/B: Regarding q as a
variable, we have
n(n1)/2

i=0
dimH
2i
(G/B)q
i
= P
n
(q) =
(q
n
1) (q
2
1)(q 1)
(q 1)
n
.
The observation that Betti numbers of G/B over C are determined by the number of points on G/B
over nite elds is deep; it led to Weils conjectures, which were eventually proved by Deligne.
4
A complex projective variety is a closed subset of some projective space P
N
(C) dened by polynomial equations. It
turns out that G/B P
N
(C) for N = 2
n(n1)/2
1.
55
5.3 The Burnside Transfer Theorem
If the order of a nite group G factors as [G[ = mp
r
with p a prime not dividing m, it is natural to ask
if the group G itself has a corresponding factorization as G M

P, where P is a Sylow p-subgroup
of G and M G has order m. For example if G = C
mp
r is cyclic then G C
m
C
p
r . Since Sylow
p-subgroups are not normal in general groups, the only reasonable hope would be that M G and
G M P. Such a subgroup M, if it exists, is called a normal p-complement.
Normal p-complements do not always exist. For example, S
4
has no normal p-complement for either
p = 2 or p = 3.
Lemma 5.9 Suppose [G[ = mp
r
, where p m. Then G has a normal p-complement if and only if the
elements of G of order prime to p form a subgroup of G.
Proof: Suppose the elements of G of order prime to p form a subgroup M G. Then M is normal in
G, since conjugation preserves orders of elements. Hence MP MP is a subgroup of G. To show
that MP = G, we must show that [M[ = m. By Cauchys theorem, p does not divide [M[ since M
has no elements of order p. It therefore sufces to show that G/M is a p-group. If this fails, then G/M
has an element xM of order d > 1, where p d. Then x
d
M, so x
d
has order k, where p k. That
means the order of x divides kd, and p kd, so x M, a contradiction. Hence G/M is a p-group, and
G = MP as claimed.
Conversely, let M be a normal subgroup of G of order m. Since p m, every element of M has order
prime to p. Suppose g G has order k, with p k. Then the order of gM in G/M has order dividing
k. But G/M is a p-group, so gM = 1, meaning that so g M. Therefore M is exactly the set of
elements of G of order prime to p, and this set is a subgroup of G.
Example 1: Let G = D
n
be dihedral of order 2n. Write n = m2
r
, where m is odd. Let M C
m
be
the subgroup generated by a rotation of order m. Then M is a normal 2-complement in G. The Sylow
2-subgroups are isomorphic to D
2
r , and we have
D
n
C
m
D
2
r ,
where the reections in D
2
r act by inversion on C
m
.
Example 2: Let G = A
4
. The subgroup of elements of order prime to 2 forms a subgroup K
C
2
C
2
, and A
4
= K P, where P is a Sylow 3-subgroup, acting on M via an element of order
three in Aut(M) = GL
2
(2). Thus, K is a normal 3-complement. However, A
4
has no normal 2-
complement, since the elements of order three lie in no proper subgroup of A
4
.
Having a normal p-complement is equivalent to having a surjective homomorphism
f : G P.
In general, a group G does not admit nontrivial homomorphisms onto a proper subgroup H G.
Suppose, however that H is abelian, and let n = [G : H]. Let x
1
, x
2
, . . . , x
n
be representatives for the
56
cosets in G/H, and let : G S
n
be the action of G on G/H. Thus, each g G gives a permutation

g
S
n
such that
gx
i
H = x
g(i)
H, 1 i n.
This means that for each i we have an element h
i
(g) H such that
gx
i
= x
g(i)
h
i
(g).
Thus, for each i we have a function h
i
: G H given by
h
i
(g) = x
1
g(i)
g x
i
.
Each individual function h
i
is not a group homomorphism. However, we have
Lemma 5.10 If H is an abelian subgroup of G then the function T : G H given by
T(g) =
n

i=1
h
i
(g)
is a group homomorphism which does not depend on the choice of coset representatives x
i
.
Proof: Let x

i
be another set of coset representatives for G/H. Then x

i
= x
i
k
i
for some elements
k
i
H. The new functions h

i
(g) are given by
h

i
(g) = (x

g(i)
)
1
g x

i
= k
1
g(i)
x
1
g(i)
g x
i
k
i
= k
1
g(i)
h
i
(g)k
i
= h
i
(g)k
1
g(i)
k
i
,
since H is abelian. Taking the product we get
n

i=1
h

i
(g) =
n

i=1
h
i
(g)k
1
g(i)
k
i
= T(g),
since

k
g(i)
=

k
i
. Hence T(g) is independent of the choice of coset representatives x
i
.
If g, y are two elements of G, we have
T(gy) =
n

i=1
x
1
gy(i)
gyx
i
=
n

i=1
(x
1
gy(i)
gx
y(i)
) (x
1
y(i)
yx
i
)
=
n

i=1
(x
1
g(y(i))
gx
y(i)
) T(y)
=
n

i=1
(x
1
g(i)
gx
i
) T(y) = T(g)T(y),
again since the product over
y
(i) equals the product over i. Hence T : G H is a homomorphism.
57

The homomorphism T in Lemma 5.10 is called the Transfer map. We can compute T(g), for xed
g G, as a product over double cosets as follows. Let = g and choose a -orbit O = xH in
G/H, and set a = [O[. Let x
1
= x, x
2
= gx, x
3
= g
2
x, . . . , x
a
= g
a1
x. Then
h
1
= h
2
= = h
a1
= 1 and h
a
= x
1
gx
a
= x
1
g
a
x.
The last term is the only contribution fromO to the product in T(g). Note that if acts freely on O then
g
a
= 1, so x
1
g
a
x = 1. Doing this for each orbit, we nd that if xH, yH, zH, . . . are representatives
of the distinct -orbits on G/H, having sizes a, b, c, . . . , then
T(g) = (x
1
g
a
x) (y
1
g
b
y) (z
1
g
c
z) .
Each term in parentheses lies in the abelian group H, so the product can be taken in any order, and the
only terms contributing to T(g) come from -orbits which are not free.
Now suppose p is a prime dividing [G[ and P is a Sylow p-subgroup of G. The center Z(P) is
nontrivial, and abelian, so we have the transfer homomorphism
T : G Z(P).
Lemma 5.11 If x, y Z(P) are conjugate in G, then they are conjugate in N
G
(P).
Proof: If x
g
= y, then y Z(P) Z(P)
g
= Z(P) Z(P
g
), so P, P
g
C
G
(y). Thus, P and P
g
are
two Sylow p-subgroups of C
G
(y), so there is h C
G
(y) such that P = P
gh
. Then gh N(P) and
x
gh
= y
h
= y, so x and y are conjugate in N(P), as claimed.
Remark: Two elements x, y in a subgroup H G are fused in G if they are conjugate in G, but not
necessarily conjugate by an element of N
G
(H). A result like Lemma 5.11 is therefore said to control
fusion.
Now, if g Z(P), each term x
1
g
a
x in T(g) is conjugate in G to g
a
hence is conjugate in N
G
(P) to
g
a
, by Lemma 5.11.
Let us now assume that that
N
G
(P) = C
G
(P).
This is equivalent to assuming that P = Z(P) (that is, P is abelian) and that the conjugation action of
N
G
(P) on P is trivial. Consider the transfer map
T : G P.
Then x
1
g
a
x = g
a
for each term in T(g), so that
T(g) = g
a+b+c+
= g
m
,
where a, b, c, . . . are the sizes of the P-orbits on G/P and m = [G : P]. Since p m, the map g g
m
is an automorphism of the abelian group P. Hence the transfer map T : G P is surjective; we even
have T(P) = P. This proves the following.
58
Theorem 5.12 (Burnside Transfer Theorem) Let Gbe a nite group and let P be a Sylowp-subgroup
of G such that N
G
(P) = C
G
(P). Then G has a normal p-complement. In fact, G = M P, where M
is the kernel of the transfer map T : G P.
Example 3: Let G be a group of order 728 = 2
3
7 13 and consider the 7-factorization
[G[ = 7 (1 + 7k).
Assume the Sylow 7-subgroups of G are not normal in G. Then, since n
7
= 1 + 7k divides 8 13 and
k > 0, we must have n
7
= 8. Hence [N
G
(P)[ = 7 13. Since 7 , 1 mod 13, it follows from Prop.
9.6 that N
G
(P) is abelian, so that N
G
(P) = C
G
(P). Hence G contains a normal 7-complement M of
order 8 13.
Corollary 5.13 Let p be the smallest prime dividing [G[. If the Sylow p-subgroups in G are cyclic,
then G has a normal p-complement.
Proof: Let P be a Sylow p-subgroup of G and assume that P is cyclic of order p
r
. Then
N
G
(P)/C
G
(P) Aut(P) (Z/p
r
Z)

.
The latter group has order p
r1
(p 1). Since P is abelian, we have P C
G
(P), so that the order of
N
G
(P)/C
G
(P) is prime to p, hence must divide p 1. But p is the smallest prime dividing [G[, so no
prime can divide N
G
(P)/C
G
(P). Hence N
G
(P) = C
G
(P) and G has a normal p-complement, by the
Burnside Transfer Theorem.

Example: Suppose [G[ = p

1
p
2
p
k
is a product of distinct primes p
1
> p
2
> > p
k
. Each Sylow
p
i
-subgroup P
i
Gis cyclic. By Cor. 5.13 the Sylow p
k
-subgroup P
k
has a normal complement G
k
of
order p
1
p
2
p
3
p
k1
, so G = G
k
P
k
. Repeat with G replaced by G
k
, etc. We get G = P
1
P
2
P
k
with subgroups G
i
= P
1
P
i1
, such that
1 = G
0
< G
1
< < G
k
< G
k+1
= G
and G
i
G
i+1
and [G
i+1
/G
i
[ = p
i
for 0 i k.
Corollary 5.14 Let p be the smallest prime dividing [G[. If the Sylow p-subgroups in G are iso-
morphic to C
p
C
p
then either G has a normal p-complement or the following holds: p = 2 and
[N
G
(P)/C
G
(P)[ = 3 and G has a unique conjugacy class of elements of order two.
Proof: The idea is similar to that of Cor. 5.13, so we sketch the proof and leave the details as an
exercise. In this situation we have
N
G
(P)/C
G
(P) Aut(P) = GL
2
(p),
which leads to [N
G
(P)/C
G
(P)[ dividing p + 1. If N
G
(P) ,= C
G
(P) then p = 2 and the result follows
from Lemma 5.11.

The second possibility in Cor. 5.14 occurs for the alternating group A
5
.
59
5.4 Simple groups
The Sylow and Burnside Transfer Theorems can be used to narrow the possible orders of small simple
groups, and to prove uniqueness of simple groups of a given order.
Throughout this section P is a Sylow p-subgroup of a simple group G with p-factorization
[G[ = p
r
(1 + pk),
where = [N
G
(P) : P] and 1 + pk = n
p
= [G : N
G
(P)].
The rst lemma is a variant of the Burnside Transfer Theorem with a weaker hypothesis, and weaker,
but still useful result.
Lemma 5.15 If P is abelian then no non-identity element g P is centralized by N
G
(P).
Proof: The proof of Thm. 5.12 shows that the transfer T : G P is nontrivial on any non-identity
element of P which is centralized by N
G
(P).
Lemma 5.16 We have k 1. If = 1 then r 3.
Proof: If k = 0 then G has a normal Sylow p-subgroup. If r 2 then the Sylow p-subgroups P G
are abelian, hence 2 by Lemma 5.15.
Lemma 5.17 If k = 1 then r = 1 and [ (p 1).
Proof: Let X be the set of Sylow p-subgroups of G. The action of G on X embeds G S
p+1
. Since
p
2
(p + 1)! we have r = 1.
The condition k = 1 also implies that P is transitive on X P. If g C
G
(P) xes some Q
X P, then
gx
Q =
x
Q for all x P, so g is trivial on X, hence g = 1, since G is simple.
Hence C
G
(P) acts freely on X P, so C
G
(P) = P and N
G
(P)/P = N
G
(P)/C
G
(P) embeds in
Aut(P) (Z/pZ)

. Therefore [ (p 1).
We study the minimal case:
Proposition 5.18 Let G be a simple group of order p 2 (1 +p). Then this is the p-factorization of G
and we have
1. The normalizer of a Sylow p-subgroup of G is dihedral of order 2p.
2. p 5 mod 24.
60
Proof: Since p only divides [G[ once, the p-factorization of [G[ is p (1 + kp) for some 2 and
k 1. Hence 2 (1 + p) = (1 + kp), so 2 = (2 k)p, which implies = 2 and k = 1.
Now N(P)/P is the unique subgroup of order two in Aut(P) = C
p1
, hence acts by inversion on P.
This proves that N(P) D
p
. Hence N(P) = a s where a has order p and s has order two and
sas = a
1
. Moreover, N(P) contains p involutions a
i
sa
i
, for 1 i p.
The set X of Sylow p-subgroups of G has cardinality [X[ = n
p
= p + 1 and the conjugation action of
G on X gives a homomorphism
: G S
p+1
whose image lies n A
p+1
since G is simple.
Since (a) normalizes P and no other Sylow p-subgroup, the cycle type of (a) is 1p. This element
generates its own centralizer in S
p+1
. One element which inverts (12 . . . p) is a product of two-cycles
which reects this p-cycle about its center. This involution xes P and at least one P
1
,= P, so
X = P,
x
P
1
: x P. Since x ,= x
1
for all 1 ,= x P, it follows that the involutions in G have
cycle type 1
2
2
(p1)/2
. Since these permutations must be even, we have p 1 mod 4.
Now the 2-factorization of [G[ is
2
2

2
(1 + 2k
2
).
And a Sylow 2-subgroup Q is isomorphic to C
2
C
2
. The centralizer C(s) of an involution s N(P)
must permute its two xed points P, P
1
on X, but s generates its own centralizer in N(P), so C(s)
acts transitively on P, P
1
. Hence [C(s)[ = 4 and C(s) = Q is a Sylow 2-subgroup of G and
C(Q) = Q.
The quotient N(Q)/Q is a nontrivial subgroup of Aut(P) = GL
2
(2) of order three. Hence N(Q)
A
4
, acting transitively on Q 1, so G has exactly one conjugacy class of involutions and there are
p(p + 1)/2 of them. Also G has
n
2
= p
_
p + 1
6
_
Sylow 2-subgroups. In particular, we must have p 1 mod 3.
Having established that
p 1 mod 4 and p 1 mod 3,
we conclude from the Chinese Remainder theorem that p 5 mod 12.
Each of the
_
p+1
2
_
involutions in G can be conjugated into Q and all involutions in Q are conjugate
by N(Q). Since Q is the full centralizer of its involutions, it follows that s can normalize no other
2-Sylow but Q, hence s has cycle type 12
(n
2
1)/2
on the set of 2-Sylows, so (n
2
1)/2 is even. Writing
p = 12k + 5, it follows that (n
2
1)/2 = (3k + 2)(4k + 1) is even, so k is even and p 5 mod 24.

If p = 5 such a simple group exists, namely A

5
has 5-factorization
[A
5
[ = 60 = 5 2 (1 + 5).
61
Perhaps this is the only simple group of order 2p(p+1). For the next possibility p = 29 we would have
[G[ = 29 2 (1 + 29) = 5724,
whose 5-factorization shows that G is not simple.
5.4.1 The simple group of order 60
The smallest group order divisible by three primes p < q < r for which the Sylow p-subgroup is not
cyclic is 2
2
3 5 = 60. We have proved that such a simple group exists, namely the alternating group
A
5
. We now prove that this simple group is the unique one of its order.
Corollary 5.19 Any simple group G of order 60 is isomorphic to the alternating group A
5
.
Proof: The 2-factorization is 60 = 2
2
(1 + 2k), so = 3 or = 5. But a Sylow 2-subgroup
P G is isomorphic to C
2
C
2
, by Prop. 5.13, so Aut(P) = GL
2
(2) has order six. Hence = 3,
and n
2
= 5, giving an injective homomorphism : G S
5
, whose image is contained in A
5
, since G
is simple. Since G and A
5
have the same order, is an isomorphism G A
5
.
The group PSL
2
(5) has order 5 (5
2
1)/2 = 60 and is simple. See Thm. 3.21. Hence Cor. 5.19
implies that
PSL
2
(5) A
5
. (19)
An explicit isomorphism
PGL
2
(5) S
5
(20)
which restricts to an isomorphism (19) may be obtained as follows.
The group PGL
2
(5) acts faithfully by fractional transformations
ax+b
cx+d
on the projective space P
1
(5) =
F over the eld F of ve elements. Thus, PGL
2
(5) S
6
and elements of PGL
2
(5) may be
regarded as permutations of , 0, 1, 2, 3, 4. From the Bruhat decomposition, PGL
2
(5) is generated
by three elements a, b, w, where
a(x) = 2x, b(x) = x + 1, w(x) = 1/x
as fractional transformations. As permutations, we have
a = (1 2 4 3), b = (0 1 2 3 4), w = (0)(2 3).
Consider the 222 cycle
0
= (0)(1 4)(2 3) S
6
. This is the unique 222-cycle xed under conjuga-
tion by both a and w, which follows from the fact that 1 + 4 = 2 + 3 = 0. We take the conjugates of

0
under b, obtaining
1
,
2
,
3
,
4
, where
i
= b
i

0
b
i
. Explicitly, we have
5

0
= (0)(1 4)(2 3)

1
= (1)(2 0)(3 4)

2
= (2)(3 1)(4 0)

3
= (3)(4 2)(0 1)

4
= (4)(0 3)(1 2).
5
As an aside, note that these 222-cycles partition the 15 transpositions of S
6
into ve sets of three.
62
The set =
i
: i Z/5Z is closed under conjugation by all of PGL
2
(5). To see this, we note rst
that is closed under b by construction, and since a
0
a
1
=
0
, it follows that a
i
a
1
=
2i
. So far
we could have made similar 2
k
-cycles for any prime p = 2k + 1. But since p = 5, it happens that w
interchanges
1

3
and
2

4
. As
0
has been chosen to be xed by w, it follows that is indeed
closed under conjugation by PGL
2
(5), so we have a homomorphism : PGL
2
(5) S
5
sending
a (1 2 4 3)
b (0 1 2 3 4)
w (1 3)(2 4).
Since PSL
2
(5) is simple, it follows that is injective, hence is an isomorphism by orders, and it must
restrict to an isomorphism PSL
2
(5)

A
5
.
5.4.2 The simple group of order 168
Proposition 5.20 Let G be a group of order 168 = 7 3 8 such that no Sylow subgroup of G is normal
in G. Then G PSL
2
(7).
Proof:
6
The idea is to nd a copy of the Bruhat decomposition in G and use this to determine the
multiplication table. For p 2, 3, 7, let n
p
be the number of Sylow p-subgroups of G.
The 7-factorization of G has the form 7 (1 + 7k) = 7 24. Since n
7
> 1, we must have n
7
= 8.
Let P be a Sylow 7-subgroup. Its normalizer H = N
G
(P) has order [H[ = 21, and H = PQ, where
Q is a Sylow 3-subgroup of G contained in H.
(i) H is a maximal proper subgroup of G.
Suppose H K G. Then H = N
K
(P), so there are [K : H] Sylow 7-subgroups of K and
[K : H] = 1 + 7 divides [G : H] = 8. Hence [K : H] = 1 or 8, meaning K = H or K = G.
(ii) H is the unique nonabelian group of order 21
If H were abelian, we would have H N
G
(Q) ,= G, forcing H = N
G
(Q) by maximality. But then
n
3
= [G : H] = 8, which is impossible since 8 , 1 mod 3. Therefore H is nonabelian. We have seen
there is a unique nonabelian group of order 21; it is isomorphic to a Borel subgroup of PSL
2
(7).
(iii) G has 28 Sylow 3-subgroups
We have seen that Q is not normal in H. Since [H : Q] = 7 is prime, it follows that N
H
(Q) = Q, so
H has 7 Sylow 3-subgroups, any two of which generate H. If these were all of the Sylow 3-subgroups
of G then they would generate a normal subgroup of G, implying H G. But H has only one Sylow
7-subgroup, and G has a Sylow 7-subgroup not contained in H, contradicting the conjugacy of Sylow
7-subgroups. Hence n
3
> 7, with n
3
1 mod 3 and n
3
[ 56. The only possibility is n
3
= 28.
(iv) The normalizer K = N
G
(Q) is isomorphic to the symmetric group S
3
.
From the previous step, we have [K[ = 168/28 = 6. Any Sylow subgroup is characteristic in its
normalizer, so N
G
(K) = K. Hence K has 28 conjugates. If K were abelian, it would be cyclic. If
6
This proof comes from [Suzuki vol 1], with some simplications.
63
this were true, each conjugate of K would contain a distinct pair of generators of order six, giving 56
elements of order six, along with 56 = 2 n
3
elements of order 3 and 48 = 6 n
7
elements of order 7.
This makes 160 elements, forcing n
2
= 1, a contradiction.
Let t be an element in K of order two.
(v) The group H P acts freely on the double coset HtP, and G = H HtP.
Since [H[ = 21, we have t / H, so P
t
,= P. Since P is the unique Sylow 7-subgroup of H, we
have H P
t
= 1. It follows that H P acts freely on HtP. Hence [HtP[ = 21 7 = [G[ [H[.
Therefore H and HtP are distinct and exhaust G.
The preceding step shows that every element g G H can be decomposed as g = pqtp

for unique
elements p, p

P and q Q. We wish to calculate this decomposition for certain elements in GH.

Let P

= P 1. Then tP

t GH since H P
t
= 1.
Lemma 5.21 The intersection P tPtPt consists of a pair of distinct mutually inverse elements:
P tPtPt = u, u
1
P

.
Let s Q be the unique element such that u
s
= u
2
. Then P

= u, u
1
, u
2
, u
2
, u
4
, u
4
and the
elements of tP

t decompose in PQtP as
tut = u
1
tu
1
tu
2
t = u
4
s
2
tu
4
tu
4
t = u
2
stu
2
tu
1
t = utu tu
2
t = u
4
stu
4
tu
4
t = u
2
stu
2
.
(21)
Proof: Since Q = 1, s, s
2
, we have a partition
PQtP = PtP PstP Ps
2
tP.
Since tP

t PQtP, we have a partition

tP

t = A
0
A
1
A
2
,
where A
i
= tP

t Ps
i
tP. Since s normalizes P, we have
[tP

t]
s
= tsP

s
1
t = tP

t,
and
[Ps
i
tP]
s
= Ps
1
s
i
tsP = Ps
i2
tP = Ps
i+1
tP.
It follows that A
s
i
= A
i+1
, with subscripts read modulo 3. As [tP

t[ = [P

[ = 6, it follows that
[A
i
[ = 2 for each i. And as each A
i
is closed under inversion, we must have
A
0
= tu
1
t, A
1
= s
1
tu
1
ts, A
2
= stu
1
ts
1
,
from which the lemma follows.
We now show how the product of two elements g, g

G is determined.
64
Case 1: g, g

H. Here the product gg

is determined by the known structure of H.

Case 2: g H and g

HtP. Here g

= htp and gg

= (gh)tp where gh is again computed in H.

Case 3: g HtP and g

H. Here g = htp and gg

= ht(pg

). Since pg

H = QP, we
can write pg

= qp

for some q Q and p

P. Then t, q K = S
3
, so tq = q
1
t so we have
gg

= ht(pg

) = ht(qp

) = hq
1
p

, with hq
1
H. Hence gg

can again be computed from the

structure of H.
Case 4: This nal case is where g and g

are both in HtP. Here g = htp and g

= h

tp

. Write
h

= p
1
q
1
, with p
1
P, q
1
Q. Then
gg

= htp h

tp

= ht(pp
1
)q
1
tp

= h t(pp
1
)t q
1
1
p

. (22)
If pp
1
= 1 this is a product in H. If pp
1
,= 1 then pp
1
P

= u, u
1
, u
2
, u
2
, u
4
, u
4
, so the
decomposition of t(pp
1
)t in PQtP is given by Lemma 21. Thus the product gg

is determined.
We have shown that there is only one isomorphism class of groups of order 168 having no normal
Sylow subgroups. As PSL
2
(7) is one such group, the theorem is proved.
The group G = GL
3
(2) also has order 2
3
(2
3
1)(2
2
1)(21) = 168. The upper and lower triangular
matrices in G are distinct Sylow 2-subgroups. The elements
_
_
0 1 0
1 1 0
0 0 1
_
_
,
_
_
1 0 0
0 0 1
0 1 1
_
_
(23)
generate distinct Sylow 3-subgroups. The elements
u =
_
_
0 0 1
1 0 1
0 1 0
_
_
, v =
_
_
0 0 1
1 0 0
0 1 1
_
_
(24)
generate distinct Sylow 7-subgroups.
7
From Prop. 5.20 it follows that we have another exceptional
isomorphism
PSL
2
(7) GL
3
(2). (25)
This makes clear, for example, that the Sylow 2-subgroups of PSL
2
(7) are D
4
, and that PSL
2
(7) acts
faithfully on two sets of seven elements: the seven lines and the seven planes in (Z/2Z)
3
.
These two actions imply that PSL
2
(7) has an outer automorphism given by (g) =
t
g
1
, the inverse
transpose in GL
3
(2). Indeed, since the normalizer of a Sylow 7-subgroup is G
21
(see step (ii) above),
it follows that G has exactly two conjugacy-classes of elements of order 7. The elements u, v in (24)
7
The matrices u, v were found by factoring the cyclotomic polynomial

7
(x) = 1 +x +x
2
+x
3
+x
4
+x
5
+x
6
= (1 +x +x
3
)(1 +x
2
+x
3
)
over Z/2Z and using rational canonical form.
65
have tr(u) = 0 and tr(v) = 1, so they are not conjugate. Thus, the trace function distinguishes the two
classes of elements of order 7. Since
(u) =
_
_
1 0 1
1 0 0
0 1 0
_
_
has tr((u)) = 1, it follows that (u) is not conjugate to u, so is not inner. Hence there are in fact
two non-conjugate isomorphisms (25).
We can see them explicitly as follows. To give a homomorphism PSL
2
(7) GL
3
(2) is to give an ele-
mentary abelian 2-group E of rank 3 and an action of PSL
2
(7) on E by automorphisms. Now, PSL
2
(7)
acts by fractional transformations of the projective line P
1
(F) = F , by which we view PSL
2
(7)
as a subgroup of the symmetric group S
8
, permuting the points in P
1
(F) = , 0, 1, 2, 3, 4, 5, 6. We
will nd E as a subgroup of S
8
normalized by PSL
2
(7).
Since F
2
= 2, the Bruhat decomposition implies that PSL
2
(7) is generated by three transformations
a(x) = 2x, b(x) = x + 1, w(x) = 1/x.
As permutations of , 0, 1, 2, 3, 4, 5, 6, these are elements of S
8
given by
a = (1 2 4)(3 6 5), b = (0 1 2 3 4 5 6), w = (0)(1 6)(2 3)(4 5).
The subgroup a, b is the Borel subgroup of PSL
2
(7) xing and is the G
21
from step (ii). To nd E
in S
8
, we note that the matrices (23) and (24) permute the seven nonzero vectors in (Z/2Z)
3
in cycle
types [331] and [7], respectively. So E 1 must consist of seven commuting elements e
1
, . . . , e
7
of order two permuted in the same way. Exactly one of them, say e
1
, is xed by a and the remaining
e
i+1
= b
i
e
1
b
i
for 1 i 6. From our experience with PGL
2
(5) we guess e
1
has the form
e
1
= (0)(1 x)(2 2x)(4 4x),
for some x 3, 5, 6. But the element
e
2
= be
1
b
1
= (1)(2 x + 1)(3 2x + 1)(5 4x + 1)
must commute with e
1
. If x = 6 these two elements would be
(0)(1 6)(2 5)(4 3) and (1)(2 0)(3 6)(5 4),
which do not commute. Therefore x 3, 5. In fact both of these choices work, and we get two
subgroups E = e
i
, E

= e

i
, where e
0
= e

0
is the identity of S
8
and
8
e
1
= (0)(1 3)(2 6)(4 5) e

1
= (0)(1 5)(2 3)(4 6)
e
2
= (1)(2 4)(3 0)(5 6) e

2
= (1)(2 6)(3 4)(5 0)
e
3
= (2)(3 5)(4 1)(6 0) e

3
= (2)(3 0)(4 5)(6 1)
e
4
= (3)(4 6)(5 2)(0 1) e

4
= (3)(4 1)(5 6)(0 2)
e
5
= (4)(5 0)(6 3)(1 2) e

5
= (4)(5 2)(6 0)(1 3)
e
6
= (5)(6 1)(0 4)(2 3) e

6
= (5)(6 3)(0 1)(2 4)
e
7
= (6)(0 2)(1 5)(3 4) e

7
= (6)(0 4)(1 2)(3 5)
8
As an aside, note that these [2222] cycles partition the 28 transpositions of S
8
into seven sets of four.
66
Each set E, E

contains 7 commuting involutions normalized by a, b. To see they are closed under

multiplication we need only check that e
1
e
2
= e
4
, e

1
e

2
= e

6
and conjugate these equations by a. Hence
E and E

are indeed elementary abelian 2-groups of rank 3 normalized by a, b. The remarkable thing
is that both sets e
i
and e

i
are also normalized by the permutation w = ( 0)(1 6)(2 3)(4 5), so
they are normalized by the entire subgroup a, b, w S
8
. Thus, we recover the two isomorphisms
PSL
2
(7) GL
3
(2) explicitly inside S
8
.
5.4.3 Simple groups of order 720
We begin with a few more lemmas to help narrow the cases.
Lemma 5.22 If H is a group of order p
r
q
s
, where p and q are primes and r, s 2 then H is not simple
Proof: We may assume p > q. If H is simple then it has p-factorization p
r
q
s
= p
r
(1 + kp)
wth k 1. Since r 2, the Sylow p-subgroups are abelian, so > 1. Hence = q, s = 2 and
q = 1 + kp > p, a contradiction
Remark: The restrictions on r and s are unnecessary. Using character theory, Burnside proved that
any nite group whose order is divisible by just two primes is not simple.
Lemma 5.23 If [G[ = m p
r
where p is a prime not dividing m, then m > 7.
Proof: Since G is simple, the action on Sylow p-subgroups embeds G A
m
, so m 5, 6, 7 and
m p
r
divides m 3 2. If m 6 then mp
r
< 60 so G is not simple. If m = 7 then p = 5, 3, 2 and
r 1, 2, 3 respectively. and the 7-factorization is (1 + 7k) with > 1 and k > 0. There are no
values of p and r satisfying these conditions.
Lemma 5.24 If [G[ N then p <
_
N/2.
Proof: From Lemma. 5.16, we have p 2 p
r
. Hence
p 2 (1 + p) [G[ N,
so p
2
< p(1 + p) N/2, and the estimate follows.
Now let G be a simple group of order [G[ 720. Let p be the largest prime dividing [G[, with
p-factorization
[G[ = p
r
(1 + kp).
We will assume p 5 because groups of order 3
a
2
b
are not simple, by Burnsides p
a
q
b
theorem (which
uses character theory and is not yet in these notes). By Cor. 5.24, we have p <

360 < 19 so
p 17. By Prop. 5.18, we exclude the case (r, , k) = (1, 2, 1). This further excludes p = 17. Since
p
r
(1 + p) [G[ 720 we have r = 1 for p = 13, 11, r 2 for p = 7, 5, r 4 for p = 3.
We note also:
67
1. If r = 1 then (, p 1) > 1.
2. If = 2 then k is odd and k 3 by Cors. 5.13 and 5.18.
3. If k = 1 then r = 1 since p
2
does not divide (p + 1)!
4. If r 2 then 2 so 1 + pk 720/2p
r
.
5. No prime divisor of 1 + kp can be larger than p.
6. If k is even then either 4 [ or is odd.
Since p
r
(1 + p) [G[ 720 we have r = 1 for p = 13, 11, r 2 for p = 7, 5, r 4 for p = 3.
For p = 13 the only surviving case is
[G[ = 13 4 (1 + 13) = 728,
which we showed cannot be the order of a simple group, see Example 3 of section 5.3.
For p = 11 it the only surviving case is
[G[ = 11 5 (1 + 11) = 660,
which is the order of the simple group PSL
2
(11).
For p = 7, we have r 2. If r = 2 then 2 so 1 + 7k (720/2 49) < 8, forcing k = 0. Hence
r = 1 and 2. And if k = 1 then [ 6, and if k is even then 4 [ .
The surviving 7-factorizations are
7 3 (1 + 7) = 2
3
3 7 = 168
7 6 (1 + 7) = 2
4
3 7 = 336
7 4 (1 + 2 7) = 2
2
3 5 7 = 420
7 2 (1 + 5 7) = 2
3
3
2
7 = 504
7 2 (1 + 7 7) = 2
2
5
2
7 = 700
The case [G[ = 168 is the order of a simple group PSL
2
(7), and this is the unique simple group of
order 168 (see Prop. 5.20).
The case [G[ = 336 has N
G
(P) P C
6
, the ax + b group over Z/7Z. An element n C
6
of order
six xes a point Q X P and is free on X P, Q, hence has cycle type [61] so is an odd
permutation. Hence G has a subgroup of index two. This occurs in nature: The group PGL
2
(7) has
order 336 and contains PSL
2
(7) with index two.
The case [G[ = 420 has an involution s C
G
(P). Suppose there exists Q X P and x P
such that
s
Q =
x
Q. Then sx N
G
(Q) so x
2
Q so P = Q. Hence s interchanges the two P-orbits
in X P and has cycle type [2
7
1] on X, which is an odd permutation, so G has a normal subgroup
of index two.
68
The case [G[ = 504 is the order of a simple group PSL
2
(8).
The case [G[ = 700 is ruled out by Cor. 5.14.
For p = 5 we have k = 1, 3, 7 since 5 is the largest prime dividing 1 + 5k. If k = 1 then r = 1 and
= 2, 4 by Lemma 5.17. If k = 3 then 5
r
720/16 = 45 so r = 1 and 9 is even. Since groups
of order 5 2
n
are not simple, by Lemma 5.23 we have = 6. If k = 7 then 5
r
720/36 = 20 so
r = 1 and = 2, 4.
Thus the surviving 5-factorizations are
5 2 (1 + 5) = 2 3 5 = 60
5 4 (1 + 5) = 2
3
3 5 = 120
5 6 (1 + 3 5) = 2
5
3 5 = 480
5 2 (1 + 7 5) = 2
3
3
2
5 = 360
5 4 (1 + 7 5) = 2
3
3
2
5 = 720
The case 5 2 (1 + 5) = 60 arises from the simple group A
5
= PSL
2
(5).
In the case 5 4 (1+5) = 120 the normalizer of a Sylow 5-subgroup contains a 4-cycle in S
6
, which is
odd. Hence G contains a normal subgroup of index two. This actually occurs: The group S
5
has order
120 and contains A
5
of index two.
In the case 5 6 (1 + 3 5) = 480, we have N
G
(P) of order 30, so N
G
(P) = g, s where g, s have
order 15 and 2 respectively, and g
3
= P. There are four groups of order 30 according to the possible
actions of s on g = C
3
C
5
by inverting one or both or no factors. Note that all four groups occur
inside the normalizer of a 15-cycle in S
15
. By the Transfer Theorem, s must invert the C
5
-factor (which
is P). Now look at the embedding of N
G
(P) in S
15
via its action on XP. Since g
3
acts with cycle
type [5 5 5], it follows that g has cycle type [15], so g acts freely and transitively on X P. Hence
the action of s on XP is equivalent to its action on C
3
C
5
. If s inverts both factors then its cycle
type is [1 2
7
] which is odd. Therefore s centralizes the C
3
factor and N
G
(P) C
3
D
5
.
9
Hence the
centralizer C
G
(Q) of a Sylow 3-subgroup has order at least 30. In the 3-factorization 3
3
n
2
we must
have
3
= 5 2
a
with a 2, lest C
G
(Q) = N
G
(Q). This forces n
3
= 4, so G is not simple
The case 5 2 (1 + 7 5) = 360 arises from the simple group A
6
.
The case [G[ = 720 is harder, but still elementary; in the next section we show there are no simple
groups of order 720.
9
Cole (Simple groups from Order 201 to Order 500, Am. J. Math, vol 14, no. 4 1892, 378-88), seems to have missed
this possibility.
69
Assuming that result, we have shown that the only possible orders 720 of nonabelian simple groups
G are 60, 168, 360, 504 and 660. Moreover, there exists a simple group of each of these orders, namely
simple group G [G[
A
5
PSL
2
(5) 60
PSL
2
(7) GL
3
(2) 168
A
6
PSL
2
(9) 360
PSL
2
(8) 504
PSL
2
(11) 660
In fact, there is exactly one simple group having each of these orders, but we have only proved this for
[G[ = 60 and [G[ = 168.
5.4.4 Almost-simple groups of order 720
Suppose that G is a simple group of order 720 = 8 9 10. Writing the order this way makes it
plausible that G should act 3-transitively on a set with 10 elements, as PGL
2
(9) has this order and
acts 3-transitively on its set of 10 Sylow 3-subgroups. Since PGL
2
(9) is not simple, it may make
sense to look in that direction for a contradiction. So we begin by studying the 3-local structure of our
hypothetical simple group Gof order 720. However, since there does exist a simple group of order 360,
the 2-local structure must eventually be decisive. It is here that Burnsides sketch (which begins with
the 5-local structure)
10
seems to have a serious gap. Our treatment of the 3-local part is based on that
of Derek Holt
11
which I have recast to better highlight his essential point, with different arguments in
places. Then we will have enough information for a correct 2-local argument in the spirit of Burnside.
From the Sylow and Burnside Transfer theorems, the possible 3-factorizations of [G[ are
[G[ = 3
2
2 40, or [G[ = 3
2
8 10. (26)
In the former case, where n
3
(G) = 40, the normalizer of a Sylow 3-subgroup P acts by an involution
on P with trivial xed points, and normalizes every subgroup of P.
Holts crucial observation about the 3-local structure is as follows.
Lemma 5.25 Every subgroup of order three in G is contained in just one Sylow 3-subgroup of G.
Proof: Let Q < G be a subgroup of order three, with normalizer N = N
G
(Q), and let P be a Sylow
3-subgroup of G containing Q. The 3-factorization of N is
[N[ = 3
2
n
3
(N),
where = [N N
G
(P) : P] and n
3
(N) is the number of Sylow 3-subgroups in N. Since the Sylow
3-subgroups are abelian, any such subgroup containing Q must lie in N. So the lemma is equivalent to
the assertion that n
3
(N) = 1.
10
Notes on the theory of groups of nite order, Bull. London Math. Soc. 1894
11
http://sci.tech-archive.net/Archive/sci.math/2006-12/msg07456.html
70
If n
3
(G) = 40 then N
G
(P) contains an element inverting P, hence normalizing Q, so > 1. If
n
3
(G) = 10 then again > 1, lest we have at least [N
G
(P) : N N
G
(P)] = [N
G
(P) : P] = 8
conjugates of Q in P, whereas the group P of order 3
2
can have at most four subgroups of order
three. Thus, in either case we have > 1. We also have n
3
(N) < 10, lest G S
4
. It follows that
n
3
(N) 1, 4.
Assume that n
3
(N) = 4. Then = 2, lest [G : N] 5, so N has 3-factorization
[N[ = 3
2
2 4.
Let P = P
1
, P
2
, P
3
, P
4
be the Sylow 3-subgroups in N, and let X =
g
Q : g G be the set of
G-conjugates of Q. We have [X[ = [G : N] = 10. The group G acts on X by conjugation and we
consider the xed points of Q:
X
Q
= Q

X : Q < N
G
(Q

).
Since a group of order three admits no automorphism of order three, we have
Q < N
G
(Q

) Q < C
G
(Q

) Q

< C
G
(Q) Q

< N,
in which case Q

< P
i
for some i. Therefore
X
Q
=
4
_
i=1
X(P
i
), (27)
where, for any Sylow p-subgroup P

of G we dene X(P

) := Q

X : Q

< P

to be the
set of conjugates of Q which are contained in P

. Note that
g
X(P

) = X(
g
P

) for any g G. In
particular, the sets X(P
i
) all have the same cardinality. Since P is abelian, N
G
(P) acts transitively on
X(P) (by the same argument used for Lemma 5.11) and the stabilizer of Q in N
G
(P) has cardinality
[N N
G
(P)[ = 3
2
2, so m = [N
G
(P) : N
G
(P) N] 1, 4, according to the two possibilities for
N
G
(P) in (26). Finally, since P
i
is generated by any two elements of X(P
i
), we have X(P
i
)X(P
j
) =
Q for i ,= j. It now follows from (27) that [X
Q
[ = 1 + 4(m1). Since [X
Q
[ [X[ = 10 we must
have m = 1, so X
Q
= Q and the Q-orbits in X have sizes 1, 3, 3, 3.
As Aut(Q) = C
2
, the centralizer C
G
(Q) has order 36 or 72, with Sylow 2-subgroup R of order 4 or
8. As R xes Q and G is simple, R acts faithfully on X Q, permuting the three Q-orbits therein,
giving a homomorphism R S
3
. If r R preserves a Q-orbit Q
1
, Q
2
, Q
3
in X then r normalizes
each Q
i
, since r commutes with Q. Hence the image of R in S
3
is nontrivial, so some r R maps to
a 2-cycle in S
3
. This means r has cycle type [1
3
2
3
] on X Q and cycle type [1
4
2
3
] on X. Thus r is
an odd permutation on X, contradicting the simplicity of G. It follows that n
3
(N) = 1 and the lemma
is proved.
Let Y be the set of Sylow 3-subgroups of G and let P Y . The lemma implies that P acts simply
transitively on Y P, so [Y [ 1 mod 9. This rules out n
3
(G) = 40, so we must have n
3
(G) = 10
and [N
G
(P)[ = 3
2
8. Regarding G as a subgroup of A
10
via its action on Y by conjugation shows that
P cannot be cyclic because the normalizer of a 9-cycle in S
10
has order 3
2
6, hence cannot contain
N
G
(P). Thus, we nd that P C
3
C
3
.
71
Choose P

in Y distinct from P and let

H = N
G
(P) N
G
(P

)
be the normalizer of P

in N
G
(P). Since N
G
(P) acts transitively on Y P, it follows that [H[ = 8,
so H is a Sylow 2-subgroup of N
G
(P) and N
G
(P) = P H. I claim that H acts freely on P 1
by conjugation. For by the lemma again, any nonidentity element t P has cycle type [1333] in Y ,
whose centralizer in A
10
has order 3
3
6/2 = 81 and intersects H trivially.
12
Let s H be an involution (an element of order two). Since s acts freely on P 1, it must act
by inversion, hence s is the unique involution in H. The only groups of order 8 containing a unique
involution are Q
8
and C
8
. In the latter case H would be generated by an [8]-cycle on Y which is odd,
contradicting the simplicity of G. Hence H Q
8
.
13
Since Q
8
does not embed in S
n
for n < 8, the
faithful action of H on Y P, P

must also be free and transitive. Hence the elements of order four
in H have cycle type [1
2
4
4
] on Y , and s has cycle type [1
2
2
4
].
If x P 1 and H H
x
,= 1, then s H H
x
, say s = h
x
for some h in H. But then h is an
involution, so h = s, meaning that x and s commute. This contradicts H acting freely on P 1,
so H H
x
= 1. It follows that there are exactly nine N
G
(P)-conjugates of s in H, and these are all
of the involutions in N
G
(P). From the cycle type of s on Y , we see that N
G
(P) and N
G
(P

) are the
only conjugates of N
G
(P) containing s. As there are ten conjugates of N
G
(P), there are 9 10/2 = 45
conjugates of s in G. Hence the centralizer S = C
G
(s) is a Sylow 2-subgroup of G. By the same
argument there are 45 conjugates of H in G, so S = N
G
(H) is also the normalizer of H in G.
I claim that distinct involutions have distinct centralizers. For suppose t ,= s is an involution with
C
G
(t) = C
G
(s). Then t N
G
(H) H so preserves the xed point set Y
H
= P, P

, but t cannot
normalize P or P

, so t must switch P and P

. But t is a product of 2-cycles and is even, while 10/2

is odd, so t must have at least two xed-points on Y . This means t N
G
(P

) N
G
(P

) for some
pair P

, P

Y P, P

. But t cannot act trivially on Y P, P

lest it be a 2-cycle on Y . As H,
which centralizes t, must preserve the xed points of t in Y P, P

, this contradicts the transitivity

of H on Y P, P

.
14
Now take h H of order 4 and consider the action of H on the set Z of Sylow 2-subgroups of G. We
have h
2
= s. Since s is contained in just one Sylow 2-subgroup, namely C
G
(s), the cycle type of s on
Z is 12
22
. Hence the cycle type of h is [14
11
], which is an odd permutation, contradicting the simplicity
of G.

The argument is difcult for several reasons. First, 720 = 2 360, and there does exist a simple group of
order 360, namely the alternating group A
6
. Secondly, A
6
PSL
2
(9) (see section 8.5.2), so SL
2
(9) is
a group of order 720 surjecting onto A
6
. Thirdly, there are three groups of order 720 containing A
6
with
index two (see below). All these groups of order 720 irt with the simple group A
6
, but themselves
just fail to be simple, for different reasons, which is why it is hard to rule them out.
12
Since [H[ = 8 = [P 1[, it follows that H is also transitive on P 1, so G has a unique conjugacy class of
elements of order three, and this class has 80 elements. However we do not need this.
13
We do not actually need to know that H = Q
8
and not C
8
.
14
A similar argument shows that all involutions are conjugate in G, but we do not need this.
72
The three groups containing A
6
with index two can be seen as follows. We have seen that automorphism
group of S
6
is Aut(S
6
) = S
6
C
2
. It follows that Out(A
6
) C
2
C
2
. By the Correspondence
Theorem, there are three subgroups of Aut(A
6
) of order 720, containing A
6
. One of these is S
6
. Via
the isomorphismA
6
PSL
2
(9), another one is PGL
2
(9). The third group is the Mathieu Group M
10
,
which is part of a family of highly transitive simple (or almost simple, in the case of n = 10) subgroups
M
n
S
n
.
To see M
10
more explicitly, we start with PGL
2
(9), which we think of as the group of permutations of
P
1
(F
9
) given by
PGL
2
(9) =
_
z
az + b
cz + d
: ad bc ,= 0.
_
.
The eld F
9
of 9 elements is built from F
3
= Z/3Z just as C is built from R, namely,
F
9
= x + iy : x, y F
3
, i
2
= 1,
and just like C, F
9
has an automorphism x + iy = x iy. We dene the group
PL
2
(9) =
_
z
az + b
cz + d
: ad bc ,= 0
_

_
z
a z + b
c z + d
: ad bc ,= 0
_
.
This group contains PGL
2
(9) with index two; a nontrivial coset representative is simply z z. It
turns out that
PL
2
(9) Aut(A
6
). In this viewpoint, M
10
is the subgroup of PL
2
(9) given by
M
10
=
_
z
az + b
cz + d
: ad bc is a square in F

9
_

_
z
a z + b
c z + d
: ad bc is a nonsquare in F

9
_
.
Thus, M
10
contains PSL
2
(9) with index two; a nontrivial coset representative is z i z.
The subgroup S
6
is generated by PSL
2
(9) and z z. In this guise an outer automorphism of S
6
is
conjugation by z iz.
6 Solvable and nilpotent groups
A group G is solvable if it has a chain of subgroups
1 = G
0
< G
1
G
2
G
3
G
n
= G (28)
where each G
i
is normal in G
i+1
with abelian quotient G
i+1
/G
i
.
If H G then G is solvable if and only if both H and G/H are solvable (exercise). It follows that G
is not solvable precisely when there exist subgroups H K G with K/H nonabelian simple. In this
sense solvable groups are diametrically opposed to nonabelian simple groups.
Examples of Solvable Groups:
73
i) Any nite group G with [G[ < 60 is solvable, because there are no simple groups of order < 60.
In particular S
n
is solvable for n 4. However, S
n
is not solvable when n 5, because it contains the
simple group A
5
.
ii) The dihedral group D
n
is solvable for every n 2 because it has a chain of subgroups 1 < C
n
<
G and G/C
n
= C
2
.
iii) For any eld F, the subgroup B of upper triangular matrices in GL
n
(F) is solvable. For example
if n = 3 one can take the series
1 <
_
_
1 0
0 1 0
0 0 1
_
_

_
_
1
0 1
0 0 1
_
_

_
_

0
0 0
_
_
= G,
with the obvious generalization to arbitrary n.
iv) Any group G of order pq, where p and q are primes, is solvable. For if p q then G has a
normal subgroup C
q
with quotient C
p
. More generally, any group whose order is divisible by at most
two primes is solvable. This is Burnsides p
a
q
b
theorem, whose original proof uses character theory
and still appears to be the most accessible proof to non-experts.
v) A group G is solvable if and only if it has the factorization property: If [G[ = m n for relatively
prime integers m, n, then G = MN for subgroups M, N of G of orders m, n. This is P. Halls
generalization of Burnsides theorem.
vi) Any nite group of odd order is solvable. This is the famous Feit-Thompson theorem from 1963,
whose proof is much more difcult than the previous two theorems.
vii) Let f(x) be a polynomial with rational coefcients having n distinct roots. The Galois group
G
f
is the subgroup of S
n
consisting of those permutations of the roots which preserve all polynomial
relations among them. The group G
f
is solvable precisely when the roots of f can be expressed in
terms of rational and radical expressions in the coefcients of f, as in the quadratic, cubic or quartic
formulas. This was discovered by E. Galois, and is the origin of the term solvable. It explains why
there is no general formula for a quintic polynomial: S
5
is not solvable.
viii) Same situation, where now the coefcients of f lie in the eld Q
p
of p-adic numbers for some
prime p. Then G
f
is always solvable. In fact, G has a canonical chain of subgroups G
1
G
0
G, where
G
1
is a p-group and G
0
/G
1
, G/G
0
are both cyclic.
Solvable groups occur naturally in many areas of mathematics. Being opposite to nonabelian simple
groups, they are a very natural class of groups to study. However, some of the most interesting theorems
about solvable groups are very difcult to prove. For example, the proof of the Feit-Thompson theorem
takes 255 pages.
Among the solvable groups are the nilpotent groups, for which the rst few interesting theorems are
easy, including a version of Halls theorem above.
The denition is as follows. For any group G, the ascending central series
15
of G is the chain of
15
Also called the upper central series.
74
subgroups of G:
Z
1
Z
2
Z
3
(29)
dened inductively as follows: Z
1
= Z(G) is the center of G, and given Z
i
, dene Z
i+1
to be the
unique subgroup containing Z
i
such that Z
i+1
/Z
i
= Z(G/Z
i
) is the center of G/Z
i
. We say G is
nilpotent if Z
c
= G for some c 1. The minimal such c is the nilpotence class of G.
A nontrivial nilpotent group must have nontrivial center, lest all Z
i
= 1. If G is nilpotent of class c
then G/Z(G) has class c 1.
If G is nilpotent and H G then H and G/H are nilpotent, but not conversely: The symmetric group
S
3
has trivial center, so its ascending central series has Z
i
= 1 for all i. Hence S
3
is not nilpotent.
However, its subgroup A
3
and quotient S
3
/A
3
= C
2
are both nilpotent.
Direct products of nilpotent groups are nilpotent. So any direct product of p-groups is nilpotent. We
will see that all nilpotent groups are of this form.
Examples of Nilpotent Groups:
i) The abelian groups are the nilpotent groups of class 1. The nilpotent groups G of class 2 are those
for which G/Z(G) is abelian.
ii) Any nite p-group is nilpotent, because any p-group has nontrivial center, hence Z
i
,= Z
i+1
, so
eventually Z
i
has the same order as G. If [G[ = p
k
then the nilpotence class of G is at most k 1, since
every group of order p
2
is abelian.
iii) The dihedral group D
n
is nilpotent if and only if n is a power of 2 (use 5. in Thm. 6.1 below).
If n = 2

and r = st is the product of two generating reections s and t, then r has order 2

and Z
i
is
cyclic, generated by r
2
i
. As D
2
/r
2
C
2
C
2
is abelian, it follows that D
2
has nilpotence class
equal to , the maximum possible for a group of order 2
+1
. There are exactly two other families of
groups of order 2
+1
having maximal nilpotence class: the generalized quaternion group Q
2
+1 and the
quasidihedral group QD
2
+1.
iv) For any eld F, the subgroup U
n
of strictly upper triangular matrices in GL
n
(F) is nilpotent of
class n 1. For example if n = 4 the ascending central series is
_

_
1 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_

_
<
_

_
1 0
0 1 0
0 0 1 0
0 0 0 1
_

_
<
_

_
1
0 1
0 0 1
0 0 0 1
_

_
= G,
with the obvious generalization to arbitrary n. The matrices in U
n
are of the formI +A, where A
n
= 0.
This may be the origin of the term nilpotent.
The main theorem on nite nilpotent groups is the following collection of characterizations.
Theorem 6.1 For a nite group G the following are equivalent.
1. G is nilpotent.
75
2. Any proper subgroup H of G is properly contained in its normalizer N
G
(H).
3. Every maximal subgroup of G is normal in G.
4. Every Sylow p-subgroup of G is normal in G.
5. G is the direct product of its Sylow p-subgroups.
6. For every divisor m of [G[ there is a normal subgroup M G with [M[ = m.
Proof: (1 2): By induction we assume the result for all groups of smaller order than G. Let H be
a proper subgroup of G. Since G is nilpotent its center Z is nontrivial and normalizes H. If Z is not
contained in H then H ,= HZ, done. Assume Z < H. Let G = G/Z, H = H/Z and let : G G
be the projection. Then H is a proper subgroup of G, so by induction H ,= N
G
(H). One checks that
H =
1
(H) and N
G
(H) =
1
(N
G
(H)),
hence H ,= N
G
(H), by the Correspondence Theorem.
(2 3): Apply 2 to a maximal proper subgroup of G.
(3 4): Suppose the Sylow p-subgroup P of G is not normal. Then N
G
(P) is a proper subgroup of
G, contained in a maximal proper subgroup M of G which is normal in G, by 3, and P is a Sylow
p-subgroup of M. For any g G, the conjugate P
g
is another Sylow p-subgroup of M, so P
g
= P
m
for some m M. Then g N
G
(P)m M.
16
Since g was arbitrary, we have shown that G = M,
contradicting M being proper.
(4 5): Induct on the number of primes dividing [G[. Let [G[ = p
r
1
1
p
r
n
n
. By 4, the Sylow p
i
-
subgroup P
i
is unique and any product of Sylow subgroups is a normal subgroup of G. In particular
H := P
1
P
n
1
is normal in G. By induction H = P
1
P
n1
, so the primes dividing the orders
of elements of H are in p
1
, . . . , p
n1
. So H P
n
= 1 and P
n
is normal in G. Hence
G = H P
n
= (P
1
P
n1
) P
n
= P
1
P
n
.
(5 6): Let [G[ = p
r
1
1
p
r
n
n
as above. Then m = p
s
1
1
p
sn
n
, for s
i
r
i
. The p-group P
i
has a
subgroup Q
i
of order p
s
i
. By 5, the subgroup M = Q
1
Q
n
is a normal subgroup of G with
order m.
(6 4): Take m to be the full power of p dividing G.
(5 1): Each p-group is nilpotent and a direct product of nilpotent groups is nilpotent.
Part 5 of Thm. 6.1 implies that in a nilpotent group two elements of relatively prime orders commute.
This explains why D
n
is not nilpotent for n odd: there is a rotation of odd order inverted by a reection.
An example of a nilpotent group which is not at rst glance a direct product of p-groups is the cyclic
group C
n
, when n has multiple prime divisors. If n = p
r
1
1
p
r
k
k
for distinct primes p
1
, . . . , p
k
then
C
n
= C
p
r
1
1
C
p
r
k .
16
This kind of transitivity proof is called the Frattini Argument. It is very similar to the proof of Lemma 5.11.
76
For a nonabelian example, let R be any commutative ring and Dene U
3
(R) to be the group
U
3
(R) =
_
_
1
0 1
0 0 1
_
_
,
where the entries are arbitrary elements in R. Then U
3
(R) is nilpotent. If we take R = Z/nZ then
applying the Chinese Remainder theorem to each matrix entry gives an isomorphism
U
3
(Z/nZ) U
3
(Z/p
r
1
1
Z) U
3
(Z/p
r
k
k
Z).
Nevertheless, Thm. 6.1 shows that the study nite nilpotent groups essentially reduces to p-groups.
7 p-groups, a second look
In this section p is a prime.
Up to isomorphism, there is one group of order p, namely C
p
. There are two groups of order p
2
, namely
C
p
C
p
, in which x
p
= 1 for every element x, and G C
p
2, which contains an element of order p
2
.
7.1 Groups of order p
3
Up to isomorphism there are three abelian groups of order p
3
:
C
p
C
p
C
p
, C
p
2 C
p
, C
p
3.
Assume from now on that G is nonabelian with [G[ = p
3
and let Z = Z(G) be the center of G. As
G/Z cannot be cyclic, we have
Z C
p
, and G/Z C
p
C
p
.
The latter is abelian, so the commutator subgroup [G, G] lies in Z. But [G, G] ,= 1 since G is non-
abelian, so in fact
[G, G] = Z.
Hence for any xed x G the commutator with x gives a map f : G Z, sending y [y, x] =
yxy
1
x
1
. As yxy
1
= [y, x]x, it follows that the conjugacy class of x G Z is xZ. Hence G has
p conjugacy classes inside Z and p
2
1 conjugacy classes outside Z.
Any maximal subgroup A G has order p
2
, is normal in G and contains Z = [G, G], since G/A C
p
is abelian. Hence the maximal subgroups of G correspond to the subgroups of G/Z C
p
C
p
,
which has p +1 subgroups, corresponding to the points in the projective line over Z/pZ. The maximal
77
subgroups A
0
, A
1
, . . . A
p
form the part of the subgroup lattice of G over Z.
G
A
0
A
1
. . . A
p
Z
The maximal subgroups A
i
may belong to different isomorphism classes, either C
p
C
p
or C
p
2. If
p = 2 then D
4
contains both kinds, while Q
8
contains C
4
but not C
2
C
2
. We will see that the opposite
holds for odd primes p.
First we need a commutator lemma:
Lemma 7.1 Let G be a group, let x, y G, and let [y, x] = yxy
1
x
1
be the commutator. Suppose
[y, x] commutes with x and y. Then for all n N we have
(xy)
n
= x
n
y
n
[y, x]
n(n1)/2
.
Proof: We expand:
(xy)
n
= (xy)(xy) (xy). (30)
We want to put all the ys to the right of all the xs. The commutator is the price to pay for replacing
yx by xy:
yx = [y, x]xy,
which we can write as
yx = xy[y, x],
since x, y commute with [y, x]. The left-most y in (30) moves past n 1 xs, so contributes [y, x]
n1
.
The new left-most y then contributes n 2 xs, and so on. Thus, we get
(xy)
n
= x
n
y
n
[y, x]
(n1)+(n2)++1
= x
n
y
n
[y, x]
n(n1)/2
,
as claimed.
Next we need a lemma about GL
2
(p).
Lemma 7.2 Any element of order p in GL
2
(p) is conjugate to
_
1 1
0 1
_
.
Proof: Let g GL
2
(p) have order p. By Sylows theorem, g is conjugate to an element
_
1 b
0 1
_
of the
Sylow p-subgroup U =
_
1
0 1
_
of GL
2
(p), with b ,= 0. Conjugating by
_
1 0
0 b
_
, we arrive at
_
1 1
0 1
_
.
78

Now return to our group G of order p

3
, and assume p > 2. We have seen that all commutators lie in Z,
so the conditions of Lemma 7.1 hold for any x, y G. Since p is odd, it divides p(p 1)/2 and since
[Z[ = p we have [y, x]
p(p1)/2
= 1. Hence for all x, y G we have
(xy)
p
= x
p
y
p
.
This means that the map x x
p
is a group homomorphism; we it simply by p : G G. Clearly
Z ker p. Since G is not cyclic we have x
p
2
= 1 for all x G, so imp ker p. Now Z cannot be the
whole of ker p, lest [ ker p[ = p and [ imp[ = p
2
. Hence there is an element y
0
ker p outside Z. The
subgroup A = y
0
, z generated by y
0
and z is then a normal subgroup of G isomorphic to C
p
C
p
.
Now there are two cases.
Case 1: ker p = G.
This means g
p
= 1 for all g G. Choose any element x G outside of A. Then x acts on A via an
element of order p in Aut(A) GL
2
(p). By Lemma 7.2 there are generators y, z of A such that
G = x A,
where x acts on A by
xzx
1
= z, xyx
1
= yz.
The structure of G is thus determined uniquely. We have
G = z
c
y
b
x
a
: a, b, c Z/pZ, (31)
with multiplication
(z
c
y
b
x
a
) (z
f
y
e
x
d
) = z
c+f+ae
y
b+e
x
a+d
.
The Heisenberg group
H :=
_
_
1
0 1
0 0 1
_
_
(32)
is a nonabelian group of order p
3
and exponent p, is therefore isomorphic to the abstract group G in
(31). Indeed, one checks that sending
z
c
y
b
x
a

_
_
1 a c
0 1 b
0 0 1
_
_
is an explicit isomorphism G

H.
Case 2: ker p ,= G.
Here G has an element y of order p
2
. Choose such a y and let Y = y C
p
2. Choose any x
0
in A but
not in Y . Then G is again a semidirect product:
G = x
0
Y,
79
with some action of x
0
on Y . Let R be the ring Z/p
2
Z. Then
R

Aut(Y ),
where r R corresponds to the automorphism y y
r
. The elements of order p in R lie in 1 +pR, so
x
0
yx
1
0
= y
1+rp
for some r R. Let s be the inverse of r in R and let x = x
s
0
. Then xyx
1
= y
1+p
,
which determines the group structure on G. We have
G = y
b
x
k
: b R, k Z/p,
and the map sending
y
b
x
k

_
(1 + p)
k
b
0 1
_
is an isomorphism from G to the group of matrices
__
a b
0 1
_
: a 1 + pR, b R
_
. (33)
7.1.1 Automorphisms of the Heisenberg group
Let p > 2 and let H be the Heisenberg group (32), with center
Z :=
_
_
1 0
0 1 0
0 0 1
_
_
.
The automorphism group Aut(H) preserves Z and therefore acts on H/Z (Z/pZ)
2
. Thus we get a
map
Aut(H) Aut(H/Z) GL
2
(p).
Proposition 7.3 This mapping ts into a split exact sequence
1 H/Z Aut(H) Aut(H/Z) 1
and Aut(H) (Z/pZ)
2
GL
2
(p) is the afne group of the plane over Z/pZ.
Proof: Let
x =
_
_
1 1 0
0 1 0
0 0 1
_
_
, y =
_
_
1 0 0
0 1 1
0 0 1
_
_
, z =
_
_
1 0 1
0 1 0
0 0 1
_
_
.
Then
z
c
y
b
x
a
=
_
_
1 a c
0 1 b
0 0 1
_
_
,
80
and H has the presentation
H = x, y, z[ x
p
= y
p
= 1, [x, y] = z, xz = zx, yz = zy.
From this (or matrix multiplication) we derive the formulas
(z
c
y
b
x
a
)(z
c

y
b

x
a

) = z
ab

+c+c

y
b+b

x
a+a

,
(z
c
y
b
z
a
)
1
= z
abc
y
b
x
a
,
[z
c
y
b
x
a
, z
c

y
b

x
a

] = z
ab

ba

.
The isomorphism H/Z (Z/pZ)
2
, sends (the coset mod Z of) z
c
y
b
x
a
(a, b). We now prove
exactness of the sequence
1 H/Z

Aut(H) Aut(H/Z) 1, (34)
where : h
h
Aut(H), with
h
(h

) = hh

h
1
, is the action of H on itself by inner automor-
phisms. Since H/Z is abelian,
h
induces the trivial automorphism on H/Z. Conversely, suppose
Aut(H) is trivial on H/Z. Then there are c, c

F
p
such that
(x) = z
c
x, (y) = z
c

y.
Let h = y
c
x
c

. One checks that

h
(x) = z
c
x,
h
(y) = z
c

y, so =
h
. Hence the sequence (34) is
exact in the middle.
We next prove that Aut(H) Aut(H/Z) = GL
2
(p) is surjective. Let A =
_
a b
c d
_
GL
2
(p). The
above commutation formula shows that
[y
b
x
a
, y
d
x
c
] = z
det(A)
.
Since every non-identity element has order p (here we are using p > 2), the elements y
b
x
a
, y
d
x
c
, z
det A
satisfy the relations of x, y, z. Hence there is a homomorphism : H H such that
x

= y
b
x
a
, y

= y
d
x
c
, z

= z
det A
.
We must show that is injective (hence is bijective). Since det A is invertible in F
p
, ker Z = 1.
Hence ker injects into the kernel of A in G/Z, which is trivial. This nishes the proof of exactness
in (34).
However, the function A just described is only a set-theoretic splitting; it is not a group homo-
morphism. In fact we could modify the denitions of x

and y

by multiplying by arbitrary elements

of Z and the relations of H would still be satised. It turns out we can make such a modication to get
a group-theoretic splitting, namely if we redene to be
x

= z
ab/2
y
b
x
a
, y

= z
cd/2
y
d
x
c
, z

= z
adbc
(35)
(note that ab/2, cd/2 are taken in Z/pZ, where 2 invertible) then if A, B GL
2
(p) correspond to ,
via (35), one can check that
(x

= x

, (y

= y

,
so that the new map A is a group homomorphism, where Aut(H) acts via the right on H.
There is one more detail to check, that the action of Aut(H/Z) on H/Z in the semidirect product
arising from the splitting just described coincides with the action (by right multiplication) of GL
2
(p)
on (Z/pZ)
2
. We leave this as an exercise.
81
7.2 Higher powers of p
Let N(p
r
) be the number of isomorphism classes of groups of order p
r
. We have found
r N(p
r
) groups
1 1 C
p
2 2 C
p
C
p
, C
p
2
3 5 C
3
p
, C
p
2 C
p
, C
p
3, U
3
(p), (33)
The case of p
3
relied in a straightforward way on the classication for p
2
. One might imagine that
all p-groups could be classied inductively in a similar way. The case of p
4
is much harder, but still
accessible to non-experts, at least for p = 2.
We give here the classication of groups of order 16
17
. There are 14 groups of order 16:
ve abelian groups: C
4
2
, C
4
C
2
2
, C
4
C
4
, C
8
C
2
, C
16
;
two direct products: D
4
C
2
, Q
8
C
2
.
the dihedral group D
8
= C
8
1
C
2
18
the quasidihedral group QD
16
= C
8
3
C
2
;
another semidirect product C
8
5
C
2
;
the unique nonabelian semidirect product C
4
C
4
;
the unique nonabelian semidirect product (C
2
C
2
) C
4
;
the unique nontrivial semidirect product Q
8
C
2
;
the generalized quaternion group Q
16
;
Of these groups only D
8
, QD
16
and Q
16
have maximal class 3; these groups have centers of order two,
and their quotients by their centers are D
4
. We will return to this.
It was also known in the 19
th
century that for p > 2 there are 15 groups of order p
4
.
So far, it seems the number of p-groups of a given order does not depend much on p. However, in the
last decade it has been found
19
that
N(p
5
) =
_

_
2p + 61 + 2(3, p 1) + (4, p 1) if p > 3
67 if p = 3
51 if p = 2,
17
For a proof, see The groups of order sixteen made easy, Marcel Wild, Bulletin of the A.M.S. 2005.
18
The notation C
n
k
C
m
means the generator of C
m
acts by the k
th
power on C
n
.
19
See the talk by B. Eick The classication of p-groups by coclass, http://homeweb1.unifr.ch/ciobanul/pub/beamer.pdf.
82
and
N(p
6
) =
_

_
3p
2
+ 39p + 344 + 24(3, p 1) + 11(4, p 1) + 2(5, p 1) if p > 3
504 if p = 3
267 if p = 2.
We also have mentioned that
N(2
10
) = 49 487 365 422.
It seems hopeless to classify p-groups whose order is a given power of p, but the fact that these re-
markable values of N(p
r
) have recently been found indicates great progress in the classication of
p-groups.
A big breakthrough came when people looked at groups of maximal nilpotence class, and found trees
of innite families of groups.
7.3 Projective limits and pro-p groups
Suppose we have a sequence of groups (or rings) X
n
, indexed by positive integers n, along with
homomorphisms f
n
: X
n
X
n1
for each n 2. Thus we have an innite sequence of groups
(or rings) and maps:
X
n
fn
X
n1
X
3
f
3
X
2
f
2
X
1
.
The projective limit of the system (X
n
, f
n
)
20
is the set of sequences
lim

n
X
n
:= (x
1
, x
2
, . . . ) : x
n
X
n
, f(x
n
) = x
n1
n 2,
under componentwise group (or ring) operations.
The projective limit X = lim
n
X
n
comes with a system of canonical projection maps

n
: X X
n
,
n
(x) = x
n
.
The quotients X
n
may be regarded as successive approximations to, or shadows of the single group X.
If all the X
n
are nite groups, the projective limit X is called pro-nite. If p is a prime and all the X
n
are p-groups, then X is called a pro-p group. The single group X describes an innite family X
n
of
nite p-groups.
The simplest example is when X
n
= Z/p
n
Z and f
n
is the natural projection
f
n
: Z/p
n
Z Z/p
n1
Z.
With these maps, the inverse limit
Z
p
:= lim

n
Z/p
n
Z
20
This is also called the inverse limit.
83
is the ring of p-adic integers, whose additive group is a pro-p group.
For a non-abelian example, let X
n
= D
2
n be the dihedral group of order 2
n+1
. Since the quotient
of D
2
n by its center is D
2
n1, we have quotient maps f
n
: X
n
X
n1
. The successive shadows
D
2
, D
4
, D
8
of the 2-adic dihedral group
D := lim

n
D
2
n
can be seen in the sequence of subgroup lattices

If (X
n
, f
n
) and (Y
n
, g
n
) are two projective systems, and we have for each n a homomorphism
h
n
: X
n
Y
n
such that h
n1
f
n
= g
n
h
n
then we get a homomorphism
h : lim

n
X
n
lim

n
Y
n
, (x
n
) (h(x
n
)).
Let Y
n
= z C

: z
2
n
= 1 with maps g
n
(z) = z
2
. Let h
n
: Z/2
n
Z Y
n
be the isomorphism
h
n
([x]) = exp(2i/2
n
).
Then we have an isomorphism
h : Z
2

Y := lim

n
Y
n
such that multiplication by 1 on Z
2
corresponds to inversion on Y . It follows that the 2-adic dihedral
group is a semidirect product
D = Z
2
C
2
,
where the nontrivial element of C
2
acts on Z
2
by negation.
84
7.4 Toward the classication of p-groups
Let G have order p
n+1
and maximal nilpotence class n. The ascending central series for G must have
the form
Z
1
< Z
2
< < Z
n1
< Z
n
= G,
where [Z
i
[ = p
i
for i < n. In particular the center Z
1
= Z(G) has order p. The group H = G/Z(G)
has order p
n
and nilpotence class n 1, which is again maximal. We indicate this relation between G
and H by an arrow G H.
Let ((p, 1) be the directed graph whose vertices are the isomorphism classes of nite p-groups of
maximal nilpotence class, having edges G H when G/Z(G) H, as above.
We will draw this graph for p = 2. Let
X
n
= D
2
n, Y
n
= Q
2
n+1, Z
n
= QD
2
n+1
be the dihedral, generalized quaternion and quasidihedral groups of order 2
n+1
. For small n we have
X
1
= C
2
2
, Y
1
= C
4
and Z
2
= Y
2
= Q
8
. Since the quotient of each of X
n
, Y
n
, Z
n
by its center is X
n1
,
the graph ((2, 1) is
X
4

X
3

X
2

X
1
Y
4

Y
3

Y
2

Y
1
Z
4

Z
3

This graph is a tree with exactly one innite path, over which the projective limit is the 2-adic dihedral
group D. The other groups are contained in branches B
n,k
consisting of all groups to the left of and
distance at most k from X
n
. The groups in ((2, 1) have different orders and nilpotence class, but they
all have the same coclass: If Ghas order p
r
and nilpotence class c then its coclass is dened to be r c.
For any prime p and integer c 1 one can dene a coclass graph ((p, c) whose vertices are all the
p-groups of coclass c with an edge between G and H if there exists N G with G/N H. The
p-groups of a given coclass c are then classied, at least qualitatively, via the structure of the graph
((p, c), about which the following is known.
1. There are only nitely many innite paths in ((p, c).
2. The projective limit over each innite path in ((p, c) is an innite pro-p group whose nite
quotients all have coclass c.
3. Every innite pro-p group with all quotients of coclass c arises from an innite path in ((p, c).
4. The groups in ((p, c) lie on branches, and these eventually become periodic.
85
For more details, see the talks online by Bettina Eick and her collaborators, along with the book The
structure of groups of prime-power order by Leedham-Green and McKay, two of the pioneers in this
new era of group theory.
8 Presentations of Groups
It is often convenient to represent group elements as words in a few symbols, having certain relations.
For example, the cyclic group C
n
can be expressed as
C
n
= a [ a
n
= 1,
which is called a presentation of C
n
. Here there is only one generator a, and one relation a
n
= 1.
Now suppose G is a group generated by two elements a, b of order two. Let C = ab be the subgroup
of G generated by ab. Since a and b have order two, we have a(ab)a = ba = (ab)
1
, and one checks
that every element of G can be written either as (ab)
i
or a(ab)
i
for some i Z. Thus, C has index two
in G and we have
G = C aC.
The multiplication in G is then completely determined by the order of C. If C is innite, then we can
describe G via the presentations
G = a, b [ a
2
= b
2
= 1. (36)
If [C[ = n < , we have an additional relation (ab)
n
= 1, so we write
G = a, b [ a
2
= b
2
= (ab)
n
= 1. (37)
These groups Gare isomorphic to the dihedral groups D

and D
n
, respectively, where a and b manifest
as reections about adjacent lines. However, the presentations (36) and (37) express G independently
of any manifestation. This allows us to identify Gin other settings, wherever we nd a group generated
by two elements of order two.
So we would like to view groups as generated by abstract symbols like a, b, c, . . . with certain relations
among the symbols such that every possible relation in the group is completely determined by the given
relations. To explain this precisely, we start with free groups, which are groups with no relations at all.
8.1 Free Groups
Let S be a set. The Free Group on S is the set F(S) consisting of an element e, and all formal
expressions s
n
1
1
s
n
2
2
s
n

where 1, the n
i
are nonzero integers, and s
1
, . . . , s

are elements of S
such that s
i
,= s
j
if [i j[ = 1. We multiply two such expressions according to the following rules.
e is the identity element of F(S).
For all s S we have s
n
s
m
= s
n+m
if n + m ,= 0 and s
n
s
n
= e.
86
For distinct s, t S we have s
n
t
m
= s
n
t
m
.
Inductively, we have
(s
n
1
1
s
n
2
2
s
n

)(t
m
1
1
t
m
2
2
t
m
k
k
) =
_

_
s
n
1
1
s
n
2
2
s
n

t
m
1
1
t
m
2
2
t
m
k
k
if s

,= t
1
s
n
1
1
s
n
2
2
s
n

+m
1

t
m
2
2
t
m
k
k
if s

= t
1
and n

+ m
1
,= 0
(s
n
1
1
s
n
2
2
s
n
1
1
) (t
m
2
2
t
m
k
k
) if s

= t
1
and n

+ m
1
= 0
Proposition 8.1 Let f : S G be a map from a set S to a group G. Then there is a unique extension
of f to a group homomorphism

f : F(S) G.
Proof: We dene

f(e) = 1 and

f(s
n
1
1
s
n
2
2
s
n

) = f(s
1
)
n
1
f(s
2
)
n
2
f(s

)
n

.
This is a group homomorphism because the multiplication in G obeys the same rules as those above
for F(S), with s
i
and t
i
replaced by f(s
i
) and f(t
i
).
If f

is another extension of f then for all s S and integers n > 0 we have

f

(s
n
) = f

(s s
. .
n terms
) = f

(s) f

(s)
. .
n terms
= f

(s)
n
= f(s)
n
=

f(s)
n
,
while if n < 0 we have s
n
= (s
n
)
1
, so by what we just proved, we have
f

(s
n
) = f

((s
n
)
1
) = f

(s
n
)
1
=

f(s
n
)
1
=

f((s
n
)
1
) =

f(s
n
).
Finally, let s
1
, . . . s

S and take nonzero integers n

1
, . . . n

. Multiplying in F(S) we have

s
n
1
1
s
n
2
2
s
n

= (s
n
1
1
) (s
n
2
2
) (s
n

),
so
f

(s
n
1
1
s
n
2
2
s
n

) = f

(s
n
1
1
) f

(s
n
2
2
) f

(s
n

) =

f(s
n
1
1
)

f(s
n
2
2
)

f(s
n

) =

f(s
n
1
1
s
n
2
2
s
n

),
and therefore f

=

f. Hence

f is the unique extension of f to a homomorphism from F(S) to G.
Corollary 8.2 If S and T are sets and f : S T is a function, then f extends uniquely to a group
homomorphism

f : F(S) F(T).
Proof: By Prop. 8.1 the inclusion map : T F(T) extends uniquely to a homomorphism :
F(S) F(T).
Corollary 8.3 If S and T are sets of the same cardinality and f : S

T is a bijection then f extends
uniquely to a group isomorphism

f : F(S) F(T).
87
Consequently, for each positive integer n there is a unique (up to isomorphism) free group on n gener-
ators. We often denote this group by F
n
.
Free groups abound in topology. Here are several related examples.
Example 1: If n = 1 then F
1
Z. This is the fundamental group
1
(C) of a circle C.
Example 2: Suppose [S[ = 2, say S = s
1
, s
2
. We can visualize F(S) = F
2
as the fundamental
group of the space X consisting of two circles C
1
and C
2
touching at a point p . Here s
i
is the
counterclockwise (say) loop once around C
i
starting at p, while s
1
i
loops around C
i
in the opposite
direction. A word like s
3
1
s
2
2
s
1
is a loop going three counterclockwise times around C
1
then two
clockwise times around C
2
, then once counterclockwise time around C
1
. The generalization to a
bouquet of n circles all touching at a single point is the obvious one.
Example 3: Let X be a two-sphere punctured at n + 1 points. Then X retracts onto a bouquet of
n circles, so
1
(X) F
n
. More generally, let X be a closed surface of genus g, punctured at n + 1
points. Then
1
(X) F
2g+n
.
Example 4: Let Gbe a torsion-free subgroup of SL
2
(Z) of nite index. Then Gacts on the upper-half
plane 1 = z Z : z > 0 by linear fractional transformations:
_
a b
c d
_
z =
az + b
cz + d
.
The quotient X = 1/G is a nitely punctured Riemann Surface with
1
(X) G. Therefore is free.
Example 5: A tree is a contractible one dimensional simplicial complex (informally, a graph without
loops). Any group acting freely on a tree is free.
Example 6: Any subgroup of a free group is free. The proof may be sketched thus: Given a free group
F, one constructs a tree T
F
on which F acts freely. Then any subgroup F

< F also acts freely on T

F
,
hence F

is free as well.
8.2 Generators and Relations
Let S be a set, and let R be a subset of F(S). Let N(R) be the smallest normal subgroup of F(S)
containing R. More precisely, N(R) is the intersection of all normal subgroups of F(S) containing R.
We dene a group
S [ R := F(S)/N(R).
This is the largest group with generators S and relations R, in the following sense.
Let G be any group and let S be a subset of G. By Prop. 8.1, the identity map : S S extends
uniquely to a group homomorphism
: F(S) G.
88
Proposition 8.4 (Mapping Property) Let R F(S) be a subset contained in ker and let :
F(S) S [ R be the quotient map. Then there is a unique homomorphism
S,R
: S [ R G
such that
S,R
= .
Proof: By assumption, the elements of R in F(S) belong to the kernel of . Hence N(R) ker , so
induces a homomorphism S [ R G, which is the desired map
S,R
.
Usually we take S to be a generating set for G, in which case the map
S,R
in Prop. 8.4 is surjective. If
the
S,R
is an isomorphism, we say that S [ R is a presentation of G, or more informally, that G is
generated by S with the relations R. Given a group G with generators S, there is always some subset
R F(S) such that
S,R
is an isomorphism. For example, we could take R to be the full kernel of .
In practice, we would like S and R to be small and simple. However, it has been proved that there is
no algorithm that decides, for a given group G with generating set S and relations R, whether or not

S,R
is an isomorphism. Indeed this is so even for G = 1! It takes a combination of work, skill and
luck to nd a presentation for a given group G, but once found, a presentation can be very useful.
Example 1: For 1 n < , the dihedral group D
n
has presentation with generating set S = a, b
and relations R = a
2
, b
n
, abab. We usually write R in terms of equations that hold in the group, as
follows
D
n
= a, b [ a
2
= b
n
= 1, aba = b
1
(38)
To prove that (38) is correct, let H = a, b [ a
2
= b
n
, aba = b
1
. Let r D
n
be a reection and let
t S
n
be a rotation of order n. Then rtr = t
1
. Hence by the mapping property, we have a surjective
homomorphism : H D
n
such that (a) = r and (b) = t. This is always the easy step towards
verifying a presentation. The tricky part is to prove that is injective. In this case, we note that the
relations in H allow us to write every element in the form b
i
or ab
i
,
21
with 1 i n. This shows that
[H[ 2n = [D
n
[, so is injective and (38) is proved.
The same presentation works for D

; we just drop the relation b

n
= 1:
D

= a, b [ a
2
= 1, aba = b
1
. (39)
To prove this we again let H = a, b [ a
2
= 1, aba = b
1
. Let r, r

be reections about adjacent

parallel lines ,

, so that t := rr

is the translation by twice the distance from to

. Then rtr = t
1
,
so we have a surjection : H D

such that (a) = r and (b) = t. Now H and D

are innite
groups, so we cannot count orders as we did previously. But still every element of H can be expressed
in the form b
i
or ab
i
for some i Z, and since t
i
,= 1 ,= rt
i
in D

, it follows that is injective.

8.3 A presentation of the symmetric group
A presentation for a given nite group G can often be found as follows. We rst nd a set S of
generators of G, and some relations R among the generators that appear to determine all the relations
in G. Let = S [ R. We then have a surjection : G. To prove that is an isomorphism
we try to show that [[ [G[. Suppose we nd a subgroup H whose order is known, say by
21
Strictly speaking, we should write something like

b
i
and a

b
i
, where a and

b are the images of a and b in H.
89
induction. It then sufces to show that [/H[ [G[/[H[. This will succeed if we can nd a set of
cosets aH, bH, . . . of size at most [G[/[H[
We illustrate with the group G = S
n+1
.
Lemma 8.5 The group S
n+1
is generated by
1
, . . . ,
n
, where
i
= (i i + 1).
Proof: We use induction on n. For n = 1 we have S
2
= e, (1 2
}
=
1
, so the result holds in
this case. Assume that S
n
is generated by
1
, . . . ,
n1
, and let G =
1
, . . . ,
n
= S
n
,
n
. Let
S
n+1
be arbitrary, and set k = (n + 1). The 2-cycle

k+1

n1

n1

k+1

k
= (k n + 1)
belongs to G, and (k n + 1) xes n + 1. Since S
n
is the stabilizer of n + 1 in S
n+1
, and S
n
G, it
follows that (k n + 1) G, so G. Hence G = S
n+1
.
For 1 i, j n dene
m
ij
=
_

_
1 if i = j
2 if [i j[ > 1
3 if [i j[ = 1.
(40)
Let
n
= S, R be the group with generating set S = s
1
, . . . , s
n
and relations
R = (s
i
s
j
)
m
ij
: 1 i, j n.
More explicitly, these relations are
s
2
i
= 1 and s
i
s
i+1
s
i
= s
i+1
s
i
s
i+1
for all 1 i n
and
s
i
s
j
= s
j
s
i
if [i j[ 2.
Proposition 8.6 The map s
i

i
extends to an isomorphism
n

S
n+1
so we have the presentation
S
n+1
s
1
, . . . , s
n
[ (s
i
s
j
)
m
ij
= 1 1 i, j n.
Proof: The 2-cycles
i
S
n+1
satisfy the same relations as the generators s
i

n
, so there is a
surjective homomorphism
:
n
S
n+1
,
such that (s
i
) =
i
for all 1 i n. It sufces to show that [
n
[ (n + 1)!.
For n = 1, we have
1
= s
1
[ s
2
1
= 1 S
2
, so the result is true. Assume that [
n1
[ n!. Let
t
1
, . . . , t
n1
be the generators of
n
. Sending t
i
s
i+1
sends
n
onto the subgroup H
n
generated
by s
2
, . . . , s
n
. Note that [H[ n!.
90
Consider the n + 1 cosets
H
0
= H, H
1
= s
1
H, H
2
= s
2
s
1
H, , . . . , H
n
= s
n
s
n1
s
1
H.
I claim that for 1 i n we have
s
i
H
i
= H
i1
, s
i
H
i1
= H
i
, s
i
H
j
= H
j
if j ,= i, i 1. (41)
The rst two are clear from the denitions of H
i
and the relation s
2
i
= 1. For the last equation, suppose
rst that j i 2. Then
s
i
H
j
= s
i
s
j
s
j1
s
1
H = s
j
s
j1
s
1
s
i
H = s
j
s
j1
s
1
H = H
j
.
Suppose next that j i + 1. Then (writing s
i
for the generator which moves)
s
i
H
j
= s
i
s
j
s
j1
s
i+1
s
i
s
i1
s
1
H
= s
j
s
j1
s
i
s
i+1
s
i
s
i1
s
1
H
= s
j
s
j1
s
i+1
s
i
s
i+1
s
i1
s
1
H
= s
j
s
j1
s
i+1
s
i
s
i1
s
1
s
i+1
H
= s
j
s
j1
s
i+1
s
i
s
i1
s
1
H
= H
j
,
as claimed. Since s
1
, . . . , s
n
generate
n
, it follows that
n
preserves the set H
0
, H
1
, . . . , H
n

n
/H. By transitivity, this containment must be equality. Hence [
n
/H[ n + 1 and [
n
[ (n +
1)[H[ (n + 1)n! = (n + 1)!.
8.4 Coxeter groups and reection groups
A Coxeter system is a pair (G, S) where G is a group with generating set S G and presentation
G S[R where the relation set R contains s
2
for all s S and the remaining words in R have the
form (ss

)
m(s,s

)
for distinct elements s, s

S and integers m(s, s

) 2. In particular, the elements of

S all have order two.
We also write m(s, s

) = if ss

has innite order, but this does not appear in R. A group G is a

Coxeter group if G is generated by a set S G of elements of order two such that (G, S) is a Coxeter
system. The rank of the Coxeter system (G, S) is the cardinality of the set S.
The Coxeter diagram of (G, S) is the graph with vertex set S, and edge set s, s

: m(s, s

) 3,
with each edge s, s

is labelled by the integer m(s, s

). Thus, two vertices s, s

have no edge between

them exactly when s, s

commute, and

s
m
-
s

means that ss

has order m = m(s, s

) 3. The most common Coxeter systems have small values of

m(s, s

), where the following alternate notation is used

if m(s, s

) = 3

s
==
s

if m(s, s

) = 4

if m(s, s

) = 6
91
A Coxeter system (G, S) is irreducible if S is not a disjoint union S = S

with m(s

, s

) = 2 for
all s

and s

. In other words (G, S) is irreducible if its graph is connected.

The classication of nite irreducible Coxeter groups is given below. Each has a label of the form X
n
where the rank n is the number of vertices in the Coxeter graph. Beware that A
n
is S
n+1
and not the
alternating group. And D
n
is not the dihedral group D
n
, but is the semidirect product of S
n
with the
group
+
C
n
2
= (
1
, . . . ,
n
) 1
n
:

i
= 1.
Coxeter label Coxeter diagram Group structure
A
n
- - - S
n+1
B
n
-- - == C
n
2
S
n
D
n

+
C
n
2
S
n
G
2
D
6
F
4
-==- D
4
S
3
E
6

order = 51840
E
7

order = 2903040
E
8

order = 696729600
I
2
(m), m ,= 2, 3, 4, 6
m
- D
m
H
3

5
A
5
C
2
H
4

5
(SL
2
(5) SL
2
(5)) C
2
Coxeter groups are a generalization of Euclidean geometry to higher dimensions, in the following
sense.
Let V = R
n
with the usual dot product u v. A reection on V is an element r GL
n
(R) xing a
hyperplane H pointwise and negating the line perpendicular to H. If u is a unit vector perpendicular
to H then r is given by
r(v) = v 2(u v)u.
A real reection group of rank n is a subgroup G of GL
n
(R) generated by reections. We say G is
irreducible if no proper subspace of V is preserved by G.
It turns out that nite Coxeter groups are nite reection groups. More precisely we have the following.
22
22
For proofs, and much more information about Coxeter groups and reection groups, see Bourbaki, Lie groups and Lie
algebras chapters 4,5,6.
92
Theorem 8.7 Let (G, S) be an irreducible Coxeter system of rank n. Then there is an injective homo-
morphism : G GL
n
(R) such that the following hold.
1. (s) is a reection for each s S.
2. The image (G) GL
n
(R) is an irreducible reection group of rank n.
3. Every irreducible reection group in GL
n
(R) is (G) for a unique irreducible Coxeter system
(G, S) of rank n.
8.5 Presentations of alternating groups
We can use the presentation of S
n+1
to get a presentation of A
n+1
, as follows. Let

i
= (1 2)(i i + 1) =
1

i
, for 2 i n.
Now let
n
be the group with generators a
2
, . . . , a
n
and relations
a
3
2
= a
2
1
= 1 for 3 i n,
and
(a
i
a
1
j
)
m
ij
= 1 for all i ,= j,
where m
ij
are as in (40).
Proposition 8.8 The map a
i

i
extends to an isomorphism
n

A
n+1
, so we have the presenta-
tion
A
n+1
a
2
, . . . , a
n
: a
3
2
= a
2
3
= = a
2
n
= 1, (a
i
a
1
j
)
m
ij
= 1 i ,= j [2, n].
Proof: We outline the proof, leaving some calculations to the reader. The elements b
i
= a
1
i

n
satisfy the same relations as the elements a
i
. Hence there is an automorphism :
n

n
such that
(a
i
) = a
1
i
.
Let

n
be the set
n
with multiplication
(,
i
)(

,
j
) = (
i
(),
i+j
)
23
In

let s
1
= (, 1) and s
i
= (, a
i
) for 2 i n. Then the s
i
generate
n
and satisfy the
same relations as
i
in S
n+1
. Hence we have a surjection : S
n+1

n
such that (
i
) = s
i
.
Reciprocally, the elements s
1
,
2
, . . . ,
n
S
n+1
satisfy the same relations as , a
2
, . . . , a
n
in

n
.
Hence there is map :

n
S
n+1
which is the inverse of . One checks that (
n
) = A
n+1
and this
completes the proof.
23
That is,

n
= , see section ().
93
8.5.1 A presentation of A
5
The presentation of A
n
in Prop. 8.8 has many relations, which can be inefcient. The following
well-known presentation of A
5
has fewer relations.
Proposition 8.9 We have A
5
a, b [ a
5
= b
3
= (ab)
2
= 1.
Proof: Let = a, b [ a
5
= b
3
= (ab)
2
= 1. We rst nd elements , A
5
obeying the same
relations as a, b. We may assume = (1 2 3 4 5), and = (i j k) is a 3-cycle such that has
order two. In A
5
, this means is a 221 cycle, so has a unique xed-point. Making this xed-point
1, we have = (1 5 k). To have this be the only xed-point of we must have k = 3. Indeed,
(1 2 3 4 5)(1 5 3) = (2 3)(4 5). Hence = (1 5 3) works, and we have a homomorphism : A
5
sending a , b . The image of contains elements of orders 2, 3, 5, hence is divisible by 30,
but A
5
is simple, hence has no subgroups of order 30, so is surjective.
It remains to show that [[ 60. Consider the subgroup A = a . Since a
5
= 1 and (a) = ,=
1, it follows [A[ = 5. It now sufces to show that [/A[ 12. This will be achieved if we can exhibit
a set of 12 cosets gA which is closed under multiplication by a and b.
The orbits of A on /A have size 1 or 5, since 5 is prime. The group B = b acts freely on /A, since
3 and 5 are relatively prime. We number the cosets according to a-orbits as follows.
1 = A
2 = bA, 3 = abA, 4 = a
2
bA, 5 = a
3
bA, 6 = a
4
bA,
7 = a
4
ba
4
bA, 8 = ba
4
bA = b 6, 9 = aba
4
bA, 10 = a
2
ba
4
bA, 11 = a
3
ba
4
bA,
12 = ba
3
ba
4
bA = b 11,
We do not need to know if these cosets are distinct, since we only seek and upper bound on [/A[.
That 1, . . . , 12 are indeed distinct will result from the proof that they are closed under multiplication
by a and b.
In the following diagram, a solid arrow i j means that a i = j, and i means a i = i. We will
show that the b-action is given by the dashed arrows, where i j means a i = j = b i.
4

9
/

3
.

10

8

12

11

94
Since (ab)
2
= 1 and b
1
= b
2
, we have the relation aba = b
2
, hence
b 2 = b
2
A = abaA = abA = 3.
Since b 1 = 2 and b has order three, we have
b 3 = b
3
1 = 1,
so 1, 2, 3 is a b-orbit. Inverting the relation aba = b
2
, we get b = a
1
b
1
a
1
. Using the known b
orbit 1, 2, 3 we compute
b 4 = a
1
b
1
a
1
4 = a
1
b
1
3 = a
1
2 = 6.
Since b 6 = 8, it follows that 4, 6, 8 is a second b-orbit. Then we have
b 5 = a
1
b
1
a
1
5 = a
1
8 = 7,
b 9 = a
1
b
1
a
1
9 = a
1
6 = 5,
so 5, 7, 9 is a third b-orbit. Finally,
b 10 = a
1
b
1
a
1
10 = a
1
7 = 11,
and since b 11 = 12, it follows that 10, 11, 12 is the fourth and nal b-orbit.
It remains only to check that a xes 12:
a 12 = aba
3
b a
4
b1 = aba
3
b a5 = aba a aba5 = b
1
ab
1
5 = b
1
a9 = b
1
10 = 12.
This completes the proof that the cosets 1, . . . , 12 exhaust [/A[. It follows that [/A[ = 12 and
A
5
as claimed.
Corollary 8.10 Let G be a group containing nontrivial elements x, y satisfying the relations
x
5
= y
3
= (xy)
2
= 1.
Then the subgroup of G generated by x, y is isomorphic to A
5
.
Proof: By Prop. 8.9 and the Mapping Property Prop. 8.4, there is a homomorphism : A
5
G
sending a x, b y. And is nontrivial since x, y are nontrivial. Since A
5
is simple, is injective.
Hence A
5
is isomorphic to the image of , which is the subgroup generated by x, y, z.
95
8.5.2 The exceptional isomorphism PSL
2
(9) A
6
Here is an illustration of the use of Cor. 8.10. The eld F
9
of nine elements can be constructed from the
eld F
3
= Z/3Z just as C is constructed fromR, namely F
9
= a+bi : a, b F
3
with multiplication
determined by the rule i
2
= 1. In SL
2
(9) the matrices
X =
_
1 + i i
1 1
_
, and Y =
_
1 1 i
0 1
_
satisfy X
5
= I, Y
3
= I, (XY )
2
= I. Hence the images x, y of X, Y in PSL
2
(9) satisfy
x
5
= y
3
= (xy)
2
= 1
and therefore x, y generate a subgroup H PSL
2
(9) with H A
5
. Since [ PSL
2
(9)[ = 9(9
2
1)/2 =
360 and [H[ = 3 4 5 = 60, we have [PSL
2
(9) : H] = 6. The action of PSL
2
(9) on the six cosets of
H gives a homomorphism
: PSL
2
(9) S
6
,
which is injective with image in A
6
, since PSL
2
(9) is simple. Since [A
6
[ = 3 4 5 6 = 360, it follows
that gives an isomorphism
: PSL
2
(9)

A
6
,
which is one of the exceptional isomorphisms between linear and permutation groups.
8.6 The Platonic Groups
The groups A
4
, S
4
and A
5
have presentations
A
4
a, b [ a
3
= b
3
= (ab)
2
= 1
S
4
a, b [ a
4
= b
3
= (ab)
2
= 1
A
5
a, b [ a
5
= b
3
= (ab)
2
= 1.
(42)
We veried this presentation of A
5
above and the others can be done in the same way. The triples
(2, 3, 3), (2, 3, 4), (2, 3, 5) also arise from the Platonic solids, as follows.
Let G be a nite group acting on a set X, with the following two properties:
1. For all x X the stabilizer G
x
is nontrivial.
2. Any nonidentity element of G has exactly two xed-points in X.
What can we say about G? Let [G[ = n and let there be r-orbits O
1
, . . . , O
r
in X, and let m
i
be the
order of the stabilizer of a point in O
i
. By the Burnside Counting formula we have
r

i=1
1 =
1
n

gG
[X
g
[ =
1
n
([X[ + 2(n 1)) =
1
n
_
r

i=1
n
m
i
+ 2(n 1)
_
,
96
so
r

i=1
_
1
1
m
i
_
= 2
2
n
. (43)
We have r > 1, lest
1 >
1
m
i
= 2
2
n
1.
And since 1 1/m
i
1/2, we have
r
2
2
_
1
1
n
_
,
implying r 3. Hence r = 2 or r = 3.
If r = 2 we have
1
m
1
+
1
m
2
=
1
n
+
1
n
,
which means that m
1
= m
2
= n, so X has just two elements and G acts trivially on X.
We arrive at r = 3, and
1
m
1
+
1
m
2
+
1
m
3
= 1 +
2
n
.
Index so that m
1
m
2
m
3
. We cannot have m
1
3, lest the left side be 1. So m
1
= 2 and
1
m
2
+
1
m
3
=
1
2
+
2
n
.
We cannot have m
2
4, lest the left side be 1/2. If m
2
= 2 then n = 2m
3
. If m
2
= 3 we have
1
m
3
=
1
6
+
2
n
,
so that m
3
= 3, 4, 5 with n = 12, 24, 60, respectively. To summarize, we have the following possibili-
ties:
r m
1
m
2
m
3
n
2 n n n
3 2 2 m 2m
3 2 3 3 12
3 2 3 4 24
3 2 3 5 60
Now assume that G is a nite group acting by by rotations on the two-dimensional sphere S
2
. Let
X = x S
2
: G
x
,= 1. Then X consists of antipodal pairs x, x on the axes of rotation of
the non-trivial elements of G. For the rst two rows of the table above we have G cyclic or dihedral,
respectively.
97
For (m
1
, m
2
, m
3
) = (2, 3, 3), G has 3 elements of order 2 and 8 elements of order three. Since a
2-Sylow subgroup of G has order four and G has no elements of order four, the 2-Sylow must be
K
4
, and is unique, hence normal in G. But the 3-Sylows are not unique, hence are not normal. It
follows that G A
4
. A 3-Sylow is the stabilizer G
x
of a G-orbit x, y, z, w in S
2
and permutes
y, z, w transitively. Hence each of y, z, w have the same distance from x. Likewise G
y
permutes
x, z, w transitively, so these points all have the same distance from y. It follows that x, y, z, w are the
vertices of a tetrahedron, whose symmetry group is G.
For (m
1
, m
2
, m
3
) = (2, 3, 4), G has 6 + 3 elements of order two, 8 elements of order three and 6
elements of order four. Hence G has four 3-Sylow subgroups and we have a homomorphism : G
S
4
. If P and Q are distinct 3-Sylows 1hen [N(P) N(Q)[ 2, so [ ker [ 2. If [ ker [ = 2 then
ker is central in G and im = A
4
. Since there are eight involutions outside ker we would have
at least 4 involutions in A
4
, which is not the case. So ker = 1 and : G S
4
is an isomorphism.
As above, one can show that the six points in S
2
with stablizer C
4
form the vertices of an octahedron
whose symmetry group is G.
For (m
1
, m
2
, m
3
) = (2, 3, 5), we have [G[ = 60. We show that G is simple. There are 24 elements of
order ve, hence G has six Sylow 5-subgroups. Let N be a non-trivial normal subgroup of G. If [N[
is divisible by 5 then all six 5-Sylows are in N, so [N[ 1 + 24, so [N[ 30. Therefore N contains
an element of order two. But G has fteen conjugate elements of order two, so [N[ 25 + 15 > 30,
hence N = G, and we have proved that G is simple. By Cor. 5.19 it follows that G A
5
. There is a
G-orbit of 12 points in S
2
whose stabilizers have order 5. One can show that these are the vertices of
an icosahedron.
9 Building new groups from old
9.1 Automorphisms
Recall that an automorphism of a group G is an isomorphism f : G G from G to itself. The set
Aut(G) of automorphisms of G forms a group under composition, with identity element I
G
, given by
I
G
(g) = g for all g G.
There are various kinds of automorphisms; some automorphisms come from G itself: For each g G,
let c
g
: G G be the function given by c
g
(x) = gxg
1
. It is easy to check that c
g
Aut(G) and that
c : G Aut(G)
is a group homomorphism. In general, the homomorphism c is neither injective nor surjective (see
examples below). The image Inn(G) = c
g
: g G of c is the group inner automorphisms of G.
In general, the kernel of c is the center Z(G) of G, and c induces an isomorphism
G/Z(G) Inn(G) Aut(G).
You can check that if Aut(G), then
c
g

1
= c
(g)
g G.
98
Therefore Inn(G) is a normal subgroup of Aut(G); the quotient
Out(G) := Aut(G)/ Inn(G)
is the outer automorphism group of G. All of these groups t into the exact sequence
1 Z(G) G
c
Aut(G) Out(G) 1.
Typically, outer automorphisms of G arise from conjugation in a larger group

G, in which G

G. This
is why
Aut(A
n
) = S
n
,
for n ,= 6 (see table below). Often one needs to know which subgroups of G are normalized by

G.
This leads to the notion of characteristic subgroup: We say that a subgroup H G is characteristic
in G if (H) = H for every Aut(G). The center Z(G) and commutator [G, G] are examples of
characteristic subgroups in any group G.
If H is characteristic in G then H is normal in G, but not conversely. For example, if G = C
2
C
2
then every subgroup is normal, but Aut(G) = GL
2
(2) moves the subgroups of order two G, so these
are not characteristic. In G = Q
8
the subgroups i, j, k have index two, hence are normal in G,
but there is an automorphism Aut(Q
8
) of order three, sending i j k i. Hence none of
these subgroups are characteristic.
Examples of Automorphism groups
G Z(G) Aut(G) Inn(G) Out(G)
Z Z C
2
1 C
2
Z
n
Z
n
GL
n
(Z) 1 GL
n
(Z)
C
n
C
n
(Z/nZ)

1 (Z/nZ)

C
n
p
C
n
p
GL
n
(p) 1 GL
n
(p)
S
n
, n ,= 2, 6 1 S
n
S
n
1
S
6
1 S
6
2 S
6
C
2
A
n
, n ,= 2, 3, 6 1 S
n
A
n
C
2
A
6
1 S
6
2 A
6
C
2
C
2
D
4
C
2
D
4
D
2
C
2
Q
8
C
2
S
4
D
2
S
3
The notation Aut(S
6
) S
6
2 means that Aut(S
6
) ts into an exact sequence
1 S
6
c
Aut(S
6
) C
2
1,
99
and similarly for A
6
. We will examine this exceptional case in the next section.
9.1.1 Automorphisms of S
n
An automorphism of a group G permutes the conjugacy classes in G, and the inner automorphisms
preserve each conjugacy class. If Aut(G) and X, Y are conjugacy classes in G such that (X) =
Y , then [X[ = [Y [ and the elements in Y have the same order as the elements in X.
Suppose is an automorphism of the symmetric group S
n
, for n 2. Then sends the class X of
2-cycles in S
n
to another class Y of elements of order two such that [Y [ = [X[ = n(n 1)/2. There
is 1 k n/2 such that the elements in Y have cycle type [2
k
1
n2k
]. One possibility is k = 1, which
means Y = X, as occurs for the inner automorphisms.
Suppose that k 2. There are n!/(k!2
k
(n 2k)!) elements in S
n
with cycle type [2
k
1
n2k
], so we
must have
n(n 1)
2
= [X[ = [Y [ =
n!
k!2
k
(n 2k)!
.
We rewrite this equation as
k(2k 2)(2k 3) 2 1 = (n 2)(n 3) (n 4)(n 5) (n 2k + 2)(n 2k + 1).
As 2k 2 n 2 and 2k 4 n 4 etc, we must have n = 2k and nd that
n = 2(n 3)(n 5) 3 1 2(n 3),
which implies that n = 6 and k = 3. We note that the classes [21111] and [222] in S
6
both have 15
elements. We have proved the following.
Lemma 9.1 If S
n
has an automorphism which does not preserve the class of 2-cycles, then n = 6
and sends the class of 2-cycles to the class of 222-cycles.
We next investigate the case k = 1.
Lemma 9.2 Suppose S
n
preserves the class of 2-cycles. Then is inner.
Proof:
24
For each 2 r n there are a
r
, b
r
1, . . . , n such that (1 r) = (a
r
b
r
). The order of
a
r
and b
r
is not determined, and we will exploit this ambiguity.
We single out r = 2, and set a = a
2
, b = b
2
, so that (1 2) = (a b). Let r 3. Since (1 r)(2 r) =
(1 2 r) has order three, we must have
(1 r) (1 2) = (a
r
b
r
)(a b)
24
This is a rewrite of the proof by I. Segal, Bull. AMS 1940, vol 46, p. 565.
100
also of order three. This means the intersection a
r
, b
r
a, b consists of a single element. Hence
either a
r
a, b and b
r
/ a, b or vice-versa. Let us switch a
r
and b
r
if necessary so that
a
r
a, b and b
r
/ a, b for all r 3.
It appears that whether a
r
= a or a
r
= b could depend on r. Suppose that for some r, s 3 we have
a
r
= a and a
s
= b. Then
(1 2 r) = (1 r) (1 2) = (a b
r
) (a b) = (a b b
r
),
and
(1 2 s) = (1 s) (1 2) = (b b
s
) (a b) = (a b
s
b).
Now (1 2 r)(1 2 s) = (1 r)(2 s) has order two, so (a b b
r
)(a b
s
b) must also have order two. But this is
impossible, for if b
r
= b
s
then (a b b
r
)(a b
s
b) = e, while if b
r
,= b
s
then (a b b
r
)(a b
s
b) = (a b
s
b
r
)
has order three.
This contradiction shows that either a
r
= a for all r 3 or a
r
= b for all r 3. We now switch a
and b, if necessary, so that a
r
= a for all r 3. Now is conjugation by the permutation sending
1 a, 2 b, and r b
r
for all r 3.
9.2 Semidirect Products (external view)
Let G and H be groups, and suppose we are given a homomorphism : G Aut(H), sending g G
to the automorphism
g
Aut(H). This is called an action of G on H via . We can then form a new
group H

G, as follows. As a set, H

G = H G is the direct product of the two sets G and H.

The multiplication is given by
(h, g) (h

, g

) = (h
g
(h

), gg

), h H, g G.
Note that H

G = H G as groups exactly when is the trivial homomorphism.

We have injective homomorphisms
: H H

G, : G H

G,
given by (h) = (h, 1) and (g) = (1, g). It is common to identify H = (H), and G = (G), but for
clarity and brevity at this stage, we write H

= (H) and G

= (G). Please check that the following

hold:
1. H

G = H

;
2. H

= 1;
3. H

(H

G) and (H

G)/H

G.
4. For g G and h H, we have (g) (h) (g)
1
= (
g
(h)).
101
These formulas spell everything out completely, but they are cumbersome to use in practice, so one
resorts to a more compact notation, such as the following. We identify H = (H) and G = (G), we
suppress , and we write HG instead of H

G, with multiplication rule

hg h

= h
g
(h

) gg

.
Thus, the semidirect product construction makes it so that conjugation by g on H is the given auto-
morphism
g
. This makes the multiplication rule easy to remember, and is shows how to recognize a
semidirect product.
Theorem 9.3 Let G be a group with subgroups H, K G having the following properties:
1. H G;
2. G = HK;
3. H K = 1.
Then G H

K, where : K Aut(H) is given by

k
(h) = khk
1
for all k K and h H.
Proof: The product map H K G, given by (h, k) hk, is a homomorphism from H

K to
G, which is surjective by part 2 and injective by part 1.
Example 1: The groups D
4
, A
4
and S
4
The group D
2
= C
2
C
2
has automorphism group
Aut(D
2
) = GL
2
(2) = S
3
.
The subgroups of S
3
are isomorphic to C
1
, C
2
, C
3
and S
3
itself. Hence we have semidirect products of
D
2
with each of these subgroups of S
3
, where is the inclusion map. In fact, we have
D
2
C
1
D
2
D
2
C
2
D
4
D
2
C
3
A
4
D
2
S
3
S
4
.
As the right side is a chain of subgroups of S
4
, it sufces to verify the last line. Let K < S
4
be the
subgroup xing 4. Then K S
3
and K does not contain any 22 cycle, so K D
2
= 1. Hence
S
4
= D
2
K.
Example 2: Dihedral groups
In the dihedral group D
n
of order 2n, let r be a rotation of order n, and let s be a reection. Then
D
n
= r

s C
n

C
2
,
102
where
s
(r) = r
1
. In particular, we have D
4
C
4
C
2
. We have also seen in example 1 that
D
4
D
2
C
2
. This shows that a semidirect product decomposition need not be unique.
Example 3: Nonabelian groups of order p
3
Let p be a prime. The group C
p
C
p
has automorphism group
Aut(C
p
C
p
) = GL
2
(p).
In GL
2
(p) we have the subgroup
U
2
(p) =
__
1 y
0 1
_
: y Z/pZ
_
C
p
so we can form the semidirect product (C
p
C
p
) U
2
(p), which is nonabelian of order p
3
. On the
other hand the group
U
3
(p) =
_
_
_
_
_
1 x z
0 1 y
0 0 1
_
_
: x, y, z Z/pZ
_
_
_
has the subgroup with y = 0 isomorphic to C
p
C
p
, the subgroup with x = z = 0 isomorphic to
U
2
(p), and we have
U
3
(p) (C
p
C
p
) U
2
(p).
This group is called the Heisenberg group over Z/pZ, because its commutator relations mimic the
Uncertainty Principle.
If p = 2, then U
3
(2) D
4
. The quaternion group Q
8
also has order 2
3
, but it cannot be expressed as
a nontrivial semidirect product, because every nontrivial subgroup of Q
8
contains the center Z(Q
8
) =
1, so condition 3 above is never satised.
Example 4: Afne transformations
Let V be a vector space over a eld k. An afne space over V is a set X on which V acts freely and
transitively. We write the action as (v, x) x + v, where v V and x X. If we choose any point
x
0
X, we get a bijection V X sending v x
0
+ v. However, there is no canonical point, or
origin in X, so we cannot identify X with V in any canonical way. This means we cannot add points
in X, because that requires an origin. We can subtract points in X, but the result is a vector in V .
Namely, for x, y in X, we dene x y to be the unique vector v V such that x = y + v.
An afne transformation of X is a mapping f : X X for which there exists

f GL(V ) such that
f(x) f(y) =

f(x y), x, y X.
Under composition, the set of afne transformations of X forms a group A(X).
Each v V corresponds to the translation t
v
(x) = x + v, which is an afne transformation with

t
v
= I
V
. One can check that f t
v
f
1
= t
f(v)
for all f A(X). Hence V is a normal subgroup of
A(X). Now choose an arbitrary point x
0
X, and let G
x
0
= f A(X) : f(x
0
) = x
0
. I claim
that
A(X) = V

G
x
0
, (44)
103
where (g) = g for g G
x
0
. We have already observed that part 1 of Thm. 9.3 holds. Part 3 holds
because V acts freely on X. For part 2, let f A(X) be any afne transformation. Then f(x
0
)
is some point in X, so we can write it as f(x
0
) = x
0
+ v for a unique v V . Then the afne
transformation g = t
1
v
f xes x
0
, and we have f = t
v
g with g G
x
0
, as required by part 2. This
proves the claim (44). Finally, I claim that we have an isomorphism
G
x
0
GL(V ), g g GL(V ).
For if g G
x
0
then g(x
0
+ v) = x
0
+ g(v), which implies that g g is an injective homomorphism.
Finally, if g
0
GL(V ), then the mapping x
0
+v x
0
+g
0
(v) is an afne transformation, proving the
claim. We have thus shown that
A(X) V GL(V ).
Note, however, that the subgroup of A(X) which is isomorphic to GL(V ) is non-canonical: it de-
pends on the choice of x
0
. If V is nite dimensional and we choose a basis of V and use it to identify
V = F
n

__
1 0
v I
n
_
: v F
n
_
and GL(V ) = GL
n
(F), we can write more explicitly:
A(X)
__
1 0
v g
_
: v F
n
, g GL
n
(F)
_
.
Example 5: Parabolic subgroups
Let V be a nite dimensional vector space over a eld F. Fix a subspace U V . In the group GL(V )
the subgroup
G
U
= g GL(V ) : gU = U
stabilizing U is called a parabolic subgroup. We will show that the parabolic subgroup G
U
is a semidi-
rect product.
Each g G
U
induces a linear transformation on the quotient vector space V/U, namely g (v +U) =
(g v) + U. Let R
U
= g G
U
: g = I
V/U
be the subgroup of G
U
acting trivially on V/U. Note
that g R
U
if and only if gv v U for all v V . One checks that R
U
is normal in G
U
. To
nd the complement, choose a subspace W V complementary to U, so that V = U W, and let
L
U
= g G
U
: gW = W be the subgroup of GL(V ) stabilizing both U and W. Then L
U
acts on
R
U
by conjugation, and we have
G
U
= R
U
L
U
.
Example 6: Frobenius groups
The essence of the proof of above is the following fact:
If a group G acts transitively on a set X and G has a normal subgroup K acting freely and transitively
on X, then G = K

G
x
, where x is any point in X and
g
(k) = gkg
1
for all g G
x
, and k K.
This is a weak result, because the existence of the marvellous subgroup K is a strong hypothesis, which
we would like to avoid. It is easy to check that the nontrivial elements of K are exactly those elements
104
of G which have no xed-points in X. So the essential question is: when do these elements form a
subgroup of G?
A nite group G is called a Frobenius group if G acts transitively on a nite set X and the following
two properties hold:
Every nontrivial element of G xes at most one point in X.
Some nontrivial element of G xes at least one point in X.
Equivalently, every stabilizer G
x
is nontrivial and any two stabilizers G
x
, G
y
intersect trivially.
Theorem 9.4 (Frobenius) If G is a Frobenius group then the set
K = k G : k x ,= x x X 1
is a subgroup of G and for any x X we have G K G
x
.
This only known proof of this theorem uses character theory, which you will learn next semester.
9.2.1 Groups of order p
2
q
Let p and q be distinct primes. In this section our aim is to classify groups G of order p
2
q, where p and
q are distinct primes. We begin with a more general situation.
A pq-group is a group G whose order is of the form p
a
q
b
for some positive integers a, b. Such a group
factors as G = PQ, where P and Q are Sylow p- and q-subgroups of G. Indeed, the sets P Q and
G have the same cardinality, and the product map P Q G, sending (x, y) xy is injective since
P Q = 1.
We now make the additional assumption that some Sylow subgroup is normal in G. If P G then G is
a semidirect product G P Q, with respect to some action of Q on P. It may happen that different
actions of Q on P give isomorphic groups P Q.
Lemma 9.5 If , : Q Aut(P) are two homomorphisms, then we have
P

Q P

Q
if and only if there is g Aut(Q) such that is conjugate to g in Aut(P).
Proof: Let F
0
: P

Q P

Qbe an isomorphism. Since P is the unique Sylowp-subgroup of both

sides, we have F
0
(P) = P. Since F
0
(Q) is another Sylow q-subgroup of P

Q, we may compose
F
0
with an inner automorphism of P

Q to obtain another isomorphism F : P

Q P

Q
105
with the property that F(P) = P and F(Q) = Q. Thus we have restrictions f = F[
P
Aut(P) and
g = F[
Q
Aut(Q).
Let x P and y Q and consider the conjugation yxy
1
in P

Q. On one hand, yxy

1
=
y
(x)
P, so
F(yxy
1
) = f(
y
(x)).
On the other hand, we have
F(yxy
1
) = F(y)F(x)F(y)
1
= g(y)f(x)g(y)
1
=
g(y)
(f(x))
in P

Q. Hence f
y
=
g(y)
f, or
f
y
f
1
= g(y) (45)
so that and g are conjugate in Aut(P), as claimed.
Conversely, if f Aut(P) and g Aut(Q) satisfy (45) then one checks that The map F : P

Q
P

Q given by F(xy) = f(x)g(y), for x P and y Q, is a group isomorphism.

Let Hom(Q, Aut(P)) be the set of all homomorphisms : Q Aut(P). The group Aut(P)
Aut(Q) acts on Hom(Q, Aut(P)) as follows. Given Aut(P), Aut(Q), the transform (, )
of a homomorphism Hom(Q, Aut(P)) is the new homomorphism Q Aut(P) given by
[(, ) ]
y
=

1
(y)

1
, for all y Q.
Here we take
1
to make this a left group action. The Lemma may now be rephrased as follows.
Corollary 9.6 The isomorphism classes of groups of the form G = P Q are in bijection with the
orbits of Aut(P) Aut(Q) on Hom(Q, Aut(P)) under the action just described.
Assume now that Q is cyclic. If we choose a generator y of Q, a nontrivial homomorphism : Q
Aut(P) is determined by the automorphism
y
Aut(P) generating a subgroup Q
y
:=
y

Aut(P) of order dividing [Q[. An automorphism of Q changes y to some power y
j
where q j and
Q
y
j = Q
j
y
= Q
y
. Hence we have
Corollary 9.7 If Qis cyclic , the isomorphism classes of groups of the formG = P Qare in bijection
with the conjugacy-classes in Aut(P) of subgroups of order dividing [Q[.
Nowsuppose [G[ = p
2
q and as above let P, Qbe Sylowpand qsubgroups of Grespectively. Recall
the p-factorization is the expression [G[ = p
2
n
p
, where = [N
G
(P) : P] and n
p
= [G : N
G
(P)] is
the number of Sylow p-subgroups of G. If = 1 then Q G, by the Burnside Transfer Theorem. In
this case the q-factorization is q p
2
1. The possible p and q-factorizations are tabulated below, where
none means no such combination is possible.
106
q p
2
1 q p p q 1 p
2
conditions
p
2
q 1 P Q P Q P Q
p
2
1 q QP none none p [ q 1
conditions q [ p 1 q [ p
2
1
We see that some Sylow subgroup of G is normal. Hence we may apply Cor. 9.6, after possibly
interchanging P and Q. The case G = P Q is equivalent to G being abelian, for which there are two
possible groups: G C
p
2 C
q
or G C
2
p
C
q
according as P C
p
2 or C
2
p
= C
p
C
p
, respectively.
The group G must be abelian unless p [ q 1 or q [ p
2
1. From now on we assume one of these
conditions holds.
Case 1: p [ q1. Then G QP, so we calculate the orbits of Aut(Q)Aut(P) on Hom(P, Aut(Q)).
Since Aut(Q) C
q1
is abelian, these are just the orbits of Aut(P) on Hom(P, C
q1
).
If P = C
p
2 we may apply Cor. 9.7. The groups G in this case correspond to subgroups of C
q1
of
order 1, p or p
2
, the latter occuring only if p
2
[ q 1.
If P = C
2
p
then Aut(P) = GL
2
(p) and C
q1
has a unique subgroup of order p, so the groups G in this
case correspond to GL
2
(p)-orbits in Hom(C
2
p
, C
p
), of which there are two: zero and nonzero.
Combining the two possibilities for P, we see that when p [ q 1 the number of groups of order p
2
q is
ve if q
2
[ p 1 and four otherwise.
Case 2: q [ p
2
1. Here we have G = P Q. Since Q is cyclic, we may apply Cor. 9.7 to see that the
groups of order p
2
q correspond to conjugacy-classes of subgroups of Aut(P) of order 1 or q.
If P = C
p
2 then Aut(P) = Z/p
2
Z

C
p
C
p1
. If q [ p 1 then Aut(P) has exactly one subgroup
of order q, giving two groups in this case, one abelian, one nonabelian. If q p 1 then there is only
the abelian group C
p
2 C
q
.
If P = C
2
p
then Aut(P) = GL
2
(p). This breaks into three subcases.
i) q = 2. In this case p is odd and GL
2
(p) has three conjugacy classes of subgroups of order dividing
two, generated by
_
1 0
0 1
_
,
_
1 0
0 1
_
,
_
1 0
0 1
_
.
The three groups are respectively
C
p
C
p
C
2
, D
p
C
p
, S[D
p
D
p
],
where the latter group is the subgroup of D
p
D
p
generated by C
p
C
p
and an involution inverting
both factors under conjugation.
ii) 2 < q [ p + 1. In this case q p 1 and GL
2
(p) has a unique subgroup of order q, up to conjugacy
(see 3.4.1), whence two groups in this case.
iii) 2 < q [ p 1. In this case we have a subgroup V = C
q
C
q
of the diagonal matrices and all
subgroups of GL
2
(p) of order q can be conjugated into V . Moreover, two such subgroups of V are
107
P = C
p
2 P = C
2
p
condition
1 1 p q 1 and q p
2
1
2 2 p [ q 1 and p
2
q 1
3 2 p
2
[ q 1
2 3 q = 2
1 2 2 < q [ p + 1
2
q+5
2
2 < q [ p 1.
Figure 1: The number of groups of order p
2
q
conjugate if and only if one is transformed into the other by switching the factors in V . Regarding
V additively, the groups G in this case correspond to lines [x, y] in the projectivization of V , modulo
the involution [x, y] [y, x]. This involution xes the lines [1, 1] and [1, 1] and acts freely on the
remaining lines. Counting the abelian case, we get 1 + 2 +
1
2
(q 1) =
1
5
(q + 5) isomorphism classes
of groups in this case.
All cases are summarized in Figure 1. For each condition on p, q we write the number of groups of
order p
2
q in the form N

+ N

, where N

(resp. N

) is the number of groups with the given p, q

condition having P = C
p
2 (resp. P = C
p
C
p
).
9.3 Extensions
The essential problem of extension theory is:
Given two groups A, B, to nd all groups G having A as a normal subgroup with quotient G/A B.
Informally, we are asking how many groups we can build with A at the bottom and B at the top.
For example, if A = C
4
and B = C
2
, then C
2
C
4
, C
8
, D
4
and Q
8
are all the groups G containing a
normal subgroup isomorphic to C
4
with cyclic quotient isomorphic to C
2
.
This is a very difcult problem. It becomes easier if we specify the maps involved.
Denition 9.8 Let A and B be groups. An extension of B by A is a triple (G, , ) where G is a
group, : A G is an injective homomorphism and : G B is a surjective homomorphism with
im = ker. Two extensions (G, , ) and (G

) are equivalent if there exists an isomorphism

f : G

G such that f

= and f =

.
Thus, an extension of B by A is given by an exact sequence
1 A

G

B 1, (46)
and two extensions (G, , ), (G

) of B by Aare equivalent if there is an isomorphismf : G

G
making the following diagram commutative:
108
G

1
G

It is possible for G and G

to be isomorphic for inequivalent extensions (G, , ), (G

) (see exer-
cise...).
We say the extension (G, , ) of B by A is split if there exists a homomorphism s : B G such that
s = id
B
is the identity map on B. The map s, if it exists, is called a section, or a splitting of the
extension.
If (G

) is an extension of B by A equivalent to (G, , ) via an isomorphism f : G

G and
s : B A is a splitting of (G

), then f s is a splitting of (G, , ). Hence the quality of being

split depends only on the equivalence class of the extension.
Proposition 9.9 For an extension (G, , ) of B by A, the following are equivalent.
1. The extension (G, , ) is split.
2. There is a subgroup B

G which is mapped isomorphically onto B via .

3.
25
There is a homomorphism : B Aut(A) such that the extension (G, , ) is equivalent to
(A

B,

), where

(a) = (a, 1) and

(b) = (1, b).

Proof:
(1 2:) Assume (G, , ) is split, and let s : B Gbe a section. Then s is injective, since s = I
B
,
so s is an isomorphism from B onto the subgroup B

= s(B) G, and it is easy to check that maps

B

isomorphically onto B.
(2 1:) Assume there is a subgroup B

G which is mapped isomorphically onto B via . Let

s : B B

be the inverse of the isomorphism [

B
: B

B. It is easy to check that s is a section, so

the extension (G, ) is split.
(3 1:) Suppose (G, , ) is equivalent to (A

B,

) as in 3, via an isomorphismf : A

B

G.
Then s(b) = f(1, b) denes a section of (G, , ).
(1, 2 3:) Assume (G, , ) is split, and let s : B G be a section. Since both and s are injective,
there is a unique homomorphism : B Aut(A) such that by (
b
(a)) = s(b)(a)s(b)
1
. The map
f : A

B

G given by f(a, b) = (a) s(b) is an isomorphism giving the equivalence asserted in
3.
25
I thank Andew Yarmola for suggesting this formulation.
109
There is one situation where an extension is guaranteed to split.
Theorem 9.10 Suppose A and B are nite groups with relatively prime orders. Then any extension of
B by A is split.
Proof: See [Isaacs Finite Group Theory p.79 Theorem 3.8].
The simplest example of a non-split extension is given by
1 C
2
C
4

C
2
1,
where is the squaring map and C
2
is viewed as the subgroup of squares in C
4
. Note that is the
unique surjection C
4
C
2
. To see that this extension is nonsplit, note that C
2
is the unique subgroup
of order two in C
4
. Thus, both groups of order four are extensions of C
2
by C
2
. One of them, C
2
C
2
,
is a split extension, while the other C
4
, is nonsplit.
Something similar happens with non-abelian groups of order eight: both D
4
and Q
8
are extensions of
C
2
by C
4
. The former is split and the latter is nonsplit (see exercise ()).
9.4 Metacyclic groups and extensions
A metacyclic group is a group G having a cyclic normal subgroup A with cyclic quotient B = G/A.
Any cyclic group C is metacyclic. Indeed, we can take A to be any subgroup of C. Then A is cyclic
and so is B = C/A.
The classication of metacyclic groups is simpler if we specify the groups A and B in advance. For
this we use the language of extensions: A metacyclic extension is an extension
1 A

G

B 1. (47)
where A and B are cyclic. In this section we classify metacyclic extensions with G nite. So we x
cyclic groups A C
m
and B C
n
, as well as generators , of A and B, respectively.
Proposition 9.11 Let (G, , ) be a metacyclic extension of B by A. Let a = () and choose an
element b G such that (b) = . Then
1. Every element of G can be written uniquely as as a
i
b
j
, for i Z/mZ and some integer 0 j <
n.
26
2. The group G has the presentation
G a, b [ a
m
= e, bab
1
= a
q
, b
n
= a
r
(48)
for some elements q, r in Z/mZ such that
q
n
= 1, and qr = r. (49)
26
Note that b need not have order n, so we cannot write j Z/nZ.
110
3. Let (G
1
,
1
,
1
) be another extension of B by A, let a
1
=
1
() and choose b
1
G
1
such that

1
(b
1
) = . Let q
1
, r
1
be as in part 2, for a
1
, b
1
. Then the extensions (G
1
,
1
,
1
) and (G, , )
are equivalent if and only if q
1
= q and r
1
= r + (1 + q + + q
n1
)k, for some k Z/mZ.
Proof: For part 1, let x G is an arbitrary element, we have (x) =
j
for some j Z, and also
(b
j
) =
j
, so xb
j
ker = a, which means that x = a
i
b
j
for some i Z/mZ and integer j.
Since (b
n
) = (b)
n
=
n
= 1, we have b
n
ker = a, so b
n
= a
r
, for some r Z/mZ. Hence in
the expression x = a
i
b
j
we may replace j by its remainder when divided by n.
We are given that has order m and has order n. Since is injective this means a has order m.
And since a G we have bab
1
= a
q
for some q (Z/mZ)

. We have proved that G satises the

relations (48). These calculations also show that the group a, b [ a
m
= e, bab
1
= a
q
, b
n
= a
r

has order at most nm, hence is isomorphic to G.

To see that q, r satisfy (49), note that b
n
= a
r
commutes with a. Hence a = b
n
ab
n
= a
(q
n
)
, which
implies q
n
= 1. Also b commutes with b
n
= a
r
, so we have a
r
= ba
r
b
1
= (bab
1
)
r
= a
qr
, so that
r = qr, proving (49).
For part 3, let (G
1
,
1
,
1
) be another extension of B by A, set
1
() = a
1
and choose b
1
G
1
such that

1
(b
1
) = . Applying part 2 to this extension, we get relations analogous to (48) and (49), namely
a
m
1
= e, b
1
a
1
b
1
1
= a
q
1
1
, b
n
1
= a
r
1
1
, (50)
along with
q
n
1
= 1, and q
1
r
1
= r
1
. (51)
Suppose the extensions (G, , ) and (G
1
,
1
,
1
) are equivalent. This means there is an isomorphism
f : G
1
G such that f
1
= and f =
1
. The former relation means that f(a
1
) = a while the
second implies that f(b
1
) is another lift of in G. Hence we have
f(b
1
) = a
k
b,
for some k Z/mZ. And since f(a
1
) = a, we have
a
r
1
= f(a
r
1
1
) = f(b
1
)
n
= (a
k
b)
n
= a
(1+q++q
n1
)k+r
,
where the last equality follows by induction from the relations bab
1
= a
q
and b
n
= a
r
. Hence we
have r
1
= (1 + q + + q
n1
)k in Z/mZ, as claimed.
Finally, we have
a
q
1
= f(a
q
1
1
) = f(b
1
a
1
b
1
1
) = f(b
1
) a f(b
1
)
1
= a
k
b a b
1
a
k
= a
k
a
q
a
k
= a
q
,
so q
1
= q in Z/mZ.
Conversely, if r
1
= (1 + q + + q
n1
)k in Z/mZ then the above calculations show that a and
a
k
b satisfy the relations of a
1
and b
1
, so there is a surjective homomorphism f : G
1
G such that
f(a
1
) = a and f(b
1
) = a
k
b. These equations imply that f
1
= and f =
1
. Finally, f is an
isomorphism since [G[ = mn = [G
1
[.
111

We next prove that such extensions exist, whenever the conditions (49) are satised.
Proposition 9.12 Suppose q, r are elements of Z/mZ satisfying q
n
= 1 and qr = r. Then there exists
an extension
1 A

G

B 1
and an element b G with (b) = , such that
bab
1
= a
q
, and b
n
= a
r
.
Proof: We will construct G as a quotient of a split extension. Let C be a cyclic group of order mn,
choose a surjective homomorphism : C B, and let be a generator of C such that f() = .
Since q
n
= 1, we have q
mn
= 1, so there is a homomorphism
: C (Z/mZ)

= Aut(A), such that () = q.

Form the semidirect product

G = A

C, where we have the relation

1
=
q
. We observe that

n
=
(q
n
)
= and
r

1
=
qr
=
r
. Thus, commutes with
n
and commutes with
r
.
Since

G is generated by and , this implies that the cyclic subgroup H
r
=
r

n
is contained in
the center of

G. In particular, H
r


G, so we can form the quotient group
G
r
:=

G/H
r
.
Let
r
: A G
r
be the projection of A

G to G
r
. This is injective since A H
r
= 1. Set
a =
r
() = H
r
and b = H
r
, both elements of G
r
. Then a and b satisfy the relations
a
m
= 1, bab
1
= a
q
, b
n
= a
r
.
Let :

G B be the composition
:

G
p
C

B,
where p is the natural projection G = AC C. We have (
i

j
) =
j
. In particular, (
r

n
) =

n
= 1. Thus, induces a surjective map
r
: G
r
B such that
r
(a
i
b
j
) =
j
. We see that

r
(b) = and that
r
(A) ker
r
. On the other hand, if
r
(a
i
b
j
) = 1 then n [ j, say j = nk, and we
have a
i
b
j
= a
i
a
rk

r
(A). This shows that ker
r
=
r
(A) and completes the proof that (G
r
,
r
,
r
) is
the desired extension.
The conditions on r and q can be understood more simply if we regard q as the endomorphism
of Z/mZ given by multiplication by q. Likewise we view q 1 and q
n
:= 1 + q + + q
n1
as endomorphisms of Z/mZ. The condition rq = r means that r ker(q 1). The condition
q
n
= 1 means that imq
n
ker(q 1). Part 3 of the proposition means that the equivalence class of
the extension (G, ) depends only on the class of r in ker(q 1)/ imN
q
. In this language, the two
propositions may be then summarized as follows.
112
Theorem 9.13 Fix generators of A C
m
and of B C
n
and let q (Z/mZ)

satisfy q
n
= 1.
Then there is a bijection from the subquotient ker(q 1)/ imq
n
of Z/mZ to the set of equivalence
classes of extensions (G, , ) of B by A such that any lift of in G acts on (A) by the power q. To the
class of r ker(q 1)/ imq
n
corresponds the equivalence class of the extension (G
r
,
r
,
r
) where
G
r
= a, b [ a
m
= e, bab
1
= a
q
, b
n
= a
r
,

r
() = a and
r
(b) = . This extension splits iff the class of r in ker(q 1)/ imq
n
is zero.
We have now classied all metacyclic extensions, and have shown that every metacyclic group is
isomorphic to one of the groups
G(m, n, q, r) = a, b [ a
n
= e, bab
1
= a
q
, b
n
= a
r
,
where q, r Z/m satisfy q
n
= 1 and (q 1)r = 0. However, the same group can appear in different
extensions. For example, if gcd(j, n) = 1 then ker(q
j
1) = ker(q 1) on Z/mZ
G(m, n, q
j
, r) G(m, n, q, r)
via a a, b b
j
. Hence the group G(m, n, q, r) depends only on the subgroup of (Z/mZ)

gener-
ated by q.
113