P Groups

The Theory of p-Groups
David A. Craven
Hilary Term, 2008

Contents
1 Motivation 1
1.1 Soluble Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 The Number of Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Sylow Structure of Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Introduction to the Structure of p-Groups 5

2.1 Commutators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 The Frattini Subgroup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Some Automorphism Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3 Extraspecial Groups 19
3.1 Central Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Alternating Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Extraspecial Groups of Order p1+2n . . . . . . . . . . . . . . . . . . . . . . . 23
4 Maximal Class and p-Rank 28

4.1 Cyclic Subgroups of Index p . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.2 The 2-Groups of Maximal Class . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Groups of p-Rank 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5 Fixed-Point-Free Automorphisms 33
6 The Critical Subgroup Theorem 38
7 How are p-Groups Embedded in Finite Groups? 42
i
Chapter 1
Motivation
1.1 Soluble Groups

A nilpotent group G is a finite group that is the direct product of its Sylow p-subgroups.
Theorem 1.1 (Fitting’s Theorem) Let G be a finite group, and let H and K be two
nilpotent normal subgroups of G. Then HK is nilpotent.
Hence in any finite group there is a unique maximal normal nilpotent subgroup, and every
nilpotent normal subgroup lies inside this; it is called the Fitting subgroup, and denoted by
F(G).
Theorem 1.2 Let G be a finite soluble group. Then
CG (F(G)) ⩽ F(G).
Note firstly that if G is not a soluble group, we may well have F(G) = 1, and even if
F(G) ̸= 1, then we need not have CG (F(G)) ⩽ F(G).
Recall that for any group G and subgroup H, we have that NG (H)/ CG (H) is isomorphic
to a subgroup of Aut(H); since CG (F(G)) is a normal nilpotent subgroup of the finite soluble
group G, this implies that any soluble group has a normal nilpotent subgroup K such that
G/K is a group of automorphisms of K.
Proposition 1.3 Let G be a finite nilpotent group, and let G = P1 × P2 × · · · × Pn , where

the Pi are the Sylow p-subgroups of G. Then
Aut(G) = Aut(P1 ) × Aut(P2 ) × · · · × Aut(Pn ).
This focuses attention on the structure of p-groups and the automorphisms of p-groups.
1
1.2 Lie Algebras
Definition 1.4 A Lie ring is a set R with two binary operations—addition and the Lie
bracket—such that
(i) (R, +) is an abelian group;
(ii) the bracket operation distributes over addition;
(iii) [x, x] = 0 for all x ∈ R; and
(iv) [[x, y], z] + [[y, z], x] + [[z, x], y] = 0 for all x, y, z ∈ R.
If F is a field, and R is an F -vector space, with a[x, y] = [ax, y] then R is a Lie algebra.
To every finite p-group one can associate a Lie ring L(G), and if G/G′ is elementary
abelian then L(G) is actually a Lie algebra over the finite field GF(p).
Proposition 1.5 Let ϕ be an automorphism of the finite p-group G. Then ϕ induces an

automorphism on L(G), and if ϕ has order prime to p, then the induced automorphism has
the same order.
There is a correspondence between a subset of Lie algebras over GF(p), called p-restricted
Lie algebras, and p-groups. Thus studying one is equivalent to studying the other.
1.3 The Number of Groups

Lemma 1.6 Let g(n) denote the number of groups of order n.
(i) g(p) = 1 for p a prime.
(ii) if p < q, then g(pq) = 1 if q ̸≡ 1 mod p, and g(pq) = 2 otherwise.
(iii) g(p2 ) = 2.
(iv) g(p3 ) = 5.
From this we can see that the number of groups of order n depends more on the prime
structure of n then on its size. We can make this explicit with the following table of n against
g(n).
2
n g(n) n g(n) n g(n) n g(n)
1 1 11 1 21 2 31 1
2 1 12 5 22 2 32 51
3 1 13 1 23 1 33 1
4 2 14 2 24 15 34 2
5 1 15 1 25 2 35 1
6 2 16 14 26 2 36 14
7 1 17 1 27 5 37 1
8 5 18 5 28 4 38 2
9 2 19 1 29 1 39 2
10 2 20 5 30 4 40 14
The result g(32) = 51 should make one believe that if one picks a group G of order at
most n at random, then as n tends to infinity, the probability that G is a p-group tends to
1, and even more that G is a non-abelian 2-group with probability 1. This looks true, but
there is still no proof of it yet.
Hence we should be interested in p-groups if only for the fact that almost all groups are
p-groups!
1.4 Sylow Structure of Groups

Theorem 1.7 (Sylow’s Theorem) Let G be a finite group and let pn be the p-part of |G|.
Then G possesses a single G-conjugacy class of subgroups of order pn of length congruent to
1 modulo p, and every p-subgroup is contained in one of them.
This implies that there are always pi -subgroups Pi of largest possible order for the various
primes pi . Let π be a set of primes, and define a π-subgroup in the obvious way; that is, a
π-subgroup is a subgroup whose order is divisible only by primes present in π. If G is a finite
group and n is the π-part of |G|, then a subgroup of order n is called a Hall π-subgroup.
Theorem 1.8 (Philip Hall’s Theorem) Let G be a finite group. Then G is soluble if
and only if, for all sets of primes π, the group G contains a Hall π-subgroup. In this case,
all Hall π-subgroups of are conjugate, and any π-subgroup is contained within one of them.
Thus Sylow’s Theorem is special, in the sense that in an arbitrary group, not only are
we not guaranteed Hall π-subgroups, if the group is insoluble there is guaranteed to be sets
of primes for which they don’t exist. What this means is that the only Hall subgroups that
we are really guaranteed in a finite simple group, for example, are the Sylow p-subgroups.
3
The structure of the Sylow p-subgroups of a finite group place considerable constraints
on the structure of the finite group itself. For example, the following theorem characterizes
all groups with abelian Sylow 2-subgroup.
Theorem 1.9 (Walter) Let G be a group with abelian Sylow 2-subgroups. Then there is
a normal subgroup K and a normal subgroup H with K ⩽ H, such that K has odd order,
H has odd index, and H/K is a direct product of an abelian 2-group and simple groups with
abelian Sylow 2-subgroups, namely:
(i) the projective special linear group PSL2 (q), q ≡ 3, 5 mod 8;
(ii) the projective special linear group PSL2 (2n ) for n ⩾ 3;
(iii) the twisted Dickson group (or Ree group) R(32n+1 ) where n ⩾ 1; and
(iv) the first Janko group J1 .
In the final chapter we will look at how the Sylow p-subgroups of a finite group can be
embedded in it.
4
Chapter 2
Introduction to the Structure of

p-Groups
Definition 2.1 A central series is a chain of normal subgroups
1 = H0 ⩽ Hi ⩽ · · · ⩽ Hr = G
such that Hi /Hi−1 ⩽ Z (G/Hi−1 ) for all 1 ⩽ i ⩽ r.
It can be shown that a finite group is nilpotent if and only if it possesses a central series.
In fact, this is traditionally the definition of a (possibly infinite) nilpotent group.
Definition 2.2 Let G be a group. Define
(i) G(1) = G′ , the derived subgroup, and G(r) = [G(r−1) , G(r−1) ];
(ii) Z1 (G) = Z (G), and Zr (G) by Zr (G) / Zr−1 (G) = Z (G/ Zr−1 (G)), the upper central
series; and
(iii) γ1 (G) = G, γ2 (G) = G′ , and γr (G) = [γr−1 (G), G] the lower central series.
Lemma 2.3 Suppose that
1 = H0 ⩽ Hi ⩽ · · · ⩽ Hr = G
is a central series for G. Then Zi (G) ⩾ Hi and γi (G) ⩽ Hr−i+1 for all i.
This lemma implies that if c is the smallest integer such that Zc (G) = G then γc+1 (G) = 1
and γc (G) ̸= 1, and any central series has length at least c. This integer c is called the
(nilpotence) class of a nilpotent group.
5
Proposition 2.4 Let G be a group of order pn . Then G is nilpotent, and if c denotes its
class, then 0 ⩽ c ⩽ n − 1, c = 0 if and only if G is trivial, and c = 1 if and only if G is
abelian.
Thus there is an easy characterization of p-groups of class 1. However, ‘most’ p-groups are
of class 2, in the sense that as n → ∞, the number of p-groups of class 2 gets unmanageably
large. Notice, however, that there is a largest possible class for each order.
Definition 2.5 Let G be a finite p-group, of order pn . If c denotes the class of G, then the
coclass of G is the quantity n − c.
Having failed completely to classify p-groups by class, we can try to classify them by
coclass. In particular, we ask the question ‘can we classify the groups of coclass 1?’
Definition 2.6 Let G be a finite abelian group. Then G is called elementary abelian if
every non-identity element has order p.
The elementary abelian groups are actually the groups Cp × Cp × · · · × Cp , where Cn is

the cyclic group of order n. If the elementary abelian group P has order pn , then the rank of
P is n. The p-rank of a finite group is the maximum of the ranks of all elementary abelian
p-subgroups.
Having failed completely to describe the p-groups by class, how about trying to classify
them by rank?
Lemma 2.7 Let G be a non-abelian group of order p3 . Then Z (G) has order p, and G/ Z (G)
is elementary abelian.
We define an extraspecial group to be a p-group for which G/ Z (G) is elementary abelian,

and Z (G) has order p. Extraspecial groups appear frequently in representation theory; can
we classify them?
2.1 Commutators
We start with the very basic results in p-group theory.
Definition 2.8 Let x and y be elements of a group G. Then the commutator [x, y] is given
by
[x, y] = x−1 y −1 xy.
The commutator subgroup or derived subgroup is the subgroup
G′ = [G, G] = ⟨ [x, y] : x, y ∈ G⟩.
6
If H and K are subgroups of G, then [H, K] = ⟨ [h, k] : h ∈ H, k ∈ K⟩.
Lemma 2.9 Let G be a group. If ϕ is any homomorphism from G, then for all x, y ∈ G,
[x, y]ϕ = [xϕ, yϕ]. Consequently, G′ is a fully invariant subgroup of G.
Proof: Let ϕ be a homomorphism from the group G. Then for any two elements x and y
of G, we have
([x, y])ϕ = x−1 y −1 xy ϕ = (xϕ)−1 (yϕ)−1 xϕyϕ = [xϕ, yϕ],

and so if ϕ : G → G is an endomorphism, then G′ ϕ ⩽ G′ , as required.
Lemma 2.10 Let G be a group, and H any normal subgroup of G.
(i) The quotient group G/G′ is abelian (the quotient group G/G′ is called the abelianiza-
tion of G);
(ii) If G/H is also abelian then G′ ⩽ H.
Proof: Suppose that x and y are two elements of G. Then [x, y] = g for some element g of
G′ . But then from the definition of [x, y], we have that x−1 y −1 xy = g, whence xy = yxg,
and so since 1 ∈ G′ , xy and yx are in the same coset of G′ . Therefore
(xG′ )(yG′ ) = (xy)G′ = (yx)G′ = (yG′ )(xG′ ),
proving assertion (i).

If G/H is abelian, then for any two elements x and y of G, we have that xy and yx are
in the same coset of H, whence xy = yxh for some element h of H. In a similar fashion to
the proof of (i), we have x−1 y −1 xy = h ∈ H, and so since all generators of G′ lie inside H,
we have G′ ⩽ H, proving (ii).
We will give a notation to extended commutators – by this we mean an object like

[[x, y], z] – which we denote by [x, y, z]. We extend this notation by induction, so that

[x1 , x2 , x3 , . . . , xn−1 , xn ] = [x1 , x2 , . . . , xn−1 ], xn .
Lemma 2.11 Let G be a group, and x, y, z be elements of G.
(i) [xy, z] = [x, z]y [y, z] = [x, z][x, z, y][y, z].
(ii) [x, yz] = [x, z][x, y]z = [x, z][x, y][x, y, z].
(iii) [x, y] = [y, x]−1 .
7
(iv) (Hall–Witt’s Identity) [x, y −1 , z]y [y, z −1 , x]z [z, x−1 , y]x = 1.
Proof: To prove (i), notice that
[xy, z] = y −1 x−1 z −1 x(zyy −1 z −1 )yz = y −1 (x−1 z −1 xz)yy −1 z −1 yz = [x, z]y [y, z],
and
[x, z][x, z, y] = x−1 z −1 xz[x−1 z −1 xz, y] = (x−1 z −1 xzz −1 x−1 zx)y −1 x−1 z −1 xzy = y −1 x−1 z −1 xzy = [x, z]y ,
giving [x, z][x, z, y][y, z] = [x, z]y [y, z]. The proof of (ii) is similar.
The proof of (iii) is obvious: [y, x]−1 = (y −1 x−1 yx)−1 = x−1 y −1 xy = [x, y].
To prove Witt’s Identity is harder: let u = xzx−1 yx, v = yxy −1 zy and w = zyz −1 xz.
Then
[x, y −1 , z]y = [x−1 yxy −1 , z]y = y −1 (yx−1 y −1 xz −1 x−1 yxy −1 z)y = (x−1 y −1 xz −1 x−1 )(yxy −1 zy) = u−1 v,
and similarly [y, z −1 , x]z = v −1 w and [z, x−1 , y]x = w−1 u, giving
[x, y −1 , z]y [y, z −1 , x]z [z, x−1 , y]x = u−1 vv −1 ww−1 u = 1,
as required.
The Hall–Witt Identity has the following consequence, which is often of use in group
theory.
Theorem 2.12 (Three Subgroup Lemma) Let X, Y and Z be three subgroups of a

group G, and let N be a normal subgroup of G. If [X, Y, Z] and [Y, Z, X] are both contained
within N , then so is [Z, X, Y ].
Proof: Let x ∈ X, y ∈ Y and z ∈ Z. Since [X, Y, Z] and [Y, Z, X] are both contained in N ,
then [x, y −1 , z]y and [y, z −1 , x]z are elements of N (since N is normal), and so
−1
[x, y −1 , z]y [v, z −1 , x]z = [z, x−1 , y]x ∈ N,
by Witt’s Identity. Since N is normal, we can conjugate by x−1 to get [z, x−1 , y] ∈ N . But
by writing x′ = x−1 , we have [z, x′ , y] ∈ N for all z ∈ Z, x′ ∈ X and y ∈ Y . Since [Z, X, Y ]
is generated by such elements, [Z, X, Y ] ⩽ N .
Proposition 2.13 Let H and K be subgroups of a group G.
(i) [H, K] ⩽ K if and only if H ⩽ NG (K).
(ii) If H and K normalize one another then [H, K] P HK, and [H, K] ⩽ H ∩ K.
8
(iii) If K is a normal subgroup of G and H ⩾ K then [H, G] ⩽ K if and only if H/K ⩽
Z (G/K).
Proof: Suppose that H ⩽ NG (K). Then h−1 k −1 h = k ′ ∈ K and so h−1 k −1 hk = k ′ k ∈ K.

Thus [H, K] ⩽ K. Conversely, suppose that [H, K] ⩽ K, but that H ̸⩽ NG (K). Then
there exist h and k such that h−1 k −1 h ∈
/ K. Then certainly h−1 k −1 hk ∈
/ K, contradicting
[H, K] ⩽ K.
Now suppose that H and K are mutually normalizing subgroups. Then for h ∈ H, k ∈ K
and [h′ , k ′ ] ∈ [H, K], we have that h 7→ hg is an endomorphism of G and so by Lemma 2.9,
[h′ , k ′ ]hk = [(h′ )hk , (k ′ )hk ] ∈ [H, K],
and thus [H, K] P HK. By part (i), [H, K] ⩽ H and [H, K] ⩽ K, yielding the second
assertion of (ii).
Finally, suppose that [H, G] ⩽ K. Then for all g ∈ G and h ∈ H, h−1 g −1 hg ∈ K; i.e.,
hgK = ghK, or that h ∈ Z (G/K). Conversely, if hK ∈ Z (G/K), then for all g ∈ G,
ghK = hgK, and so [h, g] ∈ K. This means that [h, G] ⊆ K, and thus if this is true for all
h ∈ H then [H, G] ⩽ K as required.
Corollary 2.14 Suppose that H, K and L are normal subgroups of G. Then
[HK, L] = [H, L][K, L], [L, HK] = [L, H][L, K].
Theorem 2.15 (Fitting’s Theorem) Suppose that G is a group, and H and K are nilpo-
tent normal subgroups, of classes c and d respectively. Then HK is a nilpotent subgroup of
G, of class at most c + d.
Proof: Consider the commutator
X = [HK, HK, . . . , HK],
where there are c + d + 1 terms. By Corollary 2.14, this can be written as

Y
X= [M1 , M2 , . . . , Mc+d+1 ].
Mi ∈{H,K}
Let A = [M1 , M2 , . . . , Mc+d+1 ] be one of these multiplicands. Notice that if a of the Mi are
equal to H, then A ⩽ γa+1 (H). Since there are c + d + 1 of the Mi , either c + 1 of them are
H or d + 1 of them are K. Either way, since γc+1 (H) = γd+1 (K) = 1, we have that A = 1,
and so X = 1, proving that γc+d+1 (HK) = 1, as required.
9
2.2 The Frattini Subgroup
Definition 2.16 Let G be a finite group. The Frattini subgroup is the intersection of all
maximal subgroups of G. It is denoted by Φ(G).
This subgroup has the curious property that it contains all of the elements of G that do
not generate G, in a sense that will be made precise now.
Definition 2.17 Let G be a group, and x ∈ G. Then x is said to be a non-generator if

whenever G is generated by x and a set X, then G = ⟨X⟩.
What this says is that you can remove all non-generators from a generating set and still
generate the group. We now give the strange-looking result.
Proposition 2.18 Let G be a group. Then Φ(G) is the set of all non-generators of G.
Consequently, if G = H Φ(G) for some H, then H = G.
Proof: Suppose that x is a non-generator of G, and let M be a maximal subgroup of G.

Then ⟨M, x⟩ ⩾ M , and so, since M is a maximal subgroup of G, ⟨M, x⟩ = M or ⟨M, x⟩ = G.
If ⟨M, x⟩ = G, then, since x is a non-generator, ⟨M ⟩ = G, a clear contradiction. Thus
⟨M, x⟩ = M ; i.e., x ∈ M . This is true for all maximal subgroups, and so x ∈ Φ(G).
Conversely, suppose that x is an element of Φ(G). Let G be generated by some set X,
together with x, so G = ⟨X, x⟩. Denote by N the group ⟨X⟩. If N ̸= G, it is contained
in a maximal subgroup, say M . But x ∈ M , since x is an element of Φ(G) ⩽ M , and so
⟨X, x⟩ ⩽ M < G, a contradiction. Thus ⟨X⟩ = G for all sets X for which ⟨X, x⟩ = G; i.e.,
x is a non-generator, as required.
Finally, notice that G = H Φ(G) = ⟨H, Φ(G)⟩, so G = ⟨H⟩ = H, as needed.
There is an interplay between the Fitting and Frattini subgroups, which we will examine
briefly now. The following result has several interesting corollaries.
Theorem 2.19 Let G be a finite group, and suppose that N is a normal subgroup of G
containing Φ(G). If N/ Φ(G) is nilpotent, then N is nilpotent.
Proof: A finite group is nilpotent if and only if its Sylow p-subgroups are normal. We will
show that the Sylow p-subgroups of N are normal, thus proving that N is nilpotent. Let P
be a Sylow p-subgroup of N . Then P Φ(G)/ Φ(G) is a Sylow p-subgroup of G/ Φ(G), since
if |P | = pd and | Φ(G)| = pe a, where p ∤ a, then
P Φ(G)
= pd−e ,
Φ(G)
10
which is the power of p dividing N/ Φ(G). Now N/ Φ(G) is nilpotent, so P Φ(G)/ Φ(G),
being a Sylow p-subgroup of N/ Φ(G), is normal. If a Sylow p-subgroup of G is normal, it is
also characteristic; thus P Φ(G)/ Φ(G) char N/ Φ(G), and so
P Φ(G) char N P G,
showing that P Φ(G) P G.

Now we can use the Frattini Argument: P is a Sylow p-subgroup of N , so is certainly a
Sylow p-subgroup of P Φ(G), and therefore
G = NG (P )P Φ(G).
Now P ⩽ NG (P ), so NG (P )P = NG (P ). Thus G = NG (P ) Φ(G). Now Proposition 2.18

proves that G = NG (P ), so P P G, which clearly implies that P P N , as required.
This theorem gives us several important corollaries, so in this sense it is very useful. The
first is a result of Frattini.
Corollary 2.20 The Frattini subgroup of a finite group is nilpotent.
Proof: Take N = Φ(G) in Theorem 2.19.
Corollary 2.21 Let G be a finite group. Then Φ(G) ⩽ F(G).
Proof: Φ(G) is a normal nilpotent subgroup of G.
Corollary 2.22 If G is a finite group and G/ Φ(G) is nilpotent, then G is nilpotent.
Proof: Take N = G in Theorem 2.19.
Corollary 2.23 If G is a finite group, then F(G/ Φ(G)) = F(G)/ Φ(G).
Proof: Any normal nilpotent subgroup N/ Φ(G) of G/ Φ(G) lifts to a nilpotent subgroup
N of G, and so N ⩽ F(G), showing F(G/ Φ(G)) ⩽ F(G)/ Φ(G). Certainly F(G)/ Φ(G) is
nilpotent, so F(G)/ Φ(G) ⩽ F(G/ Φ(G)), and we have the result.
So Theorem 2.19 is indeed very useful.
Proposition 2.24 Let G be a finite p-group. Then G/ Φ(G) is elementary abelian, and if
H is another normal subgroup of G such that G/H is elementary abelian, then Φ(G) ⩽ H.
11
Proof: Firstly notice that every maximal subgroup of a p-group is normal, and of index p.
This means that if M is a maximal subgroup of G, then G/M is cyclic of order p. Hence
G′ ⩽ M for all maximal subgroups M ; consequently G′ ⩽ Φ(G), and so G/ Φ(G) is abelian.
Also, since G/M has order p (for M a maximal subgroup of G), we know that (M x)p = M
for all x ∈ G; i.e., xp ∈ M for all x ∈ G and all maximal subgroups M . Thus xp ∈ Φ(G),
and so if Φ(G)x ∈ G/ Φ(G), then Φ(G)x has order p, proving that G/ Φ(G) is elementary
abelian.
Now suppose that G/H is elementary abelian of order pn . Then G/H is generated by n
cosets Hxi of G/H, each of order p. We know then that
G/H ∼
= ⟨Hx1 ⟩ × · · · × ⟨Hxn ⟩.
Now, this group has n maximal subgroups, Hi /H, each generated by {Hxj : j ̸= i}. Since
this is a direct product, the intersection satisfies
\
Hj /H = 1.
1⩽j⩽n
This means that the intersection of all Hj is H (where Hj is the corresponding subgroup in
G to Hj /H, the preimage of Hj /H). But the Hj are maximal subgroups of G/H, and hence
of G. This clearly implies that their intersection contains Φ(G): hence
\
H= Hj ⩾ Φ(G),
1⩽j⩽n
as we wanted.
This has the following consequence.
Proposition 2.25 Let Gp denote the group generated by the set {g p : g ∈ G}; i.e., the
smallest group containing all elements of order p. Then Φ(G) = G′ Gp .
Proof: Since Φ(G) contains all xp , as we saw in the proof of Proposition 2.24, Gp ⩽ Φ(G).
Also, G′ ⩽ Φ(G) since G/ Φ(G) is abelian: thus G′ Gp ⩽ Φ(G).
To prove the converse, notice that G/G′ Gp is elementary abelian: it is abelian certainly,
since G′ ⩽ G′ Gp . Also, xp ∈ Gp ⩽ G′ Gp for all x ∈ G, and so every element of G/G′ Gp has
order either 1 or p. Thus G/G′ Gp is elementary abelian, and so G′ Gp ⩽ Φ(G) by Proposition
2.24.
The subgroup Gp is important enough to be given a special notation. In fact, we can

generalize these notions.
12
Definition 2.26 Let G be a finite p-group. Then the subgroup Ωi (G) is the subgroup
generated by all elements of order dividing pi ; that is,
i
Ωi (G) = ⟨ g : g p = 1⟩.
i
The subgroup ℧i (G) is the subgroup generated by all elements of the form g p ; that is,
i
℧i (G) = ⟨ g p : g ∈ G⟩.
The Proposition 2.25 can be written as Φ(G) = G′ ℧1 (G). In fact, the subgroups Ωi (G)
and ℧i (G) are all characteristic in G: this is true since the elements by which they are
generated are left fixed by any automorphism of G.
Quickly, we notice the following lemma.
Lemma 2.27 Let G be a finite group with p dividing |G|. Then Ω1 (G) ̸= 1.
Proof: Obvious from Cauchy’s Theorem.
This seems a rather obvious lemma, but it can come in very handy.
Now we prove two important, yet not difficult, results on finite p-groups. These tell us
that the Frattini subgroup is even more interesting than we had previously thought.
Theorem 2.28 (Burnside Basis Theorem) Let G be a finite p-group, and suppose that
|G : Φ(G)| = pd . If G/ Φ(G) is generated by elements Φ(G)xi , for 1 ⩽ i ⩽ d, then G is
generated by the xi . Furthermore, any generating set of G contains a subset Y such that
G = ⟨Y ⟩ and G/ Φ(G) is generated by the images of the elements of Y .
Proof: Suppose that ⟨x1 , . . . , xd ⟩ is not the whole of G, say it is H. Since we are in a finite
p-group, we have maximal subgroups, and so H is contained within some maximal subgroup
M . Now Φ(G) ⩽ M , and so
⟨Φ(G)x1 , . . . , Φ(G)xd ⟩ ⩽ M/ Φ(G) < G/ Φ(G).
This contradicts the fact that the cosets Φ(G)xi generate G/ Φ(G), so
G = ⟨x1 , . . . , xd ⟩.
Now suppose that X is any generating set. Then the images of X under the quotient
G → G/ Φ(G) must generate G/ Φ(G). Now we can pick a subset {y1 , . . . , yd } of d elements
of X such that
G/ Φ(G) = ⟨Φ(G)y1 , . . . , Φ(G)yd ⟩,
and thus G = ⟨y1 , . . . , yd ⟩.
13
The Burnside Basis Theorem tells us that G can be generated with d elements, and this
is also the smallest number of elements you can generate G with. We are led to the following
definition.
Definition 2.29 A group is said to be d-generator if G = ⟨X⟩ for some subset X ⊆ G of

size d. A group is said to be finitely generated if G is generated by a finite subset X of G.
Then the only 1-generator groups are cyclic, for example. Dihedral groups, symmetric
and alternating groups, the quaternion group, and direct products of two cyclic groups are
2-generator. Q is an example of a group that is not finitely generated.
The second of the promised results is the Hall–Burnside Theorem. This deals with
automorphisms of p-groups. Notice that if H is a characteristic subgroup of G, then any
automorphism ϕ induces an automorphism of G/H, by permuting the cosets as
Hx 7→ H(xϕ).
Theorem 2.30 (Hall–Burnside Theorem) Let G be a finite p-group, and A be a sub-

group of Aut(G), with p ∤ |A|. If every element of A acts as the identity on G/ Φ(G), then
A = 1.
Proof: Suppose that q is a prime dividing |A|, and let ϕ be an element of A of order q.
Now ϕ acts as the identity on G/ Φ(G), so ϕ acts on each coset of Φ(G). The orbits of ϕ
are each of length either 1 or q, which tells us that Φ(G)x contains a fixed point; i.e., there
is an element in each coset of Φ(G) which is left fixed by ϕ. Now take their images under
the quotient map. This is clearly a generating set of G/ Φ(G), since it is the whole quotient
group! We can now choose a basis of G/ Φ(G), and apply the Burnside Basis Theorem, to
get that G is generated by elements that are fixed by ϕ. Hence ϕ = 1, and we are done.
2.3 Some Automorphism Groups

Proposition 2.31 Let G denote the elementary abelian group of order pn . Then Aut(G) ∼
=
GLn (p), the group of n × n matrices over GF(p).
Proof: Notice that there are pn − 1 elements of order p in G. Suppose that G is generated
by x1 , . . . , xn . An automorphism of a finite group is determined uniquely by its action on
the generators of a group, so if we know where to send the xi , we have nailed down our
automorphism. Write ϕ for this automorphism.
We also need to see that any element of G can be expressed as a product
Yn
xbi i ,
i=1
14
where the bi are unique integers between 0 and p − 1. Then G cannot be generated by fewer
than n elements, so we cannot ‘waste’ a generator by mapping it to somewhere that we can
already express in terms of the other generators. Borrowing a term from linear algebra, we
would like the images of the generators to be ‘linearly independent’.
Notice that x1 can be sent to any element of order p, so there are pn − 1 choices for x1 ϕ.
Now we have to decide what to do with x2 ; we cannot send it into ⟨x1 ⟩, since we would then
be wasting a generator, and so there are pn − p choices for x2 ϕ. Then ⟨x1 , x2 ⟩ has order p2 ,
and so there are pn − p2 choices for x3 ϕ, and so on, until we get
| Aut(G)| = (pn − 1)(pn − p) . . . (pn − pn−1 ),
which is the order of GLn (p). So if we can find a homomorphism from Aut(G) to GLn (p),
and show that it is injective, we will be done.
Using the fact that any element of G can be expressed as a multiple of the basis elements,
(ϕ)
we proceed to write down a matrix for ϕ: let Aϕ = (ai,j ), where
n
X (ϕ)
xj ϕ = ai,j xi .
i=1
So Aϕ is uniquely determined. Then
(x1 , x2 , . . . , xn )Aϕ = (x1 ϕ, x2 ϕ, . . . , xn ϕ).

(ϕ)
The function Φ : Aut(G) → GLn (p) given by ϕ 7→ Aϕ is injective, since the coefficients ai,j
are uniquely determined. We must show that it is a homomorphism. If ϕ and ψ are two
elements of Aut(G), then
n
!
X (ϕ)
(xi )(ϕΦ)(ψΦ) = ai,j xi ψ
i=1
n X
X n
(ϕ) (ψ)
= ai,j ai,j xk
i=1 k=1
n n
!
X X (ϕ) (ψ)
= aai,j ai,j xk
k=1 i=1
n
X (ϕψ)
= ai,j xk
k=1
= (xi )(ϕψ)Φ,
so (ϕψ)Φ = (ϕΦ)(ψΦ) as required. Thus Aut(G) ∼

= GLn (p).
Proposition 2.32 Let G denote the cyclic group of order n. Then Aut(G) is abelian, and
has order ϕ(n), where ϕ denotes Euler’s ϕ-function.
15
Proof: Let G = ⟨x⟩. Then an automorphism of G must send x to another generator of G,
which obviously must have order n, and so it reduces to finding out how many elements of
Cn have order n. If n and m are coprime, with 1 ⩽ m ⩽ n, then the first integer k for which
xmk = 1 is k = n. Hence, if m and n are coprime, then xm has order n. Conversely, let d
denote gcd(m, n), and suppose that xm has order n. Then since (xm )n/d = 1 (since mn/d is
divisible by n, n ⩽ n/d; this clearly implies that d = 1, and so m and n are coprime.
We have proved that xm has order n if and only if m and n are coprime, and hence
| Aut(G)| = ϕ(n), since Euler’s ϕ-function is simply the amount of numbers m ⩽ n that are
coprime to n.
To see that Aut(G) is abelian, notice that all automorphisms are of the form x 7→ xm ; if
ϕ : x 7→ xm and ψ : x 7→ xk are two automorphisms, then
x(ϕψ) = (xϕ)ψ = xm ψ = xmk = xkm = xk ϕ = x(ψϕ),
and so Aut(G) is abelian.
We have the following improvement to the previous proposition in the case where the
cyclic group is of prime order.
Proposition 2.33 Let G ∼

= Cp = ⟨x⟩. Then Aut(G) is cyclic of order p − 1.
Proof: We already know that Aut(G) is abelian of order p − 1 (since every number less than
p is coprime to p), so we simply need to show that Aut(G) is cyclic. To see this, we will
notice that Aut(G) is the same as multiplying the non-zero integers modulo p. Then since
the integers modulo a prime form a field, Aut(G) is cyclic.
Consider two automorphisms ϕm : x 7→ xm and ϕk : x 7→ xk , where m and k lie between
1 and p − 1. Then ϕm ϕk is given by
ϕmk : x 7→ xmk ,
so we get a homomorphism from Aut(G) to the multiplicative group of the integers modulo
p by
Φ : Aut(G) → (Z/pZ)⋆ , Φ : ϕm 7→ m.
[Here, F ⋆ = F \ {0} denotes the multiplicative subgroup of F .] Hence Aut(G) is cyclic of

order p − 1, as required.
Before we analyze the structure of Aut(G) further, note that ϕ(pn ) = pn−1 (p − 1).
Proposition 2.34 Let G = ⟨x⟩ be a cyclic 2-group, and write A = Aut(G).
16
(i) If G = C4 , then A = C2 , and is generated by x 7→ x−1 .
(ii) If G = C2n where n ⩾ 3, then A = C2n−2 × C2 , and is generated by ϕ : x 7→ x−1 and

ψ : x 7→ x5 .
Proof: The proof of (i) is obvious, and so we examine (ii). Note that
n−2 n−2 i
52 = (1 + 4)2 ≡1 mod 2n , 52 ̸≡ ±1 mod 2n ,
j
if 0 < i < n − 2. Hence the automorphism x 7→ x5 has order 2n−1 . Since 52 is not congruent
to −1 modulo 2n either, there is no power of ϕ that is equal to ψ, whence they form a
generating set for A, as ⟨ϕ, ψ⟩ has then correct order.
In particular, notice that there are exactly three subgroups of Aut(C2n ) of order 2.
Proposition 2.35 Let G be a cyclic p-group with p odd, and write A = Aut(G). Then A
is cyclic.
Proof: Since |A| = pn−1 (p − 1), where |G| = pn , if we can prove that A contains an element
of order p − 1 and the Sylow p-subgroup of A is cyclic, then we are done.
We firstly claim that the automorphism ϕ : x 7→ xp+1 is a generator for the Sylow
p-subgroup of A. To see this, notice that
i i
(1 + p)p ≡ 1 mod pi+1 , (1 + p)p ̸≡ 1 mod pi+2 ,
and hence ϕ has order pn−1 , as we need.

To prove that A contains an element of order p − 1, recall that this is true if n = 1. Next,
we have a surjective homomorphism A → Aut(Cp ) given by the following: if x is a generator
for G and y is a generator for Cp , then the function
(x 7→ xa ) 7→ (y 7→ y a mod p )
is a homomorphism, and clearly is surjective. Hence A contains an element of order a multiple

of p − 1 to map onto a generator of Aut(Cp ), and hence has an element of order p − 1.
In particular, notice that there is exactly one subgroup of order p, generated by x 7→

pn−1 +1
x .
With the information in Propositions 2.34 and 2.35 we can determine Aut(G) for any
cyclic group G.
17
Corollary 2.36 Let G be a cyclic group, and write |G| = 2n pn1 1 . . . pnr r . For n = 0 or n = 1,
we have !
r
Y
Aut(G) = Cpni −1 (pi −1) .
i
i=1
If n ⩾ 2, then we have
r
!
Y
Aut(G) = C2 × C2n−2 × Cpni −1 (pi −1) .
i
i=1
18
Chapter 3
Extraspecial Groups
Definition 3.1 Let G be a finite p-group. Then G is defined to be special if either G is

elementary abelian or G is of class 2 and G′ = Φ(G) = Z (G) is elementary abelian. If G is
a non-abelian special group with | Z (G) | = p, then G is said to be extraspecial.
Example 3.2 The dihedral and quaternion groups D8 and Q8 are extraspecial. More gen-
erally, if G is a non-abelian group of order p3 then G is extraspecial.
We now give some examples of p-groups that we will use in this chapter and the next.
We will actually give the definition of more groups than we need for this chapter, because it
is sometimes useful to have them all in one place.
Definition 3.3 The dihedral group D2n is given by the generators and relations
D2n = ⟨ x, y : xn = y 2 = 1, xy = x−1 ⟩ = ⟨ a, b : a2 = b2 = 1, (ab)n = 1⟩.
The quaternion group Q4n is given by
Q4n = ⟨ x, y : x2n = y 4 = 1, xy = x−1 , y 2 = xn ⟩
The semidihedral or quasidihedral group SD2n is given by

n−1 n−2 −1
SD2n = ⟨ x, y : x2 = y 2 = 1, xy = x2 ⟩.
The modular p-group Modn (p) is given by the generators and relations
n−1 n−2
Modn (p) = ⟨ x, y : xp = y p = 1, xy = x1+p ⟩.
Define
p1+2
+ = ⟨ x, y, z : xp = y p = z p = 1, [x, z] = [y, z] = 1, [x, y] = z⟩.
19
In the case where n = 3, we also write Mod3 (p) = p1+2
− . We will not prove that these
groups are well-defined or unique up to isomorphism. The groups D2n are of order 2n, and
are split extensions of Cn by C2 . The groups SD2n are of order 2n , and are split extensions
of C2n−1 by C2 . The groups Q4n are non-split extensions of C2n by C2 . The groups Modn (p)
have order pn and are a split extension of Cpn−1 by Cp . Finally, the group p1+2
+ has order p3 .
Notice that the groups D2n , Q2n , SD2n and Modpn all have a cyclic subgroup of index p.
Lemma 3.4 There are five non-isomorphic groups of order p3 , namely the three abelian
groups, D8 and Q8 if p = 2 and p1+2
± when p is odd.
Proof: The three abelian groups are obvious, and so we only need to consider the non-
abelian groups. Firstly, let p = 2, and let G be a non-abelian group of order 8. Since any
group of exponent 2 is abelian, G contains an element x of order 4. If G \ ⟨x⟩ contains an
element of order 2, then G is a split extension of C4 by C2 , and since G is non-abelian, the
homomorphism C2 → Aut(C4 ) = C2 is non-trivial, and hence an isomorphism. Thus there
is exactly one non-abelain split extension of C4 by C2 , and this is D8 .
Thus suppose that G \ ⟨x⟩ contains only elements of order 4, and let y be one of these.
It is clear that G possesses a single element of order 2, and so x2 = y 2 . Also, since ⟨x⟩ has
index 2 and is hence normal in G, xy is either x or x−1 . If xy = x−1 then G = Q8 , as defined
above. If xy = x then x and y commute and so G is abelian.
Now let p be odd, and suppose that G contains an element x of order p2 , and write
X = ⟨x⟩. Since G is non-abelian, CG (X) = X, and so if y ∈ G \ X, then y induces a
non-trivial automorphism of X, and as y p ∈ X, this automorphism has order p. By Lemma
2.35, Aut(X) possesses a unique subgroup of order p, generated by x 7→ x1+p , so that y can
be chosen so that xy = xp x = xz for z = xp of order p, lying in Z (G). Since y p ∈ X, we
must have y p = xp α for some α; write g = yx−α . Then
g p = (yx−α )p = y p x−αp = 1,
(by the fact that G is of class 2 and G/ Z (G) is elementary abelian) so that the extension
splits, and we see that G ∼
= Mod3 (p).
Now let G be a group of exponent p, and choose a subgroup of index p. Let x be a
non-central element of this subgroup, and y be an element not in this subgroup. Then
[x, y] = z ∈ Z (G), and so [x, z] = [y, z] = 1, yielding the presentation of p1+2
+ .
This has dealt with the groups of order p3 . To deal with the larger extraspecial groups,
we need central products and alternating forms.
20
3.1 Central Products
This section will outline the construction of a central product. Let G and H be groups with
isomorphic centres Z. Then we aim to construct a group with centre Z and a quotient of
G × H.
Proposition 3.5 Let G be a finite group, and let G1 , . . . , Gr be subgroups of G. The

following are equivalent:
(i) G = ⟨Gi : 1 ⩽ i ⩽ r⟩ and [Gi , Gj ] = 1 for i ̸= j; and
(ii) the map ϕ given by
ϕ : G1 × G2 × · · · Gr → G, ϕ(x1 , x2 , . . . , xr ) 7→ x1 x2 . . . , xr
is a surjective homomorphism and, if Hi denotes the subgroup of the domain of ϕ

consisting of all elements in the ith co-ordinate, then Hi ϕ = Gi , and Gi ∩ ker ϕ = 1.
Proof: Suppose that G is generated by the Gi , and [Gi , Gj ] = 1. Clearly, since G is

generated by the Gi , the map ϕ is surjective, and it is a homomorphism since Gi and Gj
commute. Again, certainly Hi ϕ ⩽ Gi , and since G is a finite group, this map is a bijection.
Finally, suppose that Gi ∩ ker ϕ ̸= 1, and let 1 ̸= x ∈ ker ϕ. Again, we see that Gi ∩ ker ϕ = 1
trivially.
Conversely, suppose that (ii) holds. Since ϕ is a surjective homomorphism, every element
of G is of the form g = x1 x2 . . . xr , and so G is generated by the Gi . Now suppose that
gi ∈ Gi and gj ∈ Gj , and consider [gi , gj ]. Let hi and hj be the preimage of gi and gj in Hi
and Hj ; since Gi ∩ ker ϕ = 1, this element is uniquely determined. Then [hi , hj ] = 1, and so
[gi , gj ] = [hi , hj ]ϕ = 1,
as required.
If a group G satisfies either (and hence both) of the conditions in the proposition above,
then G is said to be a central product of the groups Gi .
Theorem 3.6 Let {Gi : 1 ⩽ i ⩽ r} be a family of subgroups such that Z (Gi ) = Z (Gj ),
and such that
Aut(Z (Gi )) = AutAut(Gi ) (Z (Gi ))
for all i. Then there exists up to isomorphism a unique group G that is the central product
of the Gi , such that, if Hi denotes the subgroup isomorphic with Gi lying in G, then Z (Hi ) =
Z (Hj ) for all i and j.
We will not prove this theorem here.
21
3.2 Alternating Forms
Let F be a field, and let V be a vector space over F . Then a form ϕ is simply a map
ϕ : V × V → F ; we often require that ϕ is bilinear, so that
ϕ(ax1 + x2 , by1 + y2 ) = abϕ(x1 , y1 ) + bϕ(x2 , y1 ) + aϕ(x1 , y2 ) + ϕ(x2 , y2 ),
for xi , yj ∈ V and a, b ∈ F .
Definition 3.7 Suppose that ϕ is a bilienar form on the vector space V , which is over a
field F .
(i) If ϕ(x, y) = ϕ(y, x), then ϕ is said to be symmetric.
(ii) If ϕ(x, y) = −ϕ(y, x), then ϕ is said to be skew-symmetric.
(iii) If ϕ(x, x) = 0, then ϕ is said to be alternating.
If a bilinear form is alternating then it is skew-symmetric, and if char F ̸= 2 then a

skew-symmetric bilinear form is alternating.
Definition 3.8 Suppose that V is a vector space over F , and that ϕ is a bilinear form.
If two vectors v and w have the property that ϕ(v, w) = 0, then v and w are said to be
orthogonal, and it is written v ⊥ w. Write v ⊥ for the set of all vectors w such that v ⊥ w.
A vector v, is called singular if v ⊥ = V , and V is singular and non-singular according as V
contains a singular vector or not.
Theorem 3.9 Let V be a vector space over a field F , supporting a non-singular alternating
form. Then dimF (V ) = 2n is even, and there exists a basis u1 , v1 , u2 , v2 , . . . , un , vn , such that
(i) (ui , uj ) = (vi , vj ) = 0, and
(ii) (ui , vj ) = δij .
This theorem requires a few lemmas from which it will become clear. If U is a subspace
of the vector space V with form ϕ, then write U ⊥ for the set
\
u⊥ .
u∈U
Lemma 3.10 Suppose that v ∈ V , where V is a vector space with form ϕ. Write θ : y 7→
ϕ(y, v). Then v ⊥ = ker θ, and dim v ⊥ ⩾ n − 1, with equality exactly when v ∈ V ⊥ .
This will be used to prove the next lemma in the chain.
22
Lemma 3.11 Let V be a non-singular vector space, and let U be a subspace of V . Then
dim U ⊥ = codim U .
Proof: We proceed by induction on m = dim U , where n = dim V . We can assume that U

is non-trivial, so let x ∈ V \ U ⊥ , which exists since ϕ is non-singular. The space X = U ∩ x⊥
is of dimension m − 1, whence by induction we have
dimF X ⊥ = n − m + 1.
Suppose that v is an element of U \ X; then U ⊥ = X ⊥ ∩ v ⊥ , and since x ∈ X ⊥ \ v ⊥ , the

space U ⊥ is of codimension 1 in X ⊥ , yielding dim U ⊥ = n − m, as required.
Lemma 3.12 Let U be a subspace of the vector space V , and suppose that ϕ is non-singular.
Then ϕ|U is non-singular if and only if U = U ⊕ U ⊥ .
Proof: This follows easily from Lemma 3.11.
Now let V be a vector space with an alternating bilinear form, and proceed by induction
on dimF V ; choose the non-trivial element u1 arbitrarily. Since f is non-singular, V contains
a vector w1 such that ϕ(u1 , w1 ) ̸= 0, so let v1 = (ϕ(u1 , w1 ))−1 w1 . Then ϕ(u1 , v1 ) = 1,
and V1 = ⟨u1 , v1 ⟩ is a non-singular subspace of V . Thus, by Lemma 3.12, we see that
V = V1 ⊕ V1⊥ , which by induction is a sum of 2-dimensional spaces of the form stated in
Theorem 3.9, whence we are done.
3.3 Extraspecial Groups of Order p1+2n

Let A = Mod3 (p) and B = p1+2
+ . Suppose that G and H are extraspecial p-groups, and form
their central product X = G ∗ H. Then Z (X) has order p, and since G × H has a derived
subgroup equal to its centre, so does X, and in fact this is the Frattini subgroup. Hence X
is also extraspecial. In fact, these are the only extraspecial groups. Before we prove this, we
need some specific central products. We will not prove this proposition completely, although
we will prove most of it.
Proposition 3.13 Let A and B be defined as above. Then
(i) D8 ∗ D8 ∼
= Q8 ∗ Q8 ∼
̸= D8 ∗ Q8 , and
(ii) A ∗ B ∼
=A∗A∼
̸= B ∗ B.
23
Proof: Firstly suppose that p = 2; we will prove that D8 ∗ D8 and Q8 ∗ Q8 are isomorphic.
Suppose that P = Q8 ∗ Q8 ; then P = ⟨x1 , y1 , x2 , y2 ⟩, with ⟨x1 , y1 ⟩ centralizing ⟨x2 , y2 ⟩, with
xiyi = x−1 2 2
i , and xi = yi = z, the central element of order 2. Let H1 = ⟨x1 , x2 y1 ⟩, and let
H2 = ⟨x2 , x1 y2 ⟩. Then x1 y2 and x2 y1 are of order 2, and conjugate x2 and x1 respectively
into their inverses, proving that Hi ∼= D8 , and that Q8 ∗ Q8 ∼
= D8 ∗ D8 .
Now we show that Q8 ∗ Q8 and D8 ∗ Q8 are different, by proving that they have different
numbers of elements of order 4. Suppose that G1 = D8 and G2 = Q8 , where
G1 = ⟨ a, b : a4 = b2 = 1, ab = a−1 ⟩, G2 = ⟨ x, y : x4 = 1, x2 = y 2 , xy = x−1 ⟩.
The elements in H = G1 × G2 of order 4 are (a±1 , g2 ) for g2 ∈ G2 , and (g1 , g2 ) where g1 ∈ G1

and g2 has order 4. In particular, there are 52 elements of order 4. The central subgroup of
order 2 that will be quotiented out to form the central product is ⟨z⟩, where z = (a2 , x2 ),
and so an element of order 4 in G = G1 ∗ G2 is the image of an element of order 4 in H.
The 12 elements (a±1 , g2 ), where g2 ∈ G2 has order 4, square to z, and so these have order
2 in G. The remaining elements h ∈ H are identified with the elements hz, and so there are
(52 − 12)/2 = 20 elements of order 4 in D8 ∗ Q8 .
Now let G1 = D8 and G2 = D8 , where
G1 = ⟨ a, b : a4 = b2 = 1, ab = a−1 ⟩, G2 = ⟨ x, y : x4 = y 2 = 1, xy = x−1 ⟩.
The elements in H = G1 × G2 of order 4 are (a±1 , g2 ) and (g1 , x±1 ), where gi ∈ Gi . In

particular, there are 28 elements of order 4. The central subgroup of order 2 that will be
quotiented out to form the central product is ⟨z⟩, where z = (a2 , x2 ), and so an element of
order 4 in G = G1 ∗ G2 is the image of an element of order 4 in H. The 4 elements (a±1 , x±1 )
square to z, and so these have order 2 in G. The remaining elements h ∈ H are identified
with the elements hz, and so there are (28 − 4)/2 = 12 elements of order 4 in D8 ∗ D8 . Hence
D8 ∗ D8 ∼
̸= D8 ∗ Q8 , proving (i).
Since B has exponent p and the elements of each subgroup commute, we see that B ∗ B
has exponent p, whereas A ∗ B and A ∗ A have exponent p2 . Now suppose that G = A ∗ A,
with generators x1 , y1 , x2 , and y2 , where {x1 , y1 } centralizing {x2 , y2 }, o(xi ) = p2 , o(yi ) = p,
and xiyi = x1+p
i . Finally, ⟨xpi ⟩ = Z (P ), and we can suppose that xp1 = xp2 . Let a = x2 x1−1 ; we
have
ap = (x2 x−1 p
1 ) = 1,
and y2 does not centralize a. Hence ⟨a, y2 ⟩ is a non-abelian group of order p3 , and is of
exponent p, and so is isomorphic with B. It turns out that there is another non-abelian
subgroup of order p3 that centralizes the subgroup isomorphic with B, and so it is a central
24
product involving B and another group, which has to be A, as we know that it cannot be
B. Hence all parts are proved.
Now we use our knowledge of alternating forms to prove the following result.
Theorem 3.14 Let G be an extraspecial p-group.
(i) If p is odd, then either G ∼

= B ∗n or G ∼
= A∗n , and in either case, |G| = p2n+1 .
(ii) If p = 2, then either G ∼

= D8∗n or G ∼
∗(n−1)
= Q8 ∗ D8 , and in either case, |G| = 22n+1 .
Proof: By Proposition 3.13, we see that all we need to show is that G is the central product
of non-abelian groups of order p3 , and then we are done. Thus let G be any extraspecial
group, and let Z = Z (G). Identify Z with GF(p), so that if Z = ⟨z⟩, then z i is associated
with i ∈ GF(p). If x and y are elements of G, then [x, y] ∈ GF(p).
Writing Ḡ = G/Z, we see that commutation of elements induces a map ϕ : Ḡ × Ḡ →
GF(p). If x lies in G, write x̄ for the image of x (a coset of Z) in Ḡ. Notice that if
z1 , z2 ∈ Z (P ) = Φ(P ), then [xz1 , yz2 ] = [x, y], that [xx′ , y] = [x, y][x′ , y] since [x, y] ∈ Z (P ),
and so the map is bilinear.
Write ϕ(x̄, ȳ) = [x, y] ∈ GF(p). Thus ϕ becomes a bilinear form, and since ϕ(x, x) = 0,
the form ϕ is alternating. If x̄ ̸= 1, so that Zx ̸= Z, then there exists y ∈
/ Z such that
[x̄, ȳ] ̸= 0, and so ϕ is non-singular.
We can view Ḡ, which is elementary abelian, as a vector space, and can write
Ḡ = Ḡ1 ⊕ · · · ⊕ Ḡn ,
where Ḡi = ⟨x̄i , ȳi ⟩ has dimension 2, and ϕ(x̄i , ȳi ) = 1 and all Gi and Gj are orthogonal for
i ̸= j. Hence, in particular, dimGF(p) Ḡ = 2n is even.
Let G1 , . . . , Gn be preimages of Ḡ1 , . . . , Ḡn in G. Then G1 , . . . , Gn are non-abelian groups
of order p3 , generating G, such that any two of them intersect in Z (G). Also [Gi , Gj ] = 1 if
i ̸= j. Thus G is a central product
G = G1 ∗ G2 ∗ · · · ∗ Gn ,
as required.
Theorem 3.15 Let P be extraspecial of order p2n+1 .
(i) If α ∈ Aut(P ) induces the identity on P/ Φ(P ), then α is inner.
25
(ii) Let AutC (P ) = {α ∈ Aut(P ) : α|Z(P ) = 1} P Aut(P ), and put OutC (P ) = AutC (P )/ Inn(P ).
Then
Outc (P ) ,→ Sp2n (p),
and we get an isomorphism if and only if exp(P ) = p.
Proof: Let B be the group of all such automorphisms. Consider β ∈ B, and its action
on a ‘basis’ for P (i.e., a minimal generating set {g1 , . . . , g2n }). Then the total number
of automorphisms is at most p2n , which is | Inn(P )| = |P/ Z (P ) |. But B ⩽ Inn(P ), so
B = Inn(P ).
Now let α ∈ Autc (P ). Then for x, y ∈ P ,
[xα, yα] = [x, y]α = [x, y] ∈ Z (P ) ,
so α induces an automorphism on P/ Φ(P ) that preserves the form [ , ], induced by commu-

tation. Now consider
ψ : Autc (P ) → Sp2n (p).
Now ker ψ = Inn(P ) by the previous part. If exp(P ) = p, then effectively one can choose
any element of Sp2n (P ) and inflate. Otherwise, we need to preserve the pth-power map, so
that in particular Autc (P ) is not transitive on the non-zero elements of P/ Φ(P ).
Notice that if exp(P ) = p2 , then Outc (P ) is an orthogonal group. [If p is odd, then
only one of the two orthogonal groups occurs (since one of the two extraspecial groups is of
exponent p). However, if p = 2, then both groups are of exponent p2 , and so both types of
orthogonal group occur.]
We now prove a generalization of a special case of a theorem of P. Hall: this theorem
discussed p-groups in which every normal abelian subgroup is cyclic.
Theorem 3.16 Let P be a non-abelian p-group in which every characteristic abelian sub-
group is central and cyclic. Then P = Z (P ) E, where E is extraspecial.
Proof: The proof of this will go in stages. Firstly, we will show that cl(P ) = 2. Let
γi = γi (P ) be the ith term in the lower central series, so that γi+1 = [γi , P ]. Then the Three
Subgroup Lemma implies
[γi , γj ] ⩽ γi+j .
[This is a general theorem about groups.] Suppose c = cl(P ) ⩾ 3. Then 2(c − 1) ⩾ c + 1, so

that
[γc−1 , γc−1 ] ⩽ γ2c−1 ⩽ γc+1 = 1.
Thus γc−1 is a characteristic abelian subgroup, so is central, a contradiction.
26
Next we show that Φ(P ) ⩽ Z (P ). It is enough to show that Φ(P ) is abelian, since it
is already characteristic. We have Z (Φ(P )) ⩽ Z (P ) since Z (Φ) (P ) is characteristic and
abelian. Now Φ(P ) is characteristic, and so hits the centre non-trivially. Suppose that Φ(P )
is non-abelian: if we put P̄ = P/Z, then Φ(P ) is a non-trivial normal subgroup of P/Z, so
hits the centre of P̄ non-trivially. Let N ⩾ Z such that |N̄ | = p and

N̄ ⩽ Z P̄ ∩ Φ(P ).
Now N is a normal abelian subgroup, so either N is cyclic or N ∼

= Z × N1 for some N1
of order p. If N is cyclic, then |P : CP (N )| ⩽ p, because a subgroup of N of index p lies
in Z (P ). The same is true in the p-rank 2 case. In any case, CP (N ) contains a maximal
subgroup of P . Thus,
CP (N ) ⩾ Φ(P ),
i.e., N ⩽ Z (Φ(P )) = Z, a contradiction.

We now prove that P ′ = Ω1 (Z (P )). For x ∈ P , we have xp ∈ Φ(P ) ⩽ Z (P ), and for
x, y ∈ P (since P has class 2),
[x, y]p = [xp , y] = 1.
Thus (P ′ )p = 1.
Finally, we prove the conclusion. Notice that if | Z (P ) | = p, then P is extraspecial. Now
suppose that | Z (P ) | ⩾ p2 . Put P̄ = P/P ′ , the abelianization of P . Since P ′ ⩽ Φ(P ) ∼
= Cpn ,
Φ(P ) is cyclic. Thus Φ(P ) = Φ(P̄ ). But P̄ is abelian, and so
P̄ = A1 × · · · × Ar ,
with the Ai cyclic; then |Ai | = p except for A1 (since Φ(P ) is cyclic). If |A1 | ⩾ p2 , then
℧1 (P̄ ) = ⟨ ḡ p : ḡ ∈ P̄ ⟩ = ℧1 (A1 ) char P̄ .
Notice importantly that ℧1 (A1 ) = Φ(P̄ ). In any case consider Ω1 (P̄ ) and its lift Ω
\
1 (P̄ ) to
P ; call this P1 .
We can apply the same arguments to P1 as to extraspecial groups. Now P1 char P , so
Z (P1 ) char P , implying that Z (P1 ) is cyclic. So the form [ , ] on Ω1 (P̄ ) induced by commu-
tation has a 1-dimensional singular subspace. Hence P1 = Z1 E, where Z1 = Z (P1 ) ∼ = Cp 2
and E is extraspecial.
If |A1 | = p, we are done. If not, then E P P and P = E CP (E) by Theorem 3.15(i).
Now, looking at P̄ , the rank of CP (E) is 1, so CP (E) is cyclic, and CP (E) = Z (P ).
27
Chapter 4
Maximal Class and p-Rank
4.1 Cyclic Subgroups of Index p

Recall the definitions of the dihedral, quaternion, semidihedral and modular groups from the
previous chapter.
Theorem 4.1 Let G be a p-group with a cyclic subgroup of index p. If p is odd then G is
isomorphic with one of the following:
(i) Cpn ;
(ii) Cpn−1 × Cp ; or
(iii) Modn (p);
If p = 2 then G is isomorphic with one of the groups
(i) C2n ;
(ii) C2n−1 × C2 ;
(iii) Modn (2) for n ⩾ 4;
(iv) D2n for n ⩾ 3;
(v) SD2n for n ⩾ 4; or
(vi) Q2n for n ⩾ 3.
Proof: We may assume that G is a non-abelian group with an element x of order pn−1 . Let
y denote an element of G \ X. Since y p ∈ X, the automorphism induced by conjugation by
y is of order p.
28
n−2
Suppose firstly that p is odd. Then y can be chosen so that xy = x1+p , and G = ⟨x, y⟩.
Now
(xp )y = (xy )p = x,
and so xp ∈ Z (G). Since G/ Z (G) cannot be cyclic and G is non-abelian, we see that
Z (G) = ⟨xp ⟩. Since G is 2-generator, we find that Φ(G) = Z (G), and so G has class 2.
Write y p = xαp for some α, and set z = yx−α . Since G/ Z (G) is elementary abelian, we have
z p = (yx−α )p = y p x−αp = 1,
(since if G has class two and G′ = Φ(G) then xp y p = (xy)p ) and so ⟨x, z⟩ = Modn (p). Thus
we have proved the result for p odd.
Now suppose that p = 2. If y ∈ G \ X, then xy is one of the following three elements:
n−2 n−2 −1 n−2
x1+2 ; x−1 ; and x2 . Since x2 is the unique element of order 2 in X, write a for this
element. Then xy = xδ aε , where δ = ±1 and ε is equal to either 0 or 1.
Since y 2 = x2α for some α, the element y 2 centralizes both y and x, and so lies in the
centre of G. We will compute the order of the centre of G in each of the three cases for δ
and ε. If δ = −1 and ε = 0, we clearly have that | Z (G) | = 2, and with slightly more work,
one can see that | Z (G) | = 2 in the case where δ = −1 and ε = 1. Thus y has order 2 or 4
in either of these cases.
If o(y) = 2, then the extension splits, and G is isomorphic to Modn (2), D2n , and SD2n in
each of the three cases outlined above, so we assume that o(y) = 4. If (δ, ε) = (−1, 0), then
G = Q2n , so we are left with the case where o(y) = 4 and xy = x−1 a. Since a = y 2 ∈ Z (G),
we have
(yx)2 = y 2 xy x = zx−1 zx = 1,
and so there is an element of order 2, proving that the extension does split.
Hence, we are left with the case where xy = xa, where y 2 = x2α . Consider the element
yxβ , for some β; then
n−2 ) n−3 )
(yxβ )2 = y 2 (xβ )y xβ = x2α xβ(1+2 xβ = x2α+2β(1+2 .
One can choose β to satisfy 2α + 2β(1 + 2n−3 ) ≡ 0 mod 2n−2 , proving that yxβ has order 2,
giving a split extension Mod2 (n).
Notice that the centre of Modn (2) has order 4, and that the derived subgroup of Modn (2)
has order 2.
29
4.2 The 2-Groups of Maximal Class
Theorem 4.2 Let G be a 2-group of maximal class, with |G| ⩾ 8. Then G is either dihedral,
semidihedral, or quaternion.
Proof: Since G is of maximal class, | Zi (G) / Zi−1 (G) | = 2 for all 1 ⩽ i ⩽ n − 1. Thus
G/ Z (G) has order 2n−1 and is of maximal class n − 2. Hence G/ Z (G) is dihedral, semidihe-
dral, or quaternion. Let x̄ denote an element of order 2n−2 in Ḡ = G/ Z (G), and let X̄ = ⟨x̄⟩.
Write X for its preimage in G. Since X P G, and Z (G) has order 2, Z (G) ⩽ X, whence
X̄ = X/ Z (G) is cyclic, so that X is abelian. We know that X has an element of order 2n−2 ,
so either X ∼
= C2n−2 × C2 or it is cyclic. If X is cyclic, then G contains a maximal cyclic
subgroup, and we are done by Theorem 4.1, so we assume that X is not cyclic. In this case,
X = ⟨x, z⟩, where Z (G) = ⟨z⟩. Then
n−3
⟨x2 ⟩ = Ω1 (X) ∩ ℧1 (X) char X P G,
n−3
whence x2 ∈ Z (G) = {1, z}, contrary to the fact that X = ⟨x, z⟩, yielding the result.
In fact, the 2-groups of maximal class can be characterized in another way.
Theorem 4.3 Let G be a 2-group. Then G/G′ is of order 4 if and only if G is of maximal
class.
Proof: Let G be a 2-group of maximal class. The quotient G/G′ has order at least 4 since
the rank of G/Φ(G) is the number of generators of G, which is at least 2. Since G has class
n − 1, γn (G) = 1, and since γi (G) > γi+1 (G) for all i < n, we must have that |G : γ2 (G)| ⩽ 4.
Thus suppose that G is a 2-group such that |G/G′ | = 4. Since 1 ̸= G′ P G, we have
1 ̸= z ∈ Z (G) ∩ G′ . Write Ḡ = G/⟨z⟩, and note that |Ḡ/(Ḡ)′ | = 4. Hence Ḡ is of maximal
class, by induction, and so | Z (G) | ⩽ 4. If | Z (G) | = 2, then G has maximal class, and so
we assume that | Z (G) | = 4.
Let x be an element of G whose image x̄ in Ḡ has order 2n−2 ; and write X for the
preimage of X̄ = ⟨x̄⟩. If X is cyclic, then G contains a cyclic subgroup of index 2, and
we easily see that G has maximal class, so we assume that X = ⟨x, z⟩ is isomorphic with
n−3
C2n−2 × C2 . As with the previous theorem, we get x2 ∈ Z (G), and so Z (G) is a Klein
2n−3
four-group, generated by x = y and z.
Now consider H = G/⟨y⟩. This is again of maximal class, and the image of X in H
is isomorphic with C2 × C2n−3 . Since no group of maximal class has a non-cyclic abelian
subgroup of order 8, this forces n = 4.
30
Hence we have a group G with Z (G) = G′ of order 4, and G = ⟨Z (G) , a, b⟩ for some a, b.
Thus
G′ = ⟨[a, b]⟩
is cyclic, a contradiction.
4.3 Groups of p-Rank 1

The case where p is odd is very easy.
Proposition 4.4 Suppose that p is odd, and let G be a p-group of p-rank 1. Then G is
cyclic.
Proof: Suppose that G has p-rank 1. Let H be a subgroup of index p, which also has p-rank
1. Then H is cyclic by induction, and so G has a cyclic subgroup of index p. By Theorem
4.1, G is either cyclic, isomorphic with Modn (p), or is isomorphic with Cpn−1 × Cp . The last
two are clearly not of p-rank 1, and so G must be cyclic, as required.
For the case where p = 2, we start off with Philip Hall’s original theorem, about which
we had a generalization of a special case earlier. The proof will be omitted.
Theorem 4.5 (P. Hall) Let G be a p-group in which every characteristic abelian subgroup
is cyclic. Then G is the central product of an extra-special group E and a p-group R, where
R is either cyclic, dihedral, semidihedral, or quaternion.
This is a characterization of p-groups of characteristic p-rank 1; that is, those groups

whose characteristic elementary abelian subgroups are all of size p. We will refine this by
determining those groups whose normal p-rank is 1, and finally to those whose p-rank is 1.
Theorem 4.6 Let G be a p-group with normal p-rank 1. Then G is cyclic, dihedral (of
order at least 16), semidihedral, or generalized quaternion.
Proof: Suppose that G is a group with normal p-rank 1. Then G has characteristic p-rank
1, and so G is the central product E ∗ R. If E = 1 then the theorem is true, and so we
assume that E ̸= 1.
Suppose firstly that p is odd. Then E = H ∗ K, with H isomorphic with either p1+2
− or
p1+2
+ , both of which contain a normal subgroup L isomorphic with Cp ×Cp . Since [H, K] = 1,
we see that L P E, and similarly L P G, proving that E = 1, and G is cyclic.

Hence we reduce to the case where p = 2. Certainly if E ∼
= D8 ∗ H for some subgroup H,
then E contains a normal subgroup isomorphic with C2 × C2 , and since Q8 ∗ Q8 = D8 ∗ D8
31
by Proposition 3.13, we see that E = Q8 . Thus G can only be the central product of Q8 and
a cyclic, dihedral, semidihedral or quaternion subgroup.
Let H be a normal subgroup of E of order 4, and let K be a normal subgroup of R of order
4, both containing the central element z ̸= 1. Then HK has order 8 and exponent 4, since
[H, K] = 1, and so HK is a non-cyclic, abelian normal subgroup, contrary to hypothesis.
Thus E = 1, as required.
Corollary 4.7 Let G be a group of 2-rank 1. Then G is cyclic or generalized quaternion.
Proof: Certainly dihedral and semidihedral groups are of 2-rank 2, and so the only groups
left on the list given in Theorem 4.6 are cyclic and quaternion, both of which do indeed have
p-rank 1.
32
Chapter 5
Fixed-Point-Free Automorphisms
Definition 5.1 Let ϕ be an automorphism of the group G. Then ϕ is fixed-point-free if

xϕ = x implies x = 1.
As easy examples of fixed-point-free automorphisms, we have the non-trivial automor-

phism of C3 , and the automorphism of V4 of order 3. Clearly, if a finite group G has a
fixed-point-free automorphism of order p, then |G| ≡ 1 mod p.
Lemma 5.2 Let G be a group, and let ϕ be a fixed-point-free automorphism of order n.

Then
2 n−1
x(xϕ )(xϕ ) . . . (xϕ ) = 1.
The proof of this is obvious, since the left-hand side is invariant under ϕ.
Lemma 5.3 Let G be a finite group, and let ϕ be a fixed-point-free automorphism of G. If

p is a prime dividing |G|, then ϕ fixes a unique Sylow p-subgroup P of G.
Proof: If P is a Sylow p-subgroup of G, then P ϕ is also a Sylow p-subgroup of G. Therefore,

P ϕ = x−1 P x for some x ∈ G. Then for any y ∈ G,
(y −1 P y)ϕ = (y ϕ )−1 x−1 P x(y ϕ ).
However, every element of G can be expressed as xϕ x−1 (G is finite and x 7→ xϕ x−1 is an

injection) and so choose y such that (y ϕ )y −1 = x−1 . Then P y is fixed under ϕ, as required.
Now suppose that P and P x are fixed by ϕ. Therefore, (xϕ )x−1 ∈ NG (P ). Again, since
(xϕ )x−1 is in NG (P ), we see that (as NG (P ) is ϕ-invariant) there is an element y ∈ NG (P )
such that
(xϕ )x−1 = (y ϕ )y −1 .
Since the map x 7→ (xϕ )x−1 is a bijection, x = y, and so P x = P , as needed.
33
Lemma 5.4 Let G be a finite abelian group, and suppose that H is a subgroup of Aut(G)
of the form K ⋊ ⟨ϕ⟩. Suppose that, for all k ∈ K, the element kϕ is fixed-point-free and or
prime order, and that |K| and |G| are coprime. Then K fixes some non-trivial element of G.
Proof: The elements of H \ K are of the form (kϕ)i , where k ∈ K and 1 ⩽ i ⩽ p − 1. If

x ∈ G, then
p−1
i
YY
1= x(kϕ)
k∈K i=0
p−1
i
Y Y
= x|K| x(kϕ)
i=1 k∈K
p−1
i
YY
= x|K| xkϕ
i=1 k∈K
Clearly,
i
Y
xkϕ
k∈K
is a fixed point of G under the action of k ∈ K, and since x|K| is not the identity, one of the
terms in the product must also not be the identity. Hence there is a fixed point of G under
the action of K.
Corollary 5.5 Let G be a finite abelian group, and let A be a homocyclic group of auto-
morphisms of G, all of whose non-trivial elements act fixed-point-freely. Then A is cyclic.
If G is a group, then the map x 7→ x−1 is an anti-automorphism; that is, it is a map ϕ

such that (xy)ϕ = (y ϕ )(xϕ ). If G is abelian, then all anti-automorphisms are automorphisms,
and so any abelian group of odd order has a fixed-point-free automorphism of order 2.
Lemma 5.6 Suppose that ϕ : G → G is a bijection that is both an automorphism and an

anti-automorphism. Then G is abelian.
Proof: Let x and y be elements of G. Since ϕ is a bijection, there are elements x′ and y ′
such that x′ ϕ = x and y ′ ϕ = y. Since ϕ is both an automorphism and an anti-automorphism,
we have
xy = (x′ ϕ)(y ′ ϕ) = (x′ y ′ )ϕ = (y ′ ϕ)(x′ ϕ) = yx,
as G is abelian.
Corollary 5.7 Suppose that G has a fixed-point-free automorphism ϕ of order 2. Then G

is an abelian group of odd order.
34
Proof: The map ϕ satisfies x(xϕ ) = 1, by Lemma 5.2. This implies that xϕ = x−1 , and
so this map is an automorphism. It is also an anti-automorphism, and so G possesses an
automorphism that is also an anti-automorphism; thus G is abelian.
Theorem 5.8 (B. Neumann, 1956) Let ϕ be a fixed-point-free automorphism of G, and

suppose that ϕ has order 3. Then G is nilpotent of class at most 2.
2
Proof: By Lemma 5.2, we see that xxϕ xϕ = 1 for all x ∈ G. Let y = x−1 , so that
2
yϕ yϕy = 1
for all y ∈ G. Therefore

2
[x, xϕ ] = yy ϕ xxϕ = xϕ xxϕ = 1.
Clearly if ϕ is fixed-point-free, then ϕτg is fixed-point-free, where τg is conjugation by g.

2
Therefore xg commutes with xϕ and xϕ for all g and x, and so x commutes with xg .
We now note that G can have no elements of order 3: if it were to, then since x and xϕ
commute, ⟨x, xϕ ⟩ is a ϕ-invariant elementary abelian of order 9, and contains a fixed point
under ϕ, which is demonstrably impossible.
We finish by quoting a standard result. If G is a group such that [g, x, x] = 1 for all
g, x ∈ G and G contains no 3-torsion, then G is nilpotent of class at most 2.
There is a substantial generalization of this theorem, which was Thompson’s Ph.D. thesis.
Theorem 5.9 (Thompson) Suppose that a finite group G possesses a fixed-point-free au-
tomorphism ϕ of prime order. Then G is nilpotent.
This in turn rests upon Thompson’s normal p-complement theorem.
Theorem 5.10 (Thompson, 1959) Let G be a group, and let p be an odd prime with
p| |G|; write P ∈ Sylp (G) and write J ∗ (P ) for the subgroup of P generated by all abelian sub-
groups of maximal rank. Then G has a normal p-complement if and only if both CG (Z (P ))
and NG (J ∗ (P )) have normal p-complements.
We will reduce Thompson’s Theorem 5.9 to soluble groups first. We therefore assume
that G is a minimal counterexample, and prove that G is soluble. If G is a 2-group, then
G is nilpotent, and so choose q to be an odd prime dividing |G|, and let P denote a ϕ-
invariant Sylow q-subgroup of G. Since both Z = Z (G) and J = J ∗ (P ) are characteristic in
P , they are ϕ-invariant. If either Z or J is normal in G, then ϕ induces a fixed-point-free
automorphism on G/Z or G/J, which are by induction nilpotent, and thus G is soluble.
35
The other possibility is that NG (J) and NG (Z) are both proper in G. By choice of minimal
counterexample, both CG (Z) and NG (J) are nilpotent (as normalizers and centralizers of ϕ-
invariant subgroups are ϕ-invariant), and so have normal q-complements. Therefore, G has
a normal q-complement, say Q. Since Q is characteristic (a normal Hall q ′ -subgroup is
characteristic) it is ϕ-invariant, and so is nilpotent by induction, and G is soluble.
For Thompson, this was enough, since the soluble case was proved by then. However, we
haven’t, and so let’s do that now.
Theorem 5.11 Let G be a soluble group admitting a fixed-point-free automorphism of

prime order. Then G is nilpotent.
Proof: Let ϕ be a fixed-point free automorphism of order p of the soluble group G, and
let Q be a minimal ϕ-invariant normal subgroup of G (lying in G ⋊ ⟨ϕ⟩). Then Q is an
elementary abelian q-subgroup, and clearly p ̸= q.
If G is a q-group, then G is nilpotent, so let r ̸= q be a prime dividing |G|, and let R be
the ϕ-invariant Sylow r-subgroup. Consider the group QR; if QR ̸= G, then by induction
QR is nilpotent, and so Q and R centralize each other. This is true for all r ̸= q dividing
|G|, and so CG (Q) has index a power of q. Then Z (G) ̸= 1.
Thus G = QR for some R. Let K be the subgroup of Aut(Q) induced by the action of R
on it, and let H be the semidirect product of K by ⟨ϕ⟩. Then it is clear that kϕ acts fixed-
point-freely and has the same order as ϕ itself, and so K fixes a point of R. Equivalently,
there is a non-identity element z ∈ Q such that R ⩽ CG (z); clearly, z ∈ Z (G), and by
induction G/ Z (G) is nilpotent, whence G is nilpotent.
We finish discussing fixed-point-free automorphisms with a result on the structure of

groups of automorphisms all elements of which act fixed-point-freely.
Theorem 5.12 (Burnside) Let G be a finite group and suppose that G accepts a group
A of automorphisms, each (non-trivial) element of which acts fixed-point-freely. Then |G|
and |A| are coprime, and all Sylow p-subgroups of A are of p-rank 1.
Proof: Suppose that p divides both |G| and |A|, and let ϕ be an element of A of order
p. Then ϕ fixes a Sylow p-subgroup of G, and so acts fixed-point-freely on P . However,
counting ϕ-orbits yields an easy contradiction.
Now let P be a Sylow p-subgroup of A, and let S be a subgroup of P of order p2 . We
will show that S is cyclic, proving our result. We claim that G possesses an S-invariant
Sylow q-subgroup Q, where q | |G| is a prime. If this is true, then let K = Z (Q), and
apply Corollary 5.5: then S is an homocyclic group of automorphisms of an abelian group
K, whence it is cyclic, as required.
36
It remains to prove that if P is a p-group acting on a group G with p ∤ |G|, then there is a
P -invariant Sylow q-subgroup Q for all primes q dividing |G|. To see this, let R be any Sylow
q-subgroup of G, and write H = G ⋊ P . Then, by the Frattini argument, H = NH (R)G. A
Sylow p-subgroup P̄ of NH (R) is a Sylow p-subgroup of H, and hence there is an element g
such that P̄ g = P . Then
P = P̄ g ⩽ NH (Rg ),
and so Q = Rg is a P -invariant Sylow q-subgroup, finishing the proof.
Since nilpotent groups are direct products of their Sylow p-subgroups, each of which is
clearly characteristic, we see that we need to understand fixed-point-free automorphisms of
p-groups. The nilpotency class was proved to be finite by Higman, and the bound below was
given by Kreknin and Kostrikin.
Theorem 5.13 (Higman, Kreknin, Kostrikin) Let G be a nilpotent group possessing

a fixed-point-free automorphism of order p. Then the nilpotency class is bounded by the
function h(p), where
p−1
(p − 1)2 −1 − 1
h(p) ⩽ .
p−2
37
Chapter 6
The Critical Subgroup Theorem
Proposition 6.1 Let G be a group, and ϕ be an automorphism of G. Let H = G ⋊ ⟨ϕ⟩ be

the semidirect product of G and ⟨ϕ⟩. Write Ḡ for the image of G in H, and similarly ϕ̄ for
the image of ϕ in H. Then
CG (ϕ) = CḠ (ϕ̄).
Proof: Suppose that we have the setup described in the proposition, and let x ∈ CG (ϕ).
Write x̄ for the image of x in H. We have xϕ = x, and so, inside H,
x̄ϕ̄ = (x, 1⟨ϕ⟩ ) · (1G , ϕ) = ((xϕ)1G , ϕ) = (x, ϕ),
and
ϕ̄x̄ = (1G , ϕ) · (x, 1⟨ϕ⟩ ) = ((1G ϕ)x, ϕ) = (x, ϕ).
Thus x̄ ∈ CḠ (ϕ̄).

Conversely, if x̄ ∈ CḠ (ϕ̄), then x̄ϕ̄ = ϕ̄x̄. But we calculated above what x̄ϕ̄ and ϕ̄x̄ are:
they are
x̄ϕ̄ = (xϕ, ϕ), ϕ̄x̄ = (x, ϕ).
This clearly implies that ϕ centralizes x; that is, x ∈ CG (ϕ), as required.
Notice that in a group [x, y] = x−1 xy . If we identify xy with xτy , where τy represents
conjugation by y, then we have a very good candidate for the notion of a commutator with
an automorphism. We define
[x, ϕ] = x−1 (xϕ).
Notice that [x, ϕ] = 1 if and only if x ∈ CG (ϕ), which is analogous to the statement [x, y] = 1
if and only if x ∈ CG (y), which holds for any group. Consider the group G ⋊ ⟨ϕ⟩ again, and
write X̄ for the image of X in this semidirect product. We need
[x, ϕ] = [x̄, ϕ̄].
38
The proof of this is in the exercises.
Suppose that this is true: what good does it do us? Well, all of the commutator relations
that we have will work equally well for commutators involving automorphisms. For example,
let ϕ ∈ Aut(G) and x, y ∈ G. Then
[xy, ϕ] = [x, ϕ]y [y, ϕ].
To see this, take the image of [xy, ϕ] in G ⋊ ⟨ϕ⟩, and calculate there. Since
[xy, ϕ] = [x̄ȳ, ϕ̄],
we can use the fact that the identity works for the usual definition of commutator, and then
pull back. In particular, it should be noted that the Three Subgroup Lemma still works with
groups of automorphisms instead of subgroups: this is important, as it will be needed in this
section.
We have an opportunity to test out our new notation in the next result.
Theorem 6.2 Let G be a finite group, and A ⩽ Aut(G) be a group of automorphisms.

Suppose N a normal subgroup that is A-invariant, and suppose that |N | and |A| are coprime.
Then
CG/N (A) = CG (A)N/N.
Proof: Suppose that xN is an element of CG (A)N/N ; then x ∈ CG (A), and so A acts

trivially on x, so acts trivially on xN . Thus
CG/N (A) ⩾ CG (A)N/N.
It suffices to show the reverse inclusion; that is, we need to find an element x ∈ Cg (A) lying
in every A-invariant coset gN of N . Proceed by induction on the number of factors of |A|,
noting that the case where |A| is a prime is clear: for then, gN is split up into orbits of size
1 or p, and the fact that |N | and p are coprime proves that there is an orbit of size 1.
The proof of this will be omitted.
Theorem 6.2 has an important corollary, for which we first need a defintion.
Definition 6.3 Let G be a finite group, and ϕ an automorphism of G. If
1 = H0 P H1 P · · · P Hi = G
is a series, then ϕ is said to stabilize the series if ϕ acts trivially on each Hi /Hi−1 .
39
Corollary 6.4 Suppose that G is a finite group, and A is a group of automorphisms with
|G| and |A| coprime. If A stabilizes some normal series, then A = 1.
Proof: Let
1 = H0 P H1 ⩽ · · · P Hr = G
be a series, and suppose that ϕ ∈ A stabilizes this series. Then H1 ⩽ CG (ϕ). We will show
that if Hi ⩽ CG (ϕ), then Hi+1 ⩽ CG (ϕ), proving the result, since if G = CG (ϕ), ϕ = 1. Now
if Hi ⩽ CG (ϕ), then (by Theorem 6.2)
Hi+1 /Hi = CHi+1 /Hi (ϕ) = CHi+1 (ϕ)Hi /Hi = CHi+1 (ϕ)/Hi ,
since Hi ⩽ CG (ϕ) ∩ Hi+1 . Thus
Hi+1 = CHi+1 (ϕ) ⩽ CG (ϕ),
as required.
Theorem 6.5 (Thompson’s Critical Subgroup Theorem) Let G be a finite p-group.

Then there is a characteristic subgroup C of G such that:
(i) every non-trivial p′ -automorphism of G induces a non-trivial p′ -automorphism on C;
(ii) CG (C) = Z (C);
(iii) [G, C] ⩽ Z (C) (or equivalently [G, C, C] = 1); and
(iv) C has class at most 2, and C/ Z (C) is elementary abelian.
Proof: Suppose that A is a maximal abelian normal subgroup of G, and firstly assume that
A is characteristic. Then, since A = CG (A) (proof an exercise), we have (ii). Certainly,
[G, A] ⩽ A = Z (A) ,
and so we have (iii). Any abelian subgroup satisfies (iv), and so we only need to satisfy (i).
We will delay this, however, and deal with the case where A is not characteristic.
Suppose that A is a maximal abelian characteristic subgroup, and let B be a maximal
normal abelian subgroup contianing it. Now B ⩽ CG (A) and so A ̸= CG (A); since A char G,
we see that CG (A) char G. We quotient out by A, then, and notice that CG (A)/A is non-
trivial. Let C be the preimage of the group
Ω1 (CG (A)/A ∩ Z (G/A)) = CG (A)/A ∩ Ω1 (Z (G/A)),
(which is non-trivial by Lemma 2.27).
40
Notice firstly that CG (A)/A and Ω1 (Z (G/A)) are both characteristic in G/A, and there-
fore their intersection is also characteristic: hence C char G. Now what is Z (C)? Notice that
C char G, so
Z (C) char G
and since Z (C) is a characteristic abelian subgroup that contains A (since certainly A cen-
tralizes C), A = Z (C). Then C/A ⩽ Ω1 (Z (G/A)), and so is elementary abelian; hence C is
of class 2, with C/ Z (C) elementary abelian, proving (iv).
Let ϕ denote the quotient homomorphism from G into G/A. By Lemma 2.9, we have
([G, C])ϕ = [Gϕ, Cϕ] = [G/A, C/A] ⩽ [G/A, Z (G/A)] = 1,
and so [G, C] ⩽ ker ϕ = A, proving (iii). We are left to prove (ii). Let X = CG (C); then we
must show X = A. Now
Z (C) = CC (C),
and so CG (C) ∩ C = A.
Let us work in G/A. Then X/A ∩ C/A is trivial. We are aiming, in fact, to show that
X/A itself is trivial. Now X P G, so X/A P G/A; if X/A were non-trivial, then it would
intersect Z (G/A) non-trivially. By taking Ω1 , we must show that
X/A ∩ Ω1 (Z (G/A)) = 1.
Now X = CG (C) ⩽ CG (A), since every element of G that centralizes C must centralize A.
Hence
X/A ∩ Ω1 (Z (G/A)) ⩽ CG (A)/A ∩ Ω1 (Z (G/A)) = C/A.
But X/A ∩ C/A = 1, a clear contradiction. Thus X = A, and we have proved (ii). Thus in
any case we have proved that there is a subgroup C satisfying (ii), (iii) and (iv). We will
show that this subgroup satisfies (i),
Let A be a group of p′ -automorphisms of G that act trivially on C. Then [C, A] = 1, so
certainly [C, A, G] = 1. (We are clearly planning to use the Three Subgroup Lemma here!)
Also, we know that [G, C] ⩽ Z (C), and since A acts trivially on C, A acts trivially on Z (C):
this gives [G, C, A] = 1. Then the Three Subgroup Lemma gives
[A, G, C] = 1.
Then [A, G] ⩽ CG (C) = Z (C) ⩽ C. Then we have a series
1 P C P G,
which A stabilizes. Hence, by Corollary 6.4, A = 1. Thus we have (i), and we are done.
We let C, the characteristic subgroup of G in this theorem, be called the critical subgroup.
41
Chapter 7
How are p-Groups Embedded in

Finite Groups?
Theorem 7.1 (Cayley) Let G be a group with a cyclic Sylow 2-subgroup P . Then G
posseses a normal Hall 2′ -subgroup.
Proof: This is easy: consider the regular representation. Then a generator x for P is an
odd permutation, and so G has a subgroup of index 2, the set of all even permutations. By
induction, this subgroup has a normal (and hence characteristic) Hall 2′ -subgroup, and so
G has a normal Hall 2′ -subgroup.
Theorem 7.2 Let G be a soluble group, and suppose that G possesses a V4 Sylow 2-
subgroup P . Then G possesses a normal subgroup H of odd order such that G/H is either
V4 or A4 .
Proof: Let H = O2′ (G). Then G/H is a soluble group, and so has non-trivial Fitting
subgroup, and this subgroup is a 2-group Q. If |Q| = 4 then G/Q is a subgroup of Aut(Q) =
S3 of odd order, so is either trivial or C3 , and in the first case G/H is V4 and in the second
G/H is A4 . If |Q| = 2, then Aut(Q) is trivial, and this is a contradiction.
More generally, if G is a soluble group is with abelian Sylow 2-subgroup then G/ O2′ (G)
is the extension of the abelian Sylow 2-subgroup by an odd subgroup of its automorphism
group.
Proposition 7.3 Let G be the semidirect product of H by P . If x and y are conjugate in

G then x and y are conjugate in P .
Proof: Let h be an element of H, and let x be an element of P . Then
xh = h−1 xh = h−1 xhx−1 x = (h−1 h′ )x,
42
which lies in P precisely when h = h′ , in which case h centralizes x. If g ∈ G, then g = yh
for some y ∈ P and h ∈ H, whence xg is (xy )h , and the result follows.
The converse of Proposition 7.3 is also true, and it is a result of Frobenius.
Theorem 7.4 (Frobenius’ Normal p-complement Theorem) Let G be a finite group,

and let P be a Sylow p-subgroup. Then the following are equivalent:
(i) any two elements of P that are conjugate in G are conjugate in P ; and
(ii) G possesses a normal Hall p′ -subgroup.
We will not prove this theorem here, since it involves considerable finite group theory.
We will discuss fusion briefly now, though.
Definition 7.5 Let H and K be subgroups of G with H ⩽ K ⩽ G. Then K is said to

control fusion in H with respect to G if any two elements of H that are G-conjuate are also
H-conjugate.
The following result of Burnside is well-known.
Proposition 7.6 (Burnside) Suppose that G is a finite group with an abelian Sylow p-
subgroup P . Then NG (P ) controls fusion in P with respect to G.
Proof: Let x and y be elements of P , and suppose that there is g ∈ G such that xg = y.
Thus
P g ⩽ CG (x)g = CG (xg ) = CG (y).
Thus both P and P g are Sylow p-subgroups of CG (y), whence they are conjugate by some
element h ∈ CG (y). Thus P gh = P , and so gh ∈ NG (P ), and xgh = y, as required.
As a corollary of Frobenius’ normal p-complement theorem, we get Burnside’s normal

p-complement theorem.
Corollary 7.7 (Burnside’s Normal p-complement Theorem) Let G be a finite group,

and let P be a Sylow p-subgroup of G. Suppose that P is contained within the centre of its
normalizer. Then G possesses a normal Hall p′ -subgroup.
Proof: Since P is abelian, NG (P ) controls fusion in P . Since P lies in the centre of its
normalizer, all NG (P )-conjugacy classes of P are of size 1, and so actually P controls fusion
in P . Hence, by Frobenius’ normal p-complement theorem, G possesses a normal Hall p′ -
subgroup, as required.
43
We can also easily get Cayley’s result, given earlier in this chapter.
Corollary 7.8 Let G be a group with cyclic Sylow 2-subgroups. Then O2′ (G) has index a
power of 2.
Proof: Let P denote a Sylow 2-subgroup of G, and recall that Aut(P ) is a 2-group. Since
NG (P )/ CG (P ) is an odd-order subgroup of Aut(P ), it is trivial, whence P lies in the centre
of its normalizer. Thus O2′ (P ) is a normal Hall 2′ -subgroup, as required.
To demonstrate the fusion in finite groups, we prove the characterization of simpel groups
of order 168. To do this, we need the following lemma.
Lemma 7.9 Let G be a subgroup of A7 such that |G| is a multiple of 14. Then G = A7 or
G = GL3 (2) = PSL2 (7).
Proof: This proof is a case-by-case analysis. Since 14 divides |G|, we know that G contains
a 7-cycle and a double transposition. Without loss of generality, we may assume that the
7-cycle x is (1, 2, 3, 4, 5, 6, 7), and by raising x to an appropriate power, we may assume that
y = (1, 2)(a, b).
To prove that G = A7 , it suffices to find an element of order 3 and an element of order
5, since then |A7 : G| ⩽ 4. It originally appears as though there are ten possibilities for
(a, b), but by relabelling we see that (3, 4) and (6, 7) yield isomorphic groups, as do (4, 5)
and (5, 6). Again, relabelling and raising x to a power gives that (3, 7) and (4, 6) yield
isomorphic groups, as do all four of the remaining pairs.
We reduce to the four possibilities (3, 4), (3, 5), (3, 7), and (4, 5). Three of these yield
A7 , as outlined below.
(a, b) Order 3 Order 5
(3, 4) (x3 y)2 xy
2 2
(3, 7) (x y) [x, y]
(4, 5) [x, y] xy
Since GL3 (2) (whose order is a multiple of 14) acts on the seven non-zero vectors and it is
a simple group, we must have that the remaining possibility is GL3 (2), as required.
Theorem 7.10 Let G be a simple group of order 168. Then G ∼

= GL3 (2).
Proof: We will prove that G possesses a subgroup of index 7, whence the previous lemma
will prove our result. (Note that this is equivalent to proving that G possesses a subgroup
of index at most 7.) Therefore we will assume that the index of any proper subgroup of G
is at least 8, and derive a contradiction.
44
Since G possesses no subgroups of index 3 or 7, the number of Sylow 2-subgroups is 21.
Thus a Sylow 2-subgroup is self-normalizing, and so is non-abelian. Let P and Q denote
distinct Sylow 2-subgroups, and consider P ∩ Q. If this has order 4, then both P and Q
normalize their intersection, and so the order of the normalizer of this intersection is at least
24, a contradiction. Thus the intersection of any two Sylow 2-subgroups is of order at most
2.
If the Sylow 2-subgroups of G were quaternion, then they must intersect trivially, else
the normalizer of the intersection P ∩ Q would contain both P and Q, and so have order
at least 24. However, in this case there must be 21 × 7 = 147 non-trivial elements lying in
Sylow 2-subgroups. There are clearly eight Sylow 7-subgroups, contributing 48 elements of
order 7, and so this contradiction implies that P is dihedral of order 8.
Let z denote the non-trivial central element of P . Then all conjugates of z lie in the
centre of some Sylow 2-subgroup, and so there are at most 21 of them. Conversely, if some
conjugate of z lies in two different Sylow 2-subgroups, then the centralizer of z has order at
least 24, a contradiction. Thus CG (z) = P .
We will prove that there are no other elements of order 2 in G. If this is true, then the
theorem follows easily from this: let R denote a subgroup of P isomorphic with V4 , and
consider NG (R), which we claim has order (at least) 24. Each Sylow 2-subgroup contains
two V4 subgroups, and either one of them (without loss of generality R) lies in two different
Sylow 2-subgroups, or there are 42 different V4 subgroups. However, each involution appears
in exactly two V4 subgroups, and so there are fourteen V4 subgroups. Thus R lies in two
different Sylow 2-subgroups, and so its normalizer had order at least 24, as required.
It remains to prove that there are exactly twenty-one elements of order 2, and we will
prove this by counting. We have 48 elements of order 7, as seen above. There must also be
at least eight Sylow 3-subgroups (since G possesses no subgroups of index less than 8), and
so there must be twenty-eight Sylow 3-subgroups, yielding 56 elements of order 3. Together
with the identity and 21 elements of order 2, that leaves 42 elements to find.
Let x denote an element of order 4 in P . Then CG (x) has order an odd multiple of 4,
but above we calculated the normalizers of the Sylow 3- and 7-subgroups. It was shown that
4 does not divide the order of the normalizer of either a Sylow 3- or a Sylow 7-subgroup,
and so no odd element can centralize x. Thus there are 42 elements of order 4, and we have
identified all elements of the group.
Here we see a good example of Frobenius’ normal p-complement theorem in action. The
normalizer of a Sylow 3-subgroup has order 6, and since the centralizer of an involution has
order 8, the Sylow 3-subgroups are self-centralizing. Therefore the two non-trivial elements
in a Sylow 3-subgroup are conjugate, as predicted by Frobenius’ theorem. All elements of
45
order 3 are conjugate, as are all elements of orders 2 and 4.
Finally, consider the elements of order 7. Then the normalizer of a Sylow 7-subgroup
has order 21, and since no elements of order 3 centralize an element of order 7, we see that
the Sylow 7-subgroups are self-centralizing as well. Let g be an element of order 7. Thus
the normalizing element of order 3 must make g, g 2 and g 4 conjugate (as the automorphism
group of order 3 of C7 is generated by the map g 7→ g 2 ), and make g 3 , g 5 and g 6 conjugate.
Thus there are two conjugacy classes of elements of order 7, each with 24 elements in it.
Again, this agrees with Frobenius’ theorem.
We can see that the simplicity of G causes great constraints on the fusion of conjugacy
classes of p-elements. This is used extensively in finite group theory.
46

P Groups

Uploaded by

Copyright:

Available Formats

P Groups

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

P Groups

Uploaded by

Copyright:

Available Formats

The Theory of p-Groups

Hilary Term, 2008

2 Introduction to the Structure of p-Groups 5

4 Maximal Class and p-Rank 28

6 The Critical Subgroup Theorem 38

7 How are p-Groups Embedded in Finite Groups? 42

1.1 Soluble Groups

Theorem 1.2 Let G be a finite soluble group. Then

Proposition 1.3 Let G be a finite nilpotent group, and let G = P1 × P2 × · · · × Pn , where

Aut(G) = Aut(P1 ) × Aut(P2 ) × · · · × Aut(Pn ).

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

(ii) the bracket operation distributes over addition;

(iii) [x, x] = 0 for all x ∈ R; and

(iv) [[x, y], z] + [[y, z], x] + [[z, x], y] = 0 for all x, y, z ∈ R.

Proposition 1.5 Let ϕ be an automorphism of the finite p-group G. Then ϕ induces an

1.3 The Number of Groups

(i) g(p) = 1 for p a prime.

(ii) if p < q, then g(pq) = 1 if q ̸≡ 1 mod p, and g(pq) = 2 otherwise.

1.4 Sylow Structure of Groups

(i) the projective special linear group PSL2 (q), q ≡ 3, 5 mod 8;

(ii) the projective special linear group PSL2 (2n ) for n ⩾ 3;

(iv) the first Janko group J1 .

Introduction to the Structure of

Definition 2.1 A central series is a chain of normal subgroups

such that Hi /Hi−1 ⩽ Z (G/Hi−1 ) for all 1 ⩽ i ⩽ r.

Definition 2.2 Let G be a group. Define

(i) G(1) = G′ , the derived subgroup, and G(r) = [G(r−1) , G(r−1) ];

Lemma 2.3 Suppose that

The elementary abelian groups are actually the groups Cp × Cp × · · · × Cp , where Cn is

We define an extraspecial group to be a p-group for which G/ Z (G) is elementary abelian,

G′ = [G, G] = ⟨ [x, y] : x, y ∈ G⟩.

([x, y])ϕ = x−1 y −1 xy ϕ = (xϕ)−1 (yϕ)−1 xϕyϕ = [xϕ, yϕ],

and so if ϕ : G → G is an endomorphism, then G′ ϕ ⩽ G′ , as required.

Lemma 2.10 Let G be a group, and H any normal subgroup of G.

(ii) If G/H is also abelian then G′ ⩽ H.

(xG′ )(yG′ ) = (xy)G′ = (yx)G′ = (yG′ )(xG′ ),

proving assertion (i).

We will give a notation to extended commutators – by this we mean an object like

Lemma 2.11 Let G be a group, and x, y, z be elements of G.

(i) [xy, z] = [x, z]y [y, z] = [x, z][x, z, y][y, z].

(iii) [x, y] = [y, x]−1 .

Proof: To prove (i), notice that

[x, y −1 , z]y [y, z −1 , x]z [z, x−1 , y]x = u−1 vv −1 ww−1 u = 1,

Theorem 2.12 (Three Subgroup Lemma) Let X, Y and Z be three subgroups of a

Proposition 2.13 Let H and K be subgroups of a group G.

(i) [H, K] ⩽ K if and only if H ⩽ NG (K).

Proof: Suppose that H ⩽ NG (K). Then h−1 k −1 h = k ′ ∈ K and so h−1 k −1 hk = k ′ k ∈ K.

[h′ , k ′ ]hk = [(h′ )hk , (k ′ )hk ] ∈ [H, K],

Corollary 2.14 Suppose that H, K and L are normal subgroups of G. Then

[HK, L] = [H, L][K, L], [L, HK] = [L, H][L, K].

Proof: Consider the commutator

X = [HK, HK, . . . , HK],

where there are c + d + 1 terms. By Corollary 2.14, this can be written as

Definition 2.17 Let G be a group, and x ∈ G. Then x is said to be a non-generator if

Proof: Suppose that x is a non-generator of G, and let M be a maximal subgroup of G.

showing that P Φ(G) P G.

Now P ⩽ NG (P ), so NG (P )P = NG (P ). Thus G = NG (P ) Φ(G). Now Proposition 2.18

Corollary 2.20 The Frattini subgroup of a finite group is nilpotent.

Proof: Take N = Φ(G) in Theorem 2.19.