web.mat.bham.ac.ukweb.mat.bham.ac.uk/c.w.parker//sarahs mphil.pdf · 3 amalgams and the coset graph...
TRANSCRIPT
3-local identifications of somefinite simple groups
by
SARAH ASTILL
A thesis submitted toThe University of Birmingham
for the degree ofMaster of Philosophy (Sc, Qual)
School of MathematicsThe University of Birmingham
Abstract
In this thesis we characterize the groups PSL3(3), M12, G2(3) and PSp4(3) by their 3-local
structure. We identify each group as a finite faithful completion of a certain type of rank
2-amalgam known as a weak BN -pair.
Acknowledgements
I am grateful for the help, support and encouragement first and foremost from my super-
visor Professor Chris Parker. Thanks also go to Dr Rachel Fowler, Dr Murray Clelland,
Dr Rebecca Waldecker and James Sprittles. Finally, I gratefully acknowledge the financial
support of the EPSRC.
Contents
Introduction 1
1 Some Group Theory 6
1.1 Some General Group Theoretic Results . . . . . . . . . . . . . . . . 6
1.2 Extraspecial Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3 Fixed-Point-Free Actions . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Strongly Embedded Subgroups . . . . . . . . . . . . . . . . . . . . . . 17
1.5 Automorphisms of Abelian Groups . . . . . . . . . . . . . . . . . . . 19
2 Some Representation Theory 23
2.1 Endomorphism Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Natural GL2(3)-Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Sym(4)-Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Some Modular Character Theory . . . . . . . . . . . . . . . . . . . . 37
2.6 Natural SL2(2n)-Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3 Amalgams and the Coset Graph 49
4 Background Results 53
5 An Amalgam of Type PSL3(3) 61
5.1 Some Structure of the Amalgam (A,B,C) . . . . . . . . . . . . . . . 63
5.2 A Presentation for the Amalgam . . . . . . . . . . . . . . . . . . . . . 65
6 An Amalgam of Type G2(3) 75
6.1 The Amalgam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
6.2 The Coset Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
6.3 Controlling the 3-Local Structure . . . . . . . . . . . . . . . . . . . . 89
6.4 Controlling the Centralizer of an Involution . . . . . . . . . . . . . . 96
7 An Amalgam of Type PSp4(3) 101
7.1 The Amalgam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.2 The Coset Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.3 Controlling the 3 and 2-Local Structure . . . . . . . . . . . . . . . . 109
8 Higman’s Theorem on Fixed-Point-Free Elements of Order Three 119
Appendix 137
Introduction
A p-local subgroup (p a prime) of a group is the normalizer of a non-trivial p-subgroup.
To identify a group p-locally is to determine the isomorphism type of a group given some
information about its p-local structure. Suppose we have a finite group G with a Sylow
p-subgroup S such that S has subgroups A and B which are normalized by NG(S) then
the intersection of NG(A) and NG(B) contains NG(S). The Amalgam Method aims to
characterize G given the subgroup structure associated to NG(A) and NG(B).
A rank two amalgam of finite groups, A = (P1, P2, C, φ1, φ2), is a collection of three
finite groups, P1, P2 and C, and two injective homomorphisms, φ1 : C → P1 and φ2 : C →
P2. A completion, (G,ψ1, ψ2), of the amalgam is a group, G, such that ψ1 : P1 → G and
ψ2 : P2 → G are homomorphisms with G = 〈ψ1(P1), ψ2(P2)〉 and such that φ1ψ1 = φ2ψ2.
In [7], Delgado and Stellmacher first defined weak BN -pairs of rank two. They described
a situation when a rank two amalgam (P1, P2, C) satisfied some local conditions notably
that Op(Pi) (i = 1, 2 and p a fixed prime) admitted an action from Op′(Pi)/Op(Pi) ∼=
PSL2(pni), SL2(p
ni),U3(pni), SU3(p
ni), Sz(2ni) or Dih(10) (and p = 2) or Ree(3ni),Ree(3)′
(and p = 3). They classified the possible types of amalgam and described each one. In
[23], Parker and Rowley examined the BN -pairs for which the characteristic prime, p,
is odd. They characterized finite faithful completions satisfying some p-local conditions
under a K-proper hypothesis. In this thesis we are interested in the weak BN -pairs
(P1, P2, C) such that, for i = 1 or 2, O3(Pi) admits an action from SL2(3) of which there
1
are four types. One of these, the exceptional weak BN -pair of type F3, has recently been
studied by Rachel Fowler to characterize the Thompson sporadic simple group [10]. Her
characterization relies on a K-proper hypothesis to recognize a section of the Thompson
group which is isomorphic to G2(3). In this thesis we will describe the remaining three
weak BN -pairs which have types PSL3(3), G2(3) and PSp4(3). Furthermore, we will
characterize the groups PSL3(3) and M12; G2(3); and PSp4(3) as finite faithful completions
of the respective amalgams. Notably, the characterization of G2(3) will be used in [11] to
strengthen the characterization of the Thompson sporadic simple group thereby removing
any reliance on the K-proper hypothesis.
We will examine the amalgams by studying the local group theoretic structure of the
groups which constitute the amalgam. Therefore we begin in Chapter 1 with some well
known group theoretic results which will be referred to throughout this thesis. In Chapter
2 we discuss some representation theory. In particular, we describe the SL2(3)-modules
which will appear within the group theoretic structure of the groups in the amalgams. In
Chapter 3 we will formally define an amalgam and a completion of an amalgam. We will
also associate a certain graph, called the coset graph, to a completion of an amalgam. The
coset graphs associated to G2(3) and PSp4(3) are Moufang Polygons (these are defined
and classified in [29]) and so in Chapter 4 we define such graphs and state the classification
theorem of Tits and Weiss. Also in Chapter 4, we present a character theoretic result due
to Feit and Thompson ([9]) which describes a finite group containing a self-centralizing
element of order three. We also discuss a theorem of the same nature by Smith and
Tyrer ([27]). Their theorem describes groups in which the automizer of an abelian Sylow
p-subgroup (odd prime p) has order just two. We will use both results several times in
the group recognition theorems in later chapters.
In Chapter 5 we begin to prove the main theorems of this thesis. The main hypothesis
and theorem in Chapter 5 are as follows.
2
Hypothesis A. Let G be a finite group with subgroups A ∼= B such that C := A ∩
B contains a Sylow 3-subgroup of both A and B and such that no non-trivial normal
subgroup of C is normal in both A and B. Suppose further that
(i) G = 〈A,B〉;
(ii) A/O3(A) ∼= B/O3(B) ∼= GL2(3);
(iii) O3(A) and O3(B) are natural modules with respect to the actions of A/O3(A) and
B/O3(B) respectively; and
(iv) for S ∈ Syl3(C), NG(Z(S)) = NA(Z(S)) = NB(Z(S)).
Theorem A. If G satisfies Hypothesis A, then G ∼= M12 or PSL3(3).
We may refer to the amalgam (A,B,C) as an amalgam of type PSL3(3). This amalgam is
particularly interesting as it completes to PSL3(3) and also to the sporadic simple group
M12 under the same 3-local assumption, namely condition (iv) of the hypothesis.
The main theorem and hypothesis in Chapter 6 are:
Hypothesis B. Let G be a finite group with non-conjugate subgroups A1 and A2 such
that A12 := A1 ∩ A2 contains a Sylow 3-subgroup of both A1 and A2, and such that no
non-trivial normal subgroup of A12 is normal in both A1 and A2. Suppose further that,
for i = 1, 2,
(i) |O3(Ai)| = 35;
(ii) Ai/O3(Ai) ∼= GL2(3);
(iii) O3(O3(Ai))/Z(Ai) and Z(O3(Ai))/O3(Ai)′ are natural modules with respect to the
action of Ai/O3(Ai); and
(iv) NG(O3(Ai)′) = Ai.
3
Theorem B. Let G be a group satisfying Hypothesis B. Then G ∼= G2(3).
In some sense this is the strongest amalgam theoretic result we prove since we begin
with a group containing a completion of the amalgam, as opposed to Theorem A which
characterizes just the completion itself.
The methods successfully used in Chapter 6 are extended in Chapter 7 where we
describe an amalgam of type PSp4(3). The hypothesis we assume is strong and so the
theorem is less successful than other characterizations of PSp4(3) (see for example [24]
and [16]).
Hypothesis C. Let A1 and A2 be subgroups of a finite group G = 〈A1, A2〉 such that
A12 := A1∩A2 contains a Sylow 3-subgroup of both A1 and A2 and such that no non-trivial
normal subgroup of A12 is normal in both A1 and A2. Suppose further that
(i) A1/O3(A1) ∼= Sym(4);
(ii) A2/O3(A2) ∼= SL2(3);
(iii) O3(A1) is elementary abelian of order 33 and is a faithful A1/O3(A1)-module;
(iv) O3(A2) is extraspecial of order 33 and O3(A2)/Z(O3(A2)), is a natural A2/O3(A2)-
module;
(v) NG(O3(A1)) = A1 and NG(Z(O3(A2))) = A2; and
(vi) O3(A1) normalizes no non-trivial 3′-subgroup of G.
Theorem C. Let G be a group satisfying Hypothesis C. Then G ∼= PSp4(3).
Notably condition (vi) of the hypothesis makes this theorem less successful than Theo-
rem B. However, it is possible to construct groups which have shape 26 : PSp4(3) and
4
56 : PSp4(3) which satisfy the 3-local condition (v). Certainly condition (vi) in the hy-
pothesis rules out such groups however a preferable alternative hypothesis might be to
assume O3′(G) = 1. At present it is unclear whether this alternative hypothesis would be
sufficient. The uncertainty about such groups does however raise an interesting question:
for which finite groups G does there exist 1 6= N E G such that G/N ∼= PSp4(3)? Higman
answered a similar question about SL2(2n) in [17] and so Chapter 8 together with much
of the representation theory in Chapter 2 is devoted to understanding Higman’s result.
An open question is whether the methods from Chapter 8 can be extended to PSp4(3).
We include also a short appendix which contains some elementary number theoretic
proofs concerning orders of fields. The appendix also contains a proof related to the coset
graph in Chapter 6.
Before we begin we note that all groups will be finite and all vector spaces will be
finite dimensional. We use Atlas [6] notation for groups and group extensions with the
exception of the alternating, symmetric and dihedral groups which are labeled Alt(n),
Sym(n) and Dih(n) respectively.
5
Chapter 1
Some Group Theory
1.1 Some General Group Theoretic Results
Lemma 1.1. Let p be a prime and let A be a p-group acting on a non-trivial p-group V .
Then CV (A) 6= 1 and [V,A] < V . In particular if V is an FA-module where F is a field
of characteristic p then CV (A) is a non-trivial submodule and [V,A] is a proper submodule
of V .
Proof. If we form the semidirect product S := V : A then S is a p-group with normal sub-
group V . Thus, S has non-trivial centre and Z(S)∩V 6= 1. Hence CV (A) 6= 1. Now since
S is a finite p-group, S is nilpotent and so for some integer n, 1 = [S, S, n] = [S, S, . . . , S]
(n’th term in lower central series). So suppose [V,A] = V then V = [V,A,A, . . . , A] =
[V,A, n] 6 [S, S, n] = 1 which is a contradiction.
A group, G, is said to act equivalently on two sets Ω1 and Ω2 if there is a bijection
α : Ω1 → Ω2 such that for ω ∈ Ω1, g ∈ G, α(ω · g) = α(ω) · g.
Lemma 1.2. Suppose G is a group with subgroups A and B which normalize each other.
Suppose further that both are acted on by another group X. Then the action of X on
6
AB/A is equivalent to the action of X on B/(A ∩ B). In particular X commutes with
AB/A if and only if X commutes with B/(A ∩B).
Proof. Let α be the natural isomorphism α : AB/A → B/(A ∩ B) then for x ∈ X and
Ab ∈ AB/A,
α(Abx) = α(A(bx)) = (A ∩B)bx = ((A ∩B)b)x = α(Ab)x
and so the action is equivalent.
Lemma 1.3. Let G be a group and let a, b, c ∈ G. The following identities hold.
(i) [a, bc] = [a, c][a, b]c.
(ii) [ab, c] = [a, c]b[b, c].
(iii) If p is a prime and if G is a p-group then [[a, b], c][[b, c], a][[c, a], b] ∈ [G,G,G,G].
Proof. The first two identities are easy to check the third is Lemma 5.6.1 (iv) in [13,
p209].
Definition 1.4. Let A be a group acting on a group G. The action of A on G is coprime
if |A| and |G| are coprime.
Theorem 1.5. (Coprime Action) Suppose A is a group acting on the group G and sup-
pose the action of A on G is coprime. The following hold.
(i) G = [G,A]CG(A) and if G is abelian G = [G,A]× CG(A).
(ii) [G,A] = [G,A,A].
(iii) CG/N(A) = CG(A)N/N for any A-invariant N E G.
7
(iv) If A is an elementary abelian p-group of order at least p2 then G = 〈CG(a)|a ∈
A#〉 = 〈CG(A1)|[A : A1] = p〉.
Proof. For (i) and (ii) see [20, 8.2.7 p187]. For (iii) see [20, 8.2.2 p184]. For (iv) see [20,
8.3.4 p193].
A group G is said to have a normal p-complement if Op′(G) = Op(G) or equivalently if
G has a normal subgroup N such that G is the semidirect product of N with some Sylow
p-subgroup.
Theorem 1.6. (Burnside) Let P be a Sylow p-subgroup of a group G. Suppose NG(P ) =
CG(P ). Then G has a normal p-complement.
Proof. See [20, p169].
Lemma 1.7 (Frattini Argument). Let K be a normal subgroup of a group G and let
P be a Sylow p-subgroup of K (for some prime p) then G = KNG(P ).
Let K be a normal subgroup of a group G and let R 6 G. Then G = NG(R)K if and
only if RG = rg|r ∈ R, g ∈ G = rk|r ∈ R, k ∈ K = RK.
Proof. The first result is a special case of the second since if P is a Sylow p-subgroup
of K E G then for any g ∈ G, P g 6 K implies P g = P k for some k ∈ K by Sylow’s
Theorem. So we prove the second result.
Suppose RG = RK . For any g ∈ G there exists k ∈ K such that Rg = Rk. Therefore
gk−1 ∈ NG(R) and so g ∈ NG(R)k. Hence G = NG(R)K. Now suppose G = NG(R)K
and suppose RK ⊂ RG. Then there exists g ∈ G such that Rg * RK . However g = nk
where n ∈ NG(R) and k ∈ K and so Rg = Rnk = Rk ⊆ RK which is a contradiction.
Theorem 1.8. (Gaschutz) Let p be a prime and V an abelian normal p-subgroup of a
group G. Let V 6 P ∈ Sylp(G). Then G splits over V if and only if P splits over V .
8
Proof. See [1, 10.4 p31].
Lemma 1.9. (Burnside) Let G be a group and let P be a Sylow p-subgroup of G. Suppose
A1 and A2 are normal subsets of P which are conjugate in G. Then they are already
conjugate in NG(P ).
Proof. See [20, 7.1.5 p167].
Definition 1.10. Let G be a group, p be a prime and S ∈ Sylp(G). Let A(S) be the
set of abelian subgroups of S of order pr where pr is as large as possible. The Thompson
subgroup of S is J(S) = 〈A|A ∈ A(S)〉.
Lemma 1.11. Let G be a group, p be a prime and S ∈ Sylp(G). Suppose J(S) = Z(J(S))
and suppose a, b ∈ J(S) are conjugate in G. Then a and b are conjugate in NG(J(S)).
Proof. Suppose ag = b for some g ∈ G. Notice first that it follows immediately from the
definition of the Thompson subgroup that J(S)g = J(Sg). Now J(S), J(Sg) 6 CG(b).
Let P,Q ∈ Sylp(CG(b)) such that J(S) 6 P and J(Sg) 6 Q. Again, by the definition of
the Thompson subgroup, it is clear that J(S) 6 P implies J(S) = J(P ) and similarly
J(Sg) = J(Q). By Sylow’s Theorem, there exists x ∈ CG(b) such that Qx = P and so
J(S) = J(P ) = J(Q)x = J(S)gx. Thus gx ∈ NG(J(S)) and agx = bx = b as required.
Definition 1.12. The Fitting subgroup of a group G is the group, F (G), generated by
all nilpotent normal subgroups in G
Theorem 1.13. Let G be a group and F (G) the Fitting subgroup. Then the following
hold.
(i) F (G) is the unique maximal nilpotent normal subgroup of G.
(ii) If G is soluble then CG(F (G)) 6 F (G).
Proof. See [13, Thm6.1.2, Thm6.1.3 p218].
9
1.2 Extraspecial Groups
Definition 1.14. Let p be a prime and let E be a p-group. If E ′ = Z(E) = Φ(E) is
cyclic of order p then E is said to be extraspecial.
Lemma 1.15. Let E be an extraspecial p-group. Exactly one of the following holds.
(i) p = 2 and E is a central product of n copies of Dih(8).
(ii) p = 2 and E is a central product of n− 1 copies of Dih(8) and one copy of Q8.
(iii) p 6= 2 and E has exponent p.
(iv) p 6= 2 and E has exponent p2.
We denote such groups as E ∼= 21+2n+ , 21+2n
− , p1+2n+ and p1+2n
− respectively.
Proof. See [8, Thm 20.5].
It is well known that Dih(8)∗Dih(8) ∼= Q8 ∗Q8 so the description of extraspecial 2-groups
given here is not unique.
Theorem 1.16. Suppose E is an extraspecial p-group of order p2n+1.
(i) If p = 2 and E ∼= 21+2n+ then Out(E) ∼= O+
2n(2).
(ii) If p = 2 and E ∼= 21+2n− then Out(E) ∼= O−
2n(2).
(iii) If p is odd and E ∼= p1+2n+ then Out(E) ∼= Sp2n(p)Cp−1.
(iv) If p is odd and E ∼= p1+2n− then Out(E) ∼= p2n−1
+ Sp2n−2(p)Cp−1.
Proof. See [15].
10
Corollary 1.17. Let E be an extraspecial 2-group of order 25. If E admits a group of
automorphisms of order nine then E ∼= 21+4+ .
Proof. This is clear since 9 does not divide |O−4 (2)| = 120.
Lemma 1.18. The extraspecial 2-group 21+4+ contains exactly two subgroups isomorphic
to Q8 and they commute.
Proof. Firstly 21+4+
∼= Dih(8) ∗ Dih(8) ∼= Q8 ∗ Q8 so there are at least two subgroups
isomorphic to Q8 and we need to prove there cannot be more. We let Q1 and Q2 be the
two commuting quaternion subgroups and count elements of order four in Q := 〈Q1, Q2〉.
Clearly we see six elements of order four in Q1 and six more in Q2. So suppose qi ∈ Qi
for i = 1, 2 and q1q2 is another element of order four. If q1 has order two then it is central
in Q1 and then q1q2 ∈ Q2. Therefore q1, and in the same way q2, is an element of order
four and so both q1 and q2 square to the same element in z ∈ Z(Q) ∼= C2. However, this
means (q1q2)2 = q2
1q22 = z2 = 1 which is a contradiction. Therefore there are exactly 12
elements of order four in Q. So if Q3 6 Q is a subgroup isomorphic to Q8 then Q3 must
contain an element of order four from Q1 and an element of order four from Q2. However
these two elements commute so Q3 cannot be isomorphic to Q8.
1.3 Fixed-Point-Free Actions
An automorphism of a group G is said to be fixed-point-free if it leaves only the identity
of G fixed. The following famous theorem was proved by Thompson and says that groups
admitting such an action from an automorphism of prime order are nilpotent.
Theorem 1.19. (Thompson) If G is a group which admits a fixed-point-free automor-
phism of prime order then G is nilpotent.
Proof. See [13, Thm 2.1 p337].
11
Lemma 1.20. Let φ be a fixed-point-free automorphism of order two of a group G. Then
φ inverts every element of G and G is abelian.
Proof. Define
θ : G −→ G
g 7−→ g−1φ(g).
Then θ is injective since x−1φ(x) = y−1φ(y) if and only if yx−1 = φ(y)φ(x)−1 = φ(yx−1)
if and only if yx−1 = 1 if and only if x = y. Also θ is onto since G is finite. So for each
g ∈ G there is some h ∈ G such that g = h−1φ(h) therefore φ(g) = φ(h−1)h = g−1. So φ
inverts every element of G. Let x, y ∈ G. Then
x−1y−1 = φ(x)φ(y) = φ(xy) = (xy)−1 = y−1x−1
and so xy = yx and G is abelian.
The following lemma and its proof generalizes a well known result which dates back
to Burnside the proof of which can be found in [17, Thm 8.1]. We will use the notation
a ≡ b mod H when Ha = Hb where a, b ∈ G > H.
Lemma 1.21. Suppose that p is a prime and Q is a p-group. Furthermore, suppose that
z ∈ Aut(Q) has order three and CQ(z) = 〈t〉 where t ∈ Z(Q) has prime order. Then Q
is nilpotent of class at most three. In particular [Q,Q,Q] 6 〈t〉 and for any a, b ∈ Q,
[a, bz] ≡ [az, b] ≡ [a, b]z2
mod 〈t〉.
Proof. Let Q = Q1 > Q2 > Q3 > . . . be the lower central series of Q. Then for each i,
〈t〉Qi/Qi+1 is a central subgroup of Q/Qi+1 which is acted on by z. Moreover, z can only
centralize the subgroup 〈t〉Qi+1/Qi+1.
For any a ∈ Q, aazaz2is a fixed point modulo Q2 so
aazaz2 ≡ 1 mod 〈t〉Q2.
12
It follows from the commutator relations (Lemma 1.3) and from Q2/Q3 being abelian that
[a, b][az, b][az2, b] ≡ [aazaz2
, b] ≡ 1 mod Q3 and [a, b][a, bz][a, bz2] ≡ [a, bbzbz
2] ≡ 1 mod Q3
for any a, b ∈ Q. Therefore
[a, b] ≡ [b, az][b, az2
] ≡ [bz, a][bz2
, a] mod Q3.
Rearranging gives
[a, bz][b, az] ≡ [az2
, b][bz2
, a] mod Q3.
Therefore [a, bz][b, az]Q3 is a z-invariant element of Q2/Q3 and so must be in 〈t〉Q3/Q3.
Hence
[a, bz] ≡ [az, b] mod 〈t〉Q3. (1.1)
Now we repeat these arguments modulo Q4. For a, b ∈ Q the element [a, b][a, b]z[a, b]z2
is
a fixed point modulo Q3 and so
[a, b][a, b]z[a, b]z2 ≡ 1 mod 〈t〉Q3.
Using the commutator relation again and since Q3/Q4 is abelian, for a, b, c ∈ Q,
[[a, b], c][[a, b]z, c][[a, b]z2
, c] ≡ [[a, b][a, b]z[a, b]z2
, c] ≡ 1 mod Q4.
Also,
[[a, b], c][[a, b], cz][[a, b], cz2
] ≡ [[a, b], cczcz2
] ≡ 1 mod Q4.
Therefore
[[a, b], c] ≡ [c, [a, b]z2
][c, [a, b]z] ≡ [cz, [a, b]][cz2
, [a, b]] mod Q4.
13
Rearranging gives
[c, [a, b]z2
][[a, b], cz2
] ≡ [cz, [a, b]][[a, b]z, c] mod Q4.
The corresponding element modulo Q4 is z-invariant and therefore must be in 〈t〉Q4/Q4
and so
[[a, b]z2
, c] ≡ [[a, b], cz2
] mod 〈t〉Q4,
and in a similar way
[[a, b]z2
, c] ≡ [[a, b], cz2
] ≡ [[a, b], c]z ≡ [[az, b], c] ≡ [[a, bz], c] mod 〈t〉Q4. (1.2)
Now Lemma 1.3 (iii) gives
[[a, b], c][[b, c], a][[c, a], b] ≡ 1 mod Q4 (1.3)
and
[[az, b], c][[b, c], az][[c, az], b] ≡ 1 mod Q4. (1.4)
Conjugating (1.3) by z and applying (1.2) gives
[[az, b], c][[b, c], az2
][[c, az], b] ≡ 1 mod 〈t〉Q4
which together with (1.4) gives
[[b, c], az2
][[c, az], b] ≡ [[b, c], az][[c, az], b] mod 〈t〉Q4.
Hence [[b, c], az2] ≡ [[b, c], az] mod 〈t〉Q4. Finally, using (1.2) we get
[[b, c], a]z ≡ [[b, c], az2
] ≡ [[b, c], az] ≡ [[b, c], a]z2
mod 〈t〉Q4
14
is a fixed point mod Q4 which gives [[b, c], a] ∈ 〈t〉Q4 for any a, b, c ∈ Q, however this
implies that Q3 6 〈t〉Q4 6 〈t〉Q3. Since 〈t〉 has prime order, either 〈t〉Q3 = 〈t〉Q4 and
then Q3 = Q4 (so Q has nilpotence class two), else, Q3 = 〈t〉Q4. In this second case we
get Q5 = [Q4, Q] = [〈t〉Q3, Q] = [Q3, Q] = Q4 and so Q4 = 1 and Q3 = 〈t〉. In either
case Equation 1.1 becomes [a, bz] ≡ [az, b] mod 〈t〉 for any a, b ∈ Q and applying this
congruence again replacing a and b with az2and bz
2gives [a, bz] ≡ [a, b]z
2mod 〈t〉.
Corollary 1.22. Let Q be a group and let z ∈ Aut(Q) have order three and CQ(z) = 〈t〉
where t ∈ Z(Q) has prime order. Then Q is nilpotent of class at most three.
Proof. By Theorem 1.19, Q/〈t〉 is nilpotent and so is a direct product of its Sylow p-
subgroups. Since 〈t〉 is central of prime order, Q is nilpotent and a direct product of its
Sylow subgroups. Moreover each Sylow subgroup is nilpotent of class less or equal to
three and so Q is nilpotent of class less or equal to three.
Corollary 1.23. Suppose Q is a 2-group with z ∈ Aut(Q) of order 3 and CQ(z) =
〈t〉 ∼= C2. If Q = 〈X, Y 〉 with X ∼= Y ∼= Q8 and Z(X) = Z(Y ) = 〈t〉 then [x, yz] ≡
[xz, y] mod 〈t〉 for all x ∈ X and all y ∈ Y .
In fact such a group Q has order at most 27 and there exist two isomorphism classes
of groups of this order admitting such an automorphism. They can be constructed in
Magma (see [4]) by deriving possible relations from the congruence [a, bz] ≡ [az, b] mod 〈t〉
for all a, b ∈ Q.
Lemma 1.24. (Burnside) Suppose that Q is a group and suppose that z ∈ Aut(Q) has
order three and acts fixed-point-freely on Q\1. Then Q is nilpotent of class at most two.
Proof. Apply Corollary 1.22 with t = 1.
Lemma 1.25. Suppose S is a 2-group and A,B are fours groups with A 6= B and S =
〈A,B〉. If z ∈ Aut(S) has order 3, A and B are z-invariant and CS(z) = 1, then either
S is elementary abelian of order 24 or special of order 26 with |Z(S)| = 22.
15
Proof. By Lemma 1.24, S has nilpotence class at most two. If S is abelian then S has
order 24 so assume not then S has class two. Thus commutators are central giving us
that S ′ = 〈[a, b]|a ∈ A, b ∈ B〉 and for every a ∈ A, b ∈ B,
1 = [[a, b], b] = (baba)b(abab)b = (baba)(baba) = [b, a]2,
and so S ′ is elementary abelian. Since both A and B are z-invariant, we can write A\1 =
a0, a1, a2 where a1 = az0 and a2 = az2
0 and similarly we label B\1 = b0, b1, b2. Thus,
for i, j ∈ Z3, [ai, bj][ai+1, bj+1][ai+2, bj+2] ∈ CS(z) = 1. And so
[ai, bj][ai+1, bj+1] = [ai+2, bj+2]
= [aiai+1, bjbj+1]
= [ai, bj][ai, bj+1][ai+1, bj][ai+1, bj+1].
Hence for every i, j ∈ Z3,
[ai, bj+1] = [ai+1, bj],
and so S ′ = 〈[a1, b1], [a1, b2], [a1, b0]〉. But now [a1, b1][a1, b2] = [a1, b0] and so S ′ has
order at most 22. Since S ′ is z-invariant, |S ′| = 22. Now A and B do not commute
and so A ∩ S ′ 6 A ∩ Z(S) = 1 and similarly B ∩ S ′ = 1 (the intersections could not
have order two because they are z-invariant). Therefore AS ′/S ′ ∼= BS ′/S ′ ∼= 2 × 2
and so 〈A,B〉S′ = ( A
S′ )(BS′ ) ∼= 2 × 2 × 2 × 2 and so S has order 26. Now suppose S ′ <
Z(S). Since S 6= Z(S) and Z(S) is z-invariant, Z(S) must have order 24. However,
Z(S) ∩ A = 1 so S = Z(S)A but this is a contradiction since S ′ = [Z(S)A,Z(S)A] =
[Z(S), Z(S)][A,Z(S)][A,A][Z(S), A] = 1 (see Lemma 1.3).
16
1.4 Strongly Embedded Subgroups
Definition 1.26. Let G > E be groups of even order. We say E is a strongly embedded
subgroup of G if E ∩ Eg has odd order for every g ∈ G\E.
We immediately see that a strongly embedded subgroup is necessarily self-normalizing.
The following arguments originate from Bender (see [3]) and can, in particular, be
found in Aschbacher’s Sporadic Groups [2, Lemma 7.6]. Let G and E be groups of even
order with E a strongly embedded subgroup of G. Let I be the set of involutions in G
and let Ω be the set of conjugates of E in G. We consider the action of I on Ω.
Lemma 1.27. Each i ∈ I fixes precisely one element of Ω.
Proof. Let A ∈ Ω then A is a conjugate of E and so has even order. Therefore there
exists some u ∈ I ∩A and so Au = A. Suppose Bu = B for some B ∈ Ω\A. Since B is
strongly embedded, B is self-normalizing and so u ∈ B. Therefore u ∈ B ∩ A. However
A ∩ B has odd and so u ∈ I ∩ A can fix no other element of Ω. Also, since u is an
involution, u must swap pairs in Ω and so Ω has odd order. Now, since Ω has odd order,
any involution acts on Ω and fixes at least one element. So suppose v ∈ I and Cv = C
for some C ∈ Ω. As we have seen, C is self normalizing and so v ∈ I ∩C and then v fixes
precisely one element of Ω.
Lemma 1.28. Let E 6= Eu ∈ Ω. Any involution swapping E and Eu lies in the coset
u(E ∩ Eu).
Proof. Let w be such an involution. Then Ew = Eu and so uw ∈ E. Also, (uw)−1 =
wu = (uw)u ∈ Eu and so uw ∈ E ∩ Eu. Therefore w ∈ u(E ∩ Eu).
Lemma 1.29. Let E 6= Eu ∈ Ω and set K = E ∩ Eu. The number of involutions in I
that swap E and Eu is equal to |J | where J = k ∈ K|ku = k−1.
17
Proof. Any such involution, v say, which swaps E and Eu lies in the coset u(E∩Eu) = uK.
So let v = uk for k ∈ K. Then 1 = v2 = kuk and so ku = k−1 and then k ∈ J
and v = uk ∈ uJ . Of course any element of uJ is an involution and swaps E and
Eu. Therefore uJ is the set of all involutions in uK and so |J | = |uJ | is the number of
involutions swapping E and Eu.
Theorem 1.30. Let G be a group and E a strongly embedded subgroup of G. If for some
involution t ∈ E, CG(t) = CE(t) intersects trivially with every distinct conjugate of E,
then G contains an abelian subgroup K such that E = KCG(t) and K ∩ CG(t) = 1.
Proof. Set n = [G : E]. Count the triples (v, A,B) such that v ∈ I and A,B ∈ Ω
with Av = B. Let T be the set of such triples. By Lemma 1.27, each involution fixes a
unique element of Ω and permutes n−12
pairs so |T | = |I| (n−1)2
> [G : CG(t)] (n−1)2
= [E :
CE(t)]n(n−1)2
.
Now we choose an involution u ∈ I such that |I ∩ 〈u〉(E ∩ Eu)| is maximal. Equiva-
lently we have chosen two conjugates of E which are permuted by a maximal number of
involutions. As before set K = E ∩ Eu and J = k ∈ K|ku = k−1.
Now |T | =∑
A,B∈ΩMAB where MAB is the number of involutions swapping A and B.
So MAB 6 ME,Eu = |J |. Thus |T | 6(
n2
)ME,Eu = n(n−1)
2|J |. So [E : CE(t)] 6 |J |. Now
CG(t) = CE(t) intersects trivially with every distinct conjugate of E and so CG(t)∩K = 1.
Therefore every element of K lies in a unique coset of CG(t). There are [E : CE(t)] cosets
of CE(t) in E and so this number is greater than |CE(t)k|k ∈ K| = |K|. Hence
|J | 6 |K| 6 [E : CG(t)]. So we have that
|J |n(n− 1)
26 |K|n(n− 1)
26 [E : CG(t)]
n(n− 1)
26 |T | = |I|(n− 1)
26n(n− 1)
2|J |.
Hence J = K and |I| = n[E : CG(t)] = [G : CG(t)]. Furthermore, since u inverts every
element of J = K, K is an abelian subgroup by Lemma 1.20 and from the group orders
18
we get E = KCG(t).
1.5 Automorphisms of Abelian Groups
LetA be an additive abelian group that is isomorphic to a direct product ofm cyclic groups
of order q where q = pn for some prime p. Consider the ring of endomorphisms of A,
End(A). This is the set of group homomorphisms from A to itself with ring multiplication
given by composition of maps and addition given by (θ + ϕ)(a) := θ(a) + ϕ(a) for all
homomorphisms θ and ϕ and all a ∈ A. We will identify this endomorphism ring with
a ring of matrices and then determine which matrices define automorphisms of A. A
reference for this section is [25]. From now on we will view A as a direct sum of m copies
of the integers modulo q. So any a ∈ A has the form a = (a1, a2, . . . , am) where each ai
represents an equivalence class of integers modulo q and ai an integer representative of the
class. Let π be the natural group homomorphism π : Zm → A given by π(a1, a2, . . . , am) 7→
(a1, a2, . . . , am). Consider the map
ψq : Mm(Z) −→ End(A) = End(Zmq )
M 7→ ψq(M),
where ψq(M) : (a1, a2, . . . , am) 7→ π((a1, a2, . . . , am)M).
Lemma 1.31. The map ψq is a surjective ring homomorphism.
Proof. We first show each ψq(M) is a well defined endomorphism ofA. Suppose (a1 . . . , am)
and (b1, . . . , bm) satisfy π(a1 . . . , am) = π(b1, . . . , bm). Then q|ai − bi for each 1 6 i 6 m.
19
Thus
ψq(M)(a1 . . . , am)− ψq(M)(b1 . . . , bm) = π((a1 . . . , am)M)− π((b1, . . . , bm)M)
= π(Σaimi1, . . . ,Σaimim)− π(Σbimi1, . . . ,Σbimim)
= π(Σ(ai − bi)mi1, . . . ,Σ(ai − bi)mim)
= 0A.
So ψq(M) is well defined and also clearly a group homomorphism of A as π is linear and
matrix multiplication is distributive. So ψq is a map into End(A) and it remains to prove
ψq is a surjective homomorphism. Again it is clear that ψq is a ring homomorphism as π is
linear and matrix multiplication is distributive and associative. Now any endomorphism
of A is determined by where it maps each wi := (0, . . . , 1, . . . 0), the vector whose only
non-zero entry is a 1 in the i’th position. So suppose θ ∈ End(A) and θ(wi) = (a1, . . . , am).
Then consider a matrix M = (mij) where mij = aj for each j 6 m. Do this for each
i 6 m and it follows that θ = ψq(M). Thus ψq is a surjective ring homomorphism.
Lemma 1.32. The kernel of ψq is the set K := M = (mij) ∈Mn(Z)| q|mij ∀i, j.
Proof. It is clear that for all k ∈ K and each wi := (0, . . . , 1, . . . 0) defined as before,
ψq(k)(wi) = 0A. Since A is generated by the wi’s, K ⊆ Ker(ψq). However if we suppose
that some matrix, M = (Mij), is in the kernel then ψq(M)(wi) = 0A. Therefore q divides
mij for each j. We do this for each i to see that q divides each entry in M and so M ∈ K.
Lemma 1.33. θ = ψq(M) ∈ End(A) is an automorphism of A if and only if det(M) is
prime to p.
Proof. Suppose M ∈ Mm(Z) has determinant prime to p. Consider the cofactor matrix,
N , of M . This is the matrix which satisfies MN t = det(M)I. Since M has integer entries,
so does N . Since det(M) is prime to p, we can find s ∈ Z such that sdet(M) ≡ 1 mod q.
Set ψ−1M := ψ(sN) and we see that ψM ψsNt = ψsMNt = ψsdet(M)I = 1End(A) and so ψM
is invertible.
20
Conversely suppose M ∈Mm(Z) with ψM an invertible endomorphism of A. Suppose
ψM−1 = ψN for some N ∈Mm(Z). Then
ψ(MN − I) = ψ(MN)− ψ(I) = ψMψN − 1End(A) = 0End(A).
Thus MN − I ∈ Ker(ψ) and so q divides every entry of MN − I and so the entries of
MN are equal to the entries of I modulo p. Thus
1 ≡ det(MN) ≡ det(M)det(N) mod p.
Hence det(M) is prime to p.
Lemma 1.34. Set GLm(Zq) to be the subset of Mm(Zq) containing matrices with deter-
minant prime to p. Then GLm(Zq) ∼= Aut(A) and has order p(n−1)m2|GLm(p)|.
Proof. The set GLm(Zq) is a group under matrix multiplication. Consider the map,
Ψ : GLm(Zq) −→ Aut(A),
where each (aij) ∈ GLm(Zq) Ψq((aij)). This map is a well defined homomorphism as Ψq
is and non-zero since for each such (aij), the matrix (aij) is not in Ker(Ψq). It is also
injective since if (aij), (bij) ∈ GLm(Zq) with Ψq((aij)) = Ψq((bij)) then aij − bij ∈ Ker(Ψq)
and then q|(aij − bij) for each i, j and so (aij) = (bij). Now, for any automorphism of
A we can find M = (mij) ∈ Mn(Z) such that M has determinant prime to p. Therefore
(mij) ∈ GLm(Zq) and so Ψ is surjective.
Consider the map Θ : GLm(Zq) → GLm(Zp) where for each M ∈ GLm(Zq) we restrict
the matrix entries modulo p. Then Θ(M) has non-zero determinant modulo p and so is
an element of GLm(p). Clearly Θ is a surjective group homomorphism. The kernel is the
set of matrices (mij)|mii ≡ 1 mod p ∀i and mij ≡ 0 mod p ∀i 6= j. So there are pn−1
21
choices for each entry and so p(n−1)m2elements in the kernel. Thus GLm(Zq) ∼= Aut(A)
has order p(n−1)m2|GLm(p)|.
We have seen that if A = Zmq then elements of Aut(A) can be viewed as m by m
matrices with elements in Zq which have determinant prime to p. Notice GLm(Zq)
acts on the vector space V := A/Ωn−1(A) ∼= Zmp and so there is an homomorphism
GLm(Zq) → GLm(Zp) = Aut(Zmp ). Moreover this homomorphism simply restricts matrix
entries modulo p.
Lemma 1.35. Let p be a prime and set q = pn. Let G be a group acting on an abelian
group A ∼= Zmq . Then there are homomorphisms
α : G −→ GLm(Zq),
β : G −→ GLm(Zp)
for which entries of α(g) are congruent modulo p to entries of β(g) for each g ∈ G. Fur-
thermore β is a representation affording the module A/Ωn−1(A) and if β1 : G→ GLm(Zp)
is an equivalent representation then there is a homomorphism α1 : G → GLm(Zq) for
which entries of α1(g) are congruent modulo p to entries of β1(g) for each g ∈ G.
Proof. Since G acts on A, there is a homomorphism, α, from G to Aut(A) ∼= GLm(Zq)
as described previously. Moreover we have seen that restricting matrix entries modulo p
describes the action of G on the quotient group A/Ωn−1(A) ∼= Zmp and so we have the
required representation β. So suppose β1 : G→ GLm(Zp) is an equivalent representation
then there is a matrix M ∈ GLm(Zp) such that for each g ∈ G, β1(g) = M−1β(g)M .
Since M is a matrix whose elements are integers modulo p and whose determinant is
not a multiple of p, we can view M as a matrix in GLm(Zq) simply by viewing each
integer modulo p as the same integer modulo q. Now it makes sense to consider the
homomorphism α1 : G → GLm(Zq) where α1(g) = M−1α(g)M and so entries of α1 are
congruent modulo p to entries of β1(g) as required.
22
Chapter 2
Some Representation Theory
In this chapter we discuss endomorphism rings, tensor products, some modular represen-
tation theory and natural modules. The results on natural GL2(3) and SL2(3)-modules
will be used throughout Chapters 5, 6 and 7. The natural SL2(2n)-modules will appear
again in Chapter 8 as will much of the work on tensor products and modular character
theory.
Theorem 2.1. (Maschke) Let G be a group and let F be a field whose characteristic does
not divide |G|. Then every FG-module is completely reducible.
Proof. See [19, 1.9 p4].
Lemma 2.2. Let p be a prime and let K = GF(pn). If H is a group with H ∼= K× then
any irreducible KH-module has dimension one.
Proof. Suppose V is an irreducible KH-module with representation ρ : H → GL(V ). Let
〈h〉 = H then h has order pn − 1 and so any eigenvalue for ρ(h) is a pn − 1’th root of
unity and so will lie in K. Thus we can find a corresponding eigenvector for ρ(h). The
eigenspace spanned by the eigenvector is 1-dimensional and H invariant and therefore a
KH-submodule and so V must have dimension one.
23
2.1 Endomorphism Rings
Definition 2.3. Let G be a group, F a field and let V and W be FG-modules. Define
HomFG(V,W ) to be the ring of all FG-homomorphisms from V to W . The endomorphism
ring of V is the ring EndFG(V ) := HomFG(V, V ).
If V is an FG-module and E := EndFG(V ) then V can be viewed as a vector space
over E by defining scalar multiplication as e · v := e(v) for e ∈ E, v ∈ V . Thus V is also
an EG-module.
Lemma 2.4. (Schur) If V is an irreducible FG-module then EndFG(V ) is a division ring.
If, furthermore, F is a finite field then EndFG(V ) is a field.
Proof. Let 0 6= φ ∈ EndFG(V ). Set W = φ(V ). Since φ is non-zero, W 6= 0. Hence for
all v ∈ V and g ∈ G
(φ(v))g = φ(vg) ∈ W
and so W is an FG-submodule. Since V is irreducible and W 6= 0, V = W . So
φ is onto and by the rank-nullity result φ is a non-singular linear transform on V and
thus invertible. Clearly φ−1 is a linear transform and vg = φ(φ−1(v))g = φ(φ−1(v)g)
and so φ−1(vg) = φ−1(v)g and φ−1 ∈ EndFG(V ). A quick check verifies EndFG(V ) is a
ring with usual addition and composition of functions and so it is a division ring. Now
Wedderburn’s Theorem says that a finite division ring is a field. So if F is a finite field
then HomFG(V, V ) is finite and hence a field.
Lemma 2.5. If V is a faithful irreducible FG-module where F is a finite field and G an
abelian group then EndFG(V ) is a field of order |V |.
Proof. By Schur’s Lemma, E := EndFG(V ) is a field. Since elements of F define endo-
morphisms of V which commute with the action of G, we see an embedding of F in E.
24
Also, as G is abelian, G embeds in the multiplicative group of E. We can view E as an
FG-module by defining scalar multiplication and an action of G simply by composition
of maps. We then define an FG-module homomorphism,
Φ : E −→ V
α 7−→ α(v),
where v is some fixed non-zero vector in V . This map is non-zero and so the kernel is an
ideal and so must be trivial. It follows that this map is a bijection and so |E| = |V |.
Lemma 2.6. Let G be a group and H 6 G. Let F be a finite field of characteristic p and
let V be an irreducible FG-module. Suppose χ : G → C is the character afforded by V
and finally suppose that p does not divide |H|. Then dim(CV (H)) = 〈χ|H , 1|H 〉.
Proof. By Maschke’s Theorem, V decomposes into irreducible H-modules, V = V1⊕ . . .⊕
Vn, and χ|H decomposes into irreducible characters for H, χ|H = χ1 + . . .+χn. The inner
product in H, 〈χ|H , 1|H 〉 gives us the number of times the trivial character appears in
the decomposition and similarly the number of 1-dimensional modules on which H acts
trivially. Hence dim(CV (H)) = 〈χ|H , 1|H 〉.
2.2 Tensor Products
See, for example, [1] for more explanation and proofs of results. Let G be a group, K a
field and F an extension field of K.
Definition 2.7. Let U and V be KG-modules. A K-balanced map θ : U × V → A,
where A is an abelian group, is a map which satisfies
kθT (u, v) = θT (ku, v) = θT (u, kv),
25
θ(u+ u′, v) = θ(u, v) + θ(u′, v)
and
θ(u, v + v′) = θ(u, v) + θ(u, v′)
for k ∈ K, u, u′ ∈ U , v, v′ ∈ V . An abelian group T is a tensor product of U and V over
K if there is a surjective K-balanced map θT : U × V → T such that if S is an abelian
group and θS : U×V → S is a surjective K-balanced map then there is a unique K-linear
homomorphism α : T → S such that αθT = θS.
We denote a tensor product of U and V by U ⊗K V (or U ⊗V if there is no ambiguity for
the field). The tensor product exists and is unique up to isomorphism and is a KG-module
when we define
(u⊗ v) · g := u · g ⊗ v · g
for u ∈ U, v ∈ V, g ∈ G.
Suppose ui|1 6 i 6 n is a basis for U overK and vj|1 6 j 6 m is a basis for V over
K. Then U⊗KV is a vector space overK with basis ui⊗Kvj|1 6 i 6 n, 1 6 j 6 m. Now
consider the field extension F/K. Since F is a vector space over K, we can form the tensor
product F ⊗K V . This gives a vector space over F when we define f1 · (f⊗v) := (f1f⊗v).
Notice that 1F ⊗ vi|1 6 i 6 n is an F -basis for F ⊗ V and so the dimension of this
space over F is equal to the dimension of V over K.
An irreducibleKG-module, V , is called absolutely irreducible if the FG-module F⊗KV
is irreducible for each extension field F of K.
Lemma 2.8. Let V be an irreducible KG-module. Then V is absolutely irreducible if and
only if EndKG(V ) ∼= K.
Proof. See [1, 25.8].
26
Lemma 2.9. Let U and V be KG-modules. Then the FG-modules (U⊗KF )⊗F (V ⊗KF )
and (U ⊗K V )⊗K F are FG-module isomorphic.
Proof. Consider the map
Θ : (U ⊗K F )× (V ⊗K F ) −→ (U ⊗K V )⊗K F
(u⊗ f, v ⊗ e) 7−→ (u⊗ v)⊗ fe.
We claim Θ is an F -balanced map. Let f, f1, f2, e ∈ F , u, u1, u2 ∈ U , v ∈ V and observe
Θ(u1 + u2 ⊗ f, v ⊗ e) = (u1 + u2 ⊗ v)⊗ fe
= (u1 ⊗ v)⊗ fe+ (u2 ⊗ v)⊗ fe
= Θ(u1 ⊗ f, v ⊗ e) + Θ(u2 ⊗ f, v ⊗ e)
Θ(u⊗ f1 + f2, v ⊗ e) = (u⊗ v)⊗ f1e+ f2e
= (u⊗ v)⊗ f1e+ (u⊗ v)⊗ f2e
= Θ(u⊗ f1, v ⊗ e) + Θ(u⊗ f2, v ⊗ e)
and linearity in the second variable holds in the same way. Also
f1Θ(u⊗ fv ⊗ e) = (u⊗ v)⊗ f1fe = Θ(u⊗ f1f, v ⊗ e) = Θ(u⊗ f, v ⊗ f1e)
= Θ(f1(u⊗ f), v ⊗ e) = Θ(u⊗ f, f1(v ⊗ e)).
So Θ is an F -balanced map. Also if g ∈ G then
Θ(u⊗f, v⊗e)g = (u⊗v)g⊗fe = (ug⊗vg)⊗fe = Θ((u⊗f)g, (v⊗e)g) = Θ((u⊗f, v⊗e)g).
Now, by the definition of the tensor product, there is an F -linear map α : (U ⊗K F )⊗F
(V ⊗K F ) → (U ⊗K V ) ⊗K F such that αθT = Θ where θT is the usual tensor product
map θT : (U ⊗K F ) × (V ⊗K F ) → (U ⊗K F ) ⊗F (V ⊗K F ). It is easy to check θT
27
commutes with the action of G and so we get that α is FG-linear. Since θT is surjective
by definition and Θ is also clearly surjective, α is surjective. Finally since the dimension
of (U ⊗K F )⊗ F (V ⊗K F ) over F equals the dimension of (U ⊗K V )⊗K F over F , α is
an FG-isomorphism.
Given a KG-module, V , and corresponding representation, ρ : G → GL(V ), we can
define a representation ρF : G → GL(F ⊗K V ) where for g ∈ G, ρF : g 7→ 1 ⊗ ρ(g) and
for f ⊗ v ∈ F ⊗K V , 1⊗ ρ(g) : f ⊗ v 7→ f ⊗ ρ(g)(v).
Lemma 2.10. Let F be an extension field of K and suppose θ : G → GLn(K) is a
representation with corresponding module V . Let ι be the inclusion map ι : GLn(K) →
GLn(F ). Then the representation θF : G→ GLn(F ) corresponding to the module F ⊗K V
is equivalent to the representation ιθ.
Proof. Fix a K-basis v1, . . . , vn for V and write each g ∈ G as a matrix θ(g) = (aij)
where θ(g)(vi) := ai1v1 + . . . + ainvn and each aij is in K 6 F . The module F ⊗K V
affords the representation θF : G → GL(F ⊗K V ) where g ∈ G maps to the linear
transformation 1 ⊗ θ(g) : f ⊗ v 7→ f ⊗ θ(g)(v). Choose a basis for F ⊗K V to be
1 ⊗ v1, . . . , 1 ⊗ vn. Then, with respect to this basis, the matrices representing G via
θF : G→ GL(F ⊗K V ) are exactly the matrices representing G via θ since θF (g)(1⊗vi) =
1⊗ θ(g)(vi) = 1⊗ ai1v1 + . . .+ ainvn = 1⊗ ai1v1 + . . .+ 1⊗ ainvn.
If V is a vector space over a field F and φ1, . . . , φn are F -linear transformations of V ,
we say φ1, . . . , φn are simultaneously diagonalizable if there is a basis for V , v1, . . . , vm,
such that for each 1 6 i 6 n and each 1 6 j 6 m, vj is an eigenvector for φi.
Lemma 2.11. Let V be a vector space over a finite field K = GF(p) and let φ1, . . . , φn
be commuting K-linear transformations of V with order prime to p. Furthermore let
F be a field extension of K that contains all eigenvalues for φ1, . . . , φn. Then the linear
transformations of F⊗K V , 1⊗K φ1, 1⊗K . . . , 1⊗K φn, can be simultaneously diagonalized.
28
Proof. Each φi has order, ri say, prime to p and so has eigenvalues which are all ri’th
roots of unity and so there exists some extension field, F > K, which contains every
eigenvalue for every φi. Let V F = F ⊗K V and let φFi := 1 ⊗K φi : f ⊗ v 7→ f ⊗
φi(v) be the corresponding linear transformations. Notice that the linear transformations,
φF1 , . . . , φ
Fn , all commute. Find the eigenspaces for some φ := φF
i . Let ε1, . . . , εm ∈ F
be the eigenvalues for φ and Eig(ε1), . . . ,Eig(εm) the corresponding eigenspaces in V F .
Suppose 1 6 k 6 m and v ∈ Eig(εk), then, φφFj (v) = φF
j φ(v) = φFj (εv) = εφF
j (v). So each
φFj preserves each eigenspace for φ. Now φ is diagonalizable with respect to a basis of
eigenvectors. We now show we can choose a basis for V F with respect to which every φFj
is diagonalizable. If we can find a basis for each eigenspace with respect to which each φFj
is diagonalizable then we are done. So we proceed by induction on the dimension of the
eigenspace. If Eig(εk) has dimension one then, as it is preserved by each φFj , each φF
j is
scalar. So suppose Eig(εk) has dimension n+1 and that each φFj is diagonalizable on any
eigenspace of lesser dimension. Suppose some φFj is not diagonalizable on Eig(εk). Then
the vector space Eig(εk) decomposes into eigenspaces for φFj each of which is preserved by
each linear transformation, φF1 , . . . , φ
Fn and by induction there is a basis for each for which
each φF1 , . . . , φ
Fn is diagonalizable. However this gives us a basis for Eig(εk) on which φj
is diagonalizable contrary to assumption.
In the following corollary K is a field extension of L and is therefore a vector space
over L. We use the notation GL(L,K) for the group of invertible L-linear transformations
of K.
Corollary 2.12. Let p be a prime and let L = GF(p) and K = GF(pn) for some n > 1.
Each k ∈ K× defines a linear transformation of K as a vector space over L and there is
an embedding φ : K× → GLn(L) → GLn(K). Furthermore there exists M ∈ GLn(K)
such that φ(k)M = diag(k) for all k ∈ K×.
Proof. It is clear thatK is an n-dimensional vector space over L and each non-zero, k ∈ K,
29
defines a linear transformation, λk : a 7→ ka (for each a ∈ K). Thus there is an embedding
θ : K× → GLn(L) ∼= GL(L,K) and since L 6 K, we see the inclusion map ι : GLn(L) →
GLn(K) thus φ := ιθ : K× → GLn(K). Now, the linear transformations λk|k ∈ K×
commute pairwise and are all scalar on K so by Lemma 2.11, the corresponding linear
transformations of K⊗LK (as an n-dimensional vector space over K), 1⊗L λk|k ∈ K×
are simultaneously diagonalizable. Therefore we can assume there is an image of K× in
GL(K,K⊗LK), via θK say, in which θK(k) is scalar for each k ∈ K×. Finally, by Lemma
2.10, the representation φ is equivalent to θK . Therefore, there exists M ∈ GLn(K) ∼=
GL(K,K ⊗L K) such that φ(k)M = diag(k) for each k ∈ K×.
2.3 Natural GL2(3)-Modules
In later chapters we will consider amalgams. These will be subgroups inside a larger group
and the structure of the subgroups will be described via GL2(3) and SL2(3)-modules. In
the following section set G ∼= GL2(3), G > H ∼= SL2(3) and F = GF(3).
Lemma 2.13. The following are true in G ∼= GL2(3).
(i) Any subgroup of G isomorphic to 2 × 2 is conjugate to the subgroup of diagonal
matrices.
(ii) Q8∼= O2(G) 6 H.
(iii) H ∼= SL2(3) is the unique subgroup of G at index two.
Proof. Let 2× 2 ∼= T 6 G. Consider the eigenvalues for each t ∈ T#. Each t ∈ T satisfies
the polynomial X2 = 1 and so the possible eigenvalues are 1 and −1. If t has only the
eigenvalue 1 then t is the identity. If t has only the eigenvalue −1 then t must be the
central scalar matrix diag(−1,−1). If t has eigenvalues 1 and −1 then t is conjugate to
30
the matrix diag(−1, 1). If T contains the central involution then we are done since we
only need to diagonalize one more element of T . So suppose not then we must have t ∈ T
with eigenvalues 1 and −1. Let u and v be eigenvectors for t such that t(u) = u and
t(v) = −v. Let s ∈ T\〈t〉 then s(u) = s(t(u)) = t(s(u)) so s(u) is in the eigenspace 〈u〉
similarly s(v) is in the eigenspace 〈v〉 and it follows that s must be diagonal with respect
to the basis u, v and so we can simultaneously diagonalize T . Therefore (i) holds.
Notice that
K := 〈
(0 −1
1 0
),
(1 1
1 −1
)〉
is a normal subgroup of G isomorphic to Q8 and that this subgroup is contained in H.
Furthermore G cannot have a larger normal 2-subgroup since we can find two distinct
Sylow 2-subgroups of order 24 by including the element
(1 −1
0 −1
)or
(0 −1
−1 0
)so (ii)
holds.
Now, H is a normal subgroup of G at index two so suppose K is another subgroup of
G at index two. Then H ∩K E G has at index four in G and has order 322. If H ∩K
contains a normal Sylow 3-subgroup, S say, then S char H ∩K E G implies S E G which
is a contradiction since G does not have a normal Sylow 3-subgroup. In fact G contains
four Sylow 3-subgroups namely,
⟨ 1 0
1 1
⟩ ,
⟨ 1 1
0 1
⟩ ,
⟨ 0 1
−1 −1
⟩ ,
⟨ −1 1
−1 0
⟩
which pairwise intersect trivially. Each is normalized in G by a group of order 322 and so
by Sylow’s Theorem, G has exactly four Sylow 3-subgroups. Also, by Sylow’s Theorem,
H∩K contains four Sylow 3-subgroups and so H∩K contains 104 elements of order three.
Therefore H ∩K can contain only three elements of order two and so necessarily has a
normal subgroup of order four, F say. However, H contains a Sylow 2-subgroup which is
quaternion of order eight and so F must be cyclic of order four and therefore admits no
31
automorphism of order three. This implies that CH∩K(F ) = NH∩K(F ) = H ∩K and so
F commutes with each Sylow 3-subgroup but then each Sylow 3-subgroup in H ∩ K is
normal in H ∩K which is a contradiction.
Definition 2.14. Let V be a 2-dimensional vector space over GF(3). We identify G
with GL(V ) and H with SL(V ) and call V the natural G-module and natural H-module
respectively.
Lemma 2.15. Let V be a natural G-module or a natural H-module. The following hold.
(i) For S ∈ Syl3(G), CV (S) is a subspace of V of dimension one.
(ii) Both G and H act transitively on the subspaces of V of dimension one.
(iii) Both G and H act transitively on the non-zero vectors in V .
(iv) The action of G (and H) on V is faithful.
Proof. (i) Let S ∈ Syl3(G) then S is conjugate to the group of strictly lower triangular
matrices. It is easy to check that this group of matrices fix precisely the 1-space 〈(1, 0)〉.
Thus CV (S) has dimension one.
(ii) There are four subspaces of V with dimension one. The stabilizer in G of the
space 〈(1, 0)〉 is the subgroup of lower triangular matrices which has index four in G.
Similarly the stabilizer in H of 〈(1, 0)〉 is the subgroup of lower triangular matrices with
determinant 1 which has index four in H. Therefore both G and H are transitive on the
1-spaces in V .
(iii) We repeat the same arguments to see that the stabilizer in G (and in H) of a
non-zero vector in V has index eight. Therefore both G and H are transitive on the eight
non-zero vectors in V .
(iv) This is just by definition since we identified GL2(3) with the group of automor-
phisms GL(V ).
32
The following lemma describes a situation which will occur many times when we are
dealing with groups involving natural modules.
Lemma 2.16. Suppose S is a 3-group with subgroups P 6= Q 6 S of index three. Suppose
further that t is an involution in Aut(S) which leaves P and Q invariant. Then t acts
on S/P (and S/Q) and either centralizes or inverts S/P . If t centralizes S/P then there
exists x ∈ Q\P such that t centralizes x. If t inverts S/P then there exists x ∈ Q\P such
that t inverts x.
Proof. It is clear that t acts on S/P such that for any element Py ∈ S/P , Pyt := P (yt).
Pick y ∈ Q\P then the coset Py is either inverted or centralized by t. Suppose first that
Py is centralized by t then t describes a permutation of the set Py. Furthermore elements
py ∈ Py are in Q if and only if p ∈ Q ∩ P and t preserves Q so preserves those elements.
Since t is an involution and Py∩Q is a set of odd order, |P ∩Q|, t must fix some element,
x say. Thus x is in Q\P and is centralized by t. Now suppose that Py is inverted by t
then [y, t] /∈ P . Also [y, t]t = (y−1yt)t = (y−1)ty = [y, t]−1 and so [y, t] is inverted by t.
Thus [y, t] ∈ Q\P and is inverted by t.
2.4 Sym(4)-Modules
Lemma 2.17. Suppose that X ∼= Sym(4) and that V is a faithful 3-dimensional GF(3)X-
module. Let q1, q2, q3 = O2(X)#. The following hold.
(i) There is a set of distinct 1-dimensional subspaces B := 〈v1〉, 〈v2〉, 〈v3〉 where 〈vi〉 =
CV (qi).
(ii) X has orbits of lengths 3, 4 and 6 on the 1-dimensional subspaces of V with repre-
sentatives
Type 1A - 〈v1〉,
33
Type 1B - 〈v1 + v2 + v3〉 and
Type 1C - 〈v1 + v2〉
respectively.
(iii) X has orbits of lengths 3, 4, and 6 on the 2-dimensional subspaces of V with repre-
sentatives
Type 2A - 〈v1, v2〉,
Type 2B - 〈v1 + v2, v2 + v3〉 and
Type 2C - 〈v1, v1 + v2 + v3〉
respectively.
(iv) If x ∈ X has order three then the subspace CV (x) has dimension one and lies in the
orbit of 1-spaces with representative 〈v1 + v2 + v3〉.
(v) For i 6= j, vi is negated by qj. For i, j, k = 1, 2, 3, qk negates vi + vj.
(vi) |CX(vi)| = 4, |CX(vi + vj)| = 2 for i 6= j.
(vii) If x ∈ X\O2(X) has order two then x centralizes a 1-space of type 1C.
Proof. Set Q = O2(X) then Q# = q1, q2, q3. By coprime action of Q on V ,
V = 〈CV (qi)|i = 1, 2, 3〉.
Each CV (qi) is a non-trivial subspace of V and as X acts transitively on Q#, X acts
transitively on the three subspaces CV (q1), CV (q2) and CV (q3). Thus each has the same
dimension. SinceX acts faithfully on V , each subspace is proper. Suppose each CV (qi) has
dimension two. Consider CV (q1)∩CV (q2). This must be non-trivial. Moreover any vector
fixed by q1 and q2 must be fixed by Q and so this intersection is X-invariant. However
34
this is a contradiction since X cannot act faithfully on a 1-dimensional subspace. Choose
vectors vi such that 〈vi〉 = CV (qi) then
V = 〈v1〉 ⊕ 〈v2〉 ⊕ 〈v3〉.
Clearly 〈vi〉 | 1 6 i 6 3 forms an orbit of length 3 on the 1-dimensional subspaces
of V . We also have that the subspaces 〈v1 ± v2 ± v3〉 form an orbit of length 4 on V and
the subspaces 〈vi ± vj〉, where i 6= j form an orbit of length 6.
Clearly the subspaces 〈vi, vj〉 for i 6= j give an orbit of length 3 on the 2-dimensional
subspaces of V . The subspaces 〈vi ± vj, vj ± vk〉 form an orbit of length 4. Finally the
subspaces 〈vi, vi + vj − vk〉, for i j and k distinct form an orbit of length 6. We note that
for each choice of i we have two choices for k.
We see that any element of order three cycles the set Q# and so cycles the spaces
〈v1〉, 〈v2〉, 〈v3〉. But if x ∈ X cycles 〈v1〉, 〈v2〉, 〈v3〉, it must also cycle the spaces of type
〈v1 + v2〉 of which there are six. Since spaces of type 〈v1 + v2 + v3〉 lie in an orbit of length
four, x necessarily fixes at least one of them. However if it fixes them all it acts trivially
on V which is impossible.
The space 〈vi〉 is preserved by NX(qi). So for j 6= i, qj fixes this one space. However
we have seen that CV (qi)∩CV (qj) is trivial thus qj negates vi. Now if we let k be distinct
from both i and j then qk negates vi and vj and thus the vector vi + vj.
We consider the action of X on the set of vectors. We have seen that vi lies in an orbit
with its inverse and with the non-trivial elements of the other 1-spaces of type 1A. Thus
vi lies in an orbit of length six and so |CX(vi)| = 4. Also vi + vj lies in an orbit with its
inverse and the non-trivial elements of the other 1-spaces of type 1C. Thus the orbit has
length 12 and so |CX(vi + vj)| = 2. The element of order two in X centralizing vi + vj
cannot lie in O2(X) by (iv) and so must be in the other conjugacy class of involutions.
35
Since all involutions outside of O2(X) lie in this conjugacy class, each fixes some conjugate
of vi + vj.
Lemma 2.18. Suppose that X ∼= Sym(4) and that V is a faithful 3-dimensional GF(3)X-
module. Then the subspaces of order 32 have the following forms.
(i) Type 2A - These contains two 1-dimensional subspaces conjugate to 〈v1 + v2〉 and
two conjugate to 〈v1〉. There are three such subspaces.
(ii) Type 2B - These contain one 1-dimensional subspace conjugate to 〈v1+v2+v3〉 three
1-dimensional subspaces conjugate to 〈v1 + v2〉. There are four such subspaces.
(iii) Type 2C - These contain two 1-dimensional subspaces conjugate to 〈v1 + v2 + v3〉,
one conjugate to 〈v1 + v2〉 and one conjugate to 〈v1〉. There are six such subspaces.
Proof. Let Y = 〈v1, v2〉. Then Y contains the 1-dimensional subspaces 〈v1〉, 〈v2〉, 〈v1 +v2〉
and 〈v1 − v2〉. Hence Y is of type 2A.
Now let W = 〈v1+v2, v2−v3〉. Then W contains the 1-dimensional subspaces 〈v1+v2〉,
〈v2 − v3〉, 〈v1 + v3〉 and 〈v1 − v2 − v3〉. Hence W is of type 2B.
Finally let U = 〈v1, v1 + v2 + v3〉. Then U contains the 1-dimensional subspaces 〈v1〉,
〈v1 + v2 + v3〉, 〈v2 + v3〉 and 〈v1 − v2 − v3〉. Hence U is of type 2C.
The number of each type of subgroup follows from Lemma 2.17 (iii).
There are two isomorphism types of 3-dimensional Sym(4)-modules and we give an
example of both. Firstly, let
A = 〈(123), (456), (789), (147)(258)(369), (14)(2536)〉.
36
Then A has shape 33 : Sym(4) and A 6 Alt(9). Secondly, let
B = 〈(123), (456), (789), (147)(258)(369), (14)(2536)(78)〉,
then B also has shape 33 : Sym(4). We see that B 6 Sym(9) but B Alt(9). No-
tice in both groups that we can apply Lemma 2.17 by setting v1 = (123), v2 = (456)
and v3 = (789). Notice also that in B the element (789) is inverted by the element
(14)(2536)(78) however one can check that CA((789)) = NA((789)). Also, by Lemma
2.17 (vi), |CA((789))| = |CB((789))| = 4 however in A, (789) commutes with an element
of order four, (14)(2536), and again one can check that in B, (789) commutes with an
elementary abelian group of order four. We will use Lemma 2.17 and Lemma 2.18 in
Chapter 7 where it will become clear we have a Sym(4)-module of the second type.
2.5 Some Modular Character Theory
We introduce a small amount of modular character theory which we will require in Chapter
8. Results will be stated without proof. The main reference for this chapter has been
Chapter 15 of [19].
Let A be the ring of algebraic integers and let M be a maximal ideal of A containing
Ap where p is a fixed prime. Set F to be the field A/M which necessarily has characteristic
p and set ∗ : A → F to be the natural ring homomorphism. Notice that we may have
some choice over M and hence over F but our choice makes no difference to the following
results. Finally, set U = ε ∈ A|εn = 1 for n ∈ Z with (n, p) = 1. The following hold.
(i) ∗∣∣U
: U → F× is a group isomorphism.
(ii) F is an algebraically closed field.
37
Let G be a group and let ρ : G → GLn(F ) be a representation of G affording the
character χ. Let Ω be the set of elements of G with order prime to p. We define the
Brauer character of G to be the map φ : Ω → C as follows. For each x ∈ Ω, ρ(x) has
eigenvalues ε1, . . . , εn ∈ F× as F is algebraically closed. Since ∗∣∣U
is a group isomorphism
onto F×, we can find a unique ui ∈ U such that ∗ : ui 7→ εi. Thus we can define
φ(x) = Σui. A Brauer character is said to be irreducible if the representation affording it
is irreducible. Each Brauer character is clearly constant on conjugacy classes of G in Ω
and so is an example of a class function defined on the elements of Ω.
Lemma 2.19. Suppose U and V are FG-modules affording characters χ1 and χ2 respec-
tively. Then U ⊗F V affords the character χ1χ2 defined by χ1χ2(g) = χ1(g)χ2(g) for each
g ∈ G.
Proof. If u ∈ U is a λ-eigenvector and v ∈ V is a µ-eigenvector then u⊗ v ∈ U ⊗ V is an
eigenvector and the eigenvalue is λµ. Suppose ρ1 : G→ GL(U) and ρ2 : G→ GL(V ) are
the corresponding representations. Let g ∈ G and let λ1, . . . , λn be the eigenvalues for
ρ1(g) and let µ1, . . . , µm be the eigenvalues for ρ2(g). Then λiµj|1 6 i 6 n, 1 6 j 6 m
are the eigenvalues for ρ(g) where ρ : G → GL(U ⊗ V ). Now, by the definition of the
Brauer character,
χ1(g) =n∑
i=1
ai and χ2(g) =m∑
j=1
bj
where ∗(ai) = λi and ∗(bj) = µj. The Brauer character afforded by U ⊗ V is χ(g) =∑ni=1
∑mj=1 cij where ∗(cij) = λiµk = ∗(ai) ∗ (bj) = ∗(aibj) since ∗ is a group homomor-
phism. Therefore
χ(g) =n∑
i=1
m∑j=1
aibj =n∑
i=1
ai
m∑j=1
bj = χ1(g)χ2(g).
Theorem 2.20. The irreducible Brauer characters give a basis for the space of C-valued
38
class functions defined on Ω. The number of irreducible Brauer characters of G is equal
to the number of conjugacy classes of elements of order prime to p.
Proof. This theorem is Theorem 15.10 in [19].
Lemma 2.21. Let χ be an ordinary character of G and let χ be the restriction of χ to
the elements in Ω. Then χ is a Brauer character of G.
Proof. [19, 15.6 p265]
Theorem 2.22. Let χ be an ordinary irreducible character of G such that |G|/χ(1) is
prime to p. Then χ is an irreducible Brauer character of G.
This theorem is essentially Theorem 15.29 in [19] which proves that the set χ, χ is a
block.
Let V be an n-dimensional module for G over some finite field, K, of characteristic p.
Then consider the module F ⊗K V . This is an n-dimensional module for G over F when
we define (f ⊗ v) · g = f ⊗ (v · g) for f ∈ F , v ∈ V and g ∈ G. A basis for F ⊗K V is
1 ⊗K vi|1 6 i 6 n over F where vi|1 6 i 6 n is a basis for V over K. By Lemma
2.8, if V is irreducible and EndKG(V ) = K then V ⊗K F is irreducible. So we have some
tools for classifying possible modules for G over finite fields.
2.6 Natural SL2(2n)-Modules
In all that follows let X ∼= SL2(2n) where n ≥ 2, T ∈ Syl2(X) and set L := GF(2) and
K := GF(2n). We will calculate the Brauer character table for X and so we must choose
an algebraically closed field, F , of characteristic two in which to calculate.
Lemma 2.23. The following hold in X ∼= SL2(2n).
39
(i) There are at most 2n conjugacy classes of elements of odd order.
(ii) Let S, T ∈ Syl2(X). Then S∩T = 1 and S acts transitively on the set Syl2(X)\S.
(iii) For S, T ∈ Syl2(X), NX(S)∩NX(T ) has order 2n−1 and fixes precisely two elements
in Syl2(X).
(iv) X acts sharply 3-transitively on Syl2(X).
(v) For S, T ∈ Syl2(X) and t ∈ T#, X = 〈S, T 〉 = 〈S, t〉.
Proof. The characteristic polynomial for a non-identity element, A, ofX is x2−tr(A)x+1.
Since conjugate elements have the same characteristic polynomial and values of the trace
lie in K, we have 2n possibilities. So together with the identity we have at most 2n + 1
conjugacy classes in X. As one of these classes contains elements of order 2, we have at
most 2n classes of elements of odd order.
Every Sylow 2-subgroup has order 2n and is normalized by a subgroup of X conjugate
to the Borel subgroup of lower triangular matrices. This has order 2n(2n−1) and so there
are 2n +1 Sylow 2-subgroups in X. We consider the action of X on the set Ω := Syl2(X).
Let S ∈ Ω. It is clear that no non-trivial element of S fixes any other element of Ω else
we would have a larger 2-group and since |Ω\S| = 2n = |S|, we see S is transitive on
Ω\S. Thus X is 2-transitive on Syl2(X). Consider two elements of Ω namely, U , the
upper triangular and, L, the lower triangular matrices. These intersect trivially and so
any pair of Sylow 2-subgroups intersect trivially. Notice that both are normalized by the
subgroup of diagonal matrices, D. In fact D = NX(U)∩NX(L) for the intersection can be
no larger with out having even order. Suppose D normalizes a third element of Ω, P say.
Find x ∈ U such that Lx = P . Then D,Dx 6 NX(Lx) ∩NX(U). But then D = Dx and
so x ∈ NU(D) = 1. Thus D fixes exactly 2 elements of Ω and so is transitive on Ω\U,L.
So now we have shown that a 2-point stabilizer is transitive on the remaining 2n−1 points
and that the stabilizer of three points is trivial. Hence X is sharply 3-transitive on Ω.
40
Now for any distinct pair of Sylow 2-subgroups, S and T , and any t ∈ T#, the group
Y := 〈S, t〉 contains Sylow subgroups of order 2n and also an element from each Sylow
2-subgroup. As they all intersect trivially it contains all Sylow subgroups. However, the
subgroup of X generated by all Sylow 2-subgroups must be normal in X but X is simple
and so it must be all of X.
Definition 2.24. Let R be the polynomial ring in two commuting indeterminates, x and
y, with coefficients in K. Then R is a KX-module when we set
x ·
(a b
c d
)= ax+ by
and
y ·
(a b
c d
)= cx+ dy.
Let R(1) = f ∈ R|degf = 1 ∪ 0. Then R(1) is a 2-dimensional submodule which we
will call the natural module over K.
Lemma 2.25. Let V be a natural module for X over K. The following hold.
(i) V is irreducible.
(ii) X is transitive on non-zero vectors of V .
(iii) X is 2-transitive on the 1-subspaces of V .
(iv) Elements of X with odd order act fixed-point-freely on V .
Proof. The stabilizer of a 1-space is clearly the normalizer of a Sylow 2-subgroup, i.e.
conjugate to a group of triangular matrices, which again is transitive on all remaining
1-spaces. It is then clear that V is irreducible since there can be no X-invariant subspace.
Moreover, the group of triangular matrices is transitive on the non-zero vectors in the
41
1-space and so X is transitive on non-zero vectors. Finally, the stabilizer of a non-zero
vector is a Sylow 2-subgroup and so any non-identity element of X with odd order fixes
no non-zero vector in V .
Lemma 2.26. There are at least n non-isomorphic, 2-dimensional, irreducible modules
for X over K, namely the natural module over K and its n algebraic conjugates.
Proof. If φ : X → GL(V ) is the representation of X affording a natural module, V , over
K and σ ∈ Aut(K) then, after choosing a basis for V and writing X as matrices, we
can define another representation φσ : X → GL(V ) simply by applying σ to the matrix
entries. The corresponding module we label V σ and call it a Galois twist of V . It is clear
that no choice of basis elements allow the matrices to be written purely by elements of L.
See [1, 26.3] for a proof that this means V and V σ cannot be isomorphic as KX-modules.
Since K = GF(2n) has n automorphisms, namely k 7→ k2i0 ≤ i ≤ n−1, we see n distinct
irreducible modules for X over K.
Let V1, . . . , Vn be these 2-dimensional irreducible KX-modules.
Theorem 2.27. Let V be an irreducible KX-module and let K ′ > K be a field extension.
Then the K ′X-module K ′ ⊗K V is irreducible.
Proof. Recall that an irreducibleKX-module, V is called absolutely irreducible ifK ′⊗KV
is an irreducible K ′X-module for each extension K ′ > K. The splitting field for X is the
field, S, with the property that every irreducible SX-module is absolutely irreducible.
Remark 2.8.8 in [14] says that the splitting field for X is K. Thus every irreducible
KX-module is absolutely irreducible.
Corollary 2.28. Let V be an irreducible KX-module then EndKX(V ) ∼= K.
Proof. By Theorem 2.27, V is absolutely irreducible and so by Lemma 2.8, EndKX(V ) ∼=
K.
42
Let Mi = F ⊗ Vi for 1 ≤ i ≤ n. Recall these are 2-dimensional modules over the
algebraically closed field F and by Theorem 2.27, each Mi is irreducible.
Lemma 2.29. Let x ∈ X have odd order and let χi be the Brauer character afforded by
the FX-module Mi. Then χi(x) = ε2i−1+ ε−2i−1
where ε is some fixed m’th root of unity
and m and the order of x both divide 2n − 1 or both divide 2n + 1.
Proof. Any element, x ∈ X, of odd order either normalizes a Sylow 2-subgroup and so
has order dividing 2n− 1 or is fixed-point-free on the Sylow 2-subgroups and so has order
dividing 2n+1. Let ρ1 be the natural K-representation affording the KX-module V1. The
possible eigenvalues for ρ1(x) in the algebraically closed extension field F are therefore
field elements, α and α−1, satisfying the polynomial X2n−1 − 1 or X2n+1 − 1. By Lemma
2.10, the Brauer character afforded by the module M1 for x is ε+ε−1 where ε is some fixed
m’th root of unity in C and m divides either 2n− 1 or 2n + 1. The entries of the matrices
representing the action of X on each module Vi are altered by a Galois automorphism. So,
by Lemma 2.10, the Brauer character afforded by the module Mi for x is ε2i−1+ ε−2i−1
.
Lemma 2.30. There are 2n irreducible modules for X over F . They are the 2n possible
tensor products⊗
i∈I Mi where I ⊆ 1, 2, . . . , n.
Proof. We begin by identifying an irreducible character for X in characteristic zero of
degree 2n. Consider the action of X on its Sylow 2-subgroups. Let χ be the corresponding
class function and set χ1 = χ− I where I is the trivial character. It follows from Lemma
2.23 that any element of X which normalizes a Sylow 2-subgroup either has order two and
fixes precisely one Sylow 2-subgroup or lies outside of it and fixes exactly two of them.
Thus the class function χ gives values 1 for elements of order two, 2 for elements whose
order divides 2n− 1 and 0 for elements whose order divides 2n + 1. So χ1 gives respective
values 0, 1, −1. We see that this class function is an irreducible character by taking its
43
inner product. This gives
〈χ1, χ1〉 = 22n
2n(2n−1)(2n+1)+ 0
2n + a(
12n−1
)+ b(
(−1)2
2n+1
)
= 122n−1
(2n + a(2n + 1) + b(2n − 1)) ,
where a is the number of conjugacy classes of elements whose order divides 2n − 1 and b
is the number of classes of elements whose order divides 2n + 1. Thus a+ b = 2n − 1 and
it follows that 〈χ1, χ1〉 is strictly less than 2 and so must be 1 and so χ1 is irreducible.
Now by Theorem 2.22, the restriction of χ1 to the elements of x with odd order is an
irreducible Brauer character, χ1. Clearly χ(1) = |Syl2(X)| = 2n + 1 and so χ1(1) = 2n.
Therefore we have found an irreducible Brauer character for X in characteristic two of
degree 2n.
We now begin to identify all irreducible representations of X over F . Consider the
Brauer character for some element, x, of odd order afforded by the module M1 ⊗F . . .⊗F
Mn. By Lemma 2.29 and Lemma 2.19, this will give values
n∏i=1
(ε2i−1
+ ε−2i−1
) =ε2n − ε−2n
ε− ε−1,
where ε is some fixed m’th root of unity and m divides 2n − 1 or 2n + 1. So if x has
order dividing 2n − 1 then the character value is 1 otherwise it is −1. Now we have
a direct agreement between the character induced by the module M1 ⊗ . . . ⊗ Mn and
the character χ1. This means that M1 ⊗ . . . ⊗Mn is an irreducible module. It follows
immediately that every module Mr1 ⊗ . . . ⊗Mrkwhere 1 6 r1 < r2 < . . . < rk 6 n is
irreducible since any proper submodule of Mr1⊗ . . .⊗Mrkwould give a proper submodule
of M1 ⊗ . . . ⊗Mn. If M is any X-module then the module M ⊗M is not irreducible
since it has a proper submodule 〈m⊗ n− n⊗m|m,n ∈M〉. So suppose Mr1 ⊗ . . .⊗Mrk
and Ms1 ⊗ . . . ⊗Mskare isomorphic. Then their tensor product would be reducible but
44
then there is some repeated natural module Mi. Continuing in this way we would see
Mr1 ⊗ . . . ⊗ Mrk= Ms1 ⊗ . . . ⊗ Msk
. Thus we have identified 2n distinct irreducible
modules for X over F . We have already seen that there are at most 2n conjugacy classes
of elements of odd order and so by Theorem 2.20, there are exactly 2n classes and 2n
irreducible modules.
Lemma 2.31. There are exactly 2n irreducible modules for X over K. They are all
possible tensor products (over K)⊗
i∈I Vi where I ⊆ 1, 2, . . . , n.
Proof. For any set I ⊆ 1, . . . , n, (⊗
i∈I,K Vi) ⊗K F ∼= ⊗i∈I,FMi by repeated us of
Lemma 2.9. Suppose (⊗
i∈I,K Vi) had a proper KX-submodule W . Then W ⊗K F would
be a proper FX-submodule of the irreducible module ⊗i∈I,FMi and so must be trivial.
Thus each (⊗
i∈I,K Vi) is an irreducible KX-module. If for I, J ⊆ 1, . . . , n,⊗
i∈I,K Vi
was KX-isomorphic to⊗
j∈J,K Vj then⊗
i∈I,F Mi∼=⊗
j∈J,F Mj and so by Lemma 2.30,
I = J . Therefore we have at least 2n irreducible KX-module. Suppose we could find
one more, U say. Then by Lemma 2.27, U ⊗K F is irreducible and we would have
U ⊗K F ∼=⊗
i∈I,F Mi∼= (⊗
i∈I,K Vi)⊗K F and so U ∼=⊗
i∈I,K Vi. Thus there are exactly
2n irreducible KX-modules.
The previous lemma is also a famous theorem by Robert Steinberg.
Theorem 2.32. (Steinberg) There are exactly 2n irreducible modules for X over K. They
are all possible tensor products (over K)⊗
i∈I Vi where I ⊆ 1, 2, . . . , n.
Proof. See [28, 13.1].
Lemma 2.33. (i) The only irreducible modules for X over F which admit an element
of order three acting fixed-point-freely are the modules M1, . . . ,Mn.
(ii) The only irreducible modules for X over K which admit an element of order three
acting fixed-point-freely are V1, . . . , Vn.
45
Proof. We consider the possible Brauer character values for some element g of order three
acting fixed-point-freely. Let χi be the character corresponding to Mi. Then χi(g) =
ω + ω2 = −1 where ω is a primitive cube root of unity. Thus if χ is the character value
corresponding to some irreducible FX-module, M = Mr1 ⊗ . . . ⊗Mrk, then χ(g) = 1 or
−1. By Lemma 2.6,
CM(〈g〉) = 〈χ|〈g〉 , 1|〈g〉〉 =1
3(χ(1) + χ(g) + χ(g2)).
So g acts fixed-point-freely on M if and only if χ(1) = dim(M) = 2 if and only if M = Mi
for some 1 6 i 6 n.
Suppose now that V is some irreducible KX-module admitting g fixed-point-freely.
By Theorem 2.27, F ⊗K V is an irreducible FX-module. Observe that g acts fixed-point-
freely on F ⊗K V since (f ⊗ v) · g = f ⊗ v · g = f ⊗ v if and only if v · g = v for f ∈ F
and v ∈ V . So now F ⊗K V must have dimension two and so V must be KX-isomorphic
to Vi for some 1 6 i 6 n. Finally, by Lemma 2.25, X acts fixed-point-freely on V1 and
similarly on each algebraic conjugate.
Definition 2.34. A natural module for X over L is an irreducible 2n-dimensional mod-
ule, V , such that EndLX(V ) ∼= GF(2n).
Lemma 2.35. (i) Let V be a natural module for X over L. Then V can be viewed as
a 2-dimensional vector space over K and as such is KX-isomorphic to an algebraic
conjugate of the natural module over K.
(ii) Let U be an algebraic conjugate of the natural module for X over K. Then U can
be viewed as a 2n-dimensional vector space over L and as such is LX-isomorphic
to a natural module over K.
(iii) The natural module over L is unique up to LX-isomorphism.
46
Proof. Let V be a natural module over L. Then V admits an action by its endomorphism
ring and so is a vector space over K of dimension two. Let U represent the vector space
V over K. Suppose U is not irreducible then U contains a KX-invariant proper subspace
U0 say. However this would mean that U0 was LX-invariant and so also a submodule of
V , the vector space over L, which contradicts that V is irreducible by definition. So U is
irreducible of dimension two and therefore must be the natural module over L or one of
its algebraic conjugates.
Now let U be a natural module for X over K then U is also a vector space over L. Let
V represent this restriction module then V is a vector space over L and V is irreducible
as an X-module since X is transitive on non-zero vectors of U by Lemma 2.25. We
can define an action of K on V and so K is embedded in E := EndLX(V ). By Schur’s
Lemma, E is a field. However, since all elements of E commute with K, it follows that
E = EndKX(V ) = EndKX(U) = K by Lemma 2.28. So by definition V is a natural
LX-module and so the restriction of any of the irreducible 2-dimensional modules over K
gives a natural module over L.
Let V be the natural module overK and V σ a Galois twist of V . Consider the following
map
Φ : V → V σ
(k1, k2) 7→ (k1σ, k2
σ).
Then Φ is an LX-module isomorphism since for all v = (k1, k2) ∈ V and every matrix in
47
the corresponding representation of X,
Φ
(k1, k2)
a b
c d
= ((k1a+ k2c)σ, (k1b+ k2d)
σ)
= (kσ1 , k
σ2 )
aσ bσ
cσ dσ
= Φ ((k1, k2)) ·
a b
c d
.
Also because σ is trivial on L we have Φ(lv) = lΦ(v) for all l ∈ L. Thus V and V σ are
isomorphic as LX-modules. In particular it will follow that their restrictions over L are
isomorphic as LX-modules.
48
Chapter 3
Amalgams and the Coset Graph
An amalgam of finite groups is an ordered collection of three finite groups and two injective
homomorphisms A = (A,B,C, φ1, φ2), such that φ1 : C → A and φ2 : C → B.
We say that two amalgams A1 = (A1, B1, C1, φ1, φ2) and A2 = (A2, B2, C2, θ1, θ2) are
isomorphic if there exist isomorphisms α : A1 → A2, β : B1 → B2 and γ : C1 → C2 such
that αφ1 = θ1γ and βφ2 = θ2γ. We say that A is a simple amalgam if, whenever K 6 B,
with φ1(K) E A1 and φ2(K) E A2, then K = 1.
A completion of an amalgam, A = (A,B,C, φ1, φ2), is a collection triple (G,ψ1, ψ2)
where G is a group such that ψ1 : A → G and ψ2 : B → G are homomorphisms with
G = 〈ψ1(A), ψ2(B)〉 and such that ψ1φ1 = ψ2φ2. That is that the following diagram
commutes:
A
@@
@R
C G
@@
@R
B
φ1 ψ1
φ2 ψ2
49
A completion, G, is faithful if ψ1 and ψ2 are injective. In this case we often suppress
use of monomorphisms and identify A, B and C with their images in G. In fact, when
we describe particular amalgams in this thesis we will often assume we have a group G
with subgroups A, B and C := A ∩B such that G = 〈A,B〉. Therefore we are implicitly
assuming that (G, I1, I2) is a completion of the amalgam (A,B,C, i1, i2) where i1 and i2
are the inclusion maps i1 : C → A and i2 : C → B and I1, I2 are the inclusion maps
I1 : A → G and I2 : B → G.
A completion (F,Ψ1,Ψ2) ofA is called universal if given any other completion (G,ψ1, ψ2)
there is a unique homomorphism π : F → G such that Ψiπ = ψi for i = 1, 2. This im-
plies that any completion of A is isomorphic to a quotient of the universal completion
of A and that the universal completion is itself unique up to isomorphism. Isomorphic
amalgams have the same groups as completions so to recognize possible completions we
need to recognize the possible isomorphism classes of the amalgam. A fundamental result
which has become known as the Goldschmidt Lemma (see [12, 2.7]) gives us a method for
calculating the number of isomorphism classes of a given type of amalgam.
Lemma 3.1. (Goldschmidt Lemma) Let (X, Y, Z) be an amalgam of finite groups and
set X∗ = NAut(X)(Z)/CAut(X)(Z), Y ∗ = NAut(Y )(Z)/CAut(Y )(Z) then there is a one-to-one
correspondence between isomorphism classes of type (X, Y, Z) and (X∗, Y ∗)-double cosets
in Aut(Z).
Let A = A(A1, A2, B) be a simple amalgam and let G be a faithful completion of A.
We identify A1, A2 and B with their images in G and regard φi as the inclusion map of
B into Ai. We also suppose that B = A1 ∩A2. The coset graph of the amalgam A is the
graph Γ = Γ(G,A1, A2, B) that has vertex set
V (Γ) = Aig | g ∈ G, i ∈ 1, 2,
50
and edge set
E(Γ) = Aig, Ajh | Aig ∩ Ajh 6= ∅, i 6= j.
Since G is a simple amalgam, G acts faithfully by right multiplication on Γ and this action
preserves the edge set E(Γ).
For γ ∈ V (Γ), let Gγ = StabG(γ) and let Γ(γ) = δ ∈ V (Γ) | γ, δ ∈ E(Γ) be the
set of vertices adjacent to γ in Γ. For γ, δ ∈ E(Γ), let Gγδ = StabG(γ, δ) = Gγ ∩Gδ.
The following results are well known and easy to prove.
Lemma 3.2. Let G be a faithful completion of a simple amalgam of groups (A1, A2, B)
and let Γ be the coset graph.
(i) G acts faithfully on Γ.
(ii) G has two orbits on V (Γ) and is transitive on E(Γ).
(iii) For γ ∈ V (Γ), Gγ is G-conjugate to either A1 or A2.
(iv) For γ, δ ∈ E(Γ), Gγδ is G-conjugate to B.
(v) For γ ∈ V (Γ), Gγ acts transitively on Γ(γ). In particular, |Γ(γ)| = [Gγ : Gγδ] for
any δ ∈ Γ(γ).
(vi) The graph Γ is connected.
Proof. See [22, Lemma 4.1, Lemmas 4.3 and 4.5].
Parts (ii) and (iii) of Lemma 3.2 imply that the amalgam A = A(A1, A2, B) is isomorphic
to the amalgam A′ = A′(Gγ, Gδ, Gγδ) for γ, δ ∈ E(Γ).
Definition 3.3. Let (α1, α2, . . . , αn+1) and (β1, β2, . . . , βn+1) be any two paths of n + 1
vertices in Γ where α1 and β1 lie in the same G-orbit. If there exists g ∈ G such that
ai · g = bi for each 1 6 i 6 n = 1 we say that G acts locally n-arc transitively on Γ.
51
Lemma 3.4. G acts locally n-arc transitively if and only if the stabilizer of any path of
m vertices, (α1, . . . , αm), is transitive on Γ(αm)\αm−1 for each 1 6 m 6 n.
Proof. Suppose G acts locally n-arc transitively on Γ. Let 1 6 m 6 n and let (α1, . . . , αm)
be a path of m vertices in Γ. Suppose β1, β2 ∈ Γ(αm)\αm−1 then by n-arc transitivity,
there is some g ∈ G that maps the path (α1, . . . , αm, β1) to the path (α1, . . . , αm, β2) and
so β1g = β2.
Now suppose that (α1, α2, . . . , αn+1) and (β1, β2, . . . , βn+1) are two paths of n + 1
vertices in Γ with α1 and β1 lying in the same G-orbit. By edge-transitivity, there exists
g1 ∈ G such that α1, α2g1 = β1, β2. Now α3g1 ∈ Γ(β2)\β1 so, by assumption,
there exists g2 ∈ Gβ1,β2 such that α3g1g2 = β3 and then α1, α2, α3g1g2 = β1, β2, β3.
Continuing in this way proves G acts locally n-arc transitively on Γ.
52
Chapter 4
Background Results
In this chapter we include several known results which will be used in this thesis. We
discuss theorems of Feit and Thompson, Smith and Tyrer, and Hayden which are all
character theoretic as well as a result from representation theory by McLaughlin. In
contrast, we also discuss a graph theoretic theorem of Tits and Weiss.
The following result of Feit and Thompson (see [9] and [17]) characterizes a finite
group containing a self-centralizing element of order three. The proof determines some of
the character theory of a minimal counter example.
Theorem 4.1. (Feit-Thompson) Let G be a finite group containing a subgroup, X, of
order three such that CG(X) = X. Then one of the following hold.
(i) G contains a nilpotent normal subgroup, N , such that G/N is isomorphic to Sym(3)
or C3.
(ii) G contains a nilpotent normal 2-subgroup, N , such that G/N ∼= Alt(5).
(iii) G is isomorphic to PSL2(7) ∼= PSL3(2).
When applying Theorem 4.1 we will often need the following.
53
Lemma 4.2. (i) Neither Alt(5) or PSL2(7) ∼= PSL3(2) admit an outer automorphism
of order three.
(ii) If G is a group with a central involution, t, such that G/〈t〉 ∼= PSL2(7) then G ∼=
SL2(7).
Proof. It is well known (see [6]) that Alt(5) and PSL2(7) both have outer automorphism
groups of order two. It is also well known that the Schur multiplier of PSL2(7) has order
two so if G is a group satisfying G/N ∼= PSL2(7) where N is a central subgroup of G then
N has order two and G ∼= SL2(7).
The following result of Smith and Tyrer is of a similar nature to Theorem 4.1. It
describes a group containing a p-subgroup (odd prime p) with restricted normalizer. The
situation it describes is one which will appear several times in this thesis and so the
theorem is one of the most well used here. We will use it when we have a group with an
elementary abelian p-subgroup with restricted normalizer to say that the group has either
a proper normal subgroup at index a power of p or the group is p-soluble of length one.
First we define p-soluble.
Definition 4.3. A groupG is p-soluble if every composition factor ofG is either a p-group
or a p′-group.
Consider the following series
1 E Op′(G) E Op′,p(G) E Op′,p,p′(G) E . . .
where Op′,p(G) is the preimage in G of Op(G/Op′(G)) and Op′,p,p′(G) is the preimage in
G of Op′(G/Op′,p(G)) and so on. This series defines a minimal factorization of G into p
and p′ factors (minimal in the sense that the number of factors is as few as possible). We
call this the lower p-series for G.
54
Definition 4.4. A p-soluble group G has length n if there are n factors in the lower
p-series for G which are p-groups. In particular, G is p-soluble of length one if G =
Op′,p,p′(G). Alternatively, G is p-soluble of length one if for any Sylow p-subgroup, S, of
G, Op′(G)S E G.
Theorem 4.5. (Smith-Tyrer) Let G be a finite group and let P be a Sylow p-subgroup of
G for an odd prime p. Suppose P is abelian and [NG(P ) : CG(P )] = 2.
(i) If G is perfect then P is cyclic.
(ii) If P is non-cyclic, then Op(G) < G or G is p-soluble of length one.
Proof. See [27].
This theorem is used many times in Chapters 6 and 7. It is particularly well suited to
the problems encountered in Chapter 6 since each time it is used we can rule out one of
the possibilities given by the theorem. However the group structures which we encounter
in Chapter 7 satisfy both conclusions of the Smith-Tyrer theorem and so the theorem is
less useful in those cases. It is easy to construct such groups which are both p-soluble of
length one and have a proper p-residue. For example, consider the following subgroup of
Sym(10).
G := 〈(1, 2, 3), (4, 5, 6), (7, 8, 9, 10), (1, 2)〉.
Notice first that G has a normal Sylow 3-subgroup for which the centralizer of the Sylow 3-
subgroup has index two in G. Notice also that O3′(G) = 〈(7, 8, 9, 10〉 and then O3′,3(G) =
〈(7, 8, 9, 10), (1, 2, 3), (4, 5, 6)〉 has index two in G and so G is 3-soluble of length one.
However, O3(G) = 〈(7, 8, 9, 10), (1, 2), (1, 2, 3)〉 6= G and so G satisfies both conclusions of
Theorem 4.5.
Corollary 4.6. Let G be a group and let S be a Sylow p-subgroup of G (odd prime p).
Suppose S is elementary abelian of order at least p3 and |NG(S)/CG(S)| = 2. Finally, for
some x ∈ S of order p, suppose x ∈ Z(G). Then
55
(i) G is either p-soluble of length one or Op(G) < G; and
(ii) if G is not p-soluble of length one then Op(G) has index at least p2 in G and does
not contain x.
Proof. We apply the theorem of Smith and Tyrer to G to see that either G is p-soluble of
length 1 or Op(G) < G. We can apply these arguments again to G := G/〈x〉 to see that
G is either p-soluble of length one or Op(G) < G. So suppose G is p-soluble of length one.
Let N := Op′(G) and consider its preimage in G, N say. Then 〈x〉 is a Sylow p-subgroup
of N and so N splits over 〈x〉 and therefore N = M〈x〉 say. It follows that M is a normal
subgroup of G of order prime to p. Now MS = NS is normal in G since NS is normal in
G because G is p-soluble of length one. Thus we have that G is p-soluble of length one.
So we see that G being p-soluble of length one forces G to have the same property.
We now suppose that G does not have this property then neither does G and so both
have normal subgroups of index a proper power of p. By Gaschutz’s Theorem, there is a
complement to 〈x〉 in G, L say, which is necessarily normal in G. Thus Op(G) 6 L and
〈x〉 /∈ Op(G). So consider now the preimage in G of Op(G). This preimage gives a normal
subgroup of index at least p which contains x and so is distinct from L. Thus Op(G) is
contained in the intersection of the preimage of Op(G) and L and so G/Op(G) has order
at least p2 and does not contain x.
The following theorem is of the same nature as the previous theorems and is also
proved using characters. It can be found in [16, 3.3 p545].
Theorem 4.7. (Hayden) Let G be a finite group and let T ∈ Syl3(G) be elementary
abelian of order nine. Assume that
(i) NG(T )/CG(T ) ∼= 2× 2;
(ii) CG(T ) = T ; and
56
(iii) CG(t) 6 NG(T ) for each t ∈ T#.
Then G = NG(T ).
The following theorem of Jack McLaughlin comes from the paper Some Groups Gen-
erated by Transvections [21] and will be used in the proof of Theorem B. It can be used
to describe a situation in which an elementary abelian group is normalized by a group
generated by elements which each act trivially on a maximal subgroup.
Theorem 4.8. (McLaughlin) Let V be a vector space over GF(p) for odd prime p. Sup-
pose H is a subgroup of GL(V ) such that H = 〈h1, . . . , hr〉 where [V : CV (hi)] = p for
each 1 6 i 6 r. Furthermore, suppose that H contains no non-trivial normal p-subgroup.
Then for some n > 1,
V = V0 ⊕ V1 ⊕ . . .⊕ Vn,
where each Vi is a G-module and
H = H1 × . . .×Hn
where each Hj∼= SL(Vj) or Hj
∼= Sp(Vj). Finally for i 6= j, Hj acts trivially on Vi.
Proof. See [21].
We demonstrate an application of this theorem by introducing a scenario which will occur
inside the amalgam of type G2(3).
Lemma 4.9. Let M be a group and let V := O3(M) be elementary abelian and self-
centralizing in G of order 34. Let M = M/V and let R ∈ Syl3(M) have order 32 and
let S, T be non-conjugate subgroups of R of order three and suppose [V : CV (S)] = [V :
CV (T )] = 3. Then 〈SM〉 ∼= SL2(3), 〈TM〉 ∼= SL2(3) and [〈SM〉, 〈TM〉] = 1.
57
Proof. Let H = 〈SM, T
M〉 6 M . Since V is self-centralizing in M , H is isomorphic to
a subgroup of the group of automorphisms of V , Aut(V ) ∼= GL(V ). Since V = O3(M),
by correspondence, M contains no normal 3-subgroup. Thus the intersection of all its
Sylow 3-subgroups is trivial. Clearly H is a subgroup of M containing every Sylow 3-
subgroup of M and therefore H can have no normal 3-subgroup either. Thus we can
apply Theorem 4.8 to decompose both V and H. Therefore there is an integer n such
that V = V0 ⊕ V1 ⊕ . . . ⊕ Vn and H = H1 × . . . × Hn. We have several possibilities to
consider.
(i) n = 1. Then V = V0 ⊕ V1 where V1 has order 3, 32, 33 or 34 and so H ∼= 1, SL2(3),
SL3(3), SL4(3) or Sp4(3). However H has Sylow 3-subgroups of order 32 and so
n 6= 1.
(ii) n = 2. Then V = V0 ⊕ V1 ⊕ V2 where V1 has order 3 or 32 or 33 and V2 has order 3
or 32 and so H ∼= SL2(3)× 1, SL2(3)× SL2(3) or SL3(3)× 1. The only possibility is
H ∼= SL2(3)× SL2(3) again because of the order of the Sylow 3-subgroup.
(iii) n = 3. Then V = V0⊕V1⊕V2⊕V3 where V1 has order 3 or 32 and V2, V3 both have
order 3. In this case H ∼= SL2(3)× 1× 1 or 1× 1× 1. Neither of which is possible.
(iv) n = 4. Then V = V1⊕V2⊕V3⊕V4 where each has order 3 and so H ∼= 1× 1× 1× 1
which cannot happen.
Therefore V can be decomposed as V = V1 ⊕ V2 where Vi is a 2-dimensional H-module
for i = 1, 2 and H = SL(V1) × SL(V2). We consider the elements, x, of order three in
SL(V1)× SL(V2). There are three types.
(i) x = a where a ∈ SL(V1)#. These act trivially on V2 and fix a 1-subspace of V1 and
are all conjugate in SL(V1).
58
(ii) x = ab where a ∈ SL(V1)# and b ∈ SL(V2)
#. These act non-trivially on both V1 and
V2 but fix a 1-subspace in both.
(iii) x = b where b ∈ SL(V2))#. These act trivially on V1 and fix a 1-subspace of V2 and
are all conjugate in SL(V2).
Recall that S, T 6 H fix 3-dimensional subspaces of V and so have type (i) or (iii).
However S and T are non-conjugate and so, without loss of generality, we must have
〈SM〉 = SL(V1) and 〈TM〉 = SL(V2) and H = 〈SM〉 × 〈TM〉.
We now introduce a graph called a Moufang polygon. Such a graph is bipartite and
contains circuits of length 2n for some fixed integer n as well as satisfying the so-called
Moufang condition. Theorem B and Theorem C in this thesis rely on the classification of
Moufang polygons which was completed by Jacques Tits and Richard Weiss in 2002 (see
[29]).
Definition 4.10. The diameter of a graph, ∆, is n where n is the largest possible integer
for which there exists vertices α and β in ∆ with d(α, β) = n. The girth of ∆ is the
smallest integer m where ∆ contains a circuit of length m.
Definition 4.11. A generalized n-gon is a bipartite graph which has girth n and diam-
eter 2n. In particular, a generalized hexagon is a generalized 6-gon and a generalized
quadrangle is a generalized 4-gon.
Definition 4.12. Let ∆ be a generalized n-gon in which each vertex has valency three
or more. Let Θ = (α0, . . . , αn) be a path of length n which is contained in a cir-
cuit of length 2n. We set UΘ to be the stabilizer in Aut(∆) of the set of vertices
Γ(α1),Γ(α2), . . . ,Γ(αn−1) and call UΘ the root group corresponding to Θ. If UΘ is
transitive on Γ(αn)\αn−1. Then ∆ is said to satisfy the Moufang condition.
59
Let ∆ be a generalized n-gon then ∆ is a bipartite graph so we consider the subgroup of
the automorphism group of ∆ which preserves the partition. We call this group Auto(∆).
Now let Θ = (α0, α1, . . . , αn) be a path of n + 1 vertices contained in an n-cycle and
consider the root group UΘ. The subgroup of Auto(∆) generated by all root groups is a
normal subgroup of Auto(∆) which we call G†.
Theorem 4.13. (Tits/Weiss) Let ∆ be a finite generalized quadrangle or hexagon satis-
fying the Moufang condition. Then ∆ is uniquely defined by its order. In particular, if
∆ is a Moufang quadrangle with 80 vertices then G† ∼= U4(2) ∼= PSp4(3) and if ∆ is a
Moufang hexagon with 728 vertices then G† ∼= G2(3).
Proof. See section 17 and 34 in [29]. In particular pages 378-379.
60
Chapter 5
An Amalgam of Type PSL3(3)
In [23] amalgams of type PSL3(pa) are classified when a > 1 and when p is an odd prime.
The methods used to identify completions for a > 1 do not extend to the smaller cases
of type PSL3(p). In this chapter we describe an amalgam of type PSL3(3). In PSL3(3)
there are two maximal parabolic subgroups which intersect at the normalizer of a Sylow
3-subgroup. The amalgam consists of these two parabolic subgroups and we will assume
we have a completion of this amalgam which satisfies a 3-local condition. Under such
conditions the amalgam completes to PSL3(3) and also to the sporadic simple group M12.
We can describe the amalgam in each group as follows. Consider the action of a group
G1∼= PSL3(3) on a vector space of dimension three over a field of order three. Let
A1 := StabG1(〈(1, 0, 0)〉) and B1 := StabG1(〈(1, 0, 0), (0, 1, 0)〉). Now consider the action
of G2∼= M12 on a set of twelve points Ω := 1, 2, . . . , 12. Let A2 = StabG2(1, 2, 3) and
B2 = StabG2(1, 2, 34, 5, 67, 8, 910, 11, 12). A result of Goldschmidt can be used
to prove that the two amalgams (A1, B1, A1 ∩B1) and (A2, B2, A2 ∩B2) are isomorphic.
The two groups, PSL3(3) and M12, share an isomorphic amalgam and an isomorphic
centralizer of a 3-central element but differ in their centralizer of an involution. We will
apply the theorem of Feit and Thompson (Theorem 4.1) to the involution centralizer in
61
an abstract finite completion of the amalgam which satisfies a 3-local condition. This,
together with a result of Burnside about fixed-point-free automorphisms of order three
(Lemma 1.24), will allow us to restrict the structure of the completion. Furthermore, the
Burnside lemma we use will give us a particular group relation. The techniques which will
be used in later chapters are not suitable in this amalgam since the subgroups are small.
Instead we use a computer algebra package. We will create the subgroups which constitute
the amalgam in Magma and then make the universal completion as the amalgamated
free product. Factoring the universal completion by a relation forced by Lemma 1.21 and
using the coset enumeration package will then confirm the only possible completions are
M12 and PSL3(3).
Hypothesis A. Let G be a finite group with subgroups A ∼= B such that C := A ∩
B contains a Sylow 3-subgroup of both A and B and such that no non-trivial normal
subgroup of C is normal in both A and B. Suppose further that
(i) G = 〈A,B〉;
(ii) A/O3(A) ∼= B/O3(B) ∼= GL2(3);
(iii) O3(A) and O3(B) are natural modules with respect to the actions of A/O3(A) and
B/O3(B) respectively; and
(iv) for S ∈ Syl3(C), NG(Z(S)) = NA(Z(S)) = NB(Z(S)).
Our main theorem in this chapter is the following.
Theorem A. If G satisfies Hypothesis A, then G ∼= M12 or PSL3(3).
62
5.1 Some Structure of the Amalgam (A,B,C)
We assume Hypothesis A and define some further notation. Set QA := O3(A) and QB :=
O3(B).
Lemma 5.1. The following hold.
(i) S := QAQB is a normal Sylow 3-subgroup of C and S ∼= 31+2+ .
(ii) QA ∩QB = Z(S).
(iii) C = NA(S) = NB(S) = NA(Z(S)) = NB(Z(S)) = NG(Z(S)) has index four in A
and in B.
(iv) CG(Z(S)) has index two in C.
Proof. (i) If QA = QB then A and B would share a normal subgroup which is not possible.
Also, Sylow 3-subgroups of A and B have order 33 and C contains one of them so C
contains QA and QB which necessarily normalize each other. Thus S := QAQB ∈ Syl3(C)
is normal in C. Now choose some x ∈ QB\QA then 1 6= QAx ∈ A/QA∼= GL2(3). Since QA
is a natural A/QA-module, QAx acts non-trivially on QA. In particular, S is non-abelian.
If Z(S) has index three in S then we could choose any x ∈ S\Z(S) which would necessarily
commute with 〈x, Z(S)〉 = S and S would be abelian. Therefore Z(S) must have order
three. It is then immediate that Z(S) = S ′. Furthermore S/Z(S) = ( QA
Z(S))( QB
Z(S)) is
elementary abelian so S is extraspecial. Now since QA and QB have exponent three, it
follows that S ∼= 31+2+ .
(ii) Since QA and QB are not equal and have index three in S, QA ∩ QB has order
three and is a normal subgroup of S. It is therefore central in S and so QA ∩QB = Z(S).
(iii) The normalizer of a Sylow 3-subgroup in A is the preimage in A/QA∼= GL2(3)
of the normalizer of a Sylow 3-subgroup which has index four. Therefore the normalizer
63
of a Sylow 3-subgroup in A (and in B) has index four. We have NA(S) 6 NA(Z(S)) =
NB(Z(S)) 6 B and NB(S) 6 NB(Z(S)) = NA(Z(S)) 6 A and so C 6 NA(S), NB(S) 6
A ∩B = C which means C = NA(S) +NB(S) has index four in A and in B.
(iv) We have seen that C = NA(S) = NB(S) 6 NG(Z(S)). Since NG(Z(S)) =
NA(Z(S)) = NB(Z(S)) 6 A ∩ B, C = NG(Z(S)). Choose t ∈ A such that 〈tQA〉 =
Z(A/QA). Then t normalizes S and so lies in C. Also t inverts the natural module QA
and so t inverts Z(S) which is not central in C. Therefore CC(Z(S)) has index two in
C.
Lemma 5.2. If t ∈ A is an involution such that tQA ∈ Z(A/QA) then CA(t) ∼= GL2(3)
and there is x ∈ CA(t) of order three such that 〈x〉 = Z(P ) for some P ∈ Syl3(B).
Proof. Let t ∈ A be an involution such that tQA ∈ Z(A/QA) ∼= Z2. Then tQA commutes
with each Sylow 3-subgroup of A/QA and in particular commutes with QAQB/QA∼=
QB/(QA∩QB). By Lemma 2.16, there is some x ∈ QB\QA that commutes with t. Notice
that A has at most 27 Sylow 2-subgroups of A and since t is in the centre of a Sylow
2-subgroup, at most 27 A-conjugates of t. However, t commutes with an element of order
three, x ∈ QB\QA, and so there can be at most nine A-conjugates of t in A. Moreover t
inverts QA and so the set tQA = tq|q ∈ QA has order nine and therefore 〈t〉QA = 〈t〉A. By
a Frattini argument (Lemma 1.7), A = NA(〈t〉)QA = CA(t)QA and so by an isomorphism
theorem,
CA(t) ∼=CA(t)
CA(t) ∩QA
∼=CA(t)QA
QA
=A
QA
∼= GL2(3).
Finally, since x ∈ Q#B and QB is a natural B/QB-module, Lemma 2.15 (ii) implies that
〈x〉 is conjugate to Z(S) in B. Thus there exists P ∈ Syl3(B) such that 〈x〉 = Z(P ).
Lemma 5.3. Let u ∈ C be an involution then CS(u) has order three. In particular, if u
is an involution in G normalizing a Sylow 3-subgroup P of G then |CP (u)| = 3.
64
Proof. We can find a complement, T ∼= 2 × 2, to S in C by Sylow’s Theorem and
TQA/QA 6 A/QA∼= GL2(3) and TQB/QB 6 B/QB
∼= GL2(3). By Lemma 2.13, any
subgroup of GL2(3) isomorphic to 2×2 contains the central element. So choose t ∈ T such
that tQA ∈ Z(A/QA) and choose s ∈ T such that sQB ∈ Z(B/QB). We have seen that t
inverts each element of QA and centralizes some element x ∈ QB\QA. Similarly s inverts
QB and centralizes some element y ∈ QA\QB. In particular, t 6= s and therefore T = 〈t, s〉.
Now we see that st centralizes the group QA∩QB = Z(S) and inverts the groups 〈x〉 and
〈y〉. So suppose CS(st) had order nine. Then S = CS(st)〈x〉 = CS(st)〈y〉 and therefore
x = zy for some z ∈ CS(st). However this implies y−1z−1 = x−1 = xst = zstyst = zy−1
and so [y, z] = z2. Therefore z ∈ S ′ = Z(S) (Lemma 5.1). However this implies that
z2 = [y, z] = 1 and so z = 1 and x = y which is not possible. Hence each element of T#
centralizes just an element of order three in S.
Lemma 5.4. Let t ∈ A be as in Lemma 5.2 then CG(t)/〈t〉 contains a self-centralizing
element of order three.
Proof. We have seen that there is an element of order three, x ∈ CA(t), such that 〈x〉 is
the centre of some Sylow 3-subgroup in B. By Lemma 5.1 (iv), CG(x) has order 332 and
is contained in B. More precisely, CG(x) = P 〈t〉 for some P ∈ Syl3(B). Now Lemma
5.3 implies that CP (t) has order three and so CG(x) ∩ CG(t) = 〈x, t〉. Therefore x〈t〉 is a
self-centralizing element of order three in CG(t)/〈t〉.
5.2 A Presentation for the Amalgam
Lemma 5.5. A ∼= B ∼= AGL2(3)
Proof. If V is a 2-dimensional vector space over GF(3) then AGL2(3) is defined to be the
set of affine transformations φ + w where w ∈ V and φ ∈ GL(V ) and for each v ∈ V ,
φ+ w : v 7→ φ(v) + w. There are 9|GL(V )| = 2433 such transformations.
65
Notice that by Lemma 5.2, we can choose an involution t ∈ A such that CA(t) ∼=
GL2(3) and so D := CA(t) is a complement to QA in A. We will define a map, Θ, from
A to AGL2(3) as follows. Each a ∈ A can be written uniquely as a = bc where b ∈ QA
and c ∈ D. Now, for each v ∈ QA∼= V , define Θ(bc) : v 7→ b(vc−1
). We check that Θ is a
homomorphism. Let bc and b1c1 be in A such that b, b1 ∈ QA and c, c1 ∈ D. There exists
b2 ∈ QA such that b2 = cb1c−1 and so bcb1c1 = bb2cc1. Therefore Θ(bcb1c1) = Θ(bb2cc1)
and for v ∈ V ,
Θ(bb2cc1)(v) = bb2(vc−11 c−1
) = bbc−1
1 (vc−11 c−1
) = b((b1vc−11 )c−1
) = Θ(bc)(Θ(b1c1)(v)).
So Θ is a homomorphism. Suppose bc ∈ Ker(Θ) then for each v ∈ QA, bvc−1= v and
so bv2c−1
= v2. However bv2c−1
= b(vc−1)2 = b(b−1v)2 = b−1v2 and it follows that b = 1.
Therefore vc = v for all v ∈ QA. However QA is a natural A/QA-module so only the
identity acts trivially (see Lemma 2.15 (iv)). Therefore the kernel of this map is trivial
and since |A| = |AGL2(3)|, the map is an isomorphism.
There are two faithful representations of AGL2(3) in PSL3(3) which we shall see.
Consider an amalgam of type (X,Y, Z) where X ∼= Y ∼= AGL2(3) and Z ∼= NX(S)
where S ∈ Syl3(X). BothX and Y have an elementary abelian normal 3-subgroup, O3(X)
and O3(Y ) respectively, of order 9. Suppose φ1 and φ2 define embeddings of Z into X and
Y respectively. The image of Z in X contains O3(X) and the image of Z in Y contains
O3(Y ). If H is some completion of (X, Y, Z) we can ask if the corresponding images in H
share the same 3-radical subgroup. We easily see instances of both possibilities and so can
immediately identify at least two isomorphism classes of this amalgam. Firstly suppose
X = Y . Such an amalgam cannot be isomorphic to the following amalgam of subgroups
66
in H ∼= PSL3(3). Let H act on the vector space of dimension three over GF(3) and set
X = StabH(〈(1, 0, 0)〉)
=
α β 0
γ δ 0
ε ζ η
∣∣α, β, γ, δ, ε, ζ ∈ GF(3), η = (αδ − βγ)−1 ∈ GF(3)\0
and
Y = StabH(〈(1, 0, 0), (0, 1, 0)〉)
=
η 0 0
ε α β
ζ γ δ
∣∣α, β, γ, δ, ε, ζ ∈ GF(3), η = (αδ − βγ)−1 ∈ GF(3)\0
.
Let A represent the isomorphism class of this amalgam (X,Y, Z).
Lemma 5.6. Let X ∼= AGL2(3) and Z = NX(S) where S ∈ Syl3(X). Then the following
hold.
(i) Both Z and X have trivial centres.
(ii) CX(Z) = 1 and NX(Z) = Z.
(iii) Z/S ∼= 2× 2.
(iv) CZ(S) = Z(S)
Proof. The representation of AGL2(3) in PSL3(3) makes checking most of these results
trivial when we identify Z with the group
α 0 0
δ β 0
ε ζ γ
∣∣α, β, γ, δ, ε, ζ ∈ GF(3), αβγ = 1
.To show NX(Z) = Z we observe that S E Z so Z 6 NX(Z) 6 NX(S) = Z.
Lemma 5.7. Let X ∼= AGL2(3) and Z ∼= NX(S) where S ∈ Syl3(X).
67
(i) Inn(Z) ∼= Z and |Aut(Z)/Inn(Z)| = 2.
(ii) Aut(X) = Inn(X) ∼= X.
Proof. (i) By Lemma 5.6, Z has trivial centre so we identify Z ∼= Inn(Z) with its image
in Aut(Z). Let Aut(Z) ≥ R > Z be such that R/Z has prime order q. Suppose first that
CR(S) > CZ(S) = Z(S) then CR(S)/CZ(S) has order q. Since Z is soluble and R/Z is
cyclic of prime order, R is soluble. The Fitting subgroup of R contains all nilpotent normal
subgroups of R so S 6 F (R). By Lemma 1.13, CR(F (R)) 6 F (R) so F (R) > CR(S).
Since F (R) is nilpotent, it is a direct product of its Sylow subgroups so as no element
of order two in Z commutes with S, we have F (R) ∩ Z = S. Therefore it follows that
|F (R)| = 33q. If q 6= 3 then R contains a normal subgroup, N say, of order q and then
[N,Z] 6 N ∩ Z = 1 but then N would act trivially on Z, thus q = 3. Let T be a Sylow
2-subgroup of Z then T acts on the normal subgroup CR(S). Moreover, this action is
coprime so CR(S) = [CR(S), T ]CCR(S)(T ). However [CR(S), T ] 6 CR(S) ∩ Z = CZ(S)
and so 1 6= CCR(S)(T ). This means that some x ∈ R\Z is trivial on 〈T, S〉 = Z which
is a contradiction. So suppose instead that CR(S) = CZ(S). Then R/CR(S) embeds
into Aut(S) and in particular R/S embeds into Out(S) ∼= GL2(3) (Lemma 1.16). Now
2× 2 ∼= Z/S E R/S and any subgroup of GL2(3) isomorphic to 2× 2 is conjugate to the
group of diagonal matrices (see Lemma 2.13). So we identify Z/S with the subgroup of
diagonal matrices in GL2(3) which sits at index two in its normalizer in GL2(3). However,
this implies that q = 2. Therefore the only prime dividing Aut(Z)/Inn(Z) is two. So we
repeat the same argument since 2 × 2 ∼= Z/S E Aut(Z)/S and so |Aut(Z)/Inn(Z)| 6
2. Finally we see that Aut(Z)/Inn(Z) has order exactly two with the following outer
automorphism.
α :
α 0 0
γ αβ 0
δ ε β
7→
β 0 0
εα αβ 0
γε − αβδ γβ α
.
(ii) The centre of X is trivial so we immediately identify X ∼= Inn(X) with its image
68
inside Aut(X). Let Q := O3(X) ∼= 32 and let R be such that Aut(X) ≥ R > X and R/X
is cyclic of prime order q. By an isomorphism theorem,
Aut(Q) ∼= GL2(3) ∼=X
Q=
X
CX(Q)=
X
X ∩ CR(Q)∼=XCR(Q)
CR(Q)6
R
CR(Q)→ Aut(Q).
So R/CR(Q) ∼= GL2(3) and CR(Q)/Q must have order q. Now CR(Q) E R and has order
32q. Suppose q 6= 3 then R contains a normal subgroup, N say, of order q. But then
[N,X] 6 N ∩ X = 1 and N acts trivially on X which is a contradiction. Hence q = 3
and CR(Q) is a group of order 27. Now by a Frattini argument, R = XNR(S) and by an
isomorphism theorem.
NR(S)
Z=
NR(S)
NR(S) ∩X∼=R
X∼= C3.
In particular, NR(S) normalizes Z and so NR(S) = NR(Z). By part (i), |Aut(Z)|3 = 33
and so |NR(Z)/CR(Z)|3 6 33 and it follows that 3 divides CR(Z) and so R = X : CR(Z).
Now CR(Z) 6 CR(t) where t =
−1 0 0
0 −1 0
0 0 1
and CX(t) ∼= GL2(3). Therefore CR(Z)
normalizes O2(CX(t)) ∼= Q8 and so by coprime action we get
O2(CX(t)) = CO2(CX(t))(CR(Z))[O2(CX(t)), CR(Z)].
However CR(Z) 6 CR(Q) E R so [O2(CX(t)), CR(Z)] 6 O2(CX(t)) ∩ CR(Q) = 1. Thus
CR(Z) commutes with 〈Z,O2(CX(t))〉 = X which is a contradiction.
Lemma 5.8. There are exactly two isomorphism classes of amalgams of type (X, Y, Z)
with X ∼= Y ∼= AGL2(3) and Z ∼= NX(S) where S ∈ Syl3(X).
Proof. We will use the Goldschmidt Lemma (3.1), Lemma 5.6 and Lemma 5.7. Firstly,
X∗ = NAut(X)(Z)/CAut(X)(Z) ∼= NX(Z)/CX(Z) ∼= NX(Z) ∼= Z
and similarly Y ∗ ∼= Z giving us exactly two (X∗, Y ∗)-double cosets in Aut(C). Now the
69
Goldschmidt Lemma implies that there are exactly two isomorphism classes of amalgams
of type (X, Y, Z).
Corollary 5.9. Suppose A1 = (X1, Y1, Z1) is an amalgam with X1∼= Y1
∼= AGL2(3) and
Z1∼= NX(S) where S ∈ Syl3(X1). If O3(X1) 6= O3(Y1) then A1 is isomorphic to A.
Proof. We have seen there are exactly two isomorphism types of this amalgam and we
have seen examples of both. The first can be seen when X1 = Y1 and the second can be
seen in PSL3(3) which can, in particular, be characterized by the fact that the 3-radical
subgroups in the two groups constituting the amalgam are distinct.
Lemma 5.10. (A,B,C = A ∩B) is isomorphic to the amalgam A.
Proof. Lemma 5.5 gives A ∼= B ∼= AGL2(3) and Lemma 5.1 gives that C = NA(S) for
some S ∈ Syl3(A). If O3(A) = O3(B) then A and B would share a normal subgroup and
so O3(A) 6= O3(B) and so the amalgam (A,B,C) is isomorphic to A.
We can now identify our amalgam (A,B,C) with the amalgam A of matrices seen in
PSL3(3). In the following lemma we give a presentation for the universal completion of
the amalgam A.
Lemma 5.11. Let
RC = a3 = b3 = [a, b]3 = [a, [a, b]] = [b, [a, b]] = t2 = u2 = [t, u] = 1,
au = a, bu = b−1, at = a−1, bt = b,
RA = RC ∪ p4 = q4 = 1, p2 = q2 = t, [a, b]p = a, [a, b]q = [b, a]a,
pu = p−1, pb = pq, qb = p,
and
RB = RC ∪ r4 = s4 = 1, r2 = s2 = u, [a, b]r = b−1, [a, b]s = b−1[b, a],
rt = r−1, ra = rs, sa = r.
70
Then C ∼= 〈a, b, t, u|RC〉, A ∼= 〈a, b, p, q, t, u|RA〉, B ∼= 〈a, b, r, s, t, u|RB〉 and F :=
〈a, b, p, q, r, s, t, u|RA,RB〉 is the universal completion of A.
Proof. We make the following identifications of matrices in C to elements in the presen-
tation.
a ∼
1 0 0
1 1 0
0 0 1
, b ∼
1 0 0
0 1 0
0 1 1
, t ∼
1 0 0
0 −1 0
0 0 −1
, u ∼
−1 0 0
0 −1 0
0 0 1
.
Therefore 〈a, b, t, u|RC〉 is a group with a quotient group isomorphic to C since the el-
ements of C satisfy the relations. Coset enumeration using Magma ([4]) ensures it is
isomorphic to C. Similarly if we identify
p ∼
1 0 0
0 0 1
0 −1 0
, q ∼
1 0 0
0 −1 1
0 1 1
then coset enumeration gives 〈a, b, p, q, t, u|RA〉 ∼= AGL2(3) and again if we identify
r ∼
0 −1 0
1 0 1
0 −1 0
, s ∼
−1 1 0
1 1 0
0 0 1
then 〈a, b, r, s, t, u|RB〉 ∼= AGL2(3). By Corollary 5.9, the amalgam
(〈a, b, p, q, t, u|RA〉, 〈a, b, r, s, t, u|RB〉, 〈a, b, t, u|RC〉)
is isomorphic to A and so F := 〈a, b, p, q, r, s, t, u|RA,RB〉 is the universal completion of
the amalgam (see [26, Thm 1 p3] for justification that F is universal).
Recall Lemma 5.2 and Lemma 5.4 where we found an involution in A whose centralizer
in A was isomorphic to GL2(3). Notice that with the identifications we made in Lemma
5.11, the involution t satisfies the properties described in Lemma 5.2. For the remainder
71
of this chapter we identify A and B with the finitely presented groups and in particular
we identify t ∈ A = 〈a, b, p, q, t, u|RA〉 with the involution in Lemmas 5.2 and 5.4. We
also identify G with some quotient of F .
Lemma 5.12. Let G′ be any faithful completion of A then CG′(t) contains two (not neces-
sarily distinct) subgroups isomorphic to Q8 both of which are normalized but not centralized
by the group 〈b, u〉 ∼= Sym(3).
Proof. It is clear from the presentation that P := 〈p, q〉 and 〈r, s〉 are quaternion groups
of order eight. Notice that upr = t follows from the relations pu = p−1 and rt = r−1. So
R := 〈r, s〉pr is a quaternion group with central element t. It follows from the relations of
〈a, b〉 as an extraspecial group of order 27 that [a, b]−1 = [a−1, b] and so we notice that apr =
[a, b]p2r = [a, b]tr = [a−1, b]r = ([a, b]−1)r = b. Therefore prb = apr and so 〈rprb, sprb〉 =
〈rs, r〉pr is normalized by 〈b〉. Also pru = pur = up−1r = utpr and so 〈rpru, spru〉 =
〈r−1, sr〉pr is normalized by 〈u〉 also. We point out that any faithful completion of the
amalgam contains injective images of the groups constituting the amalgam. Therefore
the groups 〈p, q〉, 〈r, s〉 and 〈b, u〉 will always have the isomorphism types claimed.
We continue to set P = 〈p, q〉 and R = 〈x, y〉, where x = rpr and y = spr.
Lemma 5.13. CG(t) contains a nilpotent normal subgroup, N and P,R 6 N . In partic-
ular P and R generate a 2-group.
Proof. By Lemma 5.4, we can apply Theorem 4.1 to CG(t)/〈t〉. Suppose CG(t)/〈t〉 ∼=
PSL2(7) then by Lemma 4.2, CG(t) ∼= SL2(7) which only contains one element of order
two. This is a contradiction since we know that t commutes with another involution
u ∈ C. Therefore CG(t)/〈t〉 contains a nilpotent normal subgroup of order prime to 3.
Let N be the preimage in CG(t) of this normal subgroup then 〈t〉 ≤ N E CG(t) and
CG(t)/N ∼= Alt(5) or Sym(3) or C3. By Lemma 5.12, CG(t) contains the two groups
P ∼= R ∼= Q8 which are normalized in CG(t) by 〈b, u〉 ∼= Sym(3). We consider P ∩ N .
72
Notice first that D := P 〈b, u〉 ∼= GL2(3) so O3′(D) = O2(D) = P ∼= Q8. Also N ∩D is a
normal 3′ subgroup and so N ∩D = N ∩ P . Suppose P ∩N = 〈t〉 then
∣∣∣∣CG(t)
N
∣∣∣∣ > ∣∣∣∣DNN∣∣∣∣ =
∣∣∣∣ D
N ∩D
∣∣∣∣ =243
2= 233
which is not possible. So suppose |P ∩ N | = 4. The group of order three, 〈b〉, in D
acts regularly on P/〈t〉 and normalizes N and so we have a contradiction since P =
〈(P ∩ N)〈b〉〉 6 N . Thus P 6 N . We can argue in the same way with R since R is
normalized but not centralized by 〈b, u〉. Thus P,R 6 N . Since N is a nilpotent group
and P and R are 2-groups, 〈P,R〉 is a 2-group.
We have that G is a faithful completion of the amalgam A so is a quotient of the
universal completion F . Furthermore it contains the 2-group generated by P and R and
so 〈P,R〉 satisfies the congruences described in Corollary 1.23. Therefore, for any α ∈ P
and any β ∈ R, [αb, β][βb, α] ∈ 〈t〉. We make the following four sets of relations:
(i) R1 = [yb, q][qb, y] = 1, [pqb, x][xb, pq] = 1;
(ii) R2 = [yb, q][qb, y] = t, [pqb, x][xb, pq] = 1;
(iii) R3 = [yb, q][qb, y] = 1, [pqb, x][xb, pq] = t;
(iv) R4 = [yb, q][qb, y] = t, [pqb, x][xb, pq] = t.
Lemma 5.14. |G| divides 95040 or |G| divides 5616.
Proof. G is a quotient of one of the following groups
F1 = 〈a, b, p, q, r, s, t, u|RA,RB,R1〉,
F2 = 〈a, b, p, q, r, s, t, u|RA,RB,R2〉,
F3 = 〈a, b, p, q, r, s, t, u|RA,RB,R3〉,
73
F4 = 〈a, b, p, q, r, s, t, u|RA,RB,R4〉.
Coset enumeration, using Magma ([4]), gives respective group orders of 95040, 1, 5616,
1.
A final corollary proves Theorem A
Corollary 5.15. G ∼= M12 or PSL3(3).
Proof. Both M12 and PSL3(3) satisfy Hypothesis A and so are completions of the amalgam
A. Therefore each must be a quotient of one of the four groups presented in the previous
lemma. However the orders of the two non-trivial groups give the orders of M12 and
PSL3(3). It follows that F1∼= M12 and F3
∼= PSL3(3).
74
Chapter 6
An Amalgam of Type G2(3)
In Local Characteristic p Completions of Weak BN-Pairs [23] a completion of an amalgam
of type G2(pa) (odd prime p) is proven to be isomorphic to G2(p
a) under a K-proper
hypothesis and some strong p-local conditions. An amalgam of type G2(3) is an amalgam
of groups comprising two maximal 3-local subgroups in G2(3) which intersect at the
normalizer of a Sylow 3-subgroup. Here we will prove that a group, G, which contains
a faithful completion, G, of an amalgam of type G2(3) and satisfies a 3-local condition
must be isomorphic to G2(3). We will describe the structure of this amalgam as in [23]
and analyse the conjugacy classes of elements of order three which we see. This allows
us to fully determine the centralizer of an involution. We consider the coset graph given
by a completion of the amalgam and this will turn out to be a generalized 6-gon with
728 vertices. Thus we can appeal to the classification of generalized polygons by Tits and
Weiss to recognize the uniqueness of this graph and force the completion of the amalgam
to be isomorphic to G2(3). We are left to rule out the possibility of a strict containment
G > G. However knowledge of the full centralizer of an involution gives us that G is
strongly embedded in G and a contradiction follows.
Limiting the structure of the centralizers of 3-elements and proving that exactly 5
75
conjugacy classes exist allows us to limit the structure of many 3-local subgroups. Thus
a modular character theoretic result of Stephen Smith and Peter Tyrer proves useful and
we use it on three occasions (see Theorem 4.5).
The hypothesis we will assume is:
Hypothesis B. Let G be a finite group with non-conjugate subgroups A1, A2 such that
A12 := A1 ∩ A2 contains a Sylow 3-subgroup of both A1 and A2 and such that no non-
trivial normal subgroup of A12 is normal in both A1 and A2. Suppose further that, for
i = 1, 2,
(i) |O3(Ai)| = 35;
(ii) Ai/O3(Ai) ∼= GL2(3);
(iii) O3(Ai)/Z(O3(Ai)) and Z(O3(Ai))/O3(Ai)′ are natural modules with respect to the
action of Ai/O3(Ai);
(iv) NG(O3(Ai)′) = Ai.
The main theorem of this chapter is:
Theorem B. Let G be a group satisfying Hypothesis B. Then G ∼= G2(3).
6.1 The Amalgam
We assume Hypothesis B and define some further notation. Let G := 〈A1, A2〉 6 G then
G is a faithful completion of the simple amalgam (A1, A2, A12). We can thus introduce the
coset graph Γ = Γ(G,A1, A2, A12) and by naming the vertices α = A1, β = A2 ∈ V (Γ) we
get Gα := A1 = StabG(A1), Gβ := A2 = StabG(A2) and Gαβ := A12 = StabG(A1, A2).
Let γ ∈ α, β and set
76
(i) Qγ = O3(Gγ) - this has order 35 and is normal in Gγ;
(ii) Zγ = Z(Qγ) - this is also normal in Gγ and has order 33 since Qγ/Zγ is a natural
module so has order 32;
(iii) Yγ = Q′γ - this has order three and is normalized by Gγ;
and choose Sαβ ∈ Syl3(Gαβ). We also introduce the following notation for the correspond-
ing intersections Qαβ := Qα ∩Qβ, Zαβ := Zα ∩ Zβ.
We now begin to build up an understanding of the structure of the subgroups in the
amalgam (Gα, Gβ, Gαβ).
Lemma 6.1. The following hold.
(i) Sαβ = QαQβ E Gαβ.
(ii) Zαβ = Z(Sαβ) = YαYβ has order 32.
(iii) For γ, δ = α, β, Zγ 6 Qδ.
(iv) ZαZβ = Qαβ is elementary abelian of order 34.
(v) Qγ has exponent 3, for γ ∈ α, β.
These results are from Lemma 6.5 in [23]. In the following proof we use Lemma 2.15 several
times. Lemma 2.15 gives us information on the natural GL2(3)-module and in particular
that elements of order three in Gγ/Qγ act non-trivially on the natural module and that
Gγ/Qγ is transitive on vectors and 1-subspaces of the natural module (γ ∈ α, β).
Proof. (i) We have that Qα 6= Qβ as the amalgam is simple and so QαQβ ∈ Syl3(Gαβ).
Moreover Qα E Gαβ and Qβ E Gαβ and so QαQβ is normal in Gαβ and therefore is the
unique Sylow 3-subgroup of Gαβ. Thus Sαβ = QαQβ.
77
(ii) Notice first that both Yα and Yβ are central in Sαβ. Also Zα is central in Qα and
Zβ is central in Qβ and so Zαβ is central in QαQβ = Sαβ. Suppose Z(Sαβ) Qα then
Sαβ = QαZ(Sαβ). However this would mean that Sαβ/Qα acts trivially on the natural
modules Qα/Zα and Zα/Yα which is not possible by Lemma 2.15 and so Z(Sαβ) 6 Z(Qα)
and similarly Z(Sαβ) 6 Z(Qβ). Hence Z(Sαβ) = Zαβ. Also, since the amalgam is simple,
Zα 6= Zβ and Yα 6= Yβ. Thus YαYβ 6 Z(Sαβ) Zα, Zβ and so we get the order 32.
(iii) Let γ, δ = α, β and suppose that Zγ Qδ. Then
Zγ > Zγ ∩Qδ > Zγ ∩ Zδ = Z(Sγδ)
by (ii). Hence Zγ ∩Qδ = Z(Sγδ) follows by considering group orders. So,
[Zγ, Qδ] 6 Zγ ∩Qδ = Z(Sγδ) 6 Zδ.
Therefore, Sγδ/Qδ = ZγQδ/Qδ acts trivially on the natural module Qδ/Zδ which is not
possible by Lemma 2.15. Hence Zγ 6 Qδ.
(iv) We have seen that Qα 6= Qβ and that both subgroups have index three in Sαβ.
Thus Qαβ has index 32 in Sαβ and so has order 34. We have also seen that Zα 6= Zβ and
that Zα ∩ Zβ has order 32. So by an isomorphism theorem, C3∼= Zβ/Zαβ
∼= ZαZβ/Zα
and so ZαZβ has order 34. However Zα 6 Qα ∩ Qβ and Zβ 6 Qβ ∩ Qα by (iii) and so
ZαZβ = Qαβ. Now let γ, δ = α, β and consider the cosets in Zγ/Yγ. Every element
in the cosets Zγδ/Yγ has order dividing three and Gγ/Qγ is transitive on the non-identity
cosets of Zγ/Yγ by Lemma 2.15 so every element in each coset of Zγ/Yγ has order dividing
three. Hence Zα and Zβ are elementary abelian. Furthermore Zα 6 Qβ and Zβ = Z(Qβ)
by (iii) so [Zα, Zβ] = 1. Thus ZαZβ is elementary abelian.
(v) We have seen that ZαZβ = Qα ∩ Qβ is elementary abelian. Let γ, δ = α, β.
The elements of the cosets of ZγZδ/Zγ have order dividing three. We have that Gγ/Qγ∼=
78
GL2(3) acts transitively on the non-trivial elements of Qγ/Zγ and therefore the elements
of every coset of Zγ in Qγ have order dividing three. Hence Qγ has exponent three.
Lemma 6.2. (i) If u ∈ Qα\Qβ and v ∈ Qβ\Qα, [v, u, u] 6= 1 6= [v, u, v].
(ii) If z ∈ Sαβ has order three, then z ∈ Qα ∪Qβ.
(iii) Zαβ S ′αβ Qαβ and S ′
αβ 6= Zα, Zβ.
(iv) The cube of any element of order nine lies in the set ab|a ∈ Y #α , b ∈ Y
#β .
Proof. (i) Let u ∈ Qα\Qβ and v ∈ Qβ\Qα We show that [v, u, u] 6= 1, the proof for
[v, u, v] is similar. Suppose [v, u] ∈ Zα. Since u ∈ Qα and v /∈ Qα, we consider the action
of Qαv ∈ Gα/Qα# on Zαu ∈ Qα/Zα
#. Since [v, u] ∈ Zα, Qαv fixes Zαu. By Lemma
2.15, Qαv centralizes only the subspace ZαZβ/Zα = Qαβ/Zα. Thus u ∈ Qαβ which is
a contradiction since u /∈ Qβ. Now suppose [v, u, u] = 1. Then CQαβ(u) > 〈Zα, [v, u]〉
and |〈Zα, [v, u]〉| = 34. Therefore u commutes with Qαβ. So consider the action of
Qβu ∈ Gβ/Qβ# on Zβ/Yβ. Then u centralizes Qαβ > Zβ so Qβu acts trivially on Zβ/Yβ
which is not possible and therefore [v, u, u] 6= 1.
(ii) Let z ∈ Sαβ\(Qα ∪Qβ). Since Sαβ = QαQβ, we can write z = xy for x ∈ Qα\Qβ
and y ∈ Qβ\Qα. Both Qα and Qβ have exponent three so x and y both have order three
and hence z has order three or nine (no element of Sαβ can order greater than nine, if
w ∈ Sαβ had order 27 then Sαβ/Qα = 〈w〉Qα/Qα∼= 〈w〉/(〈w〉 ∩ Qα) and then 〈w〉 ∩ Qα
would contain elements of order nine). Suppose z has order three. We will perform a
calculation under this assumption for which we must first observe that S ′αβ 6 Qαβ since
Qαβ = Qα∩Qβ is normal in Sαβ and has index nine. Furthermore [S ′αβ, Qα] 6 [Qαβ, Qα] 6
Q′α = Yα and similarly [S ′
αβ, Qβ] 6 Yβ. Thus commutators of the form [y, x, x] and [y, x, y]
are central in Sαβ. Also notice that
y[y, x][y, x, y] = [y, x]y (∗)
79
and
x[y, x][y, x, x] = [y, x]x (∗∗).
We do the following calculation.
1 = xyxyxy
= x2y([y, x]y)xy
= x2y(y[y, x][y, x, y])xy by (∗)
= x2y2([y, x]x)y[y, x, y]
= x2y2(x[y, x][y, x, x])y[y, x, y] by (∗∗)
= x2y2x([y, x]y)[y, x, x][y, x, y]
= x2y2x(y[y, x][y, x, y])[y, x, x][y, x, y] by (∗)
= [x, y][y, x][y, x, x][y, x, y]2
= [y, x, x][y, x, y]2.
So [y, x, x] = [y, x, y]. However [y, x, x] ∈ Yα, [y, x, y] ∈ Yβ and Yα ∩ Yβ = 1. So [y, x, x] =
[y, x, y] = 1 and this contradicts part (i). Hence z has order 32.
(iii) Let u ∈ Qα\Qβ and v ∈ Qβ\Qα then [u, v, u] 6= 1 and [u, v, v] 6= 1. In particular,
the derived subgroup, S ′αβ is not central in Qα or Qβ and so is not contained in Zα or Zβ
but it does contain Zαβ = YαYβ = Q′αQ
′β. Thus
S ′αβ = [Qαβ〈u, v〉, Qαβ〈u, v〉] 6 [Qαβ, Qαβ][Qα, Qα][Qβ, Qβ]〈[u, v]〉 = YαYβ〈[u, v]〉
has order 33. It follows that S ′αβ lies strictly between Qαβ and Zαβ but is not equal to Zα
or Zβ.
(iv) Let z = xy be an element of order nine as before (where x ∈ Qα\Qβ and y ∈
Qβ\Qα) then (xy)3 = [y, x, x][y, x, y]2. We have seen that S ′αβ 6 Qαβ and so [y, x, x] ∈
[S ′αβ, Qα] 6 Q′
α = Yα and [y, x, y] ∈ [S ′αβ, Qβ] 6 Q′
β = Yβ. Finally [y, x, x] 6= 1 6= [y, x, y]
80
and so the result holds.
6.2 The Coset Graph
Recall that Γ = Γ(G,Gα, Gβ, Gαβ) is the coset graph of the amalgam where G = 〈Gα, Gβ〉.
Lemma 6.3. The coset graph Γ has valency four.
Proof. We have seen that the valency of the vertex α is given by [Gα : Gαβ] and the
valency of β by [Gβ : Gαβ]. We have seen also that Sαβ E Gαβ 6 NGα(Sαβ), NGβ(Sαβ). It
follows from the structure of Gα/Qα∼= Gβ/Qβ
∼= GL2(3) that both Gα and Gβ have four
Sylow 3-subgroups and so Gαβ has index at least four in both Gα and Gβ. By Lemma 6.2,
the normalizer of Sαβ preserves Qα ∪ Qβ as this is the set of all elements of order three
in Sαβ together with the identity. In particular this implies that Sαβ contains no other
subgroup isomorphic to Qα∼= Qβ. Therefore NGα(Sαβ) normalizes Qα and so must also
normalize Qβ. Thus NGα(Sαβ) 6 NG(Qβ) = Gβ (NG(Qα) = Gα and NG(Qβ) = Gβ follow
from Hypothesis B). Similarly NGβ(Sαβ) 6 NG(Qα) = Gα and so NGα(Sαβ) 6 Gβ ∩ Gα
and NGβ(Sαβ) 6 Gα ∩Gβ. Therefore Gαβ has index exactly four in both Gα and Gβ and
since G has two orbits on V (Γ) with representatives α and β, we see that every vertex in
Γ has valency four.
We extend our earlier notation to all groups Gλ where λ ∈ V (Γ).
Lemma 6.4. Let (λ, µ, ν) be a path of three distinct vertices in Γ. Then
(i) Sµν = QµQν;
(ii) Qµ = QλµQµν = Sλµ ∩ Sµν;
(iii) Qµν = ZµZν;
81
(iv) Zµ = Qλµ ∩Qµν = ZλµZµν;
(v) Zµν = YµYν; and
(vi) Yµ = Zλµ ∩ Zµν.
Proof. (i) Since G is edge transitive, this is just Lemma 6.1 (i).
(ii) Suppose Qλµ = Qµν . Notice that Gµ = 〈Gλµ, Gµν〉. Lemma 6.2 implies that Qαβ is
characteristic in Sαβ and thus normal in Gαβ. So if Qλµ = Qµν then this group is normal
in Gµ. However this is a contradiction since Qµν/Zµ is a 1-space of Qµ/Zµ and is not
preserved by Gµ. Thus Qλµ 6= Qµν and both groups commute modulo Zµ and therefore
normalize each other. Hence Qµ = QλµQµν . Also Qµ 6 Sλµ ∩ Sµν = QλQµ ∩ QµQν and
Qλµ 6= Qµν so we have Qµ = Sλµ ∩ Sµν .
(iii) This is just Lemma 6.1 (iv).
(iv) Since Qµ = QλµQµν , Qλµ∩Qµν has order 33. Since Qλµ and Qµν are both abelian
groups, their intersection commutes with Qµ. Thus Zµ = Qλµ ∩Qµν . Suppose Zλµ = Zµν
then Z(Sλµ) = Z(Sµν) is normalized by 〈Gλµ, Gµν〉 = Gµ which leads to a contradiction
as before. Thus Zµ = ZλµZµν .
(v) This is just Lemma 6.1 (ii).
(vi) We have seen that Zλµ 6= Zµν and so this follows immediately.
Lemma 6.5. Let (α, β, γ, δ) be a path of vertices in Γ. Then
Z3∼= Sαβ/Qα
∼= Qβ/Qαβ∼= Qβγ/Zβ
∼= Zγ/Zβγ∼= Zγδ/Yγ
∼= Yδ.
Proof. This follows from repeated use of Lemma 6.4 and an isomorphism theorem.
Lemma 6.6. Given any fours group T 6 Gαβ such that Gαβ = SαβT and we can choose
involutions tα, tβ ∈ T such that Qαtα ∈ Z(Gα/Qα) and Qβtβ ∈ Z(Gβ/Qβ).
82
Proof. Notice that for γ ∈ α, β, Gγ/Qγ∼= GL2(3) and GL2(3) has a central involution.
Notice also that this central involution normalizes every Sylow 3-subgroup in GL2(3).
Also, in GL2(3) every Sylow 3-subgroup is normalized by an elementary abelian group
of order four. Choose such a fours group T 6 Gγ such that Gαβ = SαβT . Of course T
contains three involutions and so we choose one of these, tγ, such that Qγtγ 6 Z(Gγ/Qγ).
This will of course hold in any choice of complement, T , such that Gαβ = SαβT .
Of course we can do a similar thing for any pair of adjacent vertices, γ, δ in Γ. Notice
however that any choice of tγ and tδ relies on having first chosen a fours group.
Lemma 6.7. Let γ, δ be adjacent vertices in Γ. The following hold.
(i) tγ inverts all non-trivial elements in Qγ/Qγδ, Qγδ/Zγ, Zγ/Zγδ, Zγδ/Yγ, Sγδ/Qδ,
Qγδ/Zδ, Zδ/Zγδ, and Yδ.
(ii) tγ centralizes Sγδ/Qγ, Yγ, Qδ/Qγδ and Zγδ/Yδ.
(iii) Given any choice of complement T ′ such that Gγδ = SγδT′, tγ is the unique element
in T ′ centralizing Yγ. Furthermore, tγ 6= tδ and T ′ = 〈tγ, tδ〉.
(iv) Gγ = CGγ (Yγ) 〈tδ〉.
Proof. We show the results hold for γ, δ = α, β and then the lemma follows by edge
transitivity.
Having chosen tα ∈ T 6 Gαβ as in Lemma 6.6 we consider the action of Qαtα on Qα.
Since Qα/Zα and Zα/Yα are natural Gα/Qα-modules, Qαtα acts on these spaces as the
matrix Diag(−1,−1) in GL2(3). So Qαtα inverts non-trivial elements. Now Yα E Gα and
CGα(Yα) has index at most two in Gα. Also Gα/Qα contains elements of order four which
square to Qαtα. So suppose tα doesn’t commute with Yα then 〈tα〉 is a complement to
CGα(Yα) in Gα. Then by correspondence Gα/Qα = (CGα(Yα)/Qα) : (〈tα〉Qα/Qα) however
83
〈tα〉Qα/Qα = Z(Gα/Qα) and so we have a direct product Gα/Qα = CGα(Yα)/Qα ×
〈tα〉Qα/Qα which is not possible.
Consider the following natural isomorphisms.
(i) Sαβ/Qα = QαQβ/Qα∼= Qβ/Qαβ;
(ii) Qα/Qαβ∼= QαQβ/Qβ = Sαβ/Qβ;
(iii) Qαβ/Zα = ZαZβ/Zα∼= Zβ/Zαβ;
(iv) Zα/Zαβ∼= ZαZβ/Zβ = Qαβ/Zβ;
(v) Zαβ/Yα = YαYβ/Yα∼= Yβ/(Yα ∩ Yβ) = Yβ;
(vi) Yα = Yα/(Yα ∩ Yβ) ∼= YαYβ/Yβ = Zαβ/Yβ.
By Lemma 1.2 we see equivalent actions of Qαtα (and of Qβtβ) on naturally isomorphic
groups. This is to say Qαtα inverts Qα/Qαβ∼= Sαβ/Qβ, Qαβ/Zα
∼= Zβ/Zαβ, Zα/Zαβ∼=
Qαβ/Zβ and Zαβ/Yα∼= Yβ and centralizes Sαβ/Qα
∼= Qβ/Qαβ and Yα∼= Zαβ/Yβ. So (i)
and (ii) hold.
Recall tα was chosen to lie inside a fours group T 6 Gαβ and such that Qαtα ∈
Z(Gα/Qα) and tβ was chosen such that Qβtβ ∈ Z(Gβ/Qβ). Notice that tβ must centralize
Yβ whereas we know tα inverts Yβ and so tα 6= tβ and therefore T = 〈tα, tβ〉. Hence
Gβ = CGβ(Yβ)〈tα〉 and in the same way tβ must invert Yα and so Gα = CGα(Yα)〈tβ〉.
Thus (iii) and (iv) hold.
For the remainder of this chapter we will be interested in the involutions tα ∈ Gα and
tβ ∈ Gβ so we will assume that we have made a fixed choice 〈tα, tβ〉 = T 6 Gαβ.
Lemma 6.8. T fixes a circuit in Γ.
84
Proof. Consider the action of tα on the graph Γ. We chose tα inside Gαβ so it certainly
fixes the edge α, β in Γ. We have seen also that each vertex in Γ has valency four. Thus
tα acts on the set Γ(β)\α which has size three. By the orbit-stabilizer theorem tα fixes
an element of this set. Let γ be the fixed vertex. We repeat the argument to see tα fixing
some other neighbour of γ, δ say. By assumption G is a finite group and so Γ has a finite
number of vertices and so tα fixes some cycle in Γ, (α, β, γ, δ, . . . , α). Of course tβ also
fixes some cycle (a, β, . . . , α) and if ε is any vertex fixed by tα then ε · tβ = ε · tαtβ = ε · tβtα
and so tβ preserves the subgraph of Γ fixed by tα. In particular this means we can choose
the circuit (α, β, γ, δ, . . . , α) to be fixed by T .
Lemma 6.9. Let Θ = (α, β, γ, δ, . . . , α) be a path in Γ fixed by T . Then T is a comple-
ment to Sλµ in Gλµ for each pair of adjacent vertices λ, µ in Θ furthermore, for each
λ ∈ Θ if we choose tλ ∈ T as in Lemma 6.6 then tα = tν for each vertex ν such that
d(α, ν) is a multiple of three.
Proof. Since T fixes the path Θ then T is a subgroup of each edge stabilizer Gλµ and
is therefore a complement as claimed. We have seen how to choose tα ∈ T such that
tα ∈ Z(Gα/Qα). We have seen also that T = 〈tα, tβ〉 and that tα is the only choice of
involution from T which centralizes Yα. We choose tδ in T uniquely such that it centralizes
Yδ. By Lemma 6.5, we get the natural isomorphism Sαβ/Qα∼= Yδ and by Lemma 1.2, we
see that tα centralizes Yδ and so tα = tδ. This argument holds for any vertex at distance
three from α and so for any vertex at distance a multiple of three.
We illustrate the way tα acts in Figure 6.1 which shows how tα acts equivalently on
naturally isomorphic groups. If tα is acting trivially on a section we label with 1 and if it
is inverting a section we label with −1.
Given a path (α, α+ 1, α+ 2, . . .) of vertices in Γ fixed by T we use Lemma 6.9 to see
tα = tα+3n for all n ∈ Z.
85
••
• ••
•
•
••
••
••
•
•
•
.............................................................................................................................................................................
......................
......................
......................
......................
......................
...................
.......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.......................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..............................................................................................................................................................................................................................................................................................................
Gα Gβ
Gαβ
1
•
• •
•
•
•
•
•
•
•
•
.............................................................................................................................................................................
......................
......................
......................
......................
......................
...................
.......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.......................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..............................................................................................................................................................................................................................................................................................................
Gγ
Gβγ
1
•
• •
•
•
•
•
•
•
•
•
.............................................................................................................................................................................
......................
......................
......................
......................
......................
...................
.......................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
.......................................................................................................................................................
..........................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
..............................................................................................................................................................................................................................................................................................................
Gδ
Gγδ
1
Qα
Zα
Yα
Qδ
Zδ
Yδ
Sαβ Sβγ Sγδ
1
1
1
1
1
1
−1
−1
−1
−1
1
Figure 6.1: Some partial subgroup structure inside G and the action of tα on sections.
86
Lemma 6.10. The elements of T# are G-conjugate.
Proof. Let T 6 S ∈ Syl2(Gβ) then SQβ/Qβ∼= S. So we consider the images of tαQβ and
tβQβ = tα+1Qβ in Gβ/Qβ∼= GL2(3). Thus we identify
tβQβ ∼
(−1 0
0 −1
), tαQβ ∼
(−1 0
0 1
).
Consider x ∼(
0 1
−1 0
). Then 〈txα〉Qβ/Qβ = 〈tαtβ〉Qβ/Qβ = 〈tα+2〉Qβ/Qβ. So by Sylow’s
Theorem, tα is conjugate to tα+2. By the same arguments tβ is conjugate to tα+3 = tα.
Lemma 6.11. G is locally 7-arc transitive on Γ.
Proof. See [7, p73 and p98] or Lemma A.4.
Lemma 6.12. For each vertex γ ∈ Γ we can choose an involution tγ ∈ Gγ such that
CGγ (tγ) has shape 3 : GL2(3).
Proof. Define tγ ∈ Gγ, as in Lemma 6.6, to be an involution such that tγQγ ∈ Z(Gγ/Qγ).
Then tγ is in the centre of a Sylow 2-subgroup of Gγ and also in some complement to Sγδ
in Gγδ for each δ ∈ Γ(γ). We have seen that the element tγ inverts the quotients Qγ/Zγ
and Zγ/Yγ and centralizes Sγδ/Qγ and Yγ. By Lemma 2.16, CSγδ(tγ) = 〈Yγ, b〉 for some
b ∈ Sγδ\Qγ has order 32 and is elementary abelian and CQγ (tγ) = Yγ. Thus CGγ (tγ) has
order 2432 and by an isomorphism theorem,
CGγ (tγ)/Yγ = CGγ (tγ)/(CGγ (tγ) ∩Qγ) ∼= CGγ (tγ)Qγ/Qγ 6 Gγ/Qγ∼= GL2(3).
So it follows from the order that CGγ (tγ)/Yγ∼= GL2(3) and so CGγ (tγ) ∼ 3.GL2(3).
Finally, since the Sylow 3-subgroup is elementary abelian and therefore splits over Yγ,
CGγ (tγ) splits over Yγ by Theorem 1.8.
Lemma 6.13. O2(CGγ (tγ))∼= Q8 is transitive on Γ(γ).
87
Proof. By Lemma 6.12, GL2(3) ∼= CGα(tα)/Yα. Since Z3∼= Yα E CGα(tα), the centralizer
of Yα in CGα(tα) has index at most two and since tβ inverts Yα the index is exactly
two. By Lemma 2.13, GL2(3) has a unique index two subgroup and so it follows that
CCGα (tα)(Yα)/Yα∼= SL2(3). Also by Lemma 2.13, O2(CGα(tα)/Yα) ∼= Q8 is contained
in CCGα (tα)(Yα)/Yα. Let L be the preimage in CGα(tα) of O2(CGα(tα)/Yα) then L ∼=
3 × Q8 and so it follows that CGα(tα) has a normal subgroup isomorphic to Q8 and by
correspondence it is the largest normal 2-subgroup. Now, this group of order eight acts on
Γ(α). If any element of order four fixes a vertex in Γ(α) then we have an edge stabilizer
containing elements of order four which cannot happen. Thus the action is transitive.
Lemma 6.14. Let (α, α + 1, α + 2, α + 3) be a path of vertices in Γ fixed by tα. Let
X := O2(CGα(tα)) and Y := O2(CGα+3(tα+3)). Then X 6= Y and if [X, Y ] = 1 then Γ is
a Moufang hexagon with 728 vertices. In particular, if [X, Y ] = 1 then |G| = 4245696.
Proof. Suppose X = Y . Let γ ∈ Γ(α). Since X fixes α and α + 3 and is transitive on
Γ(α), we can find x ∈ X such that γ · x has distance two from α + 3. But G preserves
distance so α and α + 3 lie on a circuit of length six. However this is not possible since
G acts 7-arc transitively on Γ.
So suppose [X, Y ] = 1. Since X 6= Y , we can choose p ∈ Y \X and consider the vertex
α ·p 6= α. This vertex has distance three from a+3 and therefore distance six from α and
so we will name it α+6 = α ·p. Let q ∈ X then (α+6) ·q = α ·pq = α ·qp = α ·p = α+6.
ThusX fixes α and α+6 and moves α+3 and so arguing as before we see that α, α+3, α+6
lie on a circuit of length 12. However G = 〈Gα, Gβ〉 acts 7-arc transitively on Γ and so
every vertex is contained in a 12-cycle. Now 7-arc transitivity ensures every vertex is in
a 12-cycle and so no vertex can be more than six vertices away from any other (so Γ has
diameter 6 and girth 12 so is a generalized hexagon). It follows from 7-arc transitivity
that the number of vertices at distance 0− 6 from α are 1, 4, 12, 36, 108, 324, 243 and so Γ
has 728 vertices. Since |Gα| = 2436 = 11664 and there are 728/2 = 364 images of α under
88
G, we get |G| = 11664× 364 = 4245696 as required. Finally let (α0, α1, . . . , α6) be a path
of seven vertices in Γ. Since G acts 7-arc transitively on Γ, the stabilizer of the path is
transitive on Γ(α6)\α5 and three divides the order of the path stabilizer. Therefore an
element of order three must necessarily fix⋃
16i65 Γ(αi) and so Γ satisfies the Moufang
condition.
Theorem 6.15. Let α, α + 1, α + 2, α + 3 be a path in Γ fixed by tα. If O2(CGα(tα))
commutes with O2(CGα+3(tα)) then G ∼= G2(3).
Proof. By Theorem 4.13 and Lemma 6.14 the coset graph Γ is a Moufang hexagon with 728
vertices and is uniquely defined. Notice that the stabilizer of a path of length six contains a
group of order three conjugate to Yγ for γ ∈ α, β. So the group 〈Yγ|γ ∈ V (Γ)〉 6 G†∩G
(where G† is defined in 4.13). To see that 〈Yγ|γ ∈ V (Γ)〉 = G we recall Lemma 6.4
which can be used to show that each Sylow 3-subgroup is contained in 〈Yγ|γ ∈ V (Γ)〉.
We have also seen that both Gα and Gβ are generated by their Sylow 3-subgroups. Thus
G = 〈Gα, Gβ〉 6 〈Yγ|γ ∈ V (Γ)〉 6 G. Thus G 6 G† ∼= G2(3). Now it follows from the
order of G that G = G† ∼= G2(3).
6.3 Controlling the 3-Local Structure
We have our group G containing a faithful completion of an amalgam (A1, A2, A12) =
(Gα, Gβ, Gαβ) of type G2(3). First we will find all conjugacy classes of elements of order
three. In particular we will identify an element of order three which acts fixed-point-freely
on a section of the centralizer of an involution and apply Lemma 1.25. We will use this to
force two subgroups to commute and then invoke Theorem 6.15 to recognize G2(3) as the
completion contained in G. Finally we will show that this containment cannot be proper.
First we require further notation.
89
For γ ∈ α, β, set Xγ = [Zγ, tγ] and choose elements yγ such that 〈yγ〉 = Yγ. As in
Lemmas 6.6 and 6.7 we fix T = 〈tα, tβ〉. Set tαβ := tαtβ.
We saw in Lemma 6.2 that the derived subgroup of Sαβ lies strictly between Qαβ and
Zαβ but is not equal to Zα or Zβ. SinceQαβ is elementary abelian, there are four subgroups
in Sαβ lying strictly between Qαβ and Zαβ three of which are normalized by Gαβ, Zα, Zβ
and S ′αβ. Thus the fourth is necessarily normalized by Gαβ too. Let A and B be the
two subgroups distinct from Zα and Zβ and we will not need to distinguish which is the
derived subgroup of Sαβ. We will however choose an element from each. Now by Lemma
6.7, both tα and tβ invert non-trivial elements in Qαβ/Zαβ and both A/Zαβ and B/Zαβ
are 1-dimensional subspaces so are inverted also. Thus tαβ centralizes them. In particular
this means we can choose a ∈ A\Zαβ and b ∈ B\Zαβ such that [a, tαβ] = [b, tαβ] = 1 (see
Lemma 2.16).
Lemma 6.16. Let γ, δ = α, β. Then Lγ := CG(yγ), has index two in Gγ, [Gγδ :
Lγ ∩Gδ] = 2, Lγ ∩ Lδ = Sγδ and Lγ/Qγ∼= SL2(3).
Proof. For each vertex δ we choose tδ as in Lemma 6.6. By Lemma 6.7, CG(yγ) = Lγ
has index two in NG(Yγ) = Gγ. Suppose Gγδ 6 Lγ then Sγδ char Gγδ E Lγ and so
Sγδ char Lγ. However, Lγ is normal in Gγ and so we have Sγδ E Gγ. This contradicts
that Qγ = O3(Gγ). Thus Gγδ Lγ and so LγGγδ = Gγ. By an isomorphism theorem,
Gγδ/(Lγ ∩ Gγδ) ∼= LγGγδ/Lγ = Gγ/Lγ and so Lγ ∩ Gγδ has index two in Gγδ. In the
same way Lδ ∩ Gγδ has index two in Gγδ. By Lemma 6.7, tγ /∈ Lδ but tγ ∈ Lγ ∩ Gγδ so
Lγ ∩Gγδ 6= Lδ ∩Gγδ. However Lγ ∩ Lδ 6 Gγ ∩Gδ = Gγδ so Lγ ∩ Lδ = Lγ ∩ Lδ ∩Gγδ has
index four in Gγδ. Thus Lγ ∩ Lδ = Sγδ. Finally, Lγ/Qγ has index 2 in Gγ/Qγ∼= GL2(3)
and by Lemma 2.13 (iii), GL2(3) has a unique subgroup at index two and therefore
Lγ/Qγ∼= SL2(3).
Lemma 6.17. Xγ = [Zγ, tγ] has order 32, is a complement to Yγ in Zγ and Xγ = 1 ∪
ygδ |g ∈ Gγ for γ, δ = α, β.
90
Proof. By coprime action on an abelian group, Zα = CZα(tα)×Xα. Recall from Lemma
6.7 that tα centralizes Yα and inverts Zα/Yα and Yβ. So Yα 6 CZα(tα). However if
this containment is strict then CZα(tα)/Yα is a proper subspace of Zα/Yα that tα acts
trivially on which is not possible. So CZα(tα) = Yα. Moreover tα inverts Yβ and so
yβ = [yβ, tβ] ∈ Xα. Now Zα is central in Qα, and Qα〈tα〉 E Gα by correspondence so we
see that [Zα, tα] = [Zα, Qα〈tα〉] EGα. Therefore ygδ |g ∈ Gγ ⊆ Xα. Lemma 6.16 implies
that CGα(Yβ) = Gα ∩ Lβ has index eight in Gα and so we see eight conjugates of yβ in
Xα = 1 ∪ ygβ|g ∈ Gα. A similar argument holds for Xβ.
Lemma 6.18. Zαβ contains three different G-conjugacy classes of elements of order three
and NG(Zαβ) = NG(Sαβ) = Gαβ.
Proof. Recall tα inverts yβ and tβ inverts yα so T = 〈tα, tβ〉 is transitive on the set
Ω := yαyβ, y2αyβ, yαy
2β, y
2αy
2β. Recall also that by Lemma 6.2 (iv), Sαβ contains elements
of order nine and the cube of any such element lies in Ω. Since Zαβ = Z(Sαβ), we can
apply Burnside’s Lemma 1.9 to see that elements of Zαβ are conjugate in G only if they
are conjugate in NG(Sαβ). However any element of NG(Sαβ) maps elements of order nine
to elements of order nine so no element of Ω can be conjugate to yα or yβ in G. Also
we have that yα and yβ are both G-conjugate to their inverses but not to each other
since by Hypothesis B, Gα is not conjugate to Gβ. Hence there are three G-conjugacy
classes of elements of order three in Zαβ. In particular, any element of G normalizing Zαβ
normalizes Yα and Yβ and so lies in Gαβ. Thus NG(Zαβ) = Gαβ and since Zαβ = Z(Sαβ),
NG(Sαβ) = Gαβ.
Recall we chose a ∈ A\Zαβ and b ∈ A\Zαβ such that [a, tαβ] = [b, tαβ] = 1.
Lemma 6.19. CSαβ(a) = CSαβ
(b) = Qαβ, |CGαβ(a)| = |CGαβ
(b)| = 342 and |NGαβ(〈a〉)| =
|NGαβ(〈b〉)| = 3422.
91
Proof. We show the results for a and similar arguments work for b. By Lemma 6.7,
tα inverts the quotient Qαβ/Zαβ and so inverts the 1-dimensional subspace A/Zαβ. By
Lemma 2.16, we can choose a′ ∈ A\Zαβ such that tα inverts a′. Consider CSαβ(a′) > Qαβ
and suppose this containment is strict. If Qα 6 CSαβ(a′) then a′ ∈ Z(Qα) ∩ A = Zαβ
which would contradict our choice of a′. So Qα CSαβ(a′) and similarly Qβ CSαβ
(a′).
So choose x ∈ CSαβ(a′)\Qαβ. By Lemma 6.2, x has order nine. Since tα inverts a′,
tα normalizes CSαβ(a′) so xtα ∈ CSαβ
(a′). Now Qαx ∈ Sαβ/Qα is centralized by tα so
xtαx−1 ∈ Qα ∩ CSαβ(a′) = Qαβ. Also, Qβx ∈ Sαβ/Qβ which is inverted by tα (by Lemma
6.7) so xtαx ∈ Qβ ∩ CSαβ(a′) = Qαβ. Therefore (xtαx−1)−1xtαx = x2 ∈ Qαβ and it follows
that x ∈ Qαβ which gives a contradiction. Thus CSαβ(a′) = Qαβ which has index nine
in Sαβ. Now we observe nine conjugates of a′ in A and since a′ is inverted by tα we
see a further nine. Since A\Zαβ is a set of order 18, each element is a conjugate of a′.
Thus CSαβ(a) = Qαβ. Finally since a is centralized by an involution and a′ is inverted
by an involution and the two elements are conjugate, it follows that |CGαβ(a)| = 342 and
|NGαβ(〈a〉)| = 3422.
Lemma 6.20. CG(Qαβ) = Qαβ, NG(Qαβ) = Gαβ.
Proof. Firstly by Lemmas 6.16 and 6.19,
CG(Qαβ) 6 CG(yα) ∩ CG(yβ) ∩ CG(a) = Lα ∩ Lβ ∩ CG(a) = Sαβ ∩ CG(a) = Qαβ,
and since Qαβ is abelian, CG(Qαβ) = Qαβ. Now consider O3(NG(Qαβ)). We have
NG(Qαβ) > Gαβ and since Sαβ is a Sylow 3-subgroup of G, Sαβ ∈ Syl3(NG(Qαβ)) and so
Qαβ 6 O3(NG(Qαβ)) 6 Sαβ. Suppose O3(NG(Qαβ)) = Qα, then NG(Qαβ) 6 NG(Qα) = Gα
and so NG(Qαβ) = NGα(Qαβ). However Qαβ/Zα is 1-subspace of the natural Gα/Qα-
module, Qα/Zα, and the stabilizer of a 1-space has order 12. But then NG(Qαβ) =
NGα(Qαβ) = Gαβ as claimed. Similarly if O3(NG(Qαβ)) = Qβ. So suppose instead that
O3(NG(Qαβ)) = Qαβ. In this case we can apply a result which followed from a theo-
92
rem of McLaughlin, namely Lemma 4.9. We apply it with M = NG(Qαβ), V = Qαβ,
S = Qα/Qαβ and T = Qβ/Qαβ. This implies that 〈QNG(Qαβ)α 〉/Qαβ
∼= SL2(3) and
〈QNG(Qαβ)
β 〉/Qαβ∼= SL2(3) intersect trivially and commute. This means that 〈QNG(Qαβ)
β 〉
normalizes Qα and then 〈QNG(Qαβ)
β 〉 6 Gα ∩ NG(Qαβ) = Gαβ. However, this implies
〈QNG(Qαβ)
β 〉/Qαβ 6 Gαβ/Qαβ∼= 32 : (2 × 2), a contradiction. Therefore we get that
O3(NG(Qαβ)) = Sαβ and NG(Qαβ) 6 NG(Sαβ) = Gαβ by Lemma 6.18.
Lemma 6.21. Qαβ ∈ Syl3(CG(a)) ∩ Syl3(CG(b)) and a is not G-conjugate to b.
Proof. We prove Qαβ ∈ Syl3(C) where C := CG(a) and the same proof will work for
CG(b). Suppose S ∈ Syl3(C) such that Qαβ < S then there exists some S0 6 S such that
Qαβ C S0. Then Qαβ < S0 6 NG(Qαβ) = Gαβ and S0 is a 3-group. Thus Qαβ < S0 6
Sαβ ∩ C = Qαβ by Lemma 6.19. So Qαβ ∈ Syl3(C).
Suppose now that there is g ∈ G such that ag = b. Then Qαβ, Qgαβ ∈ Syl3(CG(b)) so
there exists some h ∈ CG(b) such that Qghαβ = Qαβ. But then gh ∈ Gαβ and agh = b.
However Gαβ normalizes A and B and so agh = b ∈ A ∩ B = Zαβ which contradicts our
choice of a and b.
Lemma 6.22. Let d ∈ a, b then CG(d) 6 Gαβ. In particular |CG(d)| = |CGαβ(d)| = 342
and |NG(〈d〉)| = |NGαβ(〈d〉)| = 3422.
Proof. We prove C := CG(a) 6 Gαβ and the same proof will work for CG(b). Let R :=
O3′(C) and assume R 6= 1. We see that R is normalized by Xα = 1 ∪ yβGα as a
and Xα commute. By coprime action, R := 〈CR(y)|y ∈ Zα\1〉 = 〈CR(ygβ)|g ∈ Gα〉 by
Lemma 6.17. Consider CG(yβ)g ∩R = Lβg ∩R for any g ∈ Gα. Since Lβ is a 2, 3-group
and R has order prime to 3, this is a 2-subgroup of Lβg. Recall Lβ/Qβ
∼= SL2(3) and
so any 2-subgroup of Lβ is isomorphic to a subgroup of Q8. Now R E C and Qαβ 6 C
so Qαβ normalizes R. Also Qαβ 6 Qα = Qgα 6 Lβ
g so Qαβ normalizes Lgβ too. Thus
Qαβ normalizes Lgβ ∩R. But this is isomorphic to a subgroup of Q8 and so any subgroup
93
normalizing it centralizes the unique involution. However, this contradicts Lemma 6.20
which says CG(Qαβ) = Qαβ. Hence R = 1.
Now, because of Lemmas 6.19 and 6.20, we can apply a corollary to the theorem of
Smith and Tyrer, Corollary 4.6, to C withQαβ as its Sylow 3-subgroup. If C were 3-soluble
of length one then, since O3′(C) = 1, we would have that Qαβ E C 6 NG(Qαβ) = Gαβ
and we would be done. So let us assume not. Let N := O3(C) then C/N has order at
least 32. Recall that a was chosen to commute with the involution tαβ and that tαβ = tαtβ
inverts both Yα and Yβ. Thus tαβ ∈ N and so yα = [yα, tα], yβ = [yβ, tα] ∈ [Qαβ, tαβ] ⊆ N
and so Zαβ ∈ Syl3(N). By Lemma 6.18, NG(Zαβ) = Gαβ so NN(Zαβ) 6 Gαβ ∩C and so it
follows from Lemma 6.19 that 〈Zαβ, tαβ〉 = NN(Zαβ). Thus we apply Theorem 4.5 to see
that N is either 3-soluble of length one or O3(N) 6= N . If N has a normal 3′-subgroup
then so does C. So if N is 3-soluble of length one then Zαβ E N and so N = 〈Zαβ, tαβ〉
and then |C| = 342 and we would be done. So suppose O3(N) 6= N . Arguing as before,
tαβ ∈ O3(N) and Zαβ normalizes O3(N) and so it follows that Zαβ = [Zαβ, tαβ] 6 O3(N)
which contradicts that O3(N) 6= N . Hence C 6 Gαβ has order 342.
The remaining results now follow immediately from Lemma 6.19.
Lemma 6.23. There are five conjugacy classes of elements of order 3 in G and for γ ∈
α, β, Zγ contains only elements from the conjugacy classes with representatives yα, yβ
and yαyβ.
Proof. We have seen at least five conjugacy classes of elements of order three in G (with
representatives yα, yβ, yαyβ, a and b) and that any element of order three in a Sylow
3-subgroup, Sαβ, lies in Qα ∪Qβ. Thus it remains to prove that each of the 404 elements
of order three in Sαβ is in one of the known conjugacy classes.
First we look for conjugates of yα in Sαβ. By Lemma 6.16, CG(yα) ∩ Gβ = Lα ∩ Gβ
has index 8 in Gβ so we see 8 conjugates yα in Zβ. Of these 8, 6 lie outside Yα but inside
94
Qα. Since Zβ is normal in Gαβ but not in Gα, the normalizer in Gα of Zβ has index 4 in
Gα. Thus we see 4 conjugates of each of the 6 we saw already and so we have at least
4× 6 + 2 = 26 conjugates of yα in Sαβ and similarly at least 26 conjugates of yβ.
In particular, at this point we have counted all elements of order three in Zα. We
counted 16 conjugates of y, 2 conjugates of yα and 8 of yβ.
Now we count conjugates of yαyβ =: y. In Lemma 6.18 we saw four conjugates of y
and so it follows that CGαβ(y) = Sαβ. We saw also that NG(Zαβ) = Gαβ. If y commutes
with some 2-element, u, in Gα then u normalizes Zαβ and so must be in Gαβ. However,
no such element in Gαβ commutes with y and so no 2-element in Gα (or in Gβ by a similar
argument) commutes with y. Thus we see 16 conjugates of y in Zα and 16 in Zβ of which
4 are shared in Zαβ. Of the 16 in Zα, 12 lie outside Zαβ and since Zα is not normal in
Gβ, each has four conjugates lying inside Qβ giving 64 conjugates in Qβ and similarly 64
in Qβ of which the 28 inside Zα ∪ Zβ are shared giving 100 in total.
Finally we count conjugates of a and conjugates of b. We have seen that CG(a) = CG(b)
has order 342 and is contained in Gαβ. Therefore we see 72 conjugates in Gα and 72 in
Gβ of which the 18 in Qαβ are shared. So we have 126 conjugates of a and 126 of b.
Thus we have counted 26 + 26 + 100 + 126 + 126 = 404 elements of order three and so
can conclude that G has exactly five such conjugacy classes.
Notice that all conjugation we observed in this proof occurred within Gα ∪ Gβ. So if
any two elements of order three in Sαβ are G-conjugate then they are 〈Gα, Gβ〉-conjugate.
Thus we have the following corollary.
Corollary 6.24. Let x and y be elements of order three in G = 〈Gα, Gβ〉. Suppose x and
y are G-conjugate then there exists g ∈ G such that xg = y.
Proof. Since x and y are elements of order three in G and G is transitive on its Sylow
3-subgroups, we can assume x and y lie in the same Sylow 3-subgroup of G. So let us
95
assume x, y ∈ Sαβ. As x and y are elements of order three, x, y ∈ Qα ∪ Qβ. In Lemma
6.23 we counted all such elements of order three and observed that all conjugacy occurred
within 〈Gα, Gβ〉 and so we can find g ∈ G such that xg = y.
Lemma 6.25. NG(〈a, b〉) 6 NG(〈a〉) ∩ NG(〈b〉) 6 Gαβ and Qαβ is the unique Sylow 3-
subgroup of NG(〈a, b〉).
Proof. By Lemma 6.21, a is not conjugate to b so consider the set Ω := ab, a2b2, a2b, ab2 ⊆
〈a, b〉. Each element in Ω lies in Qαβ but not in A or B and so must lie in Zα or
Zβ. However by Lemma 6.23, no element of Ω can be conjugate to either a or b.
Thus any element normalizing 〈a, b〉 must normalize 〈a〉 and 〈b〉. By Lemma 6.22,
NG(〈a〉) 6 Gαβ and NG(〈b〉) 6 Gαβ and so NG(〈a, b〉) 6 Gαβ. By Lemma 6.19 and
Lemma 6.22, Qαβ 6 NG(〈a〉) 6 Gαβ and NG(〈a〉) has order 3422. Therefore Qαβ is a
Sylow 3-subgroup of NG(〈a〉) 6 Gαβ and is necessarily normal.
6.4 Controlling the Centralizer of an Involution
Lemma 6.26. Let S ∈ Syl3(CG(tαβ)). Then S has order 32 and |NCG(tαβ)(S)/CCG(tαβ)(S)| 6
2.
Proof. Consider CQαβ(tαβ). By our choice of a and b, 〈a, b〉 6 CQαβ
(tαβ). By Lemma
6.7, tαβ = tαtβ inverts every element in Zαβ = YαYβ. Therefore CQαβ(tαβ) has order nine.
Suppose 〈a, b〉 is not a Sylow 3-subgroup of CG(tαβ) and let S ∈ Syl3(CG(tαβ)) and suppose
〈a, b〉 < S then 〈a, b〉 C NS(〈a, b〉) 6 Qαβ ∩ CG(tαβ) by Lemma 6.25. However CQαβ(tαβ)
has order nine so 〈a, b〉 is a Sylow 3-subgroup. Furthermore if we set S = 〈a, b〉 then
by Lemma 6.25, NCG(tαβ)(S) 6 NCG(tαβ)(〈a〉) and since 〈a〉 is a group of order three it is
immediate that |NCG(tαβ)(〈a〉)/CCG(tαβ)(〈a〉)| 6 2 and so |NCG(tαβ)(S)/CCG(tαβ)(S)| 6 2.
Recall that tα, tβ and tαβ are all conjugate in G by Lemma 6.10. So for any conjugate
96
of tα, the centralizer in G contains an element from each of the four conjugacy classes of
elements of order three with representatives yα, yβ, a, and b. From now on set t := tα
and H := CG(t). Let (α, β = α + 1, α + 2, . . .) be a path in Γ fixed by 〈tα, tβ〉. We
have that tα = tα+3 and so CGα(t) ∼= CGα+3(t). Let Pα = O2(CGα(t)) ∼= Q8 and Pα+3 =
O2(CGα+3(t))∼= Q8.
Lemma 6.27. V := 〈yα, yα+3〉 is a Sylow 3-subgroup of H.
Proof. First observe that since t = tα = tα+3, Yα, Yα+3 6 H. We use Lemma 6.4 to see
that YαYα+1 = Zα,α+1 and Yα+1Yα+2 = Zα+1,α+2 and Yα+2Yα+3 = Zα+2,α+3. Also we see
that Zα,α+1Zα+1,α+2 = Zα+1 and Zα+1,α+2Zα+2,α+3 = Zα+2. Finally we see by Lemma 6.4,
that Zα+1Zα+2 = Qα+1,α+2 and so YαYα+1Yα+2Yα+3 = Qα+1,α+2. In particular Yα 6= Yα+3
since Qα+1,α+2 has order 34. Also YαYα+3 = 〈yα, yα+3〉 = V has order 32 and by Lemma
6.26 is a Sylow 3-subgroup of H.
Recall that the involution tβ ∈ H inverts Yα (by Lemma 6.7) and since tβ = tα+4
where α+ 4 ∈ Γ(α+ 3), tβ inverts Yα+3 also.
Lemma 6.28. O3′(H) = 〈Pα, Pα+3〉 is a 2-group and O3′(H)/〈t〉 admits a fixed-point-free
automorphism of order three.
Proof. By Lemma 6.26, we have |NH(V )/CH(V )| 6 2 but since tβ inverts V , the index
is exactly 2. Since V is abelian, we can apply the theorem of Smith-Tyrer, Theorem
4.5, to H. Clearly tβ ∈ O3(H) and tβ inverts yα and yα+3 and so yα = [yα, tβ], yα+3 =
[yα+3, tβ] ∈ O3(H). Therefore O3(H) contains a Sylow 3-subgroup and so O3(H) = H.
Thus H must be 3-soluble of length one. Let R = O3′(H). Choose x ∈ S such that x
is a conjugate of a. Then CG(x) has order 342 and so, as t ∈ CH(x), |CH(x)| 6 18. In
particular, x acts fixed-point-freely on R := R/〈t〉. By Corollary 1.22, R is nilpotent.
Let p be any prime dividing |R| and let P be a Sylow p-subgroup of R. By coprime
action, P = 〈CP (v)|v ∈ V #〉. Since the centralizer of every non-trivial element of V is a
97
2, 3-group, P is trivial unless p = 2. Thus R = 〈CR(v)|v ∈ V #〉 is a 2-group. We have
V = 〈yα, yα+3〉 so we consider CR(v) for each v ∈ V #. Firstly CR(yα) 6 Lα ∩ CGα(t) and
Lα ∩CGα(t) has order at most 2332. Also, CR(yα) is a 2-group so has order at most eight
and so CR(yα) 6 Pα and similarly CR(yα+3) 6 Pα+3. The remaining elements in V are
conjugates of a and b and so centralize just 〈t〉 in R. Therefore R 6 〈Pα, Pα+3〉. Finally,
since H is 3-soluble of length 1, Pα normalizes RV . Also, V normalizes Pα so by coprime
action, Pα = [Pα, V ]CPα(V ) = [Pα, V ]〈tα〉 6 RV and then Pα 6 R. Similarly Pα+3 6 R.
Thus R = 〈Pα, Pα+3〉 is a 2-group.
Lemma 6.29. PαPα+3 = O3′(H) ∼= 21+4+ and 21+4
+ : 32 : 2 ∼ CG(t) 6 〈Gα, Gβ〉.
Proof. Applying Lemma 1.25 to R = O3′(H)/〈t〉 = 〈Pα, Pα+3〉/〈t〉 forces R to be special
of order 26 or elementary abelian of order 24. Suppose the former then R has order 27
and has a characteristic subgroup, Z say, of order 23. Recall V = 〈yα, yα+3〉 is a Sylow
3-subgroup of H and so normalizes Z. However it is easy to check that the automorphism
group of any group of order eight does not contain a group of order nine. Thus CV (Z)
must be non-trivial and so some element of order three in V = 〈yα, yα+3〉 commutes with
a group of order eight, Z. The only possibilities are yα, yα+3 and their inverses. Hence
Z 6 Lα or Z 6 Lα+3 and so Z = Pα or Z = Pα+3 and then R = 〈Pα, Pα+3〉 has order
less than 27. So we must have that R is elementary abelian. Since R contains quaternion
subgroups, R is non-abelian and since Z(R) is normalized by an element of order three,
it must have order two or eight. However, the order cannot be eight since no abelian
group of order eight has an automorphism of order three. Therefore R is extraspecial and
by Corollary 1.17 and Lemma 1.18, R is the central product of Pα and Pα+3. We have
seen already using the theorem of Smith and Tyrer that H is 3-soluble of length 1 and so
R E RV E H. By a Frattini argument H = RV NH(V ) = R〈V, tβ〉. Thus H 6 〈Gα, Gβ〉.
The next result now follows immediately from Theorem 6.15.
98
Lemma 6.30. G contains a subgroup isomorphic to G2(3).
We are left with the possibility that G > G ∼= G2(3) however we know enough about
the structure of G now to rule this out. The following argument relies on the result Lemma
6.23 which implies that G has five conjugacy classes of elements of order three, moreover
if we look only at the elements of order three in G (or any conjugate of G in G) then we
see the same conjugacy classes. So if some element of order three, x say, lies in G and in
some conjugate G1 then CG(x) ∼= CG1(x).
Lemma 6.31. G ∼= G2(3).
Proof. SupposeG2(3) ∼= G = 〈Gα, Gβ〉 < G. By a Frattini argument, NG(G) = GαβG = G
so G is self-normalizing. Suppose Gg is a distinct conjugate of G for some g ∈ G. Suppose
further that |G∩Gg| is even. Since G2(3) contains just one conjugacy class of involutions
(see, for example, [6]), we can, without loss of generality, assume t ∈ G ∩ Gg. But then
t = sg for some s ∈ G. Since t and s are involutions in G, they are conjugate in G,
say sgh = th = s for some h ∈ G. However this implies gh ∈ CG(s) = CG(th) 6 G and
then g ∈ G which is a contradiction. Thus G is a strongly embedded subgroup of G. Let
x ∈ CG(t)\1. If x has order two then x is a conjugate of t and so x cannot lie in any
other conjugate of G. If x has order three then x is a conjugate of yα, yβ, a or b. In any
case CG(x) 6 G. Suppose x lies in some other conjugate of G, x ∈ G ∩ Gh say. Thus
x = yh for some y ∈ G. By Corollary 6.24, there is some g ∈ G such that yhg = xg = y.
Hence hg ∈ CG(y) = CG(xg) 6 G and so h ∈ G. Hence CG(t) intersects trivially with
any conjugate of G. Thus we can apply Lemma 1.30 to get that G contains an abelian
subgroup, K say, of order 34.7.13. In particular G contains an abelian subgroup, A say,
of order 34. Without loss of generality we assume A 6 Sαβ. If A 6 Qα and A 6 Qβ
then A = Qαβ and [Qαβ, K] = 1 which contradicts Lemma 6.20. So we may assume that
A 66 Qα. Then |A ∩Qα| = 33 and so A ∩Xα > 1. But then A contains a conjugate of yβ
99
and so K is contained in some subgroup of G conjugate to Gβ which is a contradiction.
So G can have no such abelian subgroup and hence G = G ∼= G2(3).
100
Chapter 7
An Amalgam of Type PSp4(3)
A Sylow 3-subgroup of PSp4(3) is isomorphic to a Sylow 3-subgroup of Sym(9) which
has order 34, centre of order three and for which the Thompson subgroup is elementary
abelian subgroup of order 33. Given a Sylow 3-subgroup in PSp4(3), S say, with Thompson
subgroup Q, the amalgam of type PSp4(3) comprises of the group A, the normalizer in
PSp4(3) of Q, together with B, the centralizer in PSp4(3) of a 3-central element of order
three. Furthermore these two 3-local subgroups intersect at C, the normalizer in PSp4(3)
of S. Of the three amalgams described in this thesis, this is the only one in which the two
groups, A and B, comprising the amalgam are not isomorphic. Also, in the amalgams of
type PSL3(3) and G2(3) we observed a natural GL2(3) action whereas, in the amalgam
of type PSp4(3), the elementary abelian subgroup Q is a faithful Sym(4)-module and
we see an SL2(3) action in B. We do however proceed in the same spirit as Chapter 6
since the task of recognizing PSp4(3) as a completion of the amalgam, in some respects,
seems to be similar. The centralizer of an involution, which we will determine some of
the structure of, admits an action from an elementary abelian subgroup of order nine and
the theorem of Smith-Tyrer seems well suited once again. However, things do not run as
smoothly as they did in G2(3). The problem appears to be much harder and consequently
101
we will assume more in our hypothesis. We will assume that Q normalizes no subgroup of
the completion of the amalgam with order prime to 3. This is a hypothesis employed in
several other characterizations of PSp4(3) to be found in the literature (see for example
[24] and [16]). Therefore the following theorem is certainly weaker than its predecessors.
It is a necessary condition though since groups with the structure 26 : PSp4(3) and
56 : PSp4(3) can be constructed with Magma and furthermore these constructions satisfy
the 3-local conditions imposed in Hypothesis C. This does however raise several interesting
questions. Firstly, if Qα normalizes some 3′-subgroup does this force there to be a normal
3′-subgroup as in the constructions 26 : PSp4(3) and 56 : PSp4(3)? Secondly, given
the appropriate 3-local conditions, what do the modules for PSp4(3) look like and is it
possible that we get a normal 3′-subgroup which is not elementary abelian? To answer
these question further works needs to be done perhaps involving signalizer functors. The
second question is reminiscent of a theorem of Graham Higman about a group G for which
G/O2(G) ∼= SL2(2n). We will discuss and prove his result in Chapter 8.
The hypothesis and theorem for this chapter are as follows.
Hypothesis C. Let A1 and A2 be subgroups of a finite group G = 〈A1, A2〉 such that
A12 := A1 ∩ A2 contains a Sylow 3-subgroup of both A1 and A2. Suppose further that
(i) no non-trivial normal subgroup of A12 is normal in A1 and A2;
(ii) A1/O3(A1) ∼= Sym(4);
(iii) A2/O3(A2) ∼= SL2(3);
(iv) O3(A1) is elementary abelian of order 33 and is a faithful A1/O3(A1)-module;
(v) O3(A2) is extraspecial of order 33 and O3(A2)/Z(O3(A2)), is a natural A2/O3(A2)-
module;
(vi) NG(O3(A1)) = A1 and NG(Z(O3(A2))) = A2;
102
(vii) O3(A1) normalizes no non-trivial 3′-subgroup of G.
Theorem C. Let G be a group satisfying Hypothesis C. Then G ∼= PSp4(3).
7.1 The Amalgam
Associate the amalgam (A1, A2, A12) to a coset graph, Γ. Let α and β be adjacent vertices
in Γ and so set A1 = Gα, A2 = Gβ.
Let Sαβ ∈ Syl3(Gαβ), Qαβ = Qα ∩Qβ, Zβ = Z(Qβ).
Lemma 7.1. The following hold in G.
(i) O3(Gαβ) = Sαβ = QαQβ ∈ Syl3(Gαβ).
(ii) Qαβ has order nine and Qβ∼= 31+2
+ .
(iii) Zβ = Z(Sαβ), Qα = J(Sαβ) are characteristic subgroups of Sαβ. CG(Qα) = Qα.
(iv) Sαβ ∈ Syl3(G).
(v) Gβ = NG(Zβ) = CG(Zβ).
(vi) NG(Sαβ) = Gαβ and Gαβ has index 4 in both Gα and Gβ.
(vii) If z ∈ Sαβ has order three then z ∈ Qα ∪Qβ.
(viii) Qβ is characteristic in Sαβ and S ′αβ = Qαβ.
Proof. (i) We have that Gα and Gβ intersect at a group containing a Sylow 3-subgroup
of G which has order 34. Also Qα E Gα and Qβ E Gβ. Thus the product QαQβ is normal
in Gαβ. If Qα = Qβ then the amalgam would not be simple and so QαQβ is a Sylow
3-subgroup of Gαβ.
103
(ii) Since Qα and Qβ have index three in Sαβ, their intersection Qαβ has index nine and
so has order nine. In particular, the extra special group, Qβ has an elementary abelian
subgroup of order nine. Since Qβ/Zβ is a natural Gβ/Qβ-module, Lemma 2.15 implies
that Gβ is transitive on the subgroups of Qβ with order nine. Thus Qβ is extra special of
+-type.
(iii) Suppose Z(Sαβ) Qβ then Sαβ = Z(Sαβ)Qβ. But then Sαβ/Qβ∼= Z(Sαβ) is a
Sylow 3-subgroup of Gβ/Qβ but acts trivially on the natural module Qβ/Zβ. This is a
contradiction (see Lemma 2.15) and so Z(Sαβ) 6 Qβ and so must be equal to Zβ.
Suppose Sαβ contains another elementary abelian subgroup, A, of order 33. Then
Qα and A intersect at a group containing a subgroup of order 32 which commutes with
QαA = Sαβ. However this contradicts Z(Sαβ) = Zβ. Thus Qα is the Thompson subgroup
of Sαβ and in particular is a characteristic subgroup of Sαβ.
(iv) Since Gα/Qα acts faithfully on Qα, no element commutes with all of Qα. Also,
since CG(Qα) 6 NG(Qα) = Gα, we see that CG(Qα) = Qα.
(v) Suppose Sαβ < S ∈ Syl3(G). Then Sαβ C R for some R 6 S. However, this means
Zβ C R 6 NG(Zβ) = Gβ which is a contradiction as Sαβ ∈ Syl3(Gβ). Thus Sαβ ∈ Syl3(G).
(vi) Suppose that CG(Zβ) < Gβ then the index must be 2. However this would give
an index 2 subgroup of Gβ/Qβ∼= SL2(3) which is not possible.
(vii) The normalizer of a Sylow 3-subgroup in both Gα/Qα∼= Sym(4) and Gβ/Qβ
∼=
SL2(3) has index 4. Thus by correspondence the same holds in Gα and Gβ. Now Sαβ E
Gαβ and so Gαβ has index at least 4 in both Gα and Gβ. Since Z(Sαβ) = Zβ, NG(Sαβ) 6
NG(Zβ) = Gβ. Thus NGα(Sαβ) 6 Gαβ and it follows that Gαβ has exactly index 4 in Gα
and so order 342. But then Gαβ must have index 4 in Gβ also and NG(Sαβ) = Gαβ.
Lemma 7.2. (i) If z ∈ Sαβ has order three then z ∈ Qα ∪Qβ.
(ii) Qβ is characteristic in Sαβ and S ′αβ = Qαβ.
104
Proof. (i) We have that Sαβ = QαQβ so any element in Sαβ\(Qα ∪ Qβ) has the form yx
where y ∈ Qβ\Qα and x ∈ Qα\Qβ. Notice that S ′αβ 6 Qα and so [x, y, x] = 1 as Qα is
abelian. Suppose [y, x, y] = 1, then the element Qαy ∈ Gα/Qα acting on Qα fixes the
1-space 〈[y, x]〉. By Lemma 2.17, Qαy can fix just a 1-space but it already fixes the space
Zβ and so we must have Zβ = 〈[y, x]〉.
We also have that Qβx ∈ Gβ/Qβ acts on Qβ/Zβ and since the action of Gβ/Qβ on
this module is natural, we see that Qβx fixes precisely the 1-subspace Qαβ/Zβ. However,
y /∈ Qαβ and Qβx fixes the space Zβy since [y, x] ∈ Zβ. This is a contradiction. Thus
[y, x, y] 6= 1 and [y, x] /∈ Zβ .
Now suppose yx as order three. We use the relations x[x, y][x, y, x] = [x, y]x and
y[x, y][x, y, y] = [x, y]y together with [x, y, x] = 1 in the following calculations. Also we
use that x ∈ Qα and so commutes with all commutators. Hence,
1 = yxyxyx
= y2x[x, y]xyx
= y2x2[x, y][x, y, x]yx
= y2x2[x, y]yx
= y2x2y[x, y][x, y, y]x
= y2x2y[x, y]x[x, y, y]
= y2x2yx[x, y][x, y, x][x, y, y]
= y2x2yx[x, y][x, y, y]
= [y, x][x, y][x, y, y]
= [x, y, y].
Which gives a contradiction. Thus yx has order nine.
105
(ii) If Sαβ contained another extraspecial subgroup of exponent three then we could
choose some non-central element of order three inside it that would not be in Qα ∪ Qβ
and that is not possible. Thus Qβ is a characteristic subgroup. Now Qαβ = Qα ∩ Qβ is
a normal subgroup of Sαβ and so S ′αβ 6 Qαβ. Suppose the containment was strict then
Sαβ would have order three and would be a central subgroup of Sαβ which would mean
S ′αβ = Zβ. However we have seen already that with x and y as before, [x, y] Zβ. Thus
Zβ 6= S ′αβ and so S ′
αβ = Qαβ.
7.2 The Coset Graph
Lemma 7.3. The coset graph Γ has valency four.
Proof. We have seen that the valency of the vertex α is given by [Gα : Gαβ] and the
valency of β by [Gβ : Gαβ] both of which are four by Lemma 7.1. Since G has two orbits
on V (Γ) with representatives α and β, we see that every vertex in Γ has valency four.
Lemma 7.4. (i) There is an involution tβ ∈ Gβ such that Gαβ = Sαβ〈tβ〉 and Qβtβ ∈
Z(Gβ/Qβ).
(ii) tβ fixes a circuit Θ = (α, β, . . . , α) in Γ.
(iii) For each vertex γ ∈ Θ lying in the same G-orbit as β, Qγtβ ∈ Z(Gγ/Qγ).
Proof. Notice that the centre of Gβ/Qβ∼= SL2(3) has order two and that the central
involution normalizes each Sylow 3-subgroup in Gβ/Qβ. So choose an element of order
two, tβ, as a coset representative such that Qβ〈tβ〉/Qβ = Z(Gβ/Qβ). Since tβ ∈ Gαβ =
NGβ(Sαβ), tβ fixes the edge α, β in Γ and since every vertex has valency four it follows
that tβ fixes a circuit in Γ. In particular, this means for any vertex γ ∈ Θ lying in the
same G-orbit as β, tβ ∈ Gγ and furthermore tβ normalizes a Sylow 3-subgroup of Gγ and
so Qγtβ ∈ Z(Gγ/Qγ) as before.
106
Lemma 7.5. G is locally 5-arc transitive on Γ.
Proof. See [7, p73 and p98].
Lemma 7.6. CQβ(tβ) = Zβ, 3× 3 ∼= CSαβ
(tβ) 6 Qα and CGβ(tβ) ∼= 3× SL2(3).
Proof. By the choice of tβ, we have that Qβtβ ∈ Z(Gβ/Qβ) and so tβ is in the centre of a
Sylow 2-subgroup of Gβ. Also tβ centralizes Zβ and Sαβ/Qβ. Clearly CQβ(tβ) = Zβ else
we have a non-trivial subgroup of Qβ/Zβ centralized by tβ which contradicts that this
group is a natural Qβ/Zβ-module. By Lemma 2.16, we can choose some x ∈ Qα\Qβ such
that tβ commutes with x. Hence CSαβ(tβ) = 〈x, Zβ〉 6 Qα is elementary abelian of order
nine. Consider the set tGβ
β = tgβ|g ∈ Gβ. Since Gβ has at most 34 Sylow 2-subgroups
and tβ is in the centre of one of them, this set has order at most 34. However tβ commutes
with a group of order nine, 〈Zβ, x〉 so it follows this set has order at most nine and so
〈tβ〉Gβ = 〈tβ〉Qβ . Now by a Frattini argument (Lemma 1.7), Gβ = CGβ(tβ)Qβ and by an
isomorphism theorem,
CGβ(tβ)/Zβ = CGβ
(tβ)/(CGβ(tβ) ∩Qβ) ∼= CGβ
(tβ)Qβ/Qβ = Gβ/Qβ∼= SL2(3).
The element 〈x〉 gives a complement to Zβ in CSαβ(tβ) and so by Gaschutz’s Theorem
(Theorem 1.8), CGβ(tβ) splits over Zβ and so CGβ
(tβ) ∼= 3× SL2(3).
At this point we recognize that CGβ(tβ) has a normal subgroup, X say, isomorphic to
SL2(3). Therefore Q8∼= O2(X) = O2(CGβ
(tβ)). Also, if γ is a vertex fixed by tβ at
distance two from β thenQγtβ ∈ Z(Gγ/Qγ) by Lemma 7.4 and so it follows that CGγ (tβ) ∼=
3× SL2(3) and so Q8∼= O2(CGγ (tβ)) also.
Lemma 7.7. Let (γ, α, β) be a path of vertices in Γ fixed by tβ. Let X := O2(CGβ(tβ))
and Y := O2(CGγ (tβ)). Then X 6= Y and if [X, Y ] = 1 then Γ is a Moufang quadrangle
with 80 vertices. In particular, if [X, Y ] = 1 then |G| = 25920.
107
Proof. Suppose X = Y . Each x ∈ X fixes β and γ. Notice, edge stabilizers have order
342 so do not contain elements of order four so if x ∈ X has order four then x does not fix
α. So there are elements in X which fix β and γ but move α. Therefore Γ must contain
a four cycle however this is impossible since G acts 5-arc transitively on Γ.
Since X 6= Y , we can choose p ∈ Y \X and consider the vertex β · p 6= β. This
vertex has distance two from γ and therefore distance four from β. Let q ∈ X then
β · pq = β · qp = β · p. Thus X fixes β and β · p and moves γ and so arguing as before we
see that β, γ, β ·p lie on a circuit of length 8. However, G = 〈Gα, Gβ〉 acts 5-arc transitively
on Γ and so every vertex is contained in an 8-cycle and no vertex can be more than four
vertices away from any other (so Γ has diameter 4 and girth 8). The number of vertices at
distance 0− 4 from α are 1, 4, 12, 36, 27 and so Γ has 80 vertices. Since |Gα| = 2334 = 648
and there are 80/2 = 40 images of α under G, we get |G| = 40× 648 = 25920 as required.
Finally let (α0, α1, α2, α3, α4) be a path of five vertices in Γ. Since G acts 5-arc transitively
on Γ, the stabilizer of the path is transitive on Γ(α4)\α3 and so we must have three
dividing the stabilizer of the path. Therefore an element of order three must necessarily
fix⋃
16i63 Γ(αi) and so Γ satisfies the Moufang condition.
Theorem 7.8. Let γ, α, β be a path in Γ fixed by tβ. If O2(CGβ(tβ)) commutes with
O2(CGγ (tβ)) then G ∼= PSp4(3).
Proof. By Theorem 4.13 and Lemma 7.7, the coset graph Γ is a Moufang quadrangle with
80 vertices and is uniquely defined. Notice that the stabilizer of the path of length three,
γ, α, β, δ, contains Qαβ. Thus the group G† (as defined in 4.13) contains all conjugates of
Qαβ. Notice that QαβQβδ = Qβ so Qβ 6 G† and similarly Qα 6 G†. It now follows that
G† contains every Sylow 3-subgroup of G and since Gα and Gβ are both generated by
their Sylow 3-subgroups, G 6 G†. Now it follows immediately from the order of G and
Theorem 4.13 that G = G† ∼= PSp4(3).
108
7.3 Controlling the 3 and 2-Local Structure
By Hypothesis C, Qα is a 3-dimensional faithful Sym(4)-module for Gα/Qα so we will
use Lemmas 2.17 and 2.18 and we will use the terminology introduced by referring to the
proper, non-trivial subgroups of Qα as being of type 1A, 1B, 1C, 2A, 2B or 2C. Find a,
b and c in Qα as in Lemma 2.17 where a ∼ v1, b ∼ v2 and c ∼ v3 give the subspaces fixed
by the non-trivial elements in O2(Gα/Qα). Then Qα = 〈a, b, c〉. Notice that the only type
of 1-space centralized by an element of order three in Gα/Qα has type 1B therefore Zβ
has type 1B. Choose a, b and c in such a way that Zβ = 〈abc〉.
Lemma 7.9. Qαβ = S ′αβ has type 2B.
Proof. The group Gα has four Sylow 3-subgroups which intersect at O3(Gα) = Qα. Now,
Gα is transitive on its Sylow 3-subgroups and thus on their corresponding derived sub-
groups which each have order 32. However, by Lemma 2.17, there is no orbit of subgroups
of order nine in Qα of length one. Therefore the derived subgroups must be distinct and
have type 2B.
Lemma 7.10. Sαβ has four G-conjugacy classes of elements of order three with represen-
tatives a, ab, abc, a2b2c2. In particular abc, a2b2c, a2bc2, ab2c2 and a2b2c2, abc2, ab2c, a2bc
are in disjoint conjugacy classes.
Proof. A lemma of Burnside (1.9) says that abc ∈ Zβ is conjugate to its inverse in G only
if it is conjugate to it in NG(Sαβ) = Gαβ 6 CG(Zβ). Thus abc and any other 3-central
element is not conjugate to its inverse. By Lemma 2.17, elements such as a and ab are
conjugate to their inverses and so cannot be conjugate to elements of type abc.
Subgroups of type 2B contain one subgroup of type 1B and three of type 1C. Now
consider S ′αβ = Qαβ. This group of order 32 has type 2B. Since Qβ/Zβ is a natural
Gβ/Qβ-module, Gβ is transitive on the 1-spaces and so Gβ is transitive on subgroups of
109
order 32 contained in Qβ. By Lemma 7.2, every element of order three in Sαβ lies in
Qα∪Qβ and so any element of order three in Sαβ lies in a subgroup of type 1A, 1B or 1C.
Finally, since Qα = J(Sαβ), Lemma 1.11 proves that no conjugate of ab can be conjugate
to any conjugate of a in G else they would already be conjugate in Gα. Thus there are
exactly four conjugacy classes of elements of order three in G.
In particular, notice that by Lemma 2.17 (v), abc is conjugate to a2b2c, a2bc2 and
ab2c2.
Lemma 7.11. 〈a, abc〉 = CSαβ(tβ) 6 Qα has order nine and has type 2C, CGα(tβ) ∼
3× 3 : 2× 2 and 〈a, abc, tβ〉 = CGαβ(tβ) ∼ 3× 3× 2.
Proof. In Lemma 7.6 we saw that 3× 3 ∼= CSαβ(tβ) 6 Qα. Choose some element d ∈ Qα
such that 〈Zβ, d〉 = CSαβ(tβ). Notice that Qαtβ ∈ (Gα/Qα) \ O2(Gα/Qα) (we see this is
true since Qαtβ normalizes a Sylow 3-subgroup of Gα/Qα). Now by Lemma 2.17 (vi),
Qαtβ centralizes a subgroup of Qα of type 1C and so we can choose d so that 〈d〉 has type
1C. By Lemma 2.18, 〈Zβ, d〉 has type 2B or 2C. It is easily checked that there are four
groups of order nine in Qα containing Zβ of which three have type 2C and one has type
2B. We have seen that the subgroup of type 2B is S ′αβ = Qαβ however CQβ
(tβ) = Zβ by
Lemma 7.6 and so 〈Zβ, d〉 must have type 2C. In particular this means CSαβ(tβ) has a
subgroup of type 1A and so we may as well assume 〈a〉 6 CSαβ(tβ) and then d ∈ 〈bc〉.
Now we see that
QαCGα(tβ)
Qα
6 CGα/Qα(Qαtβ) ∼= 2× 2
since Qαtβ is an involution in Sym(4) which normalizes a Sylow 3-subgroup and so com-
mutes with just a group of order four. In fact, we have equality in the previous expression
since Gα has Sylow 2-subgroups which are dihedral of order eight and so tβ at least
110
commutes with another involution. Now, by an isomorphism theorem,
2× 2 ∼=QαCGα(tβ)
Qα
∼=CGα(tβ)
CQα(tβ)
and we have seen already that CQα(tβ) ∼= 3× 3.
It is immediate that CGαβ(tβ) has order 18 and is abelian.
We have now that CSαβ(tβ) is a subgroup of type 2C and therefore contains subgroups
of type 1A, 1B and 1C. Thus, this group of order nine contains an element from each of
the four conjugacy classes of elements of order three in G. We therefore aim to restrict the
structure of the normalizer in G of this subgroup and then use the Theorem of Smith and
Tyrer once again. First we must restrict the structure of some of the 3-local subgroups
in G.
We continue to set 〈Zβ, d〉 = CSαβ(tβ) where d ∈ Qα\Qβ is chosen so that 〈d〉 has type
1C.
Lemma 7.12. 〈Qα, tβ〉 = CGβ(d) = CGα(d) = CGαβ
(d) ∼ 33.2.
Proof. By Lemma 2.17 (iv), CGα/Qα(d) has order two. Also, as d is not central in Sαβ, it
commutes with exactly Qα here. So Qαtβ ∈ CGα(d)/Qα 6 CGα/Qα(d) ∼ 2 which implies
that CGα(d) = 〈Qα, tβ〉 has order 332.
We chose d such that d ∈ Qα\Qβ and so Qβ〈d〉/Qβ is a Sylow 3-subgroup of Gβ/Qβ.
Therefore CGβ/Qβ(Qβd) ∼ 3× 2. By an isomorphism theorem,
CGβ(d)Qβ/Qβ
∼= CGβ(d)/CQβ
(d) = CGβ(d)/Qαβ.
Also, it is easily seen that CGβ(d)Qβ/Qβ 6 CGβ/Qβ
(d) = CGβ/Qβ(Qβd). Together this
gives us
CGβ(d)/Qαβ
∼= CGβ(d)Qβ/Qβ 6 CGβ/Qβ
(Qβd) ∼ 3× 2.
111
Hence CGβ(d) ∼ 33.2 and so the result follows.
Lemma 7.13. Qα ∈ Syl3(CG(d)) and CG(d) is either 3-soluble of length one or has a
normal subgroup at index nine.
Proof. Set C := CG(d) and notice that Qα 6 C. Now NC(Qα) = NG(Qα)∩C = Gα∩C =
CGα(d) ∼ 33.2. So suppose Qα is not a Sylow 3-subgroup of C then Qα < P ∈ Syl3(C)
and then Qα C NP (Qα) which is a contradiction since |NC(Qα)| = 233. By Lemma 7.1
(iv), CG(Qα) = Qα and so |NC(Qα)/CC(Qα)| = 2. Therefore C satisfies Corollary 4.6
and so C is either 3-soluble of length one or O3(C) has index at least nine. Suppose
|C/O3(C)| = 33 then O3(C) = O3′(C) and C would be 3-soluble of length one so we are
done.
Now we make full use of our strong assumption in Hypothesis C that Qα does not nor-
malize any non-trivial 3′-group in G.
Lemma 7.14. CG(d) 6 Gαβ, NG(d) 6 Gα.
Proof. Again we set C := CG(d) and we consider the two possibilities given by Lemma
7.13. First we suppose C is 3-soluble of length one. Since Qα does not normalize any
non-trivial subgroup of G of order prime to three, O3′(C) = 1. Thus C has a normal
Sylow 3-subgroup and so C = NC(Qα) = Gα ∩ C = 〈Qα, tβ〉 6 Gαβ and then C 6 Gαβ
and NG(d) 6 NG(Qα) 6 Gα as claimed. So suppose C is not 3-soluble of length one then
C has a normal subgroup, N say, at index nine. Let x be an element of order three in
N and consider CN(x). By Burnside’s normal p-complement Theorem (1.6), CN(x) has
a normal 3-complement, M say. However M is normalized by Qα and has order prime
to 3 and so must be trivial. Therefore x is self-centralizing in N and so we can apply a
theorem of Feit and Thompson (4.1) to N . Since O3′(C) = 1, O3′(N) = 1. Therefore by
Theorem 4.1, N ∼= Alt(5) or N ∼= Sym(3) or N ∼= PSL3(2). Notice N admits an action
from Zβ and CN(Zβ) = N ∩ Gβ 6 C ∩ Gβ ∼ 322. Hence CN(Zβ) = 〈x, tβ〉 can have
112
order just six. Suppose N ∼= Alt(5) or N ∼= PSL3(2). By Lemma 4.2, N does not admit
an outer automorphism of order three. Therefore 〈x, Zβ〉 ∩ CC(N) is non trivial. Clearly
neither x nor Zβ commutes with N so choose zβ ∈ Z#β such that xzβ ∈ CC(N). Then
1 = [xzβ, tβ] = txzβ
β tβ as tβ ∈ N . Therefore tβ = tz−1β
β = txβ which contradicts that x is self
centralizing in N . Thus N ∼= Sym(3) and so |C| = 6.32 which forces C 6 Gαβ. Finally,
by Lemma 2.17 (v), an element of order two in Gα inverts d. Thus NG(〈d〉) 6 Gα.
Corollary 7.15. Let d′ ∈ Qα such that 〈d′〉 is a subgroup of type 1C. Then 33 : 2 ∼
CG(d′) 6 Gα.
Proof. This is now immediate since all such elements in Qα are conjugate in Gα.
We now know the centralizers in G of elements of order three from three of the four
conjugacy classes. So we will consider CG(a) and by our choice of a we have tβ ∈ CG(a).
Lemma 7.16. CGα(a) ∼ 33 : (2× 2) and Qα ∈ Syl3(CG(a)). In particular NCG(a)(Qα) ∼
33 : (2× 2).
Proof. By Lemma 2.17, CGα/Qα(a) has order four. Also Qαtβ ∈ CGα/Qα(a). Notice that
this forces CGα/Qα(a) to be elementary abelian (since Qαtβ normalizes an element of order
three in Gα/Qα∼= Sym(4) and so no element of order four squares to Qαtβ). Now, we
clearly have CGα(a)/Qα 6 CGα/Qα(a), however, equality follows since a ∈ Qα is abelian.
Thus CGα(a)/Qα∼= 2× 2 and so CGα(a) ∼ 33 : (2× 2).
We have NCG(a)(Qα) = NG(Qα) ∩ CG(a) = CGα(a) ∼ 33 : (2 × 2). So if Qα were not
a Sylow 3-subgroup of CG(a) then we would have Qα < T for some T ∈ Syl3(CG(a)) and
then Qα C NT (Qα) 6 NCG(a)(Qα) ∼ 33 : (2 × 2). This is a contradiction and so Qα is a
Sylow 3-subgroup of CG(a) with normalizer as described.
We will apply a theorem of Hayden (Theorem 4.7) to the group CG(a) = CG(a)/〈a〉 so
we must first work to show the hypothesis of the theorem holds.
113
Lemma 7.17. Set T := Qα 6 CG(a) =: A then A and T satisfy the following.
(i) NA(T )/CA(T ) ∼= 2× 2.
(ii) CA(T ) = T .
(iii) CA(t) 6 NA(T ) for each t ∈ T#.
In particular, A = NA(T ).
Proof. By Lemma 7.16, T = Qα is a Sylow 3-subgroup of A = CG(a) and it is clear that
NA(T ) = NCG(a)(Qα)/〈a〉. We now choose some coset representatives for the elements of
T and consider their centralizers in A. We choose 〈ab, abc〉 = T and so we must consider
the groups (i) CA(ab), (ii) CA(abc), (iii) CA(a2b2c) and (iv) CA(c). We will show each is
contained in CGα(a).
(i) Suppose x ∈ CA(ab). Then [ab, x] = 1 implies [ab, x] ∈ 〈a〉 and then (ab)x ∈
ab, a2b, b. Lemma 7.10 implies that ab cannot be conjugate to b in G but is conjugate
to a2b in Gα. So there exists y ∈ Gα such that (ab)y = a2b. So suppose (ab)x = a2b
then (ab)xy−1= ab. Therefore xy−1 ∈ CG(ab) 6 Gα by Corollary 7.15 and so x ∈ Gα. If
instead (ab)x = ab then x ∈ CG(ab) 6 Gα so again x ∈ Gα.
(ii) Suppose x ∈ CA(abc). Then [abc, x] = 1 implies [abc, x] ∈ 〈a〉. Therefore (abc)x ∈
abc, a2bc, bc. By Lemma 7.10, the only possibility is (abc)x = abc. Therefore x ∈
CG(abc) = Gβ. So x ∈ CG(abc) ∩ CG(a) 6 CG(bc) 6 Gα by Corollary 7.15.
(iii) Suppose x ∈ CA(a2b2c). Then [a2bc2, x] = 1 implies [a2b2c, x] ∈ 〈a〉. Therefore
(a2b2c)x ∈ a2b2c, b2c, ab2c. Lemma 7.10 implies that (a2b2c)x = a2b2c and so x ∈
CG(a) ∩ CG(a2b2c) 6 CG(b2c) 6 Gα by Corollary 7.15.
(iv) Finally we suppose x ∈ CA(c). Then [c, x] = 1 implies [c, x] ∈ 〈a〉. Therefore
cx ∈ c, ac, a2c. Once again by Lemma 7.10, we have cx = c and so x ∈ CG(a)∩CG(c) 6
CG(ac) 6 Gα by Corollary 7.15.
114
So we have shown that CA(t) 6 CGα(a) = NCG(a)(Qα) = NA(T ) for each t ∈ T#. Our
calculations also show that if x ∈ CA(T ) then x ∈ CG(〈abc, c, a〉) = CG(Qα) = Qα by
Lemma 7.2. Therefore CA(T ) = T and NA(T )/CA(T ) ∼= 2 × 2. Thus we have shown A
satisfies the hypothesis of Theorem 4.7 and so A = NA(T ) = NCG(a)(Qα)/〈a〉.
Corollary 7.18. 33 : (2× 2) ∼ CG(e) 6 Gα for each e ∈ Qα such that 〈e〉 has type 1A.
Proof. By Lemma 7.17, we have that CG(a)/〈a〉 = NCG(a)(Qα)/〈a〉 = (CG(a) ∩ Gα)/〈a〉
and so CG(a) 6 Gα. If e ∈ Qα and 〈e〉 has type 1A then e is a conjugate of a by an
element of Gα and so the result holds.
We have now calculated the centralizer in G of every conjugacy class of element of order
three so the following result is immediate.
Corollary 7.19. If x ∈ G has order three then CG(x) is a 2, 3-group. In particular if
x ∈ Sαβ then CG(x) 6 Gα or CG(x) 6 Gβ.
We now work to restrict the structure of the group H := CG(tβ).
Lemma 7.20. S := CSαβ(tβ) is a Sylow 3-subgroup of H and |NH(S)/CH(S)| = 2.
Proof. By Lemma 7.11, S has type 2C and so contains elements of order three from each
of the four G-conjugacy classes of elements of order three. In particular the only conjugate
of d in S is d−1 so for any x ∈ NH(S)\S, x ∈ NG(〈d〉) 6 Gα. Therefore, by Lemma 7.11,
S 6 NH(S) 6 Gα ∩ H = CGα(tβ) ∼ 32 : 22 has a normal 3-subgroup of order nine. So
NH(S) = CGα(tβ) has order 2232. In particular, this means S ∈ Syl3(H) since if not then
S < T ∈ Syl3(H) and then S C NT (S) 6 NH(S) which contradicts that |NH(S)|3 = 32.
Now, CH(S) 6 CH(Zβ)∩CH(d) = CGβ(d) ∼ 33.2 by Lemma 7.12 and so CH(S) has order
at most 232. In fact this is the exact order as S commutes with itself and with tβ. Hence
|NH(T )/CH(T )| = 2.
115
Recall Lemma 7.4 implies that the involution tβ fixes a circuit in Γ, (γ, α, β, . . .)
and Qγtβ ∈ Z(Gγ/Qγ). Thus CGγ (tβ) ∼= 3 × SL2(3). Furthermore Zγ 6 CGγ (tβ) and
〈Zβ, Zγ〉 6 Qα is abelian. So 〈Zβ, Zγ〉 6 CSαβ(tβ) = S and therefore S = 〈Zβ, Zδ〉.
Lemma 7.21. 1 6= O3′(H) is nilpotent and contains A := O2(CGβ(tβ)) and B := O2(CGγ (tβ)).
Proof. By Lemma 7.20, |NH(S)/CH(S)| = 2 and so we can once again apply the theorem
of Smith and Tyrer (Theorem 4.5) to see that either H is 3-soluble of length one or
O3(H) 6= H. If H is 3-soluble of length one then X := O3′(H)S is a normal subgroup of
H. Then a Frattini argument implies that H = XNH(S). It is clear that tβ ∈ O3′(H) and
so 〈tβ, S〉 6 X ∩NH(S) and so by an isomorphism theorem, H/X ∼= NH(S)/(X ∩NH(S))
has order at most two. Therefore X is a normal subgroup of H at index at most two.
If H = X then H has a normal 3-complement and clearly A,B 6 O3′(H). So we may
as well assume X has index two in H. Suppose A X. Then by an isomorphism
theorem, 2 = |H/X| = |AX/X| = |A/(A ∩ X)| and so A ∩ X has order four. However
3 × SL2(3) ∼= CGβ(tβ) 6 H and so X > 〈(A ∩ X)CGβ
(tβ)〉 = A which is a contradiction.
Therefore A 6 X and similarly B 6 X and then clearly both quaternion groups are
contained in O3′(H).
So we are left to consider the possibility that Y := O3(H) 6= H. By Lemma 7.11 and
Lemma 7.14, CH(d) 6 Gαβ ∩H ∼= 3× 3× 2. So CH(d) = 〈S, tβ〉.
Since we know |NH(S)/CH(S)| = 2, there exists an element of even order x ∈
NH(S)\CH(S). If x ∈ CH(d) then x commutes with S. Recall, S has type 2C and
the only conjugate of d in S is d−1 (see Lemma 2.18 and Lemma 7.10). Thus x must
invert d as d−1 is the only other conjugate of d in S. Since x has even order, x ∈ Y and
therefore Y 3 dxd−1 = dxdx = d2x and so d ∈ Y . Now CY (d) = 〈S, tβ〉 ∩ Y has order 6
so if we set H := H/〈tβ〉 then d is a self-centralizing element of order three in Y . Hence
we can apply Theorem 4.1 to Y and we have three cases to consider.
Firstly, suppose Y ∼= PSL3(2). Then by Lemma 4.2, S ∩ CG(Y ) is non-trivial else we
116
would have an outer automorphism of PSL3(2) of order three which is not possible. Thus
for some s ∈ S#, and each y〈tβ〉 ∈ Y , y〈tβ〉s = y〈tβ〉. Since the coset y〈tβ〉 has order
two and s ∈ S has order three, s must centralize each element of the coset. In particular
we could choose y〈tβ〉 to have order seven. Then y has order seven and commutes with
s ∈ S# and this contradicts Corollary 7.19.
Secondly, suppose Y has a normal 2-subgroup, M , such that Y/M ∼= Y /M ∼= Alt(5).
Then by Lemma 4.2, S ∩ CG(Y/M) is non-trivial else we would have an outer automor-
phism of Alt(5) of order three which is not possible. Thus for some s ∈ S#, and each
yM ∈ Y/M , yM s = yM . Since the coset yM has order a power of 2 and s ∈ S has order
three, s must centralize each element of the coset. In particular we could choose yM to
have order five. Then y has order a multiple of five but commutes with s ∈ S# and this
contradicts Corollary 7.19.
So we are left only with the third case given by Theorem 4.1. Therefore Y has a
normal 3′-subgroup M such that Y/M ∼= Y /M ∼= Sym(3) or C3. In particular this means
that M : S is a subgroup of H at index at most two and so is normal. It follows that
H must be 3-soluble of length one and so by the previous argument we see that H has a
normal 3′-subgroup containing the necessary quaternion subgroups.
It remains only to prove that in each case O3′(H) is nilpotent. First notice that
tβ ∈ O3′(H) and CH(d) ∩ O3′(H) = 〈tβ〉 so we can apply Corollary 1.22 to O3′(H) to see
that O3′(H) is nilpotent.
Lemma 7.22. X := 〈O2(CGβ(tβ)), O2(CGγ (tβ))〉 is an extraspecial 2-group and has shape
21+4+ . In particular, [O2(CGβ
(tβ)), O2(CGγ (tβ))] = 1.
Proof. Since O3′(H) is nilpotent, it is a direct product of its Sylow subgroups and so the
two 2-groups A := O2(CGβ(tβ)) and B := O2(CGγ (tβ)) generate a 2-group. We apply
Lemma 1.25 to X/〈tβ〉 = 〈A,B〉/〈tβ〉 to get that X/〈tβ〉 is either elementary abelian
117
of order 24 or special of order 26 with centre of order four. Suppose the latter then,
by correspondence, X has a normal subgroup of order eight, Z say, containing tβ. By
coprime action, Z = 〈CZ(s)|s ∈ S#〉 so we consider the conjugacy class of each s ∈ S#.
If s = d (or d−1) then CZ(s) = 〈tβ〉 since we already saw that d acts fixed-point-freely on
O3′(H)/〈tβ〉. Consider CZ(s) 6 X ∩Gβ = A for s ∈ Z#β . Notice
CZ(s) 6 X ∩ CG(s) = X ∩Gβ 6 CGβ(tβ) ∼= 3× SL2(3)
and CZ(s) will be a normal 2-subgroup of CGβ(tβ) and so CZ(s) 6 A. If A 6 Z then
A = Z E X which is a contradiction since X = 〈A,B〉 has order 27. Suppose |A∩Z| = 4
then A∩Z is normalized by S and consequently centralized by S which is a contradiction
since CZ(d) = 〈tβ〉. Thus A ∩ Z = CZ(s) = 〈tβ〉. Similarly if s ∈ Z#γ then CZ(s) =
B ∩ Z = 〈tβ〉. So we must have Z = 〈CZ(s)|s ∈ S#〉 = CZ(a) has order eight. However
this is a contradiction since eight does not divide the order of CG(a) by 7.18. Hence
X/〈tβ〉 must be elementary abelian. Finally, since X contains non-abelian quaternion
groups, X is non-abelian and tβ ∈ Z(X) admits an automorphism of order three which
centralizes precisely 〈tβ〉 so Z(X) has order two or eight. However, the order cannot be
eight since no abelian group of order eight has an automorphism of order three. Thus X
is extraspecial of order 25. Now Corollary 1.17 together with Lemma 1.18 concludes the
result.
We can now use Theorem 7.8 to prove Theorem C.
Corollary 7.23. G ∼= PSp4(3).
118
Chapter 8
Higman’s Theorem on
Fixed-Point-Free Elements of Order
Three
Theorem 8.2 of Higman’s notes, Odd Characterisations of Finite Simple Groups [17], is
a powerful result about an action of SL2(2n) in which an element of order three acts
fixed-point-freely on a 2-group. The theorem is as follows.
Theorem 8.1. Let G be a finite group with a normal 2-subgroup, Q, such that X :=
G/Q ∼= SL2(2n) for some n > 2 and some element of order three in G acts fixed-point-
freely on Q. Then Q is elementary abelian and a direct product of natural modules.
Higman’s proof is sketchy and incomplete so we give a complete proof here. We will
use new methods of Chermak and Meierfrankenfeld found in [5] to give conditions under
which SL2(2n) modules can be decomposed and then finally we use Higman’s arguments to
prove that the subgroup Q is elementary abelian and thus an SL2(2n) module. In all that
follows let X ∼= SL2(2n) where n ≥ 2, T ∈ Syl2(X) and set L := GF(2) and K := GF(2n).
119
We must recall much of Chapter 2. In particular the definitions of the natural modules
over K and that these are the only irreducible modules over K for SL2(2n) which admit
an element of order three acting fixed-point-freely. We recall the definition of the natural
module over L also. We continue notation from Chapter 2 where we set V1, . . . , Vn to be the
irreducible 2-dimensional KX-modules. We also fixed an algebraically closed extension
field F of K over which we could define Brauer characters and we set Mi = F ⊗K Vi.
We proved these were the only 2-dimensional X-modules over an F . Recall also that by
Lemma 2.33, these were the only irreducible FX-modules admitting an element of order
three in X acting fixed-point-freely.
Lemma 8.2. Let V1, . . . , Vn be the n algebraic conjugates of the natural KX-module.
Then HomKX(Vi ⊗K Vj, Vk) = 0 for all i, j, k 6 n.
Proof. Firstly if i 6= j then we have seen in Lemma 2.31 that Vi ⊗ Vj is irreducible so the
only KH-module homomorphism Vi⊗K Vj → Vk is the zero map. So we consider the case
i = j. Suppose α : Vi ⊗K Vi → Vk is a KX-homomorphism. Since Vk is irreducible, this
map is either a zero map or surjective so we will assume α is surjective. Let U = Ker(α)
then U is a 2-dimensional submodule of V := Vi⊗KVi. Set W := 〈m⊗n+n⊗m|m,n ∈ Vi〉
then W is a KX-submodule of V . Let v1, v2 be a basis for Vi and write all elements of
X as matrices with respect to this basis. Any element of Vi can be written v = av1 + bv2
for a, b ∈ K. So for a, b, c, d ∈ K, a general element of W has the form
(av1 + bv2)⊗ (cv1 + dv2) + (cv1 + dv2)⊗ (av1 + bv2) = (ad+ bc)v1 ⊗ v2 + (ad+ bc)v2 ⊗ v1.
Thus W has dimension 1 and is spanned by the element 〈v1 ⊗ v2 + v2 ⊗ v1〉. If W U
then 1 6= WU/U V/U ∼= Vk where the isomorphism is an isomorphism of KX-modules.
This is not possible as Vk is irreducible. So W 6 U . Consider the 3-dimensional KX-
module V/W . This contains a 1-dimensional submodule U/W . The only 1-dimensional
KX-module is the trivial module (see Lemma 2.31) and so X acts trivially on U/W . A
120
basis for V := V/W is v1 ⊗ v1, v2 ⊗ v2, v1 ⊗ v2. Consider the element x :=
(0 1
1 1
)in X (represented with respect to the module Vi and the basis chosen) and let v :=
av1 ⊗ v1 + bv2 ⊗ v2 + cv1 ⊗ v2 ∈ V for a, b, c ∈ K. Now v1 · x = v2 and v2 · x = v1 + v2 and
so
v · x = bv1 ⊗ v1 + (a+ b+ c)v2 ⊗ v2 + cv1 ⊗ v2.
So if v were contained in U then it would be fixed by x and so a = b = c. Since K > L,
there is some ω ∈ K such that ω 6= 0, 1. So consider y :=
(ω 0
0 ω−1
)∈ X. Then
v1 · y = ωv1 and v2 · y = ω−1v2. So w := v1 ⊗ v1 + v2 ⊗ v2 + v1 ⊗ v2 ∈ U must be
centralized by y. However
w · y = ω2v1 ⊗ v1 + ω−2v2 ⊗ v2 + v1 ⊗ v2 6= w.
Thus there can be no such 0 6= α ∈ HomKX(Vi, Vi, Vk).
Corollary 8.3. Let U, V,W be non-trivial KX-modules admitting an element of order
three acting fixed-point-freely then HomKX(U ⊗K V,W ) = 0.
Proof. Suppose we have a surjective map, ρ : U ⊗ V → W . Every irreducible KX-
submodule of U , V and W is KX-isomorphic to Vi for some 1 6 i 6 n by Lemma 2.33.
Let W1 be an irreducible submodule of W and consider the preimage in U ⊗ V of W1.
There must exist submodules U1 6 U and V1 6 V such that ρ : U1 ⊗ V1 → W1. Suppose
U1 is not irreducible. For each proper submodule U2 of U1, ρ maps U2⊗V1 to a submodule
of W1. Since W1 is irreducible, this image is either W1 or 0. Since ρ∣∣U1⊗V1
is non-zero
we must be able to choose an irreducible submodule, U2 of U1 such that ρ : U2⊗V1 → W1.
We repeat these arguments in the case that V1 is not irreducible to find an irreducible
submodule V2 6 V1 such that ρ must map U2 ⊗ V2 to W1 where each of U2, V2,W1 is
KX-isomorphic to a natural KX-module. However, by Lemma 8.2, this is not possible
and so we have a contradiction.
121
In the following lemma we will refer to some known number theoretic results which
can be found in the appendix.
Lemma 8.4. (Chermak)[5, 4.1] If V is an LX-module satisfying [V, T, T ] = 0 and such
that CV (X) = 0 then [V,X] is a direct sum of natural modules for X.
Proof. Set B = NX(T ) and let H ∼= Cq−1 be a complement to T in B. Choose an element
s ∈ X that normalizes H but is not in B. Now, T is a 2-group and V is a module
over a field of characteristic two so CV (T ) 6= 0 and is B-invariant. In particular |H| is
coprime to |CV (T )| so CV (T ) is completely reducible as a sum of irreducible LH-modules
by Maschke’s Theorem. Let U be one of these irreducible modules. Set E := EndLH(U).
By Schur’s Lemma, E is a finite field of order 2m for some m. Moreover, as H is abelian,
there is a homomorphic image of H in the multiplicative group E×. Let H be this image
and let r = |H|. Since U is an irreducible H-module, it is also an irreducible H-module.
Now H 6 E× and so r divides |E×| = 2m − 1. Also, r divides |H| = 2n − 1. Let d be the
greatest common divisor or 2n − 1 and 2m − 1 and let g be the greatest common divisor
of n and m. Then d = 2g − 1 by Lemma A.2 and r divides 2g − 1. Since g divides m,
E contains a subfield, E0 say, of order 2g (See Lemma A.3). Also, since r divides 2g − 1
and multiplicative groups of fields are cyclic, it follows that H 6 E×0 . We can view U
as a vector space over E and also as a vector space over E0. Since E0 contains H, any
1-dimensional E0-subspace of U is H-invariant so U has dimension 1 over E0. It follows
that |U | = |E0| 6 2n.
For any C ⊆ X set UC := 〈uc|u ∈ U, c ∈ C〉 and for D 6 X set [UC , D] := 〈[uc, d] =
ucd − uc|u ∈ U, c ∈ C, d ∈ D〉. Now consider the LX-module W := UX . By the BN-
property of X, we have X = B ∪BsB and so
W = UB + UBsB = U + U sB = U + U sHT = U + U sT
because s normalizes H. Now for any u ∈ U and any t ∈ T , ust = us+[us, t] ∈ U s+[U s, T ]
122
so U sT 6 U s + [U s, T ]. Also, U s 6 U sT and [U s, T ] 6 U sT . Thus W = U + U s + [U s, T ].
Since the action of T is quadratic on V and U 6 CV (T ), U + [U s, T ] 6 CW (T ). Suppose
[U s, T ] = 0, then U commutes with 〈T, T s−1〉 = X (by Lemma 2.23) which contradicts
that CV (X) = 0. Hence W/CW (T ) ∼= U s has order |U |.
Now, [U s, T s] = 0, so U s 6 CW (T s) and therefore W = CW (T ) + CW (T s). Consider
CW (T ) ∩ CW (T s). This space commutes with 〈T, T s〉 = X and so must be trivial. Now
by an isomorphism theorem,
U s ∼=W
CW (T )=CW (T ) + CW (T s)
CW (T )∼=
CW (T s)
CW (T ) ∩ CW (T s)∼= CW (T s).
Since W is X-invariant, and therefore s-invariant, CW (T s) = CW (T )s and so CW (T ) has
order |U | and therefore |W | = |CW (T )||U s| = |U |2 and W = U ⊕ U s.
Let 0 6= v ∈ U s. If [v, t] = 0 for any 1 6= t ∈ T then v ∈ CV (T s)∩CV (t) = CV (〈T s, t〉) =
CV (X) = 0 (by Lemma 2.23). Hence CT (v) = 1 and so the set [v, t] = v · t − v|t ∈
T ⊆ [v, T ] has order |T | = 2n. Since T acts quadratically on V , [v, T, T ] = 0 and so
[v, T ] 6 CW (T ) = U . Hence |T | 6 |U |. But we already saw that |U | 6 2n and therefore
|U | = |T | = 2n and so |W | = 22n. Also, CT (v) = 1 implies that T acts regularly on
non-zero vectors in U s. Similarly T s acts regularly on non-zero vectors in U . So suppose
W1 is a proper X-invariant submodule of W . If u ∈ U# is in W1 then all of U is in W1
and then so is U s and so W = W1. Similarly if us in W1 then so is all of U s and again all
of U . So suppose u1, u2 ∈ U# and w = u1 +us2 ∈ W1. Then for t ∈ T#, wt = u1 +ust
2 ∈ W
and then w − wt = us2 − ust
2 ∈ U ∩W1 = 0 and so us2 = ust
2 which contradicts that T
acts regularly on U s. Hence W is an irreducible LX-module.
By Lemma 2.5, EndLH(U) is a field of order |U | and so EndLH(U) ∼= K. Notice that
any element of EndLX(W ) preserves CW (T ) = U and so any LX-endomorphism of W
restricts to give an endomorphism of U . Also, any LH-endomorphism of U , α say, extends
123
to an endomorphism of W = U⊕U s, α : u1+us2 7→ α(u1)+α(u2)
s. Thus EndLX(W ) ∼= K.
Also, W has dimension 2n over L and is irreducible and therefore is a natural module
over L.
Now [V,X] 6 [V, T ] + [V, T s] 6 CV (T ) + CV (T s) and CV (T ) is completely reducible
as a product of irreducible H-modules say CV (T ) = U1 ⊕ . . .⊕ Ur. Thus
CV (T ) + CV (T s) = (U1 ⊕ U s1 )⊕ . . .⊕ (Ur ⊕ U s
r )
gives a direct sum of natural modules for X and so [V,X] is a submodule of this decom-
posed module and so also decomposes.
Lemma 8.5. The only irreducible module for X over L which admits an element of order
three acting fixed-point-freely is the natural module.
Proof. Suppose V is an irreducible LX-module admitting an element of order three acting
fixed-point-freely. Set E := EndLX(V ). Then V is an irreducible EX-module and satisfies
E = EndEX(V ). Recall that F is the algebraically closed field of characteristic two over
which we defined Brauer characters. By Lemma 2.8, the FX-module F⊗EV is irreducible
and also admits an element of order three acting fixed-point-freely. By Lemma 2.33,
F ⊗E V has dimension two over F and so V has dimension two over E. Now, [V, T ] is a
proper E-subspace of V and so has dimension one and therefore [V, T, T ] is trivial. So by
Lemma 8.4, [V,X] (as an LX-module) is a direct sum of natural LX-modules. However,
some x ∈ X acts fixed-point-freely on V and so [V,X] > [V, x] = [V, x] ⊕ CV (x) = V
by coprime action. Thus, V is a direct sum of natural LX-modules. Finally, since V is
irreducible, V must be a natural LX-module.
Lemma 8.6. Let K/L be a field extension and let B be a group. Let U1 and U2 be KB-
modules such that EndLB(Ui) ∼= K for i = 1, 2. Suppose that there exists an LB-module
124
isomorphism α : U1 → U2. Then there exists β ∈ Gal(K/L) such that for every u ∈ U1
and every k ∈ K, α(ku) = β(k)α(u).
Proof. Let λi : K → EndLB(Ui) be the ring isomorphism where λi(k) is the LB-homomorphism
induced by multiplication by k. Clearly λi(k+ j) = λi(k) + λi(j) and λi(kj) = λi(k)λi(j)
for j, k ∈ K. For each k ∈ K, αλ1(k)α−1 is an LB-endomorphism of U2 and so applying
λ−12 to it gives an element of K, λ−1
2 (αλ1kα−1). Consider the map β : K → K given by
β(k) = λ−12 (αλ1(k)α
−1). We check this is a ring homomorphism. Let k, j ∈ K then
β(k + j) = λ−12 (αλ1(k + j)α−1)
= λ−12 (α(λ1(k) + λ1(j))α
−1)
= λ−12 ((αλ1(k)α
−1) + (λ1(j)α−1))
= λ−12 (αλ1(k)α
−1) + λ−12 (αλ1(j)α
−1)
= β(k) + β(j).
Also,
β(kj) = λ−12 (αλ1(kj)α
−1)
= λ−12 (αλ1(k)λ1(j)α
−1)
= λ−12 (αλ1(k)α
−1)λ−12 (αλ1(j)α
−1)
= β(k)β(j).
So β is a non-zero field homomorphism and thus an isomorphism. For l ∈ L, αλ1(l)α−1
is just multiplication of U2 by l as α and α−1 are L-linear maps. So αλ1(l)α−1 = λ2(l)
and therefore β(l) = λ−12 (λ2(l)) = l. Thus β is an automorphism of K fixing L. Now
for each u ∈ U1 we can find v ∈ U2 such that u = α−1(v). Thus α(ku) = α(kα−1(v)) =
(αλ1(k)α−1)(v) = k1α(u) where k1 ∈ K satisfies λ2(k1) = αλ1(k)α
−1 and therefore k1 =
λ−12 (αλ1(k)α
−1) = β(k) as required.
Lemma 8.7. Let T ∈ Syl2(X) and let Cp−1∼= H ≤ NX(T ) be a complement to T in
NX(T ). Then T is a KH-module. Let V,W be natural KX-modules and σ a Frobenius
125
map such that V σ and W are isomorphic as KX-modules. Let V1 = CV (T ) and W1 =
CW (T ).
(i) If α : T → HomK(V1,W1) is an LH-module isomorphism then there exists β ∈
Gal(K/L) such that for 0 6= k ∈ K, β(k2) = k−1σ(k).
(ii) If α : T → HomK(V1,W/W1) is an LH-module isomorphism then there exists β ∈
Gal(K/L) such that for 0 6= k ∈ K, β(k2) = k−1σ(k−1).
(iii) If α : T → HomK(V/V1,W1) is an LH-module isomorphism then there exists β ∈
Gal(K/L) such that for 0 6= k ∈ K, β(k2) = kσ(k).
(iv) If α : T → HomK(V/V1,W/W1) is an LH-module isomorphism then there exists
β ∈ Gal(K/L) such that for 0 6= k ∈ K, β(k2) = kσ(k−1).
Proof. Identify X with the 2-dimensional representation affording the module V . Choose
T to be the upper triangular matrices and H to be the diagonal elements. Fix the
parameterizations K× → H of H by k 7→ hk and K → T of T by l 7→ tl where
hk =
(k−1 0
0 k
), tl =
(1 l
0 1
).
So T is a KH-module when we define ktl := tkl and tl · hk := h−1k tlhk = tk2l for every
k ∈ K,hk ∈ H, tl ∈ T . In particular tl ·hk = k2 ·tl. By Lemma 2.5, we see EndLH(T ) ∼= K.
Now given our choice of representation we can identify both V and W σ−1with the
vector space (a, b)|a, b ∈ K. For (a, b) ∼ v ∈ V and hk ∈ H we have
v · hk ∼(
a b
)( k−1 0
0 k
)=(
k−1a kb
)and for (a, b) ∼ w ∈ W = V σ
w · hk ∼(
a b
)( σ(k−1) 0
0 σ(k)
)=(
σ(k−1)a σ(k)b).
126
Thus we identify V1 and W σ−1
1 with (0, b)|b ∈ K and V/V1 and (W/W1)σ−1
with (a, 0)+
〈(0, 1)〉|a ∈ K. So for each hk ∈ H,
(i) v · hk = kv for all v ∈ V1,
(ii) w · hk = σ(k)w for all w ∈ W1,
(iii) v · hk = k−1v for all v ∈ V/V1,
(iv) w · hk = σ(k−1)w for all w ∈ W/W1.
Consider the K-vector space MAB := HomK(A,B) where A = V1 or V/V1 and B = W1
or W/W1. Since A and B are 1-dimensional K-modules and elements of MAB commute
with K, each φ ∈ MAB is determined by where it maps one non-zero element of A.
Therefore, M has order |B| = 2n. Also, MAB is aKH-module when we define (f ·hk)(a) :=
f(a ·h−1k ) ·hk for f ∈MAB, k ∈ K, hk ∈ H and a ∈ A. Now K embeds into EndLX(MAB)
and MAB is a vector space over EndLX(MAB). However, |MAB| = |K| so MAB has
dimension one over K and therefore we must have that K ∼= EndLX(MAB).
We are given the LH-module isomorphism α and so we can apply Lemma 8.6 with
U1 = T , U2 = MAB, to see that there exists β ∈ Gal(K/L) such that β(k2) ·α(t) = α(k2 ·t)
for every t ∈ T and every k ∈ K. Now α(k2 · t) = α(t · hk) = α(t) · hk as α is an LH-
module homomorphism. The action we defined of H on MAB means for any a ∈ A,
(α(t) · hk)(a) = α(t)(a · h−1k ) · hk. We have four possibilities:
(i) If A = V1, B = W1 then α(t)(a · h−1k ) · hk = σ(k)α(t)(k−1a) = k−1σ(k)α(t). Thus
β(k2) = k−1σ(k).
(ii) If A = V1, B = W/W1 then α(t)(a · h−1k ) · hk = σ(k−1)α(t)(k−1a) = k−1σ(k−1)α(t).
Thus β(k2) = k−1σ(k−1).
127
(iii) If A = V/V1, B = W1 then α(t)(a · h−1k ) · hk = σ(k)α(t)(ka) = kσ(k)α(t). Thus
β(k2) = kσ(k).
(iv) If A = V/V1, B = W/W1 then α(t)(a · h−1k ) · hk = σ(k−1)α(t)(ka) = kσ(k−1)α(t).
Thus β(k2) = kσ(k−1).
Lemma 8.8. (Meierfrankenfeld)[5, 4.2] Let V be an LX-module with a submodule W
such that V/W and W are natural modules over L. Then V is completely reducible.
Proof. By Lemma 2.25, any non-identity element of X with odd order, g say, acts fixed-
point-freely on W and on V/W . Therefore g acts fixed-point-freely on V and so CV (X) 6
CV (g) = 0 and by coprime action, V = [V, g] ⊕ CV (g) = [V, g] 6 [V,X] 6 V . Thus if T
acts quadratically on V then V is completely reducible by Lemma 8.4 and we are done, so
assume not. Set V = V ⊗L K so V is a module over K with dimK(V ) = dimL(V ). Every
irreducible KX-submodule of V is a natural module over K or a Galois conjugate by
Lemma 2.33. Assume for a contradiction that there is an indecomposable KX-invariant
section V0 of V of dimension four. Again we can assume that T does not act quadratically
on V0.
Let W0 be a 2-dimensional submodule of V0 and let σ ∈ Aut(K) such that (V0/W0)σ
and W0 are isomorphic as KX-modules. Find 0 6 s < n such that σ raises elements of k
to the power 2s. Set U = CW0(T ) then U is a 1-dimensional subspace of W0 as is CW0(t)
for each t ∈ T# so U = CW0(t) also. Let R be the preimage in V0 of CV0/W0(T ) then R has
dimension three and [R, T ] 6 W0. For each t ∈ T , [R, t, t] = 0 and so [R, t] 6 CW0(t) = U .
Thus [R, T ] = U .
Since W0 is 2-dimensional, by Lemma 1.1, we get [W0, T, T ] = 0 and since U =
CW0(T ) is 1-dimensional, we get U = [W0, T ]. Similarly R/W0 = [V0/W0, T ] and so
R = W0 +[V0, T ]. Since H is transitive on the non-trivial elements of T , H is transitive on
the spaces [V0, t] for t ∈ T# as [v, th] = v·(th)−v = (v·h−1)·t·h−(v·h−1)·h ∈ [V0, t]·h. Now
128
if [V0, t, T ] = 0 for all t ∈ T then [V0, T, T ] = 0. So for some t ∈ T# (and thus all t ∈ T#)
we have [V0, t, T ] 6= 0. The space [V0, t] is invariant under T so using that [V0, t] 6 R
and [R, T ] = U we have 0 6= [V0, t, T ] 6 [V0, t] ∩ [R, T ] = [V0, t] ∩ U . Thus U 6 [V0, t] for
each t ∈ T . If [V0, t] has dimension one then [V0, t] = U but then [V0, t, T ] = 0 and so
this containment is strict. Since [V0, t, t] = 0, [V0, t] 6= [V0, T ] else [V0, T, t] = 0 for every
t ∈ T and again we would have [V0, T, T ] = 0. So we have U [V0, t] [V0, T ] = R.
Also, U 6 CV0(T ) 6 CV0(t) and [V0, t] 6 CV0(t) R. It follows from the dimensions that
CV0(t) = [V0, t] and when we observe that [V0, t] 6= CV0(T ) (else [V0, t, T ] = 0) we get
U = CV0(T ) [V0, t] = CV0(t) [V0, T ] = R.
Now R/U is a 2-dimensional KH-module so by Maschke’s Theorem it is completely
reducible. By Lemma 2.2, all irreducible submodules are 1-dimensional. Therefore, we
can find some 2-dimensional H-invariant submodule D 6 R with D ∩ W0 = U and
D + W0 = R and since CV0(T ) = U , [D,T ] 6= 0. We see immediately that [D,T ] = U
since [D,T ] 6 [R, Y ] = U . Consider the map,
α : T −→ HomK(D/U,U)
t 7−→ λt,
where λt : d+U 7→ [d, t]. This map is well defined as [U, T ] = 0. Also, since α is non-zero
and both T and HomK(D/U,U) are 1-dimensional, α is an isomorphism. By Lemma 8.7,
we see T is a KH-module with H-action t · h = h−1th and HomK(D/U,U) is a KH-
module with H-action described by (f · h)(x) = f(x · h−1) · h for f ∈ HomK(D/U,U),
129
h ∈ H and x ∈ D/U . Thus, for t ∈ T , h ∈ H, and d+ U ∈ D/U we get
(λt · h)(d+ U) = λt(d+ U · h−1) · h
= λt(d · h−1 + U) · h
= [d · h−1, t] · h
= (d · h−1 · t− d · h−1) · h
= d · h−1 · t · h− d · h−1 · h
= [d, t · h]
= λt·h(d+ U).
Therefore α(t · h) = λt·h = λt · h = α(t) · h and so α is an LH-module isomorphism.
We have D/U = D/W0 ∩ D ∼= W0 + D/W0 = R/W0 as KH-modules so we can
apply Lemma 8.7 (i) to show there exists some β ∈ Gal(K/L) = Aut(K) such that for
each non-zero k ∈ K, β(k2) = k−1σ(k). We find 0 6 b < n such that β(k) = k2b. Thus
(k2)2b= k−1k2 and therefore 2b+1 ≡ 2s−1 mod 2n−1. Since b, s < n, 2b+1−2s+1 = 2n−1
and then it follows that s = 1 and b = n− 1.
Consider the map,
γ : T −→ HomK(V0/R,R/D)
t 7−→ µt,
where µt : v + R 7→ [v, t] + D. We proceed as before. Firstly µt is well defined as
[R, T ] = U 6 D. Also, γ is non-zero as R = [V0, T ] and both T and HomK(V0/R,R/D)
are 1-dimensional so γ is an isomorphism. As before T and HomK(V0/R,R/D) are KH-
130
modules. Thus for t ∈ T , h ∈ H, and v +R ∈ V0/R we get
(λt · h)(v +R) = λt(v +R · h−1) · h
= λt(v · h−1 +R) · h
= [v · h−1, t] +D · h
= (v · h−1 · t− v · h−1 +D) · h
= v · h−1 · t · h− v · h−1 · h+D
= [v, t · h] +D
= λt·h(v +R).
Thus γ(t · h) = λt·h = λt · h = γ(t) · h and so γ is an LH-module isomorphism.
Now R/D = D + W0/D ∼= W0/(W0 ∩ D) = W0/U as KH-modules so we can apply
Lemma 8.7 (iv) to show there exists some δ ∈ Gal(K/L) such that for each non-zero
k ∈ K, δ(k2) = kσ(k−1). So, as before, we have some 0 6 d < n such that δ(k) = k2d
and then 2d+1 ≡ −2s + 1 mod 2n − 1. Therefore 2n − 1 divides 2d+1 + 2s − 1 = 2d+1 + 1
as s = 1. It follows that 2n − 1 = 2d+1 + 1 and then d = 0 and n = 2. So suppose
X ∼= SL2(4). Then T is normalized by an element of order three. Moreover, the element
of order three, x say, acts fixed-point-freely on T and fixed-point-freely on V0. Let vt be
some element of the split semidirect product V0T that is fixed by x. Then (vt)x = vt and
so v−1vx = tt−x ∈ V0 ∩ T = 1. Thus vx = v and tx = t and so CV0T (x) = CV0(x)CT (x).
This means x acts fixed-point-freely on the split semidirect product V0T . Thus by a lemma
of Burnside, 1.24, V0T has nilpotence class at most two. This gives a contradiction since
now 1 = [V0T, V0T, V0T ] > [V0, T, T ] and so T acts quadratically on V0.
Corollary 8.9. Let V be an LX-module such that an element of order three in X acts
fixed-point-freely on V . Then V is completely reducible as a product of natural modules
for L.
Proof. By Lemma 8.5, the only irreducible submodules of V are natural modules over L.
131
Suppose V is not completely reducible then there must be some submodule of dimension
4n which cannot be reduced and satisfies Lemma 8.8 which is a contradiction.
Lemma 8.10. Let G be a finite group with a normal 2-subgroup, Q, such that X :=
G/Q ∼= SL2(2n) for some n > 2 and some element of order three in G acts fixed-point-
freely on Q. Suppose G is minimal subject to Q being non-elementary abelian then Q is
abelian.
Proof. Suppose Q is non-abelian then by minimality, Φ(Q) = Q′ = Z(Q) = Ω1(Q). Set
Q = Q/Φ(Q) then Q and Q′ are LX-modules (where vector addition is group multipli-
cation and the action of X is by conjugation) admitting an element of order three acting
fixed-point-freely. Construct the KX-modules A := Q ⊗L K and B := Q′ ⊗L K then
these also admit a fixed-point-free element of order three.
We define a surjective map,
φ : A× A → B
(p⊗L k1, q ⊗L k2) 7→ [p, q]⊗L k1k2,
We need to check that φ is K-balanced (see Definition 2.7). Let p, q, r ∈ Q, k, k1, k2 ∈
K. Then
φ((pq ⊗ k1, r ⊗ k2)) = [pq, r]⊗ k1k2
= [p, r][q, r]⊗ k1k2
= [p, r]⊗ k1k2 + [q, r]⊗ k1k2
= φ((p⊗ k1, r ⊗ k2)) + φ((q ⊗ k1, r ⊗ k2))
132
and
φ((p⊗ k + k1, q ⊗ k2)) = [p, q]⊗ (k + k1)k2
= [p, q]⊗ kk2 + k1k2
= [p, q]⊗ kk2 + [p, q]⊗ k1k2
= φ((p⊗ k, q ⊗ k2)) + φ((p⊗ k1, q ⊗ k2))
and finally
kφ((p⊗ k1, q ⊗ k2)) = k[p, q]⊗ k1k2
= [p, q]⊗ kk1k2
= φ((p⊗ kk1, q ⊗ k2))
= φ((p⊗ k1, q ⊗ kk2)).
The map is bilinear in the second coordinate in exactly the same way and so φ is K-
balanced. By the definition of A⊗KA, there must be aK-homomorphism α : A⊗KA→ B
such that αθT = φ where θT : A×A→ A⊗K A is the map θT : (p, q) 7→ p⊗K q. We aim
to show the action of X on these modules commutes with α. So let x ∈ X, k ∈ K and
p, q ∈ Q. Then
φ((p⊗ k1, q ⊗ k2)) · x = [px, qx]⊗ k1k2 = φ(px ⊗ k1, qx ⊗ k2) = φ((p⊗ k1, q ⊗ k2) · x)
and
θT ((p⊗k1, q⊗k2))·x = (p⊗k1)⊗K (q⊗k2)·x = (px⊗k1)⊗K (qx⊗k2) = θT ((p⊗k1, q⊗k2)·x).
So φ and θT commute with the action of X and since αθT = φ, α commutes with the
action of X also. So α is a KX-homomorphism. However, by Corollary 8.3, such a map
must be the zero map and so Q must be abelian.
Theorem 8.11. Let G be a finite group with a normal 2-subgroup, Q, such that X :=
G/Q ∼= SL2(2n) for some n > 2 and some element of order three in G acts fixed-point-
freely on Q. Then Q is elementary abelian and a direct product of natural modules.
133
Proof. If Q were elementary abelian and hence an LX-module then by Corollary 8.9, we
would be done. Thus all rests on proving Q is elementary abelian.
Let G be a minimal counter example to the theorem. Then Q is not elementary
abelian but Q is abelian by Lemma 8.10 and so contains elements of order four. By
minimality, Q/Ω1(Q) is elementary abelian and therefore Q has exponent four. Let W be
the characteristic subgroup of Q generated by all elements of order four. Then W/Ω1(W )
is an LX-module and so is decomposable into a direct product of natural modules. Let W1
be the preimage in W of some irreducible submodule of W/Ω1(W ). Then the irreducible
submodule has dimension 2n and soW1 is a direct product of 2n cyclic groups of order four.
Since X acts on W1, there is a homomorphism Φ : X → Aut(W1) ∼= Aut(Z4 × . . .× Z4).
In Section 1.5 we investigated such groups and saw that each element of X can be viewed
as a 2n× 2n matrix with entries in Z4 and with odd determinant. We also saw how each
of these matrices acts on the natural module W1/Ω1(W1) and that the corresponding
representation simply restricts matrix entries modulo 2. By Lemma 1.35, we have the
map
α : X −→ GL2n(Z4)
and the representation affording the natural LX-module
β : X −→ GL2n(2)
where for each x ∈ X, the entries in α(x) are congruent modulo 2 to the entries in β(x).
We saw in Lemma 2.35 that given any algebraic conjugate of the natural KX-module,
we can view it as a module over L and as such it is the natural LX-module. We represent
elements of X as 2 × 2 matrices with entries in K. We then obtain a natural LX-
representation as follows. The field K is a vector space of dimension n over L and each
a ∈ K defines an L-linear transformation, namely, for each a ∈ K, λa : k 7→ ak for each
134
k ∈ K. So if we fix a basis for K over L we can make the n-dimensional matrix with
entries in L representing λa. We will call this matrix [a] for each a ∈ K. Notice that
the map K× → GLn(2) is a homomorphism and in particular this means for a, b ∈ K×
[a][b] = [b][a]. Now consider the map
β1 : X −→ GL2n(2)
such that β1 :
(a b
c d
)7→
([a] [b]
[c] [d]
). Then β1 is a natural LX-representation of X.
The natural LX-representation is unique up to equivalence. Lemma 1.35 implies that
there is a homomorphism α1 : X → GL2n(Z4) such that for each x ∈ X entries of α1(x)
are congruent modulo 2 to entries of β1(x). Consider the images of the matrices
(1 a
0 1
)for each a ∈ K. It follows that such a matrix is represented via α1 as(
I + 2A1 [a] + 2A2
0 + 2A3 I + 2A4
)
where A1, A2, A3, A4 are n× n matrices with 0, 1 entries. This matrix has order two and
so we can simplify to get (I + 2A1 [a] + 2A2
0 −I + 2[a]−1A1[a]
).
We can now choose non-zero elements a, b, c ∈ K such that a + b + c = 0 thus we follow
the same process for a, b and c and we see(I + 2A1 [a] + 2A2
0 −I + 2[a]−1A1[a]
)(I + 2B1 [b] + 2B2
0 −I + 2[b]−1B1[b]
)(I + 2C1 [c] + 2C2
0 −I + 2[c]−1C1[c]
)
is equal to the identity matrix. This gives that A1 + B1 + C1 is the zero matrix and
[a]−1A1[a] + [b]−1B1[b] + [c]−1C1[c] is the identity matrix. Now consider Corollary 2.12.
We can view [a], [b] and [c] as matrices over K and so there exists M ∈ GLn(K) such that
[a]M = diag(a), [b]M = diag(b) and [c]M = diag(c). Since A1, B1 and C1 are in Mn(2), it
135
makes sense to conjugate each by M and so we have the following.
I = IM = ([a]−1)MAM1 [a]M + ([b]−1)MBM
1 [b]M + ([c]−1)MCM1 [c]M = AM
1 +BM1 +CM
1 = 0M
which is a contradiction. Therefore, there is no minimal counter example and so Q is
elementary abelian and a direct product of natural modules over L.
136
Appendix
Lemma A. 1. Let p be a prime and let a and b be natural numbers. Then pa− 1 divides
pb − 1 if and only if a divides b.
Proof. If a divides b then we can write pb − 1 = (pa − 1)(pb−a + pb−2a + . . .+ pa + 1) and
so pa− 1 divides pb− 1. So assume that pa− 1 divides pb− 1. Choose an integer k so that
ka ≤ b < (k + 1)a. Then pa − 1 divides pka − 1 and so pa − 1 divides pb − 1− (pka − 1) =
pka(pb−ka−1). Since pa−1 is coprime to p, pa−1 divides pb−ka−1. But pb−ka−1 < pa−1
by the choice of k. So pb−ka − 1 = 0 and this means that b = ka as required.
Lemma A. 2. Let p be a prime and let m and n be natural numbers with greatest common
divisor g. Then pg − 1 is the greatest common divisor of pn − 1 and pm − 1 for any prime
p.
Proof. Since g divides m and n, pg−1 is a common divisor of both pn−1 and pm−1. Let
us assume m > n and m is not a multiple of n or else the result holds trivially. We will use
the Euclidean algorithm with pn−1 and pm−1 to find their greatest common divisor. Find
the integer k such that 0 < m− kn < n. Notice that pm − 1 = (pn − 1)pm−n + (pm−n − 1)
and also pm−n−1 = (pn−1)pm−2n +(pm−2n−1). So substituting the second into the first
gives
pm − 1 = (pn − 1)(pm−n + pm−2n) + (pm−2n − 1)
137
and we continue in this way k times to get
pm − 1 = (pn − 1)(pm−n + pm−2n + . . .+ pm−kn) + (pm−kn − 1).
This gives the first step in the Euclidean algorithm and so we have GCD(pm−1, pn−1) =
GCD(pn − 1, pm−kn − 1). We continue in this way until the process terminates and we
have that the greatest common divisor of pm − 1 and pn − 1 has the form pl − 1 for some
positive integer l. But now we have that l divides both n and m and so l divides g and
furthermore pl − 1 divides pg − 1. Thus pg − 1 is the greatest common divisor of pm − 1
and pn − 1.
Lemma A. 3. Let p be a prime and let F be a field of order pn for some natural number
n. Then any subfield of F has order pm where m divides n. Furthermore if m divides n
then there is a subfield of F of order 2m.
Proof. Suppose K is a subfield of F then K is a field of characteristic p so has order pm
for some m 6 n. Also K× is a subgroup of F×. So pm − 1 divides pn − 1. However this
can only occur when m divides n.
So suppose now that some natural number m divides n. The group F× is cyclic of
order pn−1. Since pm−1 divides pn−1, there is a unique subgroup H say of F× of order
pm − 1. We claim K := H, 0F is a subfield of F of order pm. We need to check K is
closed under addition. So let a, b ∈ K. If either a or b is zero it is clear that a + b ∈ K
so we will assume a, b ∈ H then both a and b have order dividing pm − 1 so apm= a and
bpm
= b. Notice that for any natural number r,
(a+ b)r = apm
+ rar−1b+r(r − 1)
2ar−2b2 + . . . rabr−1 + br.
So if r = pm then (a+ b)pm= apm
+ bpm
= a+ b. Thus either (a+ b)pm−1 = 1 or a+ b = 0.
If (a + b)pm−1 = 1 then a + b lies in the unique subgroup, H, of F× of order pm − 1. In
138
either case a + b ∈ K so K is closed under addition. Now we show K contains additive
inverses. We have seen that K is closed under addition so fix some non-zero k ∈ K. Then
define a map τ : K → K such that for l ∈ K, τ : l 7→ l + k. This map is clearly injective
and since K is a finite set, it must be surjective. In particular this means that there exists
some l ∈ K such that l + k = 0 and so K has additive inverses. All other axioms follow
trivially so K is a subfield of F .
Lemma A. 4. Let G be a faithful completion of an amalgam of type G2(3) and Γ the
coset graph. Then G is locally 7-arc transitive on Γ.
Proof. Let (α1, α2, α3, α4, α5, α6, α7) be a path of seven vertices. By Lemma 3.4, it
is necessary to prove that the stabilizer of the path (α1, α2, . . . , αi) is transitive on
Γ(αi)\αi−1 for each i 6 7. By the Orbit-Stabilizer Theorem, it is enough to show
|Gα1,α2,...,αi: Gα1,α2,...,αi+1
| = 3 for each i 6 7. We apply Lemma 6.4 and Lemma 6.1
several times. Firstly i = 1 is just edge transitivity which is always true in a coset
graph. When i = 2, Gα1α2 ∩ Gα2α3 has Sylow 3-subgroup Sα1α2 ∩ Sα2α3 = Qα2 and so
3 divides |Gα1,α2 : Gα1,α2,α3|. When i = 3, Gα1,α2,α3 ∩ Gα2,α3,α4 has Sylow 3-subgroup,
Qα2 ∩ Qα3 = Qα2α3 , with order 34 and so 3 divides |Gα1,α2,α3 : Gα1,α2,α3,α4|. When i = 4,
Gα1,α2,α3,α4 ∩ Gα2,α3,α4,α5 has Sylow 3-subgroup, Qα2α3 ∩ Qα3α4 = Zα3 , with order 33 and
so 3 divides |Gα1,α2,α3,α4 : Gα1,α2,α3,α4,α5|. When i = 5, Gα1,α2,α3,α4,α5 ∩ Gα2,α3,α4,α5,α6 has
Sylow 3-subgroup, Zα3 ∩ Zα4 = Zα3α4 , with order 32 and so 3 divides |Gα1,α2,α3,α4,α5 :
Gα1,α2,α3,α4,α5,α6|. When i = 6 Gα1,α2,α3,α4,α5,α6 ∩ Gα2,α3,α4,α5,α6,α7 has Sylow 3-subgroup,
Zα3α4 ∩ Zα4α5 = Yα4 , with order 3 and so 3 divides |Gα1,α2,α3,α4,α5,α6 : Gα1,α2,α3,α4,α5,α6,α7|.
When i = 7 Gα1,α2,α3,α4,α5,α6,α7 ∩Gα2,α3,α4,α5,α6,α7,α8 has Sylow 3-subgroup, Yα4 ∩ Yα5 = 1,
and so 3 divides |Gα1,α2,α3,α4,α5,α6,α7 : Gα1,α2,α3,α4,α5,α6,α7,α8|. Hence G acts locally 7-arc
transitive on Γ.
139
Bibliography
[1] M. Aschbacher. Finite group theory, volume 10 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge, second edition, 2000.
[2] Michael Aschbacher. Sporadic groups, volume 104 of Cambridge Tracts in Mathe-
matics. Cambridge University Press, Cambridge, 1994.
[3] Helmut Bender. Transitive Gruppen gerader Ordnung, in denen jede Involution genau
einen Punkt festlaßt. J. Algebra, 17:527–554, 1971.
[4] Wieb Bosma, John Cannon, and Catherine Playoust. The Magma algebra system.
I. The user language. J. Symbolic Comput., 24(3-4):235–265, 1997. Computational
algebra and number theory (London, 1993).
[5] Andrew Chermak. A class of p-local finite groups of local characteristic p. Unpub-
lished, 2007.
[6] J. H. Conway, R. T. Curtis, S. P. Norton, R. A. Parker, and R. A. Wilson. Atlas of
finite groups. Clarendon Press, Oxford, 1985.
[7] A. Delgado, D. Goldschmidt, and B. Stellmacher. Groups and graphs: new results
and methods, volume 6 of DMV Seminar. Birkhauser Verlag, Basel, 1985. With a
preface by the authors and Bernd Fischer.
140
[8] Klaus Doerk and Trevor Hawkes. Finite soluble groups, volume 4 of de Gruyter
Expositions in Mathematics. Walter de Gruyter & Co., Berlin, 1992.
[9] Walter Feit and John G. Thompson. Finite groups which contain a self-centralizing
subgroup of order 3. Nagoya Math. J., 21:185–197, 1962.
[10] Rachel Fowler. A 3-local characterization of the Thompson sporadic simple group.
PhD thesis, 2006.
[11] Rachel Fowler and Christopher Parker. A 3-local characterization of the Thompson
sporadic simple group. In preparation.
[12] David M. Goldschmidt. Automorphisms of trivalent graphs. Ann. of Math. (2),
111(2):377–406, 1980.
[13] Daniel Gorenstein. Finite groups. Chelsea Publishing Co., New York, second edition,
1980.
[14] Daniel Gorenstein, Richard Lyons, and Ronald Solomon. The classification of the
finite simple groups. Number 3., volume 40.3 of Mathematical Surveys and Mono-
graphs. American Mathematical Society, Providence, RI, 1998.
[15] Robert L. Griess, Jr. Automorphisms of extra special groups and nonvanishing degree
2 cohomology. Pacific J. Math., 48:403–422, 1973.
[16] John L. Hayden. A characterization of the finite simple group PSp4(3). Canad. J.
Math., 25:539–553, 1973.
[17] Graham Higman. Odd Characterisations of Finite Simple Groups, Lecture Notes of
University of Michigan. Ann Arbor. 1968.
[18] Christopher Hillar and Darren Rhea. Automorphisms of finite abelian groups. Un-
published.
141
[19] I. Martin Isaacs. Character theory of finite groups. Academic Press [Harcourt Brace
Jovanovich Publishers], New York, 1976. Pure and Applied Mathematics, No. 69.
[20] Hans Kurzweil and Bernd Stellmacher. The theory of finite groups. Universitext.
Springer-Verlag, New York, 2004. An introduction, Translated from the 1998 German
original.
[21] Jack McLaughlin. Some groups generated by transvections. Arch. Math. (Basel),
18:364–368, 1967.
[22] Christopher Parker and Peter Rowley. Symplectic amalgams. Springer Monographs
in Mathematics. Springer-Verlag London Ltd., London, 2002.
[23] Christopher Parker and Peter Rowley. Local characteristic p completions of weak
BN -pairs. Proc. London Math. Soc. (3), 93(2):325–394, 2006.
[24] A. R. Prince. A characterization of the simple groups PSp(4, 3) and PSp(6, 2). J.
Algebra, 45(2):306–320, 1977.
[25] Arthur Ranum. The group of classes of congruent matrices with application to the
group of isomorphisms of any abelian group. Trans. Amer. Math. Soc., 8(1):71–91,
1907.
[26] Jean-Pierre Serre. Trees. Springer-Verlag, Berlin, 1980. Translated from the French
by John Stillwell.
[27] Stephen D. Smith and A. P. Tyrer. On finite groups with a certain Sylow normalizer.
I, II. J. Algebra, 26:343–365; ibid. 26 (1973), 366–367, 1973.
[28] Robert Steinberg. Endomorphisms of linear algebraic groups. Memoirs of the Ameri-
can Mathematical Society, No. 80. American Mathematical Society, Providence, R.I.,
1968.
142
[29] Jacques Tits and Richard M. Weiss. Moufang polygons. Springer Monographs in
Mathematics. Springer-Verlag, Berlin, 2002.
143