GROUPS ACTING ON REGULAR GRAPHS

AND GROUP AMALGAMS

·"'Z • D'JO kOV1C

Dragomlr ./

Department of Pure Mathematics

University of Waterloo

Gary L. Mill er

Departments of Mathematics

and Computer Science

The University of Rochester

TR24

The work of the first author was supported in part by NRC Grant
A-5285, and that of the second author by NRC Grant A-5549 .
• •
 Groups Acting on Regular Graphs

and Group Amalgams

v . /
Dragomir Z. Djokovic and Gary L. Miller

O. INTRODUCTION

The s-transitive groups ac~ing on connected regular graphs

have been studied by several people: W.T. Tutte, A. Gardiner, N. Biggs,

R.M. Weiss, the authors and others. The purpose of this paper is two­

fold. First we establish that the problems about constructing and

existence of such groups have a simple purely group-theoretical formulation

in terms of group amalgams. Second, W~ determine the group amalgams

associated with the locally regular groups operating on cubic graphs.

In Section 1 we introduce graph-theoretical concepts that we

need and in Section 2 the basic definitions about group amalgams. Section

3 explains the connection betweenl-transitive groups and group amalgams.

Section 4 classifies finite simple amalgams of degree (3,2). It is

surprising (Theorem 4) that there are precisely 7 such amalgams. Section

5 expresses some known results about s-transitive groups in the language

of group amalgams. Section 7 explains the connection between locally

l-transitive groups and group amalgams. The next two sections deal with

the amalgams associated to locally regular groups acting on cubic graphs.

We show that there is essentially only one such amalgam for locally s­
. 4
 - 2 ­

regular groups s = 1,2,3,4,5, or 7.

We end with two remarks, in the last section, which illustrate

the use fullness of the graph-theoretical method in the study of group

amalgams.

1. GROUPS ACTING ON GRAPHS

Let G be a graph having neither loops nor multiple edges.

Let A be a group and f:A + Aut(G) a homomorphism. Then we say that

A acts on G and we write f(a)(x) = a-x for a f A and x a vertex

of G. If f is injective we say that the action is faithful and in that

case we may consider A as a subgroup of Aut(G).

From now on we consider a pair (G,A) where G is a connected

graph and A a subgroup of Aut(G). Recall that an s-arc S of G

is a map

S:{O,l, ••• ,s} + V(G) = V

(V(G) = the vertex-set of G) such that S(i) and S(i + 1) are adjacent

for 0 ~ i ~ s - 1 and S(i) + S(i + 2) for 0 ~ i s s - 2.

We say that A is s-transitive (resp. locally s-transitive)

if given any two s-arcs Sl and S2 (resp. two s-arcs Sl and S2

satisfying Sl(O) = S2(0)), there exists (l € A such that (l 0 Sl = S2'

If A is locally s-transitive (s ~ 1) then A acts transitively on

the set of edges of G [1, Lemma 1]. Thus, in this case, A is either

transitive on V or it has precisely two orbits, say V+ and V

-, and

G is bipartite with partition + ,V).

­ In the first case A is in
• •(V
 - 3 ­

fact s-transitive.

We say that A is s-regu1ar (resp. locally s-regu1ar) if

A is s-transitive (resp. locally s-transitive) and for each s-arc

S and a ~ A the equality a 0 S =S implies a = 1.

We shall say that G is s-transitive, locally s-transitive,

etc. if Aut (G) haa the corresponding property. If G is a-transitive

(s ~ 0) then it is a regular graph, i.e., every vertex of G has the

same degree (or valence). If G is locally a-transitive (s ~ 1) then

it ia either regular or bipartite with partition (V+,V-) of V and any

two vertices in V+ (resp. V-) have the same degree. A regular graph

of degree 3 is also called cubic graph.

We say that A has index s (resp. local index s) if A

is s-transitive (resp. locally s-transitive) but it is not (s+l)­

transitive (resp. locally (s+l)-transitive). It is well-known [5] that if

G is a cubic graph and A has index s (1 ~ s < 00) then A is s­

regular and s takes one of the values 1,2,3,4,5. We remark that the

corresponding statement is not true for locally s-transitive groups. For

instance if (the complete bipartite graph on 6 vertices with

three vertices in each part) then Aut (G) has a subgroup isomorphic to

C x S3 which has local index 1 and is not locally 1-regu1ar.

3
We recall that R.M. Weiss [7] has shown that if G is a cubic

graph and A is locally s-regu1ar (1 ~ s < 00) then s takes one of

the values 1,2,3,4,5,7.

 - 4 ­

2. GROUP AMALGAMS

A group amalgam is a triple (X,Y;H) where X,Y,M are groups,

H is a subgroup of both X and Y and Xn Y = H. Since X and Y

determine H we shall write sometimes (X,Y) instead of (X,Y;H).

A morphism from an amalgam (Xl,Y1;H ) to the amalgam

l
(X2 ,Y ;H ) is a pair (f ,f ) of group homomorphisms
2 2 l 2

such that fl and f2 coincide on Hl and

We say that an amalgam (X' ,Y'jH') is a subamalgam of the

amalgam (X,Y;H) if X' (resp. Y') is a subgroup of X (resp. Y) and

X' n H = y' n H = H'. We say that this subamalgam is transitive if

X'H =X and y'H = Y.

-1
Let (X,Y;H) be an amalgam and a EO: H. Let axa

(x EO: X) and f (y)

2
= aya -1 (y EO: Y). Then is an automorphism

of (X,Y;H) which is called inner automorphism. Two subamalgams (X' ,Y')

and (X",yfI) of (X,Y) are said to be conjugate if there exists an inner

of (X,Y) such that f (X')

1
= Xfl

A normal subgroup of (X,YjH) is, by definition, a subgroup

N of H which is normal in both X and Y. By Zorn's lemma, there

exists a unique maximal normal subgroup of (X,Y;H) which is called the

core of this amalgam. If the core is trivial (i.e., the identity subgroup)

then we say that the amalgam is~simple.

 - 5 ­

If K is a normal subgroup of (X,Y;H) then (X/K,Y/K;H/K)

is also an amalgam called the quotient of (X,Y;H) modulo K. If K

is the core of (X,Y;H) then this quotient is a simple amalgam.

I
If (Xi,Y i ), i = 1,2 are amalgams then a morphism

(fl,f2):(~'Yl) + (Xz,Y 2 ) is called an imbedding if both fl and f2

(X ,Y ) are called equivalent if the subamalgams (f (X ),f (Y » and

2 2 l l 2 l
(gl(~),g2(Yl» are conjugate in (X2 ,Y 2 ).

An amalgam (X,Y) is finite if both X and Yare finite

groups. The degree of an amalgam (X,Y;H) is the ordered pair (m,n)

where m (resp. n) is the index of H in X (resp. Y).

If (X,Y) = (X,Y;H) is an amalgam then we denote by

X *Y
H

the generalized free product of X and Y amalgamating the subgroup H.

An amalgam is proper if its degree (m,n) is such that m~ 2

and n;:: 2.

We remark that an amalgam (X,Y;H) may be simple without the

group X *Y being simple. This is always the case if the amalgam is

H

proper and finite.

3. AMALGAMS OF I-TRANSITIVE GROUPS

We consider pairs (G,A) where G is a connected regular

graph of valence d, and A a subgroup of Aut(G) which is I-transitive .

..
 - 6 ­

Fix a I-arc S in G, S(O) = a, S(l) = b; thus {a,b} is an edge

of G.

Define X = A(a) = the fixer in A of the vertex a. Let

Y be the subgroup of A consisting of all a A such that

•
€

{a(a),a(b)} = {a,b}. Then H = X n Y is the fixer in A of the I-arc S.

Hence we can associate to (G,A) and S the group amalgam

(X,Y;H). Since A is I-transitive this group amalgam is in fact

independent (up to isomorphism) of the choice of S. Indeed, if T is

another I-arc, T(O) = u, T(l) = v and a € A is such that a 0 S =T

then axa-1 = A(a) and aYa

-1
is the stabilizer in A of the edge {u,v}.

Theorem 1. Let (G,A), S, and (X,Y;H) be as above. Then A is

generated by X and Y, the amalgam (X,Y;H) is simple and has degree

(d,2).

Proof: The first assertion follows from [3, Proposition 1]. Let N be a

normal subgroup of (X,Y;H). Thus N cHand since X and Y generate

A, we have N 4 A. Now let x € V = V(G) and choose a € A such that

x = a(a). If S € N then

S(x) Sa(a) = a(a-lSa(a» = a(a) = x

-1

because a Sa € N c H c A(a). Hence S(x) = x for all x € V and so

S =1 and N = {l} . Thus (X,Y;H) is a simple amalgam.

X = A(a) acts transitvely on the set of d neighbours of a

and H is the fixer in X of the vertex b. Therefore (X:H) = d.

Similarly, (Y:H) = 2.
.,
 - 7 ­

Let (Gi,A ),
i
i = 1,2 be two pairs as (G,A) above. We say

that they are of the same type if the associated amalgams (Xi,Yi;H ),
i
i = 1,2 are isomorphic.

t
The converse of Theorem 1 is also valid.

Theorem 2. Let (X,Y;H) be a simple amalgam of degree (d,2). Then there

exists a pair (G,A) consisting of a connected regular graph G of

valence d and a subgroup A of Aut (G) which is I-transitive such

that the amalgam associated to (G,A) is isomorphic to (X,Y;H).

Proof: Let A be any group containing the amalgam (X,Y;H) which is

generated by X and Y and such that

-1
Y Xy n X = H (y € Y, Y ;. H).

For instance, we can take A=X * Y.

H

Define G as follows: its vertices are the cosets uX, u € A

and its edges are {uX,uyX} for u € A, Y € Y, Y i H. Since X and Y

generate A, the graph G is connected. The group A acts on G by

left translations and it is transitive on V = V(G) = A/X. Therefore, G

is a regular graph. The neighbours of the vertex X in G are the vertices

xyX where x € X and y € Y \ H is fixed. The number of these vertices

-1 -1
is equal to the index of y Xy n X = H in y Xy, i.e., it is equal to

-1 -1-1
(y Xy:H) = (y Xy:y Hy) = (X:H) = d.

Let S be the I-arc of G defined by S(O) = X, S(l) = yX.

The subgroup X of A is th~ fixer of the vertex X € V. Similarly, the
 - 8 ­

-1
fixer in A of the vertex yX is the subgroup yXy of A. The fixer

of the I-arc S is the subgroup

-1
= yHy = R. "

Hence the kernel of the homomorphism A ~ Aut(G) . is contained in H.

Since (X,Y;H) is a simple amalgam we conclude that the action is faithful

and hence we may consider A as a subgroup of Aut(G).

Since X permutes transitively the neighbours xyX (x € X)

of X € V in G, it follows that A is I-transitive. Thus we have a

pair (G,A) where G is a connected regular graph of valence d and

A ~ Aut(G) is I-transitive. The amalgam associated to (G,A) and the

I-arc S is precisely the original amalgam (X,Y;H).

Let (X,Y;H) be a simple amalgam of degree (d,2), d :::: 3.

Choose x € X \ H, Y € Y\H and put z = xy. Now define H

1
= H,
-1
Hi +l = H.1. n zH.z
1.
(i :::: 1).

Theorem 3. Let (G,A) be as before with associated amalgam (X,Y;H). Then

the index of A is the smallest integer s (or ro otherwise) such that

Proof: Let S be the I-arc used in the construction of (X,Y;H) and

8(0) = a, 8(1) = b. From our choice of x and y above, we have

x(a) = a, x(b) = c ~ b, y(a) = b, y(b) = c. Thus z(b) = xy(b) = x(a) = a,

z(a) = xy(a) = x(b) = c, i.e., in the terminology of [3], z is an A-

shunting. • j
 - 9 ­

Let us define aO = b, a
1
= a and in general a +
i 1
= z(a )
i
(i ~ 0). Note that H = H1 = A(aO,a1 ) is the fixer of the 1-arc
-1
(a ,a )· Hence zHz is the fixer of the 1-arc (a ,a ) and consequently
O 1 1 Z
-1
HZ >== H1 n zH1 z is the fixer in A of the Z-arc (a ,a ,a )' In general,
O 1 Z
Hi is the fixer of the i-arc (aO,a 1 ,···,ai )·

Let s be the index of A. For i < s the equality

(Hi :Hi~l) == d -1 is valid since Hi acts transitively on the neighbours

of a. distinct from and Hi +1 is the fixer in Hi of

1.

Since A is not (s + 1) - transitive, the same argument shows that

In view of this theorem we define the index of a simple amalgam

(X,Y;H) of degree (d,Z), d ~ 3 to be the smallest integer s such that

The subgroups H. are defined as above and it follows

1.

from Theorem 3 that this s is independent of the choice of x and y.

If (X,YjH) is a simple amalgam of degree (d,Z), d ~ 3 and

index s and if H == {l}, or equivalently the order of H is (d_l)s-l,

s
then we shall say that this amalgam is s-regular or just regular.

4. FINITE SIMPLE AMALGAMS OF DEGREE (3,Z)

We shall now define seven important amalgams, see [3, Section 7].

We denote by C (resp. D) the cyclic (resp. dihedral) group of order n

n n
(resp. Zn). The amalgams will be defined by generators and relations; it is

to be understood that for each X,Y~H one should use only the relations

which involve only the generator~ of that group_

 - 10 ­

We now list our amalgams:

3
a =y 2 = 1 , HI = {I }.

Am(2' ) = (XZ'Yi;H Z), X

2
= <a,b>, 2
=
y' <b,y>, H2 = <b>,
Z 2

a = b2 = y2 = (yb) = (ab)3 = 1­

Thus X ;;: D , y' =....

C x C , H2 ;;: C2 •
2 3 2 2 2

= <a,b>, = <y>, 224

X
2
y"
2
H2 = <b>, a =b =y =1,

2 3

y ,.;, b, (ab ) = 1.

Thus Y"2 ....

=

Am(3') = (X ,Y ;H ),
3 3 3
X3 = <a,b,c>, Y
3
= <b,c,y>, H3 = <b,c>,
y2 = a 2 = b2 = c 2 = (ab)2 = (bc)
2
= (ac)
3
= 1, yby = c.

y2 = a 2 = b 2 = c 2 = d2 = (ab)2 = (bc)2 = (cd)2 = 1,

(ac)2 = b, (bd)2 = c, (ad)3 = 1, (yc)2 = 1, yby = d.

Am(4") = (X ,Y ;H ), X = <a,b,c,d>, Y <b,c,d,y>, H4 <b,c,d>,

4 4 4 4 4= =

2 2 2 2 = (ab)2 = (bc)2 = (cd)2 = 1,

a = b = c = d
2 -1
(ac)2 = b, (bd)2 c, (ad) 3 = 1, Y = c, yby = d.

Thus y" is the so called quasi-dihedral group of order 16.

4
••
 - 11 '­

Am(S') = (X~'YS;HS)' Xs = <a,b,c,d,e>, YS = <b,c,d,e,y>, HS = <b,c,d,e>,

y2 = a 2 = b2 = c
2 = d2 = e 2 = (ab)2 = (bc)2 = (cd)2
2 22
= (de) ~ (ac) = (bd) = (ce) 2 = 1,
(ad)2 = be, (be)2 = cd, (ae)3 = 1, yby = e, ycy = d.
Thus xS
= <a,bc,cd,e> x <c>,

s = <b,c,d,e>)<f <y> (semidirect product),

Y and

'HS = <b ,e> x <c> = <b ,e> )(<d>, <b ,e> at D •

4

Theorem 4. Every finite simple amalgam of degree (3,2) is isomorphic to

one of the seven amalgams listed above.

Proof: Let (X,Y;H) be such an amalgam. By Theorem 2 there exists a

connected cubic graph G and A S Aut(G) which is l-transitive such

that (X,Y;H) is the amalgam associated to (G,A). Then A must be s-

regular for some s = 1,2,3,4,S and our assertion follows from [3, Prop­

osition IS].

Theorem S. The number of subama1gams of Am(s') or Am(S") which are

isomorphic to Am(t') or Am(t") is given in the table below.

l' 2' 2" 3' 4' 4" S'

l' 1 2 0 2 16 0 16
2' 1 0 1 0 0 0
2" 0 1 1 0 0 0
3' 1 0 0 0
4' 1 0 1
4" 0 1 1
• j
S' 1
 - 12 ­

Proof: This follows from [3, Theorems 3 and 4].

We have shown in [3] that the two sub amalgams Am(lt) are

not conjugate in Am(2t) but they are conjugate in Am(3 t ). The 16

subamalgams Am(lt) in Am(4 t ) split into 2 conjugacy classes, each

of size 8. Inside Am(5 t ) all these 16 subamalgams are conjugate.

5. FINITE SIMPLE AMALGAMS OF DEGREE (d,2), d > 3,

Very little is known about such amalgams. We shall summarize

here the most important results.

First we show that there are infinitely many finite simple

amalgams of degree (4,2). Let H be an elementary abelian group of order

n+l Let Ht
2 with a basis be the subgroup of H generated

by aO,al, ••• ,a _ l • We take X to be the semidirect product

n
2 2 4
X = Ht )(J <an ,x> where <an ,x> ;;: D ,
4
x = a
n
= 1, (xa)
n
= 1- The

action of an on H' is trivial and x acts as follows: xaix = an-1 -1..

(0 siS n - 1). Finally, let Y be the semidirect product Y = R)<l <y>

2
where <y>:;:, e2 , Y =1 and y acts on H as follows ya.y
1.
= an­i

(0 SiS n). Then (X,Y;H) is a finite amalgam of degree (4,2) and

it is simple. Indeed, let N c H be a subgroup which is normal in both

X and Y. Then

N c H' ~
II
yH'y = <al' a 2"'" a n-l > .
 - 13 ­

By repeating this argument we obtain that N = {I}.

The next Theorem we deduce from a result of R.M. Weiss [8].

Theorem 6. Let (X,Y;H) be a finite simple amalgam of degree (d,2)

where d is a prime > 3. If X is solvable then its order divides

d(d-l)2 and consequently the index of (X,Y;H) is ~ 2.

Proof: By Theorem 2 there exists a pair (G,A) where G is a connected

regular graph of valence d and A ~ Aut(G) is I-transitive such that

the amalgam associated with (G,A) is isomorphic to (X,Y;H). Now we can

proceed as in the proof of R.M. Weiss of his Satz in [8].

In the next theorem we collect some know pastial results about

amalgams of degree (d,2), d > 3.

Theorem 7. Let (X,Y;H) be a finite simple amalgam of degree (d,2),

d > 3 and index s.· Then

(i) if d =4 we have s = 1,2,3,4,7;

(ii) if d - 1 is a prime ~5 we have s = 1,2,3,4;
(iii) if d 1 is a prime ~ 3 and the amalgam is regular we have

s = 1,2,3 and if s = 2,3 then d - 1 is a Mersenne prime;

(iv) if d = np r + 1 where n < p, p a prime, and the amalgam is

regular, then s = 1,2,3,4,5,7.

The parts (i) and (ii) are due to A. Gardiner [4], (iii) to
+,":r.At"
,
theYauthor [2], and (iv) to .Weiss (6] .
 - 14 ­

6. AMALGAMS OF LOCALLY I-TRANSITIVE GROUPS

Let G be a connected bipartite graph, V = V(G) its vertex

set and (V+,V-) the corresponding partition of V. Let A S Aut(G) be

locally I-transitive. In order to avoid duplication with the case of 1­

transitive groups we assume here that A preserves both V

+ and V.
­

Thus V+ and V are the two orbits of A in V. It follows that every

vertex in V+ has the same degree d+ and every vertex in V has degree d-.

A has also two orbits in the set of all I-arcs of G. If Si'

i = 1,2 are two I-arcs then there exists an a € A such that a 0 Sl = S2
iff Sl (0) and S2(0) are both in V+ or both in V •

Fix a I-arc S such that S(O) = a € V+ and S(l) =b € V •

Let X = A(a) be the fixer in A of the vertex a and let Y = A(b).

Then H =X n Y = A(a,b) is the fixer in A of S. We associate with

(G,A) and S the group amalgam (X,Y;H). Since A is transitive on 1­

arcs T such that T(O) € V+ , it follows that this amalgam is independent

(up to isomorphism) of the choice of S.

Theorem 8. Let (G,A), Sand (X,YjH) be as above. Then A is generated

by X and Y, the amalgam (X,Y;H) is simple and has degree + ­

(d ,d ).

Proof: Let B = <X,Y>. In order to show that B =A it suffices to prove

that B is transitive on V+ and on V • Let u € V+ and we shall

prove that there exists a. € A such that a(u) = a. We shall prove this

by induction on the distance 8(a,u) = d from a to u. If d =0

then u = a and we can take. jJ. = 1. Let d > 0 and assume that our
 - 15 ­

claim is true for vertices in + whose distance from

V. a is less than

. d. If deb ,u) = d + 1 we can choose /3 E: X such that deb ,/3(u» = d -1.

Of course we still have d(a,/3(u» = d. This shows that we can assume

that d(b,u) =d - 1. Then there exists y € Y such that d(a,y(u» =d-2

and we can then use the induction hypothesis. Thus A is generated by

X and Y.

Since X acts transitively on the d+ neighbours of a and

H is the fixer in A of b, it follows that (X:H) = d+ • Similarly,

(Y:H) = d-. Finally, the simplicity of (X,Y;H) follows as in the proof

of Theorem 1.

The converse is also valid.

+ -
Theorem 9. Let (X,Y;H) be a simple amalgam of degree (d ,d ). Then

there exists a connected bipartite graph G with the underlying partition

(v+,V-) of V = V(G) and a subgroup A ~ Aut(G) which is 1-transitive

and such that the amalgam associated with (G,A) is isomorphic to (X,Y;H).

Proof: Let A be a group containing the amalgam (X,Y;H) which is

generated by X and Y. For instance we can take A =X* Y.

H

Define G as follows: its vertex-set V is the disjoint union

of V+ = G/X and V- = G/Y and the edges of G are {uX,uY} for u E: A.

-1
It is easy to check that uX and vY are adjacent iff u v E: XY. Since

X and Y generate A, G is connected and bipartite with associated

partition (V+ ,V- ). The group A acts on G by left translations and V+

and V are the two orbits of A in V. Hence any two vertices in V+

(resp. V-) have the same degree. The neighbours of the vertex X are
 - 16 ­

the vertices xY, x c X. The number of these vertices is equal to the

index (X:X n Y) = (X:H) = d+. Similarly, the vertex Y has d

neighbours.

Let 5 be the 1-arc 5(0) = X, 5(1) = Y. The fixer of the

vertex X (resp. Y) in A is the subgroup X (resp. Y) of A. Hence

the kernel of the homomorphism A ~ Aut(G) is contained in H =Xn Y.

Since (X~Y;H) is a simple amalgam we conclude that A acts faithfully

on G and hence we may consider A as a subgroup of Aut(G).

Since the group X (resp. Y) permutes transitively the

neighbours of the vertex X (resp. Y), it follows that A is locally

1-transitive. The amalgam associated to (G,A) and the 1-arc S is

precisely the original amalgam (X,Y;H).

We continue to use the notation and hypotheses from the beginning

of this section. Fix an x c X \ Hand y c Y \ H. Then we define vertices

a i
-
= x(a.),
1.
(i ~ 1),

(1 ~ 0).

Define H+ = H­ = H
1 1
+ + - -1
and Hi + 1 = Hi n YH1 y , i ~ 1;
- +-1
Hi + 1 = Hi n xH i x , i ~ 1­

Theorem 10. Use the above hypotheses and notation. Then the local index

of A is the smallest integer s (or 00 if no such s exists) such that

••
 - 17 ­

or
+ +
(Hs:H s + l ) <
r­d
1

- 1
if

if s
s is even,

is odd,

1 if s is even,
(H~ :H~+l) < {::
I if s is odd.

Proof: Note that H is the fixer in A of the I-arc (aO,al ) = (a,b) •

-1
Hence, xHx is the fixer of the I-arc (a_l,aO) and so H2 is the

fixer of the 2-arc (a_l,aO,a l )· Similarly, H+ is the fixer of

2
In general, H. is the fixer of the i-arc
1.

Let s be the local index of A. If i < s then H:

1.
acts

transitively on the neighbours of a. distinct form a.1.­ 1. If i is

1.
+ + +
even then a. e: v+ and so (Hi :Hi + l ) = d - 1. If i is odd then a e: V­
1. i

= d­
+ +
and so (Hi:H i + l ) 1.

Similarly, we have

1 if i is even,

- 1 if i is odd.

Since A is not locally (s + I)-transitive, the same argument

shows that for i =s at least one of the inequalities given in the theorem

must hold.

In view of this theorem we define the local index of a simple

ama1gam (X,;H
Y) 0 f d egree (d'4,d-) to be the smallest integer s such
 - 18 ­

that one of the inequalities in Theorem 10 is valid. It follows from our

theorem that s is independent of the choice of x and y.

If (X,Y;H) is a simple amalgam of degree + -

(d ,d ),
d+ ~ 2, -
d ~ 2 and local index s and if H+ ... H­ ... {I}, then we shall
s s
say that this amalgam is locallx s-regular or just locallx regular •

. 7. FINITE LOCALLY REGULAR SIMPLE AMALGAMS OF DEGREE (3,3).

Let (X,Y;H) be such an amalgam and let s be its local

index. By Theorem 9 there exists a bipartite connected cubic graph G

with associated partition (V+,V-) of V = V(G) and a subgroup A S Aut(G)

which preserves V+ and V such that the amalgam associated to (G,A)

is isomorphic to (X,Y;H). Say, X is the fixer of a vertex a E: v+ and

Y is the fixer of a neighbouring vertex b € V •

Since the amalgam (X,Y;H) is locally s-regular for some

positive integer s, it follows that the group A is locally s-regular

and v+ and V are two orbits of A in V. It was proved by Weiss [7]

that we must have s = 1,2,3,4,5, or 7.

Theorem 11. For each s = 1,2,3,4,5 there exists precisely one locally

s-regular finite simple amalgam of degree (3,3). For s = 7 there are

two such amalgams (X,Y;H) and (Y,X;H) which differ only in the order

of X and Y.

The proof will be given separately for each value of s.

• •
 - 19 ­

Case s = 1. The local 1-regularity implies that H = {1} and so

Case s = 2. Now H has order 2, so H t'Y

= C2 " If x e X \ H then
-1
H2 =H n xHx satisfies (H:H )
2 = 2. Hence X is a non-abelian group

of order 6 and the same is true for y, Thus we have X~ Y ~ D , H ~ C "

3 2

·Case s = 3. Let (G,A) be constructed from our amalgam as in the

t'Y

beginning of this section. Let b be the generator of the fixer A(a,b,c)

of the 2-arc (a,b;c) which is cyclic of order 2~ see Figure 1,

•
a
r
b c
Figure 1
• •
d

t'Y

The element b fixes b and its 3 neighbours and moves each vertex at
t'Y

distance 2 from b. It is a unique such element in A and hence b

belongs to the center of A(b) = y~ say.

Since a(c) =e we have sea = e. Thus (ac)2 = ec which

(ac)2. ........ 3
implies that ac is conjugate to Since (ac) € A(a,b,c) we have
3 . ......., 3 ....

either (ac) =1 or (ac) = b, The second case is impossible since ac

and (ac)2 have the same order, Hence (ac)3 = 1.

Thus A(b)

Similarly A(a) =X~ D6' Further we have

H =Xn Y = A(a,b) = <a,b> ~ C x C ,

2 2

Clearly this determines uniquely our amalgam.

 - 20 ­

Case s = 4. The fixer A(b,c) has order 8, see Figure 2.

:~g
p • q r • s
• •
a b c d e

Figure 2

Let a be in the center of A(b,c). We claim that a fixes a and d.

Indeed, say that a(d) = f, aCe) = g. Let all be in A(b,c,d,e).

Then since a and a commute it follows that a also fixes f = a(d)

and g = aCe). Thus a fixes the 4-arc (e,d,c,f,g) and since A is

locally 4-regular one must have a = 1, a contradiction.

Thus a ~ A(a,b,c,d) and so the center of A(b,c) is of

order 2 and we shall write a = q where q is the edge {b,c}. The

element is of order 2 and fixes all neighbouring vertices while it

moves every vertex at distance 2 from q. Such an element is defined

for every edge of G.

...--., ....,
Since p(q) =q we have pqp = q i. e. , p and q commute.
On the other hand per) = t and so prp = t.

We have

Also

because (pr)2 = '""­

tr = q. The, ~quality
"'­
tr = q follows from the fact that
 - 21 ­

,...
A(b,c,d) == A(b,c,f)==A(f,c,d);;; C x C and that q, r, t are the 3
2 2
non-identity elements in this group.

Now we have

A(c) = <p,q,r,s>.

Since ps(d) = p(d) = f, ps(f) = p(b) = b, and ps(b) = p(f) == d we

-3
have· (ps) € A(b,c,d). Thus the order of -ps is either 3 or 6. But
,...,...
as in the previous case we have that ps and (ps)2 are conjugate in A,

so that we must have - 3 == 1.

(ps)

Thus

,., ,... ,... ,...

<p,s> ~ D3 and <p,s> acts on <q,r> as follows:

----
pqp = q, prp
....-.,..,
= t == r q, -
-- =sqs t == rq,
......., .......-..
srs == r.
,...

Hence A(c) ~ 54 and similarly, A(b);;; 54'

Thus in this case X~ Y ~ 54 and H ~ D •

4

Case s = 5. Let a be in the center of A(c,d), see Figure 3.

w
v
u
x c d e f

z h

Figure 3
• Ii
 - 22 ­

We claim that a fixes also band e. Indeed, if a(b) =u and say

a(a) = v then let e ~ I be in A(x,a,b,c,d). Since a and e

commute it follows that e also fixes u = a(b) and v = a(a). Thus

e fixes as-arc (x,a,b,c,u,v) and so e = 1, a contradiction.

Thus a € A(b,c,d,e) and so the center of A(c,d) has order

2 or 4.

Assume that this center has order 2 and let a ~ I be its

generator. Then a(b) = b, . aCe) = e . but a must move either a or f,

say it moves a. Thus a(a) = y. Let. e 1 1 be in A(x,a,b,c,d). If

a(x) = z then since ae = ea, it follows that e also fixes y and z.

Thus e fixes all vertices at distance 2 from b and necessarily moves

all vertices at distance 3 from b. Thus e is uniquely determined by

the vertex b and we shall write e = h. Let us say that b € V+. Then

for every vertex t € V+ we have the corresponding involution t.

Let y 1 1 be in A(a,b,c,d,e). We claim that y(v) = w.

Otherwise y fixes all vertices at distance 2 from c and moves all

vertices at distance 3. This implies that y lies in the center of A(c)

and so the center of A(c,d) contains <e,y> which has order 4, a

contradiction. Thus we have proved that y(v) = w.

It follows that A(c) contains 3 involutions which fix a

4-arc with c as midpoint. We say that anyone of these three involutions

A.
is associated to c and will denote one of them by c. The analogous claim

is valid for every vertex t € V •

Choose "c = y and e" so that e(c) = c. Then e(u) = b

,..
""c) 2
• j
and so and c do not commute. But (e fixes b,c,d,e,f and so
,,'" 2
(ec) = d.IW
Therefore
"
<e,c>
"­
= A(c,d,e) ~ D4 .
 - 23 ­

"f'"W 2
Since ~(b) =u we have ~b~ =u and so (eb) = t"'foII/'V

ub. is an

element of order 2. On the other hand bee) = g, so

t'>tJ",~
beb = '"

g and
"'~ 2 ,,'" "b....) 2
(eb) = ego But and so ""
eg and (e are elements of order

4. This is a contradiction.

Thus we have proved that the center of A(c,d) must be of

order 4, and so it coincides with A(b~c,d,e). Thus it is an elementary

abelian group of order 4. Let a be the generator of A(a,b,c,d,e).

Since a belongs to the cen~er of A(c,d) it follows that a(u) = u.

Thus a fixes all vertices at distance 2 from c and moves all those

at distance 3. Thus we may write a =C since it is uniquely determined

....
by c. Similarly, we have d which generates A(b,c,d,e,f). Thus for

every vertex t there is a unique involution t associated to it in this

manner.

Now we find that

........

A(a,b,c,d) = <b,c> ~ C x C'

2 Z
.... _....
A(b,c,d) = <b,c,d> ~ C x C x C •
2 2 2

Since bee) = g we have

""I\J_
beb g and so
.......,

be :F eb.
-
Bu.t fixes
_t"<tJ ~ 2 ,..,...., ~
b,c,d,e and so belongs to <c,d>. Since (be) = bu = ge it follows
that it moves both a and f and so we must have (~)2 = ~.
Thus

A(c,d) = <b,~,d,;> = <b,e> x <c>

Now ae permutes cyclically the vertices b,d,u and so

~ ..... 3 . "

(ae) € A(b,c,d). Thus ae has order 3 or 6. The same argument as in

 - 24 ­

the previous cases shows that the order must be 3. Hence <i,e> ~
D •
3
We claim that <a,e> normalizes the four-group <~,Cd>. Indeed we have

""-I ,.,.."...",

= dad) d == cbcd
2f'<IJ
--- = ,..,...,
bd,

and so i normalizes this four-group. Similarly, e normalizes this

four-group. Consequently

(the symmetric group)

and

Similarly, A(d) ~ S4 x C •
2

Thus we have in this case

The case s = 7 will be considered in the next section.

8. END OF PROOF OF THEOREM 11

It remains to consider the case s = 7. We may assume that

our amalgam is associated with a pair (G,A) where G is a connected bi­

partite cubic graph and A S Aut(G) is locally 7-regular. If (V+,V-)

is the underlying partition of V = V(G) then A necessarily preserves

V+ and V since A cannot be 7-regular •

• •
 - 25 ­

It follows from local 7-regu1arity of A that a fixer in

6
A of a vertex in G has order 3e 2 , and the fixer of an i-arc
7-i
(1 !> i ~ 7) has order 2 • The fixers of some vertices and arcs in

Figure 4 are sketched in Figure 5.

•y x a b e f g h m

t \.
A(a) A(b) A(c) A(d) A(e) A(f) A(g)

A(a,b,c)

A(a,b,c,d,e)

A(a,b,c,d,e,f,g)

Figure 5

.,
 - 26 ­

Assume first that for every vertex U e V there exists a ~ 1

in A such that a fixes all vertices at distance :s; 3 from u. Then

local 7-regularity of A implies that a moves every vertex at distance

4 from u. Clearly a is uniquely determined by the vertex u and we

....
shall write a = u. If a € A and a(u) =v then it is clear that
0 ....0- 1
pUp. = .....v, by using the above geometric characterization of the involutions

U. Then (see Figure 4) we have

.....
A(a,b,c,d,e,f,g) == <d> .....
= C2 ,
A(b,c,d,e,f,g) == -
.... ..... C
<d,e> == 2 x C'2'

A(b,c,d,e,f) == <c,d,e> ..... C x C x C ,

2 2 2

A(c,d,e,f) = <c,d,e,f> .....

= C2 x C2 x C2 x C2 ,

because, for instance, c(f) =f implies that Cfc = f, i.e., c and f

commute.

Since b(f) == i and f(b) = j it follows that

--
bfb = i
.....
and

fb1 = j. Hence we have (bf)2 == if = bj which shows that (bf)2 fixes all
the vertices a,b,c,d,e,f,g and moves the vertex h. This implies that

(bf)2 == d. On the other hand bf € A(c,d,k,£.) and A(c,d,k,£.) = <~,d,k,l>

is elementary abelian group of order 16. This forces that· (bf)2 = land

we have a contradiction.

Thus we have proved that there exists a vertex u e V such

that if a € A fixes all vertices at distance :s; 3 from u then a = 1.

We may assume that u € V, and hence every vertex in V has this property.
• II
 - 27 ­

Now we claim that A(a,b,c,d) is non-abelian. Otherwise

the fixer of every 3-arc would be abelian. Since A(d) is generated

by the abelian subgroups A(a,b,c,d), A(b,c,d,e), A(c,d,e,f), and

A(d,e,f,g) and each of them contains the non-trivial element a of

A(a,b,c,d,e,f,g), it follows that a lies in the center of A(d). But

A(d) acts transitively on 6-arcs having d as midpoint and a fixes

one of them. Since a is central in A(d) it follows that a fixes

all of these 6-arcs and hence a fixes all vertices at distance s3

from d. But a ~ I and since d is arbitrary this contradicts the

fact established above. Thus we have proved that A(a,b,c,d) is non­

abelian.

Let a belong to the center of A(d,e). Then we claim that

a fixes every vertex at distance s2 from the edge {n,e}. Indeed,

assume that, say, a(b) = j. Let (x,y,a,b,c,d,e) be a 6-arc and let

S ~ I be an involution in A fixing this 6-arc. Since as = Sa and

a(b) = j, it follows that a fixes the 8-arc (x,y,a,b,c,j,a(a) ,a(y),a(x».

This is a contradiction since' S ~ I and A is locally 7-regular. Thus,

a must fix every vertex at distance s2 from {d,e}. Hence the center of

A(d,e) is contained in the group A(b,c~d,e,f,g). This latter group has

order 4 and is elementary abelian since it is generated by involutions.

We claim that the center of A(d,e) has order 2. Otherwise

the center of A(d,e) would coincide with A(b,c,d,e,f,g). Let a # I be

the involution fixing the vertices a,b,c,d,e,f,g. Since a belongs to

the center of A(d,e), an argument which we used earlier shows that a fixes

all vertices at distance ~3 from d. Similarly, if S # I is the

• •
 - 28 ­

involution fixing b,c,d,e,f,g,h then it fixes all vertices at distance

s3 from e. This contradicts the fact proved above about vertices in V.

Consequently, we have proved that the center of A(d,e) has order 2.

Let a +1 be the central element of A(d,e). Assume that

a moves every vertex at distance 3 from the edge p = {d,e}. Then a

is uniquely determined by this edge and we can write a = ,. A similar

statement is then valid for every edge of G since A is transitive on

A 1
the edges of G. If 6 E: and 6(p) = q then we have 6-6-
p = -q ,
....,
because of the geometric description of the involutions p. Since (see

Figure 6)

Figure 6

pu fixes all vertices at distance S 2 from the edge q and ~ +- 1, it

follows that pu = q. Now, we have pet) = sand t(p) =u and conseq­

uent1y
~

ptp = sand tpt= u .

......... 2 -2
Thus (pt) = st = rand (pt) = -­pu = q, and we have a contradiction

since r + q.
Hence we have proved that a must fix at least one vertex at

distance 3 from {d,e}. Say, we have a(h) = h. Then a is the invo1­

ution in A which fixes the 6-arc(b,c,d,e,f,g,h). Since a is central

• /I

in A(d,e), the usual argument shows that a fixes every vertex at

 - 29 ­

distance :S 3 from e. Thus Ct is uniquely d~termined by the vertex e

and we shall write Ct

,...
= e . Consequently ,we have e €
+ and for every
V

u f V+ there is an associated involution u. We repeat that every vertex

at distance :S 3 from u is fixed by u and every vertex at distance 4

from u is necessarily moved by U.

For each 6-arc S let S be the unique involution in A

which fixes S. If the midpoint S(3) = u is in v+ then we know that

,... ,...
S = u.
-1 ,...
Now assume that S(3) =u f V • Then we claim that S has

the form described on Figure 7 where the double arrows indicate the vertices
,..,
which are interchanged by S.

\
k-~l
/
-_.A:~

• j

Figure 7
 - 30 ­

Indeed, let v = S(O), w = S(l). Then, say, v(S(4» = x,

v(S(5» = y, v(S(6» = z. Now, v belongs to the center of A(v,w)
• and
,...
S E A(v,w). It follows that
,­ vS = Sv and consequently S also
....
fixes x,y,z. Since u E V , S cannot fix all vertices at distance 3
....
from u. Say S interchanges sand t. If s' = v(s) and t' = vet)

then S must also interchange a' and t'. Let z(s)::I a" and z(t) = til.

Since --
zS =
......
Sz, it follows that ""
S also interchanges S" and til. The

remaining pairs of vertices indicated by arrows on Figure 7 must be inter­

changed by S because A is locally 7-regular and S~ 1. Thus our

"'"
claim about S is proved.

It is clear from Figure 7 that there are 4 different involutions

,...
S such that S(3) =u where u E V •

From now on we shall use the notation from Figure 4. The

,vertices a,c,e,g are in V+ and hence we have canonical involutions

a,c,e,g. Let d be the involution fixing vertices a,b,c,d,e,f,g.

We have
'"
A(a,b,c,d,e,f,g) = <d> = CZ'

,.. '"
= ""e, d'"
A

Since dee) =e we have did i.e., and "'"

e cotmnute. Hence

A(b,c,d,e,f) = '"
<d,e> ~ C x C •
z Z
Similarly,

.... '" ....

A(b,c,d,e,f) = <c,d,e> ; C x C2 x CZ'

z
.... ....
since, for instance, c(e) =e implies that c and e commute.
••
 - 31 ­

Let "f E A be the involution fixing c,d,e,f,g,h,m. Since

df fixes the vertices c,d,e,f,g, we have that (df) 2 fixes also b

•
and h. Therefore (df)2 is either 1 or
....e. But "f(b) = j implies
""" """ '"
that fdf moves a and so fdf " d, i. e. , (df) 2 " 1- This proves that
(df)2 = e. Hence

A " '"
A(c,d,e,f,g) = <d,e,f>
A.
= <d,f> ; D4"

Since '"f(c) = c, the involutions c and "f commute. Thus

A A A A

A(c,d,e,f) = <c,d,e,f> = <d,e,f> x <c>

Since c(g) - nand g(c) =k we have cgc = n and gcg = k. Thus

which shows that (cg)2 fixes the vertices b,c,d,e,f,g,h. Since also

-
c gc = n:f. g we have (Cg)2 :f. 1 and consequently C"''')
cg 2 = ,..,e. Thus

<c,g> ~ D4 and

A(d,e, f)

where 0 denotes central product of groups. (see [9, p. ]) •

Let b" E: A be the involution fixing the vertices y,x,a,b,c,d,e.

" "f(b) """
i" =bfb, j" """
We have b(f) =i and =j, and put = fbf. Since
'"i(g) = n and
A

j (a) 'I a, it follows that the element

"'A 2 "''''' "A.

(bf) = if = bj

fixes b,c,d,e,f and moves .a ~

and g. Hence (bf)2ce fixes the vertices

a,b,c,d,e,f,g and so it is equal to 1 or "d. But "

band "f fix k
 - 32 ­

"" 2
and move l. Hence bf fixes k and t and consequently (bf) fixes
........

• all vertices at distance 2 from k. The same is true for ce and con­
,..." 2 ....
sequently for (bf) ceo This fact and the description of elements S on
"""''' 2~ "" 2
Figure 7 show that (bf) ce " d.
A
Therefore we have (bf) = ........
ceo Hence
A ," ""
A(c,d,e) = <b,c,d,e,f>

is a group of order 32. Omitting tildes and hats we have the following

defining relations for this.group:

2
b = c 2 = d2 = e 2 -- f2 =I,
c and e are central,
2
(bd)2 = c, (df) = e, ceo

We have, say, beg) =z and g(b) = t. Thus bgb = z and we

put 1 = gbg. Then 1 fixes k,d,e and we have

g) 2 -- zg
(b" .... = b"~-L.
2
It follows from here that (bg) fixes the vertices c,d,k,e,f,i and moves

t he ve r tices g b
g , z ,-L,. The e I emen t (b"....g)2 .c..."f fi xes thever ti ces b ,c , d ,e , f , g

and moves z. Therefore, it belongs to " ....

<d,e> and since it moves z, we

conclude that
A,..., 2,...,,"" ,..,
(bg) cfd € <e>.

By replacing "f by "fe (if necessary) we may assume that

Thus we have

A(d,eJ " " '"

= <b,c,d,e,f,g>

with defining relations the same as for A(c,d,e) plus the following:
 - 33 ­

2
(gc) ... e, (bg) 2 = cdef.

........

• The element ag permutes cyclically the vertices c,e,k. Thus

(ag)3 is in A(c,d,e) and so the order of ag is 3,6, or 12. But

Cag) 2 =: --au where u ... g(a) and so it is conjugate in A to

,..,....
ago There­

fore ag has order 3. Hence

" '" .....

A(d) = <a,b,c,d,e,f,g>

with defining relations the 'same as those for A(d,e) plus the following:

2' 2 223
a = (ab) = (ac) = (ad) = (ag) = 1,
2 2
(ae) = c, (af) = bcde.
A

Let h E A be an involution fixing a 6-arc (e,f,g,h,m, ••• ).

A'" 3
Then we have again (bh) = 1, by a similar argument. Hence
A A "" A
A(e) = <b,c,d,e,f,g,h>
with defining relations the same as those of A(d,e) plus the following:

2
(dh)2 ... eg, (fh)2 ... g, (ch) = defg.

Only the last relation needs justification. We have

•
proved in the same way as . . . 2 = ""
(Ch) dEiIg or
"2 AA '" ..... "
(Cll) == dfg. In the second case we need only replace h by hg = h'. Indeed,

we then have
 - 34 ­

This completes the proof in the case s = 7.

We summarize the result by giving the amalgam (X,Y;R) in

terms of generators and defining relations.

Thus

x = <a,b,c,d,e~f,g>,

Y = <b,c,d,e,f,g,h>,
H = X n Y = <b,c,d,e,f,g>,
2
a = b 2 = c 2 = d 2 = e 2 = f2 = g2 = h 2 = 1,
(ab)2 = (bc)2 = (cd)2 = (de)2 = (ef)2 = (fg)2 = (gh)2 = 1,
2 2
(ac) = (ce) = (eg) 2 = 1,

2 = g,

(bd)2 = c, (df) = e, (fh)2

(ad)2 = (be)2 = (cf)2 = (dg) 2 = (eh)2 = 1,

. 2
(ae)2 = c, (cg) = e,

(bf)2 = ce, (dh)2 = eg,

(af) 2 = bcde, (bg) 2 = cde f , (ch)2 = defg,

3
(ag) = (bh)3 = 1.

9. SOME REMARKS ABOUT AMALGAMS

,
Let (X,Y;R) be a simple amalgam and let (X' ,Y';H') be a

transitive subama1gam of it (see section 2 for the definition). Then we

claim that the only subgroup of R which is normalized by both X' and

y' is {1}. In particular, the subama1gam (X' ,Y' ;H') is simple.

The proof is easy: By Theorem 9 we may assume that (X,Y;H)

. ~

is the amalgam associated to (G,A) where G is a connected bipartite

 - 35 ­

graph2 with underlying partition (v+,V-) of V = V(G) and A is a locally

I-transitive subgroup of Aut (G) preserving V+ and V • Say X = A(a)

•
+
and Y = A(b) where {a,b} is an edge of G2 a ~ V , b ~ V • Let A'

be the subgroup of A generated by Xt and y' • By hypothesis A' is

transitive on V+ and on V • If N S H is normalized by both X' and

yt then N fixes a and b and also every vertex z(a) and z(b) for

z ~ At. Since V+ and V are two orbits of A' and a ~ V+ , b ~ V ,

it follows that N fixes every vertex of G. This implies that N = {I}

because A S Aut(G).

We leave to the reader to find a purely group-theoretical proof

of this result 2 which is not difficult.

Second remark is about finite simple amalgams (X,YjH) of degree

(d+,d).
­ We claim that if this amalgam is proper and a prime p divides

the order of H then or p < d • Indeed, then we have (G,A) as

in the first remark.

If a ~ H is an element of order p and if p ~ d+ and

p ~ d- then since a fixes the vertices a and b these inequalities imply

that a must fix all neighbours of a and b, etc. Thus a =1 and we

have a contradiction .
.

. ..

 - 36 ­

References

v ..
[1] I.Z. Bouwer and D.Z. Djokovic, On Regular Graphs III, J. Comb. Theory
14 (1973), 268-277.
v
[2] D.Z. Djokovi{, On Regular Graphs VI, J. Comb. Theory (to appear).
, v
[3] D.Z. Djokovit and G.L. Miller, Regular Groups of Automorphisms of
Cubic Graphs, J. Comb. Theory (submitted).

[4] A. Gardiner, Arc transitivity in graphs III (preprint).

[5] W.T. Tutte, Connectivity in Graphs, Univ. of Toronto Press 1966.

[6] R.M. Weiss, Uber s-regu1are Graphen, J. Comb. Theory 16 (1974),

229-233 •
..
[7] R.M. Weiss, Uber local s-Regu1are Graphen, J. Comb. Theory 20 (1976),
124-127.
"
[8J R.M. \veiss, Uber synnnetrische Graphen, deren Va1enz eine Primzah1 ist,
Math. Z. 136 (1974), 277-278.

[9] D. Gorenstein, Finite Groups, Harper and Row 1968.

The work of the first author was supported in part by NRC Grant
A-5285, and that of the second author by NRC Grant A-5549 •

..

• •