Eigenvalue techniques in design and graph theory

Citation for published version (APA):

Haemers, W. H. (1979). Eigenvalue techniques in design and graph theory. [Phd Thesis 1 (Research TU/e /
Graduation TU/e), Mathematics and Computer Science]. Stichting Mathematisch Centrum.
https://doi.org/10.6100/IR41103

DOI:
10.6100/IR41103

Document status and date:

Published: 01/01/1979

Document Version:
Publisher’s PDF, also known as Version of Record (includes final page, issue and volume numbers)

Please check the document version of this publication:

• A submitted manuscript is the version of the article upon submission and before peer-review. There can be
important differences between the submitted version and the official published version of record. People
interested in the research are advised to contact the author for the final version of the publication, or visit the
DOI to the publisher's website.
• The final author version and the galley proof are versions of the publication after peer review.
• The final published version features the final layout of the paper including the volume, issue and page
numbers.
Link to publication

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal.

If the publication is distributed under the terms of Article 25fa of the Dutch Copyright Act, indicated by the “Taverne” license above, please
follow below link for the End User Agreement:
www.tue.nl/taverne

Take down policy

If you believe that this document breaches copyright please contact us at:
[email protected]
providing details and we will investigate your claim.

Download date: 04. May. 2024

EIGENVALUE TECHNIQUES
IN DESIGN
AND GRAPH THEORY
EIGENVALUE TECHNIQUES
IN DESIGN
AND GRAPH THEORY

PROEFSCHRIFT

TER VERKRIJGING VAN DE GRAAD VAN DocrOR IN DE

TECHNISCHE WETENSCHAPPEN AAN DE TECHNISCHE
HOGESCHOOL EINDHOVEN, OP GEZAG VAN DE RECTOR
MAGNIFICUS, PROF. IR. J. ERKELENS, VOOR EEN
COMMISSIE AANGEWEZEN DOOR HET COLLEGE VAN
DEKANEN IN HET OPENBAAR TE VERDEDIGEN OP
DINSDAG 30 OKTOBER 1979 TE 16.00 UUR

DOOR

WILHELMUS HUBERTUS HAEMERS

GEBOREN TE EINDHOVEN

1979
MATHEMATISCH CENTRUM, AMSTERDAM
DIT PROEFSCHRIFT IS GOEDGEKEURD
DOOR DE PROMOTOREN

Prof.dr. J.J. Seidel

en
Prof.dr. J.H. van Lint
 CONTENTS

PREFACE 1

1. MATRICES AND EIGENVALUES

1.1. Introduction 4
1.2. Interlacing of eigenvalues 6
1.3. More eigenvalue inequalities 10

2. INEQUALITIES FOR GRAPHS

2.1. Induced subgraphs 15
2.2. Chromatic number 22

3. INEQUALITIES FOR DESIGNS

3 • 1 • Subdesigns 27
3.2. Intersection numbers 31

4. 4-COLOURABLE STRONGLY REGULAR GRAPHS

4.1. Introduction 37
4.2. Strongly regular graphs on 40, 50, 56, 64 and 77 vertices 40
4.3. Recapitulation 48

5. GENERALIZED POLYGONS
5.1. Introduction 50
5.2. An equality for generalized hexagons 53
5.3. Geometric and pseudo-geometric graphs for generalized polygons 60

6. CONSTRUCTIONS
6 .1. Some 2 - (71, 15, 3) designs 64
6.2. Some strongly regular graphs 71

APPENDIX I. GRAPHS AND DESIGNS 79

APPENDIX II. TABLES 85

REFERENCES 93
NOTATIONS 100

INDEX 101

SAMENVATTING 103

CURRICULUM VITAE 104

 1

PREFACE

The application of the theory of matrices and eigenvalues to combina-

torics is certainly not new. In a certain sense the study of the eigenvalues
of the adjacency matrix of a graph even became a subject of its own, see for
instance [BS], [C11], [C12], [H14] and [S2]. Also in the theory of designs,
matrix and eigenvalue methods have often been used successfully; see for
instance [C6], [C9], [H17] and [R3]. In the present monograph the starting
point is a new theorem concerning the eigenvalues of partitioned matrices.
Applications of this theorem and some known matrix theorems to matrices
associated to graphs or designs lead to new results, and new proofs of
known results. These concern inequalities of various types, including con-
clusions for the case of equality. In addition we obtain guiding-principles
for constructing strongly regular graphs or 2-designs. Let us give some
more details.
Our theorem (Theorem 1.2.3) about eigenvalues and partitionings of
matrices, which was announced in [Hl], reads as follows:

THEOREM. Let A be a complete he'f'l'flitian n x n matri:c, pa.I'titioned into m2

bZoak matrices, suah that ail diagonal bZoak matrices are square. Let B be
the m x m matrix whose i,j-th entry equals the average row sum of the i,j-th
block matrix of A for i, j = 1, •• .,m. Then the eigenvalues Al (A) <:: ... <:: An (Al
of A and the eigenvalues A1 (B) <:: ••• <:: Am(B) of B satisfy
Ai {A) <: Ai (B) <: An-m+i (A) for i = 1, ••• ,m •
Moreover, if for some integer k, O s k s m, A1 (A) = Ai(B) for i = 1, ••• ,k
and Ai(B) = >.n-m+i(A) for i = k+l, ••• ,m, then all the block matrices of A
have constant row and column sums.

The weaker inequalities Al (A) <:: >.i(B) <: An(A) were already observed by
c.c. Sims (unpublished), and have been applied successfully by HESTENES &
HIGMAN [H10], PAYNE [P4], [P6] and others. They are usually applied under
the name Higman-Sims technique. Many proofs by means of this Higman-Sims
technique can be shortened by use of our generalization. But, what is more
important, our theorem leads to new results, which we shall indicate below.
 2

Suppose G is a graph on n vertices, whose (0,1)-adjacency matrix has

eigenvalues A1 (G) ~ ••• ~ An(G). DELSARTE [D1] proved that, for strongly
regular G, the size of any coclique {independent set) cannot be larger than
- n An ~G) I {A 1 {G) - An (G)). A.J. Hoffman (unpublished) proved that this bound
holds for any regular graph G. Using the above theorem we prove that an
upperbound for the size of a coclique in any graph G is provided by

2
- n Al (G) An (G} / Cdmi n - Al (G) An (G}) ,

where dmin denotes the smallest degree in G. This generalizes Hoffman's

bound, since in case of regularity A1 (G) = dmin holds. More generally, by
use of the same methods we find bounds for the size of an induced subgraph
of G in terms of the average degree of the subgraph (Theorem 2 .1 • 2) • Apart
from the inequalities of Delsarte and Hoffman we also find inequalities of
Bumiller, De Clerck and Payne as corollaries of our result.
By applying the generalization of the Higman-Sims technique {with
m == 4) to the adjacency matrix of the incidence graph of a design, we obtain
bounds for the sizes of a subdesign in terms of the singular values of the
incidence matrix (Theorem 3.1.1). For nice designs, such as 2-designs and
partial geometries, this result becomes easy to apply, since then the sin-
gular values are expressible in the design parameters (for symmetric 2-
designs the inequality also appeared in [H4]}. Thanks to the second part of
our theorem certain conclusions may be drawn easily for the case that the
bounds are attained, for the graph case as well as for the design case.
We also prove results concerning the intersection numbers of designs,
such as the inequalities of AGRAWAL [Al] (Theorem 3.2.1) and the results of
BEKER & HAEMERS [B2] about 2 - (v,k,A) designs with an intersection number
k - A(v - kl I Ck - 1).
HOFFMAN [B13] proved that the chromatic number y (G) of a graph G
satisfies y (G) 1 - A1 (G) I An (G). To achieve this, Hoffman first proves a
~

generalization of the inequalities of Araonszajn concerning eigenvalues of

partitioned matrices. In Section 1.3 we give a new proof of these inequali-
ties using our generalization of the Higman-Sims technique. In Section 2.2
the application of these inequalities yields a generalization of Boffman's
bound {Theorem 2.2.1). For non-trivial strongly regular graphs this leads
to y (G} ~
max{l - A1 (G} I An (G) , 1 - An (G) / A2 (G}}. In Chapter 4 we use these
bounds, and many other results from the previous chapters, in order to
determine all 4-colourable strongly regular graphs. This chapter is also
 3

meant to illustrate some applications of the results and techniques obtained

in the first three chapters.
Chapter 5 is in the same spirit, but rather independent from the other
ones. The main result is the inequality of HAEMERS & ROOS [H3]: t :s; s 3 if
s # 1 'for a generalized hexagon of order (s,t), together with some addition-
al regularity for the case of equality. This is proved by rather elementary
eigenvalue methods. The same technique applied to generalized quadrangles
of order (s,t) yields the inequality of HIGMAN [B11]: t ~ s 2 if s # 1, the
result of BOSE [B7] for the case of equality and a theorem of CAMERON,
GOETHALS & SEIDEL [CS] about pseudo-geometric graphs.
Using eigenvalue methods we obtain guiding-principles for the con-
struction of designs and graphs. In Section 6 .1 we construct a 2 - ( 56, 12, 3)
design, for which the framework is provided by Theorem 3.2.4. This design
is embeddable in a symmetric 2 - ( 71, 15, 3) design. By modifying this design
we obtain eight non-isomorphic 2 - (71,15,3) designs. All these designs seem
to be new (the construction is also published in [B2]). In Section 6.2 some
ideas for the construction of strongly regular graphs are described. We
construct strongly regular graphs with parameter sets
3 2 2 3n 2n+l 2n n n n
(q +q +q+1,q +q,q-1,q+l) and (2 +2 , 2 +2 -1, 2 -2, 2 ) for prime power q
and positive integer n. Strongly regular graphs with the first parameter
set are known; however, our construction yields graphs which are not iso-
morphic to the known ones. The second family seems to be new. Special at-
tention is given to strongly regular graphs with parameter set (40,12,2,4).
Several such graphs are constructed with the help of a computer.
In the first appendix we recall some basic concepts and results from
the theory of graphs and designs (including finite geometries). This appen-
dix is meant for readers who are not familiar with the theory of designs
and graphs. The second appendix exhibits explicitly some of the designs and
graphs constructed earlier.
4

CHAPTER I

MATRICES AND EIGENVALUES

1.1. INTRODUCTION

In this chapter we shall derive some results about eigenvalues of

matrices. They provide the main tools for our investigations. We shall
assume familiarity with the basic concepts and results from the theory of
matrices and eigenvalues. Some general references are [M3], (Nl], (WS].
Let A be a square complex matrix of size n. The hermitian transpose
of A will be denoted by A*. Suppose A has n (not necessarily distinct)
real eigenvalues, which for instance is the case if A is hermitian (i.e.
A = A*). Then we shall denote these eigenvalues by

If denoted by subscripted variables, eigenvectors will always be ordered

according to the ordering of their eigenvalues. Vectors are always column
vectors. The linear span of a set of vectors u 1 , ••• ,un is denoted by
< u , ••• , un >. A basic result, which is important to our purposes, is
1
Rayley's principle (see (Nl], [WS]).

1.1.1. RESULT. Let A be a 'he:rwritian ma:t;r>i:c of size n. For some i, O s i s n,

iet u 1 , ••• ,ui be an orthonormai set of eigenvectors of A for A1 (A), ••• ,Ai(A).
Then

equaiity ho74s iff u is an eigenvector of A for Ai (A).

ii. A
*
(A)~~
i+1 *
u u

equaiity ho74s iff u is an eigenvector of A for Ai+l(A).

For the multiplicity of an eigenvalue we shall always take the geo-

metric one, that is, 1:fe maximal number of linearly independent eigenvectors
 5

(to be honest, this agreement is only of influence to the proof of the next
lemma, because throughout the remainder of this monograph we shall only
consider eigenvalues of diagonalizable matrices). Now we shall prove some
easy and well known, but nevertheless useful lemmas.

1.1. 2. LEMMA. Let M* and. N be aorrrp le;;; m

1
x m
2
matriaes. Put

Then the foLZo!iJing al'e equivalent.

i. A of 0 is an ei(Jenvalueof A of multipliaity f;
11. -A of 0 is an ei(Jenvalueof A of muitipUcity f;
2
A. .;.o is an ei(Jenvalue of MN of muitipUaity f;
iii.
2
iv. A. of O is an eigenvalue of NM of multiptiaity f.

PROOF.
1. (i) ..... (ii): let AU A.U for some matrix u of rank f. Write

and define

where Ui has mi rows for i • 1,2. Then NU 2 = A.u 1 and Mu 1 = AU 2 • This implies
AU = -;\U, Since rank u = rank u,
the first equivalence is proved.
2. {iii) ..... (iv): let MNU' = A.U' for some matrix u• of rank f. Then
NM(NU') = A.NU', and rank NU' rank U', since

rank A.U' = rank MNU' ::; rank NU' ::; rank u• ,

and ;\ 1' O. This proves the second equivalence.

3. (i) ..... (iii): because

A 2"' ~NM OJ '

0 MN

this equivalence follows immediately from the previous steps. D

The singu"lal' values of a complex matrix N, are the positive eigen-

values of
6

[:. :J ·
By the above lemma we see that they are the same as the square roots of the
' eigenvalues of NN* •
non-zero

1.1.3, LEMMA. Let

be a comp'le:x: matri:x:. If A11 is non-singu'lar, and rank A11 • rank A, then

~· For i "' 1,2, let ai j denote the j-th co1umn of Ai 2 • From

rank A= rank [A* A* J* .itI follows
11 21

-1
for some vector u. But if A11 is non-singular, then u • A a 1 ,j. Hence
-1
11
a 2 ,j = A21 A11 al,j' which proves the lemma. D

The identity matrix of size n will be denoted by In or I. The matrix

with all entries equal to one by J; a column vector of J is denoted by j;
Jn is a square J of size n; the symbol © is used for the Kronecker product
of matrices.
As a last result in this section we observe that, if K := In © Jm' then

1.2. INTERLACING OF EIGENVALUES

Suppose A and B are square complex matrices of size n and m, respec-

tively (m ~ n), having only real eigenvalues. If
 7

for all i = 1, ••• ,m, then we say that the eigenvalues of B interZaae the
eigenvalues of A. If there exists an integer k, 0 s k s m, such that

for i = 1, ••• ,k
and

then the interlacing will be called tight.

1.2.1. THEOREM. Let s be a aompZe:x: n x m matl'i:1; such that s*s = Im. Let A
be a he:r:rm:itian matl'i3: of size n. Define B := s*AS, and Zet vl' •• .,vm be an
orthonormal, set of eigenvectors of B. Then
i. the eigenvaZues of B intertaae the eigenvalues of A;
ii. if>..i (B) € {;\i(A), An-m+·(A)}
i
for some i € {1, ••• ,m}, then thel'e
e:x:ists an eigenvector v of B for >.i(B), such that sv is an eigen-
vectol' of A for Ai(B);
iii. if, for some JI.€ {O, ••• ,m}, .Ai(A) = Ai(B) for aU i • l, ••• ,R.,
then svi is an eigenveatol' of A for .Ai(A), for i = 1, ••• ,J!.;
iv. if the intertacing is tight, then SB = AS.

~· Let u 1 , ••• ,un be an orthonormal set of eigenvectors of A. For any i,

1 $ i s m, take

~i; 0 .

Thus also

This proves (i) •

If .Ai(B) = >. 1 (A), then viand S~i are eigenvectors of Band A respectively
for the eigenvalue Ai(B) = >. 1 (A). This, together with the same result
applied to - B and - A yields (ii) •
8

We shall prove (iii) by induction on I!.. If I!. = O, there is nothing to

prove. Suppose I!. > o. We have

* *ASVI!.
vl!.S
* *svl!.
vl!.s

On the other hand, SVR. E < SV , ••• ,SV1!.-l >'L, and by the induction hypothesis
1
sv1 , ••• ,svl!.-l are orthonormal eigenvectors of A for A1 (A), ••• ,Al!.-l(A). Now
1.1.1.ii yields that svl!. is an eigenvector of A for Al!.(A). This proves
(iii).
Let the interlacing be tight. By applying (iii) to A with I!. = k and to -A
with R. = m - k,
we find that Sv 1 , ••• ,svm is an orthonormal set of eigenvec-
tors of A for A1 (B), ••• ,Am(B). Write V := [v 1 ••• vm] and
D := diag(Al(B), ••• ,Am(B)). Then ASV = SVD, and BV =VD. Hence

ASV = SBV •

Because V is non-singular, (iv) has now been proved. 0

A direct consequence of the above theorem is the following theorem.

This result is known under the name Ca.uohy inequalitie8, see [H7], [M2],
[WS].

1.2.2. THEOREM. Suppo8e

ie a hermitian ma.tri:x:.
i. The eigenvalueB of A 11 interlaae the eigenvalues of A.

ii. If the interlar:dng is tight, then A 12 = o.

~·Let m be the size of A11 • Defines :=[Im oJ*, and apply 1.2.1. 0

Another consequence of 1.2.1 is the following result which was

announced in [al] (see also [H2]). This result will often be used in the
forthcoming sections.
 9

1.2.3. THEOREM. Let A be a he:rmitian matri:x: partitioned as foUows

suah that A.11. is square for i = 1, ••• ,m. Let b.1J. be the averoge row swn of
Aij' for i,j = 1, ••• ,m. Define them x m matri:x: B := (bij).

i. The eigenvalues of B interlace the eigenvalues of A.

ii. If the interlacing is tight, then Aij has constant row and colwnn
sums for i,j = 1, ••. ,m.
iii. If,-for i,j = 1, ••• ,m, Aij has constant row and colwnn swns, then
any eigenvalue of B is also an eigenvalue of A with not smaller
a muZtipZicity.

PROOF. Let n. be the size of A ..• Define

1 11

1 1 0 0 0 0
0 0 1 1 0 0
-* 0 0 0 0 0 0
s :=

0 0 0 0 1 1
nl n2 n
m

*
D := diag(n , ••• ,nm)' and S := -SD -1 .Then SS= I and --
SS= D. 2 We easily
1
-~
see that (S AS)ij equals the sum of the entries of Aij" Hence

B = -* - -2 •
S ASD

By 1.2.1.i we know that the eigenvalues of s*AS interlace the eigenvalues

of A. But B has the same eigenvalues as s*AS, since

* -1- - -1 -1
S AS = D S ASD = D BD •

This proves (i).

-1
It is easily checked that AS = S(D BD) reflects that Aij has constant row
sums. for all i,j = 1, ••• ,m. Hence 1.2.1.iv implies (ii).
10

On the other hand, i f AS = SD-1 BD and BO • l i (B) u for some matrix u and
-1
integer i, then A(SD U) = li (B)SD-1 u, and rank u = rank SD
-1
u. This proves
{iii). D

As a special case of the above theorem we have that

for i = 1, ••• ,m. These inequalities are well known and usually applied
under the name Higman-Sims technique, see [H10], [P4]. We shall also use
the name Higmans-Sims technique if we apply the more general result 1.2.3.
Also 1.2.3.iii is well known, see for instance [C12], [H9] (note that this
result remains valid for non-hermitian A). We see that 1.2.3.ii gives a
sufficient, and that 1.2.3.iii gives a necessary condition for the block
matrices of A to have constant. row and column sums. However, neither of
these conditions is both necessary and sufficient. This is illustrated by
the following partitioned matrices:

For both A and A' the eigenvalues are 2,0,0,-2, and the average row sums of
the block matrices are given by the entries of B = (~ ~). The block
matrices of A have constant row sums, whilst the interlacing is not tight.
The row sums of the block matrices of A' are not constant, whilst the
eigenvalues of B are also eigenvalues of A.

1.3. MORE EIGENVALUE INEQUALITIES

In this section we shall use interlacing of eigenvalues in order to

prove some known inequalities and equalities, which we shall use in later
sections. The ·first result is due to WEYL [W2] (see also [WS]).

1.3.1. THEOREM. Let A1 and A2 be hel'ITlitian matl'ices of size m. Then

for i = 1, ••• ,m, 0 s j s min{i-1,m-i}.

 11

PROOF. Define

and

S:=\fi[I I]*.
m m

Then A1 (A) = O, and

With 1.2.l we now have

If we replace A 1 and A 2 by - A 1 and - A 2 , we get the second inequality. D

The next theorem is due to HOFFMAN [B13].

1.3.2. THEOREM. Let A be a hermitian matri:x: of size n, partitioned as

fol.tows

~
All. Alm]
..
A= . '
A~l ••• A~

where A11 is a square matioix of size ni, for i = 1, ••• ,m. Let j 1 , •.• ,jm be
integers euah that ls ji s n1 for i = 1, ••• ,m. Then

!!!:Q2!.· Let u , ••• ,un be an orthonormal set of eigenvectors of A. Let

1
u11 , ••• ,uin. be an orthonormal set of eigenvectors of A11 for i = 1, ••• ,m.
l.
...uk"' [ulk
"'* •.• "'*
umk ]* ;. O, such that
Put k := j
1 + ••• + jm. Choose a vector
12

and

It follows from dimension considerations that we can always do so. Now

define

and

for i = 1, ••• ,m. Furthermore put

and
0 0

0
s :==

0 0

Then we have

S *S =I and (S *AS) i i for i = 1 , ••• ,m •

By 1.1.1.i and the choice of Qik' the last formula gives

for i = 1, ••• ,m •

Hence

m
(*)
i=l
r
On the other hand, sw = ~uk, s*s = I and 1 .1 .1 yield
 13

(**)

Apply~ng 1.2.1 gives

m-1 m-1
I A. cs*AS> s I A.i CA) •
i=1 i i=l

Combining (*), (**) and (***) yields the first inequality of our theorem.
Again, the second inequality follows by substituting -A for A in the first
one. 0

If the matrix A of the above theorem is positive semi-definite and

m = 2, then we have

These are the inequalities of ARONSZAJN [A3] (see also [H7]).

The following consequence of 1 • 3. 2 will turn out to be a useful tool

in computations with eigenvalues.

1.3.3. THEOREM. Suppose

is he:t'l'fl'itian of aise n. Suppose A has just 1::1Po distinct eigenvaiues, that

is,,

for some f, 1 s f < n. Let n1,and n2 be the aisea of A11 and A22 respective-
ty. Then
14

PROOF. By the Cauchy inequalities (1.2.2.i) we have

This proves the result for 1 s i < f+1-n 1 , and for f < i s n •
2
For the remaining values of i, 1.3.2 gives

which proves the required result. D

It is an easy exercise to give a direct proof of the above theorem.

The proof then could go analogously to the one of Theorem 5.1 of [CS],
where a similar result is stated.
 15

CHAPTER 2

INEQUALITIES FOR GRAPHS

2.1. INDUCED SUBGRAPBS

In this section we shall derive inequalities for induced subgraphs of

graphs, using the results of section 1.2 on interlacing of eigenvalues.

Let G be a graph on n vertices. The eigenvaZues of G are the eigen-

values of its (0,1)-adjacency matrixi we denote them by A (G) ~ ••• ~ An(G).
1
Let a 1 be an induced subgraph of G. Then by 1.2.2 (Cauchy inequalities) the
eigenvalues of G interlace the eigenvalues of G. In particular, if G1 is a
1
coclique of size a, then Aa(G) ~ Aa(G 1 ) = O, and An-a+l(G) ~ A1 CG 1 ) = O.
Bence, we have the following result, which was first observed by CVETKOVIC
[Cll] (see also [C12]).

2.1.1. 'l'BEOREM. The size of a cocZique of a graph G cannot e:x:ceed the number
of nonnegative [nonpositive] eigenvaiues of G.

Now we shall derive inequalities for induced subgraphs using the

Higman-Sims technique (1.2.3). Suppose G is a graph on n vertices of
average degree d. Let the vertex set of G be partitioned into two sets,
and let a 1 and G be the subgraphs induced by these two sets. For i = 1,2,
2
let ni b~ the number of vertices of Gi' let di be the average degree of Gi'
and let di be the average of the degrees in G over the vertices of Gi. Now
we can state the following theorem.

2.1.2. TBEO:REM. For i = 1,2

i.

ii. if equaUty hoUs on one of the sides~ then G1 and G2 are

reguZar~ and aZso the degrees in G are conatant over the
vertices of a 1 and a 2 •
16

~· If G is complete we easily see that the theorem is correct. So let

us assume that G is not complete. Let A11 , A and
22

be the adjacency matrices of G1 , G2 , and G, respectively. Put

Then the entries of B are the average row sums of the block matrices of A.
By 1.2.3.i we have

Because trace A = 0, we have A.n (A) 0. From 2 .1 .1 we know that I. 2 (A) <: 0,
s;
since otherwise G would be complete. Hence

On the other hand we know

We quickly see

This yields

With{*) and(**) this proves (i).

If equality holds on one of the sides, the interlacing must have been
tight. Hence 1.2.3.ii gives (ii). D

Now let us look at the consequences of the above inequalities for

some special cases. The size of the largest coclique and clique of G are
denoted by a(G) and w(G), respectively.
 17

2.1.3. THEOREM. If dmin and dmax are the smaUest and the Zargest degree in
the gra:ph G, respectively, then

i.

ii.

PROOF. (i) • Substitute a (G) = n1 , d1 =0 and d 1 :2: dmin in the right hand
inequality of 2.1.2.i.
(ii). Substitute w(G) = n1 , d1 = w(G) -1 and ""
d
1
~ dmax in the left hand
inequality of 2.1.2.i. 0

2.1.4. THEOREM. If G is a r>eguw graph on n vertices of degree d, then

any subgra:ph G1 of G with n

i.
1 vertices and aver>age degree d 1
satisfies
nd -n d
A (G) :2: l l ~ An(G) ,
2 n - n
1

~·If G is regular then A1 (G) = d = ""d 1 = ....a 2 = dmin = dmax. Now

2.1.2.i, 2.1.3.i and 2.1.3.ii give the required results. 0

The inequality 2.1.4.ii is an unpublished result of A.J. Hoffman

(see [C12], [H2], [L2]). In fact, the inequalities (ii) and (iii) of 2.1.4
(just as the left and the right hand inequality of 2.1.4.i) are equivalent,
because either one can be obtained from the other by using w(G) = a(G),
>. 1 (G) .. n - Al (G) - 1 and Xi (G) = - An-i+ (G) - 1
2
for i = 2, ••• ,n (G is
the complemen~ of G).

For the graph G with its subgraphs G1 and G2 , we define D(G,G 1) to be

the incidence structure whose points and blocks are the vertices of G1 and
G , respectively, a point and a block being incident if the corresponding
2
vertices are adjacent. If we have equality in any of the inequalities of
2.1.2 - 2.1.4, then 2.1.2.ii yields that D(G,G 1) is a 1-design, possibly
18

degenerate. Now let G be strongly regular. Then by use of 1.2.3.iii it is

not difficult to show that equality holds in 2.1.4 iff D(G,G ) is a 1-design.
1
If G1 is a coclique or a clique we have a criterion for D(G,G ) to be a 2-
1
design.

2.1.5. TEIEOREM. If G is a strongly regu'la:r> graph on n vertices of de{II'ee d,

then
i. a(G) s 1 + (n-d-1) I o. 2 (G) +1),

iii. if equality hoUB in (i) or (ii), and G1 is a coclique of size

a(G), or a clique of size w(G), respectively, then D(G,G ) is a
1
2-design, possibly degenerate.

PROOF. If G is strongly regular, we know (see [CS], [S4] or Appendix I)

(d - A. (G))
2 (d - An (G)) = n(d + ). 2 (G) An (G)) •

From this it follows in a straightforward way that (i) and (ii) are equi-
valent to 2.1.4.ii and 2.1.4.iii.
From the definition of a strongly regular graph we know that any two points
of D(G,G 1 ) are incident with a constant number of blocks of D{G,G 1 ).
Furthermore, equality in (i) or (ii} implies that D(G,G 1 ) is a 1-design, so
in this case D(G,G ) is then a 2-design, possibly degenerate.
1 0

The theorems 2 .1. 5 and 2 .1. 4 for strongly regular graphs are known.
They are direct consequences of the linear programming bound of DELSAR'l'E

[Dl] (see also [H2]}. They were also proved by BUMILLER [B9].
Applying 2.1.4.i to the point graph of a partial geometry (see Appen-
dix I, or [B6], [Tl]) gives the following result of DE CLER.CK [C7] (see
also [P3] for the case a~ 1: and [B9] fort= t 1 ).

2.1.6. COROLLARY. Let P be a partial geometry with parameters (s,t,B),

containing a partial subgeometry P 1 with parameters Cs 1 ,tl'B). Then
 19

~· If G and G1 are the point graphs of P and P , respectively, then

1
(see Appendix I or [Tl]) G1 is an induced subgraph of G, and

n = (s+l){st+a) / a , n1 = (s
1
+l)(s t 1 +a)/a,
1

Substitution of these values in the left hand inequality of 2.1.4.i leads

to

{s - s ) (st +a) {s t +a - s - 1) ~ 0 •
1 1 1

This proves the result. D

The next theorem gives a result in case both Boffman's bound {2.1.4.ii)
and cvetkovic • s bound (2 .1.1) are tight.

2.1. 7. THEOREM. Let G be a strongly regular graph on n vertices. Let fn (G)

denote the multiplicity of the eigenvalue An(G). Then

iii. let G1 be a cocUque,, u>hose size attains both of these bounds,

then G2, the subgraph of G induced by the remaining vertices, is
strongly regular u>ith eigenvalues
A. CG 2) = A. (G)a(G) /(n-a(G))
1 1

~·Theorem 2.1.1 implies (i); (ii) is the same as 2.1.5.i. Let A and A
2
be the adjacency matrices of·G and G2 , respectively. Then

has· just two distinct eigenvalues, A2 (G) and An (G) of multiplicity n - fn (G)
and fn(G) = a(G), respectively. From 1.3.3 it follows that
20

has three distinct eigenvalues, A2 (G), A2 (G) + An(G) and A2 {G) + An(G) +
+ a(GHA 1 {G) -A. 2 CG)) In, where the last eigenvalue is simple (has multi-
plicity one). On the other hand, 2.1.S.iii gives that G is regular of
2
degree A (G) a (G) I (n - a (G)) • This shows that A2 and A 2 have a common basis
1
of eigenvectors, and that the simple eigenvalue of i belongs to the eigen-
2
vector j. Thus A2 has the desired eigenvalues, and therefore (see [C6] or
Appendix I) G
2 is strongly regular. 0

Using 1.1.3 it is not difficult to show that D(G,G 1 ) is a quasi-

symmetric 2-design (see Section 3.2), whose block graph is the complement
of G2 • This situation has been studied by SHRIKHANDE [SS].

In proving 2.1.2 we applied interlacing to the product of eigenvalues.

We did so in order to get reasonably nice formulas. However, for non-
regular graphs the inequality for the product carries less information than
the separate inequalities. For this reason, applying the Higman-Sims
technique directly to the adjacency matrix of a given non-regular graph,
may yield better results than 2.1.2 or 2.1.3. Also, if more is known about
the structure of G or G1 , it is often possible to get better results by a
more detailed application of the Higman-Sims technique. Let us illustrate
this by the following result.

2.1.8. THEOREM. Let G be a regular graph on n vertiaes of degree d, and let

the aomp'lete bipal'tite graph Kt,m be an induaed subgraph of G. Let x 1 and
x2' x
1 ~ x 2, be the zeros of

2
(n - t -m)x + (dt +dm - 2.tm)x - tm(n - 2d) •

Then

PROOF. Without loss of generality, let G have adjacency matrix

I:
J

A= 0

~21 A"
21
 21

where the diagonal block matrices are square of sizes R., m and n , respect-
2
ively. Usin9 the BiCJmaO-Sims technique (i.2.3) we find that the ei9envalues
of

I. _.
0 m d-m

B := 0 d-R.

R.--
d-R. R.d +md - 2R.m
m-- d -
n2 n2 n2

interlace the eigenvalues of A. Clearly 1 1 (B) = 1 1 (A) = d and hence

This yields x =1 2 (B) and x = 1 3 (B). Now the interlaci09 gives the
1 2
required result. D

BUMILLER [B9] showed for stron9ly regular G and m = 1 that

one easily checks that this follows from the second inequality of the
above theorem. PAYNE [P6] proved that

2
(.t - 1) (m - 1) :s;. s ,

if G is the point graph of a 9eneralized quadrangle of order (s,t) (see

Chapter 5 or Appendix I). This follows after substituting

n = (s+1)(st+1) ~ d = s(t+l) , 1 2 (G) = s-1

in the first ,inequality of the above theorem. It should not be surprisi09
that for this case we obtain the same result as Payne, because he too uses
the BiCJmaO-Sims technique.
22

2. 2. CHROMATIC NUMBER

In this section we shall derive lower bounds for the chromatic number
of a graph in terms of its eigenvalues. The main tool is Hoffman's generali-
zation of Aronszajn's inequalities (1.3.2).
Let G be a non~void graph on n vertices. Then it follows immediately
that y(G), the chromatic number of G, satisfies

y(G)a(G) ?: n •

combining this with the upper bounds for a(G) found in the previous section
we obtain lower bounds for y(G). For instance, 2.1.3 gives

However, this is not best possible, since HOFFMAN [H13] (see also [BS],
[H2], [H14], [L2]) showed that

which, if G is not regular, is better than the above bound. If G is regular,

then the two bounds coincide. Takinq into account that a(G) is an integer
we get
2
y(G) ?: nI l n A. 1 (G)A n (G) I CA. 1 (G)A n (G) -dmi n ) J,
which is occasionally better than Hoffman's bound.

HOFFMAN [H13] proves his lower bound by use of the inequalities 1.3.2.
We shall use the same technique, but in a more profound way, in order to
obtain a generalization of Hoffman's inequality.

2.2.1. THEOREM. Let G be a g:Paph on n vertiaea with ahromatia nwnber y. Let

k be an integer satisfying O !> k !> n I y. Phen

(y - 1) 1 k+l (G) ?: -\1-k (y-1) (G)

ii.

~· Let A be the adjacency matrix of G. Then without loss of generality /

 23

where'Aii is the nix ni all-zero matrix, for i • 1, ••• ,y.

First, we assume that n1 > k for i • 1, ••• ,y. Let u , ••• ,un denote an
1
orthonormal set of eigenvectors of A. Define

Clearly the matrices uiui* and A have a canmon basis of eigenvectors. This
implies

(*) 1, ••• ,n-k •

For 1 = 1, ••• ,y, let Aii be the submatrix of Acorresponding to Aii' Since
uiu~ is positive semi-definite of rank one, l uiui* is positive semi-
k

"' i•1
definite of rank k. This yields that -Aii is positive semi-definite of
rank at most k, hence

A
n.-
k (Al.l.
.. ) = 0 for i 1, ••• ,y •
l.

Now we apply the left hand inequality of 1.3.2 with ji = ni -k. This gives

With (*) this yields

Bence

(***)

Now suppose that ni s k for some i E {1, ••• ,y}. Let L c {1, ••• ,y} be such
that ni s k iff i E L. Let A' be the n' x n' submatrix of A, obtained by
discarding all block matrices Aij with i EL or j EL. Putt:= iLj. From
k < n/y it follows that t < y, hence n' > 0. Now (***) gives
 24

(y-.t-1)/.l+k(A') + An'+k.t-k(y-l)(A');:: O •

Using n' +k.t > n and the Cauchy inequalities (1.2.2) we have

"n'+kt-k(y-1) (A') ~ "n-k(y-1} (A') ~ "n-k(y-1) (A) '

Hence

From k < n/y it follows that /.l+k (A) ;:: "n-k(y-1) (A}, hence "i+k (A) <:: o.
Thus

This proves (i).The proof of (ii) proceeds analogouslyz'but also follows

I
from the above by replacing A by - A. 0

We see that the second inequality of the above theorem for k = 0 is

Hoffman's bound. In Chapter 4 we shall need a sharpened version of this
inequality (see [H13]):

y-2
- l An-i(G) <:: t. 1 (G) ,
i=O

which is in fact just formula (**) in the above proof with k = O, and A
equal to minus the adjacency matrix of G.
If k > O, the above inequalities are not really bounds for y(G),
since y(G) also occurs in an index. However, this does not matter much if
we use these inequalities for estimating the chromatic number of a given
graph. It is also not difficult to derive proper bounds from these in-
equalities. The next results illustrate this.

2.2.2. COROLLARY• Let fn(G) denote the muitipLicity of the eigenvaiue

An (G). Then

y(G) <: min{l +f (G), 1 ->. (G) I t. CG)}

n n 2

PROOF. Suppose y
--
= y(G) ~ f (G).Then A (G) =A
n n n-y
+l(G). Now 2.2.1.i with
k = l gives (y - 1) t. 2 (G) <:: - An (G). This proves the required result. 0
 25

For strongly regular graphs the above results lead to the following
theorem.

2.2.3. THEOREM. Let G be a st1.'ongl-y Ngul-a1.' g1.'aph on n vel'tiaes. Suppose G

is not the pentagon 01.' a aompl.ete Y-pal'tite gz>aph. Then
y(G) 2!: max{1-A1(G) /An(G), 1-An(G) /A2(G)} •

PROOF. Due to the above results, it suffices to prove the following claim:

To achieve this, we distinguish three cases.

a) n s 28. For this case it is easily checked that all feasible parameter
sets for strongly regular qraphs which violate our claim are those of the
pentaqon and the complete y-partite qraphs.

b) :1. (G) < 2.. In this case A (G) = 1, or G is a conference graph (see
2 2
[C9] or Appendix I). I f G is a conference qraph, then A (G)
2 = - l:i + l:i Iii.,
and hence n < 25 and we are in case 1. Strongly regular graphs with
A2 (G) = 1 were detexmined by SEIDEL [s3]. They satisfy n s 28 or G is a
ladder {disjoint union of edges), the complement of a lattice (line qraph of
a Km,m ) or the complement of a trianqular graph (line graph of . a Km) • One
easily verifies that these three families of graphs satisfy our claim.

c) ~
2 {G) 2!: 2 and n > 28. If G is imprimitive (G is complete y-partite or
the disjoint union of complete qraphs), the result is obvious. So assume G

is primitive. Suppose the claim does not hold. Using l 2 (G) 2!: 2, A1 (G) < n
and

fn {G) An (G) + (n - 1 - fn (G)) A. 2 (G) + Al (G) = 0

we obtain

f! (G) < - fn (G) ln (G) I A2 (G)

This yields

f!(G) + 3fn(G) < { n + 2~ •

26

For primitive G the absolute bound (see [D2], [54]) reads

Hence' ~n < 2'/fn, i.e. n < 24. This contradicts our assumption, and there-
fore the theorem is proved. O

2.2.4. EXAMPLE. Let G be the 5chllifli graph, which is drawn in Figure 1;

two black or two white vertices are adjacent iff they are on one line, a
black vertex is adjacent to a white one iff they are not on a line, (see
[53], [H2]). Then G is strongly regular, n = 27 and

-2,

···="2.iCG) - 5,

where G denotes the complement of G. From Figure 1 we see that

a{G) 2 3 and a(G) 2 6 •

The thin vertical lines partition G into six cliques, hence y{G) ::;; 6. The
numbering gives a colouring of G with nine colours, so y(G) ::;; 9. Using our
bounds it follows that equality holds in all these inequalities. Indeed,
by 2.1.4.ii or 2.1.5.i we have a(G) ::;; 3; 2.1.1 yields a(G) ::;; 6; y(G) 2 9
follows from Hoffman's bound, and y(G) 2 6 follows from our last theorem.

FIGURE I

The chromatic number of strongly regular graphs will be the subject of

Chapter 4.
 27

CHAPTER 3

INEQUALITIES FOR DESIGNS

3.1. SUBDESIGNS

In this section we shall derive inequalities for subdesigns of de-

signs.
Let D be a design with incidence matrix N. It is clear that we cannot
apply the Higman-Sims technique (1.2.3) to N, because N does not have to
be symmetric. Instead, we apply the Higman-Sims technique to

A -[:. :] •

By definition the positive eigenvalues of A are the singular values of N.

Let o ~ o ~ ••• > 0 denote these singular values. Then we can state the
1 2
main result of this section.

3 .1.1. THEOREM. Let D be a 1 - (v ,k,r) design with b btocks. Let o1 be a

possibty degeneztate 1- Cv 1 ,k 1 ,r 1 ) subdesign of D with b 1 btoaks. Then

i.

ii. if equa.Uty hoUa~ then each point [btoak] off o1 is incident

'With a constant nunUJel' of btocks [points] of o1•

PROOF. Let N and

1

be the incidence matrices of o and D, respectively. Put

1
28

0 0 N1 N2 0 0 rl r-r 1

0 0 N3 N4 0 0 x r-x
A I'" and B :=
N1* N3* 0 0 k k-k1 0 0

N2* N4* 0 0 y k-y 0 0

where

x := b (k-k ) I (v-v 1 )
1 1 and y :'" v 1 Cr-r 1) /(b-b 1) .
Then the entries of B are the average row sums of the block matrices of A.
By 1.1.2 we know

for i = 1, ••• ,b+v, j c 1, ••• ,4. We easily have

From det B = rk (r 1 - x) Ck 1 -y) it now follows that

Now 1.2.3.i gives

With b 1k = v r 1 this yields (i).

1 1
If equality holds, then the interlacing must be tight. Thus 1.2.3.iv gives
(ii). D

From the above proof it is clear that the result also holds if o 1 is
not a 1-design, but then we have to take r 1 and k 1 to be the average row
and column sums of N •
1
For many 1-designs cr is expressible in terms of the parameters of
2
the design. For instance, cr~ = r -.>. if D is a 2 - (v,k,>.) design, and
cr; = s + t - a+ 1 if D is a partial geometry with parameters (s,t,a) (see
Appendix I, or [C6], [T1]).

We shall make explicit two consequences of the above theorem.

 29

3.1.2. COROLLARY. If a symmetric 2 - (v,k,>.) design aontains a symmetric

2 - Cv 1 ,k 1 ,>. 1 > eubdesign, possibly degenerate, then

2
~· Substitute b = v, k = r, b1 = v 1, k
1 = r 1 and a 2 = k ->. in
3.1.1.i. D

3.1.3. COROLLARY. Let X and Y be a set of points a:nd a set of lines,

respeatively, of a partial geometry with paramete:r>s (s,t,a), such that no
point of X is incident with a Une of Y. Then

(ajxj + (s+t+l-a)(s+l)}(alYI + (s+t+1-a)(t+1)) :s:

~·Substitute k 1 = r = O, b = jY!, v == jxj, k = s+l, r = t+l,

1 1 1
v = (s+l)(st+a) /a, b = (t+l)(st+a) /a and cr~ = s+t+l-a in
3.1.1.i. D

corollary 3 .1. 2 appeared in [H4]. A Baer subplane of a projective

plane (see [03]) satisfies 3.1.2 with equality. Other examples which meet
this bound (hence where 3.1.1.ii applies) can be found in [I:I4].
The bound of 3. 1. 3 can also be tight. For instance, let Q be the
partial geometry with parameters (2,4,1) (generalized quadrangle), whose
points and lines are the vertices and the triangles of the complement of
the Schlafli graph (see Example 2.2.4). There are 15 triangles which do
not have a vertex in common with a double six (the black vertices in
Figure 1) , Thus we have an empty subgeometry (no point and line are in-
cident) of Q having 12 points and 15 lines. This satisfies 3.1.3 with
equality.
If o1 is an empty design (kl= r 1 = 0), then one easily finds examples
which meet the bound of 3.1.1.i. For instance: a projective plane with a
maximal arc (see [03]); a symmetric 2-design containing an oval without
tangent blocks (see [A4]); a 2-design having a block repeated b/v times
(see [Ll]; here the inequality 3.1.1.i is Mann's inequality [M2]).
30

Although the results of this section are similar to those of Section

2.1, we did not start with a general inequality for substructures of an
incidence structure like we did for subqraphs of a graph in 2.1.2. This has
two reasons. Firstly, the formula for an arbitrary incidence structure is
more complicated than 2.1.2. The second reason is that there does not seem
to be much interest in incidence structures without any additional proper-
ties; this is certainly not true for graphs. Yet we shall give one result
for an arbitrary incidence structure, namely an inequality for the sizes of
an empty substructure.

3.1.4. THEOREM. Let D be an incidence structure with v points and b bioaks.

Let every point [block] be incident with at least r min
. blocks [k i points].
mn
Let X and Y be a set of points and a set of blocks, Pespectively, suah that
no point of X is incident with a block of Y. Then
2 2 2 2
r. k i
min m n
lxllYI s a1 a2 Cv-!Xl)(b-!YI),

wher-e a 1 and a2 denote the two lar-gest singular' values of the incidence
matr'i:t Of D.

PROOF. Let the incidence matrix of D be

where O denotes the lxl x IYI all-zero matrix. Let ri and. k 1 be the
* respectively, for i = 2,3,4. Then by
average row sums of Ni and Ni,
1.2.3.i the eigenvalues of

0 0 0 r2
0 0 r3· r4
B :=
0 k3 0 0

k2 k4 0 0

interlace the eigenvalues of

A :=
~* :] .
 31

Now with 1.1. 2 it follows that

on the other hand we have

3.2. INTERSECTION NUMBERS

If two distinct blocks of a design D have exactly p points in cOlllllon,

then p is called an intersection n1.DT/ber of D. It is obvious that an inter-
section number p of a 1 - (v,k,r} design satisfies

k ~ p ~ max{0,2k-v} •

The next result, which is due to AGRAWAL [Al], gives non-trivial bounds
for the intersection numbers of a 1-design. Like in the previous section,
the singular values of the incidence matrix of a design D will be denoted
by 01 ~ 02 ~ ••• > o.

3.2.1. THEOREM. Let D be a 1 - (v,k,r} design unth b bl.oake. Let B

1
and B 2
be distinat blocks of D. Then
2
rk - cr2 - k + ..,2 •
i. IBl n B2 IS 2 b v
2 ,

if equality ho'lds then IB1 n B3 I + IB2 n B3 I= 2 (rk - a;) Ib for

any further b l.oak B3 ,

ii. IB 1 n B2 1 ~ k - c~ ;

if equality ho'lds then IB

1
n B3 1 = IB2 n B3 1 for any further
bl.ook B3 •

PROOF. The result is obvious if b s 3, so assume,b ~ 4. Let N be the in-

cidence matrix of D, such that the first two columns correspond to the
blocks B1 and B2 • Define
32

*
A := N N

Then the off-diagonal entries of A are the intersection numbers of D, and

the row and column sums of A equal kr. Put p := ls 1 n s2 1and consider the
follo~ing partitioning of A:

A=

Define

kr-k-pl •
kr-x

Then the entries of B are the average row sums of the block matrices of A.
Clearly

rk and 1' (B) = k + p - x •

2

By 1.2.3.i we have

Hence

cr;(b-2)::; (k+p)(b-2) - 2(kr-k-p).

This yields

If equality holds, the interlacing is tight and 1.2.3.i gives that every
column sum of A equals x. This proves (i}.
12
To prove (ii} we apply 1.2.l·to A with

1 0
o]* [2 ol-'2
0 1 1 0 n-2

l
Then

k-p 0
B := S *AS
[ 0 kr-x
 33

It is easily seen that kr-x ~ k -p if b ~ 4. So A. (B} • k-p. Hence, by

2
1.2.1.i

Here equality does not have to imply that the interlacing is tight. There-
2
fore we shall use 1.2.1.ii. If a 2 • k -p = kr-x, then r = 1, p = 0 or
r = 2, p = O, b = 4, and the result is easily checked to be true. If
a;= k-p < kr-x, then 1.2.1.ii implies that S{l,O)* • (-/2,/2,0, ..• ,0}*
is an eigenvector of A for the eigenvalue k-p. Thus A* C-1,1) * = O.
12
This proves (ii). D

It is straightforward to verify that equality in (i) or (ii) for a

pair of blocks of D implies also equality for the corresponding blocks of
the complement of D.
Although Agrawal's proof of the above theorem is different fran ours,
it also uses eigenvalue techniques (in essence the Cauchy inequalities
1. 2. 2) • MAJtJMDAR [M1.] gives a proof of this theorem for the case that D is
a 2-design, using counting arguments. See also BUSH [BlO] and CONNOR [C8]
for similar results.
It is clear that our method also leads to inequalities if we consider
the intersection pattern of more than two blocks.

3.2.2. THEOREM. Let D be a 1 - (v,k,r} design with b bl.ocks. Let Y be a set

of bl.ocke which rrrutuatZ.y have p points in common. Then
i. IYI (bp - rkl ~ b (p - k} ,

2 2
ii. IYI (bp -rk+a
2) ~ b(a
2 -k+p)

~· Let N be the incidence matrix of D. We apply the Higman-Sims

technique to A :• N*N, partitioned according to Y and the other blocks of
D. Put

x := IYI k er - u - P cIYI - 1l
b - IYI
Then

rk - k + P - P IYI]

rk-x
34

carries the average row sums of the block matrices of A. Clearly

A. (B) =k-p+plYl-x= (b(k-p)+(bp-kr)IYJ)/(b-IYI).

2

From 1.2.3.i we have

This lower and upper bound for A. 2 (B) yields (i) and (ii), respectively. D

We define two blocks B 1 and B2 of a 1 - (v,k,r) design to be equivalent

if

Then from 3.2.1.ii it is clear that this indeed defines an equivalence

relation, and that the number of colll1ll0n points of two blocks only depends
on the equivalence classes of these blocks. By the use of 3.2.2 we find
bounds for the· size of the equivalence classes.

3.2.3. THEOREM. Let D be a 1 - (v,k,r) desi(Jn u>ith b bZocks. Let Y be an

equivalence class of bZoaks. Then

i. k and k - o~ cannot both be an inteI'section numbeI' of D,,

ii. if k - o~ is an inte:t'section number of D,, then

2
bo
2___
IY ls ---,,,_...;;. ,,
2
bo 2 - bk+ rk

iii. if k is an intersection number of D, then

bo 2
2
IYl s 2 •
bk - rk + o ·
2

~· Assume 2r s b; we may do so because of the remark right after

2 2
Theorem 3 • 2. 1. Suppose k - o 2 is an inter section number. Then k - o 2 ~ 0,
hence o;-k+2(rk-o~) /b s 2k(r-1) /b < k. So 3.2.1.i yields that k can-
not be an intersection number.
 35

Formulas (ii) and (iii) follow immediately from (i) and (ii) of 3.2.2 by
substitu.tion of p = k -a 22 and p = k, respectively. D

2
Suppose D is a 2 - (v,k,A) design, so a
2
= r -A = (bk -rk) / (v -1).
From 3.2.3.iii it follows that D has at most b/v repeated blocks; this is
the inequality of MANN [M2] (see also [Li]). If D has an intersection
2
number k-a 2 = k-r+A, then 3.2.3.ii implies that the size of any equi-
valence class is at most b/ (b -v + 1); this bound appeared in [B2]. This
paper also contains the next result (see also [B1]).

A 2-design with just two distinct intersection numbers is called

quasi-sy1m1etric. Consider the graph G, whose vertices are the blocks of a
quasi-symmetric 2-design D, two vertices being adjacent if the number of
points which the corresponding blocks have in common equals the larger
intersection number. We call G the b'lock graph of D. GOE'1'BALS & SEIDEL [G2]
(see also [C6]) proved that the block graph of a quasi-symmetric 2-design
is strongly regular.
Now suppose D is a 2 - (v,k,A) design with just three distinct intersection
numbers k - r + A, p 1 and p 2 (p > p ). We have already observed that the
1 2
number of points which two blocks have in common only depends on the equi-
valence classes of these blocks. For this reason the following definition
is legitimate. The class graph of D is the graph whose vertices are the
equivalence classes, two vertices being adjacent if two blocks representing
the corresponding classes have p 1 points in common.

3.2.4. THEOREM. Let D be a 2 - (v,k,A) d.esi(ln r..Jith just three intersection

numbers, k-r+A, p 1 and p 2• Then the class graph of Dis a strongly
Ngu"Lar graph on

b (k - r + A - p ) (k - r + A - P 2 >
1
n := _.,,.2---------.,,,2---...-------
Ak - k(r -A) + (r - A) + bp p - Av(pl + P2)
1 2

vertices, unth eigenvalues

36

PROOF. Let N be the incidence matrix of D. Define A := N*N. Then

(*) A2 = N* (NN* )N = N* (),J + (r -A)I)N = Ak 2J + (r -A)A •

Put Po := k -r +A. Let xj (j = O, 1, 2) denote the number of times that p j

occurs in the i-th row of A, for some i E {1, ••• ,b}. Then

(AJ) i1

on using (*) • Substitute x 2 = b - 1 - x 0 - x 1 and subtract the first equation

multiplied by (p 1 + p > from the second one. This yields
2

x (k - r + A - p 1) (k - r + A - p 2 >
0
=
= (b-1)p p - k(r-l)(pl +p 2) + k(r-A} + k 2 (>.-l) •
1 2

Hence x 0 does not depend on i and therefore all equivalence classes have
size x 0 + 1. Now n = b/ (x + 1) yields the given formula for n.
0
Now we partition A according to the equivalence classes. Let A denote the
adjacency matrix of G. Then the definition of G yields that the entries of

are the row sums (which are constant) of the block matrices of A. Since A
has three distinct eigenvalues, rk, r -A and O, it follows from 1.2.3.iii
that each eigenvalue of B is equal to rk, r -A or 0. We easily check that
rk is a simple eigenvalue of B, belonging to the all-one vector j. Now from
(**) the eigenvalues of Afollow. Hence Ahas an eigenvector j and just two
distinct eigenvalues not belonging to j. This implies (see Appendix I or
[C6]) that G is strongly regular. D

Examples of designs which satisfy the hypothesis of the above theorem

can be found in [Bl], [B3] or [MS]. For all these examples the class graph
is a complete multipartite graph. In Section 6.1 we shall give an example
for the above theorem where the class graph is primitive (not complete
multipartite or the complement). For other results on 2-designs with an
intersection number k - r +A, see [B2].
 37

CHAPTER 4

4-COLOURABLE STRONGLY REGULAR GRAPHS

4.1. INTRODUCTION

In this chapter we shall illustrate the use of the results and

techniques obtained in the previous chapters. The result will be the de-
termination of all 4-colourable strongly regular graphs.
It is obvious that a regular complete y-partite graph, and a disjoint
union of complete graphs on y vertices are strongly regular graphs with
chromatic number y. Strongly regular graphs, not belonging to one of these
two families, are called primitive.
1 2
Let G be a strongly regular graph with parameters (n,d,p11 ,p 11 >. Then
(see [CS], [c9] or Appendix I)

d = >. 1 {G) ,

Moreover, G has at most three distinct eigenvalues:

-1 <:: ).f+ (G) ••• = An (G) ,

2

where f = f 2 (G), the multiplicity of >. 2 (G), satisfies

If G is primitive, then p 2 > O, >. (G) > O, and >.n(G) < -1.
2
11

4.1.1. LEMMA. If G is a primitive strongty regutar gru:ph, not the pentagon,

then
i. d s·->.n(G) (y(G) -1) ,

iii. >.. (G) <y(G)-1.

2
38

PROOF. (i) and (ii) are quoted from 2.2.3. Since G is primitive,
0 < p~ 1 = d - A (G)An(G). Hence (i) gives y(G) -1
2 ~ ~d/Xn(G) > X2 (G). 0

~s a direct consequence of this lemma we have the following theorem.

4.1.2. THEOREM. Given y e :N, the numbe:r> of p:r>i.mitive strongiy :r>eguZ.a:r>

g't'aphs with chr>omatic number y is finite.

2
~· If the graph G is primitive, then p <?: 1 and hence by use of the
11
formulas above

By Lemma 4.1.1 we have

This completes the proof. 0

Now let us examine the case y(G) s 4.

4.1.3. LEMMA. Let G be a 4-coZ.Our>abie st:r>ongZ.y regul.ar !J:l'aph. Suppose G has

a non-integmi eigenvaZ.ue. Then G is the pentagon.

~· Since G has a non-integral eigenvalue, we have (see [C9] or

Appendix I)

Al(G) = ;cn-1), A2(G) =- ; + ;rn, A (G) • - ;

n
- Ii in,

n E 1 (mod 4) , in f. :fi •

By 2.1.5 we have a(G) s In, hence

4 ~ y(G) ~ n/a(G) ';;: n/ LIn J ,

therefore n = 16 or n s 12. Combining the restrictions for n we have n = 5,
hence G is the pentagon. 0

4.1.4. LEMMA. A 4-cotour>abZ.e p:r>i.mitive strongZ.y rieguZ.a:r> gmph has one of

the foUowing parameter sets:
 39

i. (5,2,0,1) , vii. .. (16,9,4,6) ,

ii. (9,4,1,2) , viii. (40,12,2,4),

iii. (10,3,0,1), ix (50,7,0,1),

iv. (15,6, 1,3» x. (56,10,0,2),

v. (16,5,0,2), xi. (64,18,2,6),

Vi (16,6,2,2), xii. (77,16,0,4).

PROOF. Let G be such a graph. Suppose G is not the pentagon (which has
parameter set (i)). Then by 4.1.3, the eigenvalues of G are integers. The
primitivity of G yields A (G) > O, An(G) < -1. Now 4.1.1.iii gives
2

Suppose A (G)
2
= 1. Then by 4.1.1

Straightforward computations give that the only feasible parameter sets

satisfying these conditions are (ii) - (v), (vii) and (10,6,3,4). However,
a graph G with this last parameter set satisfies a(G) s 2, therefore
y(G) ~ 5. Suppose A2 = 2. Then 4.1.1 implies

An(G) E {-2,-3,-4,-5,-6} , d s 18 •

With a little more work than for the previous case, this leads to the
feasible parameter sets (vi), (viii) - (xii) and (57,14,1,4). However,
WILBlUNK & BROUWER [W4] proved the nonexistence of a strongly regular graph
with this last parameter set. 0

For graphs with parameters (i) - (v) existence and uniqueness is known
(see [83]). Cases (i), (ii) and (iii) are the pentagon, the line graph of
K (also ca:J.].ed the lattice graph L2 (3)), and the Petersen graph,
313
respectively. It is easily seen that these three graphs have chromatic
number three. From 4 .1.1 it is clear that none of the other graphs is
3-colourable. case (iv) is the complement of the line graph of K6 (also
called the complement of the triangular graph '1'(6)), which is easily seen
to be 4-colourable. Case (v) is the Clebsch graph (see [S3]). This graph
40

is given in Figure 2, where two black or two white vertices are adjacent
iff they are not on one line, whilst a black vertex is adjacent to a white
one iff they are on one line. We almost immediately see that this graph is

FIGURE 2

4-colourable. There are precisely two nonisomorphic strongly regular

graphs with parameter set (vi) (see [S6]): the line graph of K (L 2 (4))
414
and the Shrikhande graph (see cover), both graphs are easily seen to be
4-colourable. Case (vii) is the complementary parameter set of (vi). We
quickly see that the complement of the line graph of K is 4-colourable,
414
however, the complement of the Shrikhande graph is not 4-colourable.
Imeed, the size of the largest coclique equals three. The remaining cases
are more difficult. They will be treated in the next section.

4.2. STRONGLY .REGULAR GRAPHS ON 40, 50, 56, 64 AND 77 VERTICES

In this section we shall study the feasible parameter sets for 4-

colourable strongly regular graphs, which remain from the previous section.

The first case is the parameter set (40,12,2,4). Although several

strongly regular graphs with these parameters are known (see Section 6.2),
it will turn out that no such graph has chromatic number four. To prove
this we use the following lemma.

4.2.1. LEMMA. There is no reguZar bipartite graph on 20 vertices 1iJith

eigenvaZues 4, 2, o, -2, -4 of muitipUcity 1, 6, 6, 6, 1,, respectiveZy.
 41

PROOF. Suppose G were such a bipartite graph. Let

be the adjacency matrix of G. Then N is the incidence matrix of a

1 - (10,4,4) design, D say, with singular values a 1 = 4, o = ••• = a 2.
2 7
Let s 1 and B2 be two distinct blocks of D. Then 3.2.1.i yields

Suppose s 1 and s 2 are disjoint. Let x and y be the two points of D which
are not incident with s 1 and s 2 • Let s 3 be a block through x. Using
3.2.1.ii it follows that

so s 3 is incident with y. Bence, any block incident with x is also in-

cident with y. However, this is not possible, since two points of D have
at most two blocks in common, as follows from 3.2.1.i applied to the dual
of D. So we have

This implies that B : = N*N - J - 3I is the adjacency matrix of a (strongly)

regular graph with eigenvalues -3, 1 and 3 of multiplicity 3, 6 and 1,
respectively. This is impossible, since

2 2 2 2
30 =trace B ~ 3.(-3) + 6.1 + 1.3 • D

4.2.2. THEOREM. Then e:x:ists no 4-coZ.OUPab"le st:vong"ly ngula:v gruph UJith

pa:l'<:1111ete:vs (40,12,2,4).

~· Let G be a strongly regular graph with parameter set (40,12,2,4).

The eigenvalues of A, the adjacency matrix of G, are 12, 2 and -4 of multi-
plicity 1, 24 and 15, respectively. Suppose G is 4-colourable. Then without
loss of generality
42

0 A12 A13 A14

A21 0 A23 A24

A=
A31 A32 0 A34

A41 A42 A43 a

By 2.1.5.i all block matrices are square of size 10, and by 2.1.5.iii all
row and column sums of Aij are equal to 4, for i,j = 1,2,3,4, i Y. j. Define

for i = 1,2. Let i e: {1,2}. Let Gi be the graph with adjacency matrix Ai.
Now A has just two distinct eigenvalues 2 and -4 of multiplicity 25 and 15,
respectively. Furthermore, Ai and Ai have the same eigenvalues, except for
~e one belonging to the eigenvector j, which equals 4 for Ai and -1 for
Ai. Since Gi is bipartite, Ai also has an eigenvalue -4. Now by the Cauchy
...,
inequalities (1.2.2) it follows that Ai' and hence also Ai, has at least
five times the eigenvalue 2. But Gi is bipartite, therefore Ai' and hence
also Ai' has at least five times the eigenvalue -2. Now from 1.3.3 it
follows that A3_i, and hence also Ai and Ai' has at least five times the
eigenvalue O. Since Gi is bipartite on an even number of vertices, the
multiplicity of the eigenvalue O is even, so at least six. Going backwards
through the above reasoning we conclude that the multiplicities of the
eigenvalues 2 and -2 of Ai are also at least six. Thus Ai has eigenvalues
4, 2, o, -2, -4 of multiplicity 1, 6, 6, 6, 1, respectively. Now Lemma
4.2.1 finishes the proof. D

HOFFMAN & SINGLETON [H15] showed the existence and uniqueness of a

strongly regular graph with ~arameters (50,7,0,1). So we only have to
determine whether this graph is 4-colourable or not. To do so, we shall
use a description of the Hoffman-Singleton graph (this description seems
to be folklore, since it is well known; however, I could not find a refer-
ence) based on the following result, see [Bll] or [C10].

4.2.3. RESULT. The thirtyfive "lines of PG(3,2) can be repzoesented by the

thirtyfive tripLes of a set IJJith seven eLements~ suah that two Zines
 43

intersect iff the corresponding triples 'have one element in common.

REMARK. This result is directly related to the isomorphism of the groups

PSL(4,2) and the alternating group on eight symbols (see for instance [B4]
or [C10]).

Now we construct the Hoffman-Singleton graph as follows. The vertices

are the fifteen points and the thirtyfive lines of PG(3,2). Points are
mutually non-adjacent; a point is adjacent to a line iff the point is on
that line; two lines are adjacent iff the triples, which correspond with
these lines according to the above result, are disjoint. It is an easy
exercise to check that this construction indeed gives the desired strongly
regular graph.

4.2.4. THEOREM. The Hoffma,n-Singleton graph 'has chromatic number four.

PROOF. Colour the fifteen points red. Fix two elements x and y of the 7-set
of Result 4.2.3. Colour lines blue, if they correspond to a triple contain-
ing x. Of the remaining lines, colour those yellow, whose corresponding
triple contains y, and colour the other ones green. From our definition it
is obvious that this is a correct colouring of the Hoffman-Singleton
graph. D

GEWIRTZ [Gl] showed existence and uniqueness of a strongly regular

graph with parameters (56,10,0,2). Before giving a description of the
Gewirtz graph we first prove the following.

4.2.5. PROPOSITION. If the GeUJirtz graph 'has two disjoint cocUques of size
16~ then its chromatic number equals four.

PROOF. Assume that the Gewirtz graph has adjacency matrix

where A = A* is square of size 16. We know that A has three distinct

12 21
eigenvalues, 10, 2 and -4. From 2.1.5 it follows that A12 , A21 , A31 and A32
 44

have constant row sums equal to 4. Therefore the graph G3 , whose adjacency
matrix is A , is a disjoint union of cycles. Suppose one of these cycles
33
has c vertices. Partition A33 according to the vertices of this cycle, and
. the remaining ones. This induces a partition of A into sixteen block
matrices, and the entries of the matrix

0 4 lie 6-l.,ic

4 0 lie 6-l.,ic
B :=
4 4 2 0

4 4 0 2

are the average row sums of these block matrices. We immediately see that
the eigenvalues of B are

However, we know that

Hence the eigenvalues of B interlace the eigenvalues of A tightly, thus by

1.2.3.ii all block matrices have constant row sums. Therefore, lie is an
integer. This proves that each component of G3 is a cycle of even length.
Thus G3 is bipartite, and therefore the whole graph is 4-colotirable. 0

We use the description of the Gewirtz graph given in [G2], where this
graph is obtained as the complement of the block graph of a quasi-symmetric
2 - (21,6,4) design with intersection numbers O and 2 (see Section 3. 2).

4.2.6. TBEORF.M. The Gewirtz graph has chromatic number four.

~· Let D be the quasi-sYl!)llletric 2 - (21,6,4) design. It is clear that

all blocks through a fixed point of D yield a coclique in our graph of size
16. To see that there is another coclique of the same size, disjoint from
this one, we proceed as follows. D can be obtained from. a 3 - ( 22, 6, 1) de-
sign o (the extension of PG(2,4), see for instance [C6]) by deleting one
point and all blocks through that point (i.e. Dis a residual design of D).
Take a block B of o, which is not a block of D. An elementary counting
argument (see [C2]) shows that there are 16 blocks of D which are disjoint
 45.

from B, and which mutually have 2 points in common. Hence these 16 blocks
provide a coclique of size 16 in the Gewirtz graph, which, if B has been
chosen appropiately, is disjoint from our previous coclique. Application of
4.2.5 completes the proof. D

Three 2 - (16,6,2) designs on a common point set are called Urik.ed if

any two blocks from distinct designs have 3 or 1 points in common (see [Cl]
or [M7]). Let N1 , N2 and N be incidence matrices of three linked
3
2- (16,6,2) designs. Then we know that

for i = 1,2,3. Moreover, for i,j = 1,2,3, i ~ j, the matrix

is a (0,1) matrix by definition. In fact, N.. is the incidence matrix of a

l.J
2 - (16,6, 2) design, since

2J + 4I I

by use of the above formulas. Define

for i 0,1,2,3, j = 1,2,3. Then

for i,j,k = 0,1,2,3, as follows readily from the formulas above. This
implies that

0 NOl N02 N03

NlO 0 N12 N13

A :=
N20 N21 0 N23

N30 N31 N32 0

is a symmetric matrix, which satisfies

2
A = 18I + 2A + 6 (J - I -A) •
46

Hence A is the adjacency matrix of a strongly regular graph with parameters

(64,18,2,6), which is 4-colourable. We call this graph the incidenae {fl'aPh
of the three linked designs.

4. 2. 7. THEOREM. Let .G be a strong'ly regu'la:r> g:r>aph with par>ameters

(64,18,2,6) and ahromatic number four. Then G is the inaidence graph of
three 'linked 2 - (16,6,2) designs.

~· Suppose G has adjacency matrix

0 AOl A02 A03

A10 0 A12 A13

A=
A20 A21 0 A23

A30 A31 A32 0

A has eigenvalues 18, 2, and -6 of multiplicity 1, 45 and 18, respectively.

By 2.1.5, Aij is square of size 16, andall row and column sums of Aij are
equal to 6, for i,j = 0,1,2,3, i ~ j. This implies that A and

have a common basis of eigenvectors. Using this, we obtain that

A:= A - 2I + l:!K

has eigenvalues 24, 0 and -8 of multiplicity 1, 48 and 15, respectively.

Thus
-
rank A= 16 •

For i = 0,1,2,3, put Aii := l:iJ 16 - 2I 16 • Then

-1
From rank A00 =rank A, it follows that AiO A00 A0 j = Aij' for i,j = 1,2,3,
on applying 1.1.3. By use of

A~~ = {l:iJ - 2I) -l .. (1/24)J - l:!I and AijJ = 6J

 47

this implies that for i,j • 1,2,3

This ~pletes the proof. Indeed, if i • j, then (*) implies

* • 2J +4I, showing that A01 is the incidence matrix of a 2 - (16,6,2)
A0iAOi
design, and if i ~ j, then (*) reflects that the designs represented by
A01 , A02 and A03 are linked, and that Aij is of the desired form for
i,j = 1,2,3. D

MATBON [M7] proved that there are exactly twelve non-isomorphic

triples of linked 2 - (16,6,2) designs, which lead to eleven non-isomorphic
incidence graphs. Bence there are precisely eleven non-isomorphic 4-colour-
able strongly regular graphs with parameters (64,18,2,6). It is fairly easy
to show that one of these graphs is the point graph of the known generaliz-
ed quadrangle of order (3,5) (see Chapter 5; a construction is described in
6.2.3). For completeness we list in Appendix II the systems of three linked
2 - (16,6,2) designs, which provide the ten remaining graphs; these systems
are taken from Mathon's paper. It is not known whether there are any
further strongly regular graphs with these parameters, which are not 4-
colourable.

Finally, the next theorem deals with the last set of parameters of
Lemma 4.1.4.

4.2.8. THEOREM. There e:x;ists no 4-co'Lourab'le stPcmg'ly Ngu'la:l' gruph UJith

paX'a17letePs (77,16,0,4).

PROOF. Let G be a strongly regular graph with parameters (77,16,0,4). Then

G has eigenvalues 16, 2 and -6 of multiplicity 1, 55 and 21, respectively.
Suppose G is 4-colourable, and let c be the size of the largest colour
class. Then c <!: f 77/4 l = 20. ·Let

where 0 is square of size c, be the adjacency matrix of G. Then G2 , the

graph with adjacency matrix A22 , is 3-colourable. From 2.1. 7 it follows that
c s a(G) s 21, and that c = a(G) = 21 implies that G2 is the Gewirtz graph.
48

Since the Gewirtz graph has chromatic number four, G cannot be coloured
with four colours if c = 21 {this shows that G is 5-colourable if a (G) = 21,
which is the case for the known strongly regular graph with these para-
meters; see [G2]). Suppose c = 20. Now we apply 1.3.3 to A - {2/11)J, so as
to obtain that

2
-6, "56 <A22 - IT J) -4.

Now 1.3.1 gives

On the other hand, the average row sum of A equals 592/57. Hence by
22
1.2.3.i (take m = 1) we have

Now the sharpened version of Hoffman's inequality for the chromatic number
(see Section 2.2), applied to the 3-colourable graph G gives
2

This is a contradiction, proving the theorem. 0

4. 3. RECAPITULATION,

All cases of Lemma 4. 1. 4 have been treated now. The only thing left is
to state the main theorem.

4.3.1. THEOREM. If G is a 4-coZ.ou:rubZ.e strongZ.y l'egular groph, then on6l of

the foZZouYing hoZ.ds:
i. y(G) = 2, and G is a regul,,ar> aompZ.ete bipartite groph, or a dis-
joint union of edges;
ii. y(G) = 3, a:nd G is a r-eguZar> compZ.ete 3-paritite groph, a dis-
joint union of t'Pia:ngZ.es, the pentagon, the Une (J'Paph of K ,
313
or the Petersen groph;
iii. y{G) = 4, and G is a reguZ.a:l' compZ.ete 4-partite groph, a dis-
joint union of K4 's, the compZ.ement of the Z.ine g-paph of K6 , the
Une groph of K 414, or its compZement, the Shrikhande g-paph, the
 49

Ctebsch g:ra:ph, the Hoffman-Singteton g:r>a::ph, the Gewil'ts g:r>a::ph,

o:r> one of the eteven incidence grapha of three Unked 2 - (16,6,2)
designs.
50

CHAPTER 5

GENERALIZED POLYGONS

5. 1. INTRODUCTION

A gene:.ro.'lized n-gon of order (s,t), s > O, t > O, is a 1 - (v,s+l,t+l)

design whose incidence graph has girth 2n and diameter n (see (T4], (D3],
(Fl], [H12] or Appendix I). A generalized polygon is a generalized n-gon
for some n.
Generalized n-gons were introduced by TITS [T4]. An important result
is the theorem of FEIT & HIGMAN [Fl] (see also (B12], [K2]), which states
that a generalized n-gon of order (s,t) is an ordinary n-gon (s = t = 1) or
n e {2,3,4,6,8,12}.
For a generalized polygon we speak of lines rather than blocks. We
shall often omit the adjective "generalized". If D is a polygon of order
(s,t), then we immediately see that the dual of D (points and lines inter-
changed) is a polygon of order (t,s).
Suppose N is the incidence matrix of the incidence graph of an n-gon
of order (s,s). Then N is the incidence matrix of a 2n-gon of order (1,s).
conversely, it can be proved easily that all generalized n-gons of order
(1,s), s > 1, are of this form. FEIT & HIGMAN [F1] also proved that s =1
or t =1 for a 12-gon of order (s,t), thus in a sense generalized 12-gons
are the same as generalized hexagons of order (s,s). Generalized n-gons of
order (s,t) with s > 1, t > 1, are called thick.
A generalized 2-gon is degenerate (every point is incident with every
line). It is not difficult to verify that a generalized triangle of order
2
(s,t) is a 2 - (s +s+1,s+l,1) design, which is the same as a projective
plane of orders (thus s = t). So 3-gons of order (s,s) exist for every
prime power s. For projective planes see DEMBOWSKI [D3] or HUGHES & PIPER
(B16].
Thick generalized quadrangles of order (s,t) are known to exist for
(s,t), (t,s) = (q,q), . (q,q2 ), 2 3
(q ,q ), (q-1,q+l), for every prime power q
(q ~ 2 for the last case). Constructions are due to AHRENS & SZEKERES [A2],
HALL [H6], KAN'l'OR [Kl], PAYNE [P1], [P2] and TITS [T4], see also [D3],
[T3]. HIGMAN [B11] showed that
 51

for thick quadrangles of order (s,t). Several other proofs of this inequal-
ity have been found, see [C3], [CS], [H12], [P6] and Section 3 of this
chapter; some of these proofs also lead to consequences for the case of
equality. There is an extended literature on generalized quadrangles. We
mention the survey papers [P4], [T1] and ['1'3].
Thick generalized hexagons are known to exist for the orders {s3 ,s),
3
(s,s} and (s,s J for prime power s, see [T4]. A necessary condition for
existence of a hexagon of order (s,t) is that st be a square, see [F1].
HAEMERS & ROOS [H3] showed that

for thick hexagons of order (s,t}. This inequality, together with a result
for the case of equality, will be the subject of the next section. For more
information about generalized hexagons we refer to [M4], [Rl], [St], ['1'4],
[Yl].
Thick generalized octagons of order (s,t) are only known to exist for
(t,s}, (s,t) = {~,22m), for odd m. The construction is due to J. Tits, see
[D3]. A necessary condition for existence of an octagon of order (s,t) is
that 2st be a square, see [F1]. HIGMAN [H12] showed that

for thick octagons. There is hardly any literature about octagons.

Let G be a connected graph of diameter m. For vertices x and y of G,

let p(x,y) denote the distance between x and y. For i,j = O, ••• ,m, define

pij (x,y) := I{z I p (x,z} = i & p (y,z) = j} I .

If pij(x,y) depends on i, j and p(x,y) only, then G is called diatanae
regu'lar (see [BS]), and we write p~j := pij(x,y) where k := p(x,y), and
0
di :o• pii' fo.r i,j = o, ..• ,m.
The numbers pij • t eraect"
k are called the ?..n i.on
nwnbera of G. Clearly, a distance regular graph is regular of degree d 1 ,
and a distance regular graph of diameter 2 is the same as a connected
strongly regular graph (in general, a distance regular graph of diameter m
is equivalent to a metric association scheme with m classes, see [Dl]).
For a distance regular graph G of diameter m, we define the matrices
A0 , ••• ,Am, indexed by the vertices of G, by
52

i f p(x,y) =i ,
otherwise.

m
Clearly, A0 = I, l Ai =J and A1 is the adjacency matrix of G. Moreover,
i•O

(A A.) = ./ ~x,y)
i J xy iJ

implies

Ai J = di J , for i,j 0t • • • ,m •

These equations show that A0 , ••• ,Am generate an m + 1 dimensional algebra.

This type of algebra turns out to be useful in the study of distance
regular graphs and similar configurations, see [BS], [B8], [Dl], [B12],
[W1].
The point graph of a generalized n-gon D is the graph whose vertices are
the points of o, two points being adjacent whenever they are on one line of
D. It is well known (see [D3], [Y2]) that the point graph G of an n-gon of
order (s,t) is distance regular of diameter L~nJ, and that the intersection
numbers of G can be expressed in terms of s and t (in the forthcoming sec-
tions of this chapter we shall exhibit this result for n =6 and n = 4) •
A graph G is called geometric for an n-gon if G is the point graph of
an n-gon. A graph G is called pseudo-geometria for an n-gon if G is dis-
tance regular of diameter L~nJ and its intersection numbers are such that G
could be geometric, that is, there exist integers s and t, such that the
intersection numbers of G depend on s and t as for geometric graphs.

Let D be a generalized polygon. An element of D is a point or a line

of o. A sequence of R. + 1 elements e 0 , ••• ,eR., is called a path of length R.
between e 0 and eR.' if ei is incident with ei-l for i = 1, ••• ,R. (thus in
e0 , ••• ,eR.' points and lines ~lternate). The distanae between elements e 0
and eR. of o, .denoted by A. (e0 ,e.~), is the length of the shortest path
between e 0 and eR.. Thus, if e0 and eR. are both points, then A.(e0 ,eR.) is
twice the distance between e 0 and eR. in the point graph of D.

In the next two sections we shall describe a method which for quadr-
angles and hexagons leads to the inequalities mentioned above, and to the
 53

results in case of equality (unfortunately, this method does not work for
octagons). The same method also yields a new proof of a theorem of CAMERON,
GOETHALS & SEIDEL [CS], which states that a pseudo-geometric graph for a
quadrangle of order (s,s 2 ) is geometric.

5.2. AN INEQUALITY FOR GENERALIZED HEXAGONS

For n =6 the definition of a generalized n-gon is equivalent to the

following one:

s.2.1. DEFINITION. A geneI'O.Uzed hexagon of oI'deZ' (s,t) is an inaidence

stI'UCtUPe bJith points and iines, suah that
i. each Line has s + 1 points,
ii. each point is on t + 1 Lines,
iii. two distinct iines meet in at most one point,
iv. fol' any non-incident point-line pail' x,L the:l'e is a unique path
of iength < 6 between x and L.

Throughout this section H will denote a generalized hexagon of order

(s,t). By use of the above definition it is straightforward to count the
intersection numbers p~j of the point graph of H. They are exhibited in
Table 1. The amount of work in computing these numbers can be reduced by
use of the equalities

This counting also shows that the point graph of a generalized hexagon is
distance regular.

TABLE 1

k k'. k k k k
k Pu P12 P22 P13 P23 P33

0 s (t+l) 0 s 2t(t+1) 0 0 s3t2

s2t2 2 2
1 s-1 st st(s-1) 0 s t (s-1}
2
2 1 s-1 s (t 2+t-1) st st(s-1) (t+1) st(s t-st-s+t)
3 0 t+1 (s-1)(t+1) 2 (s-1} (t+l) (t+1) Cs 2t-st-s+t) t(t-1) (s 2t-s+t)
54

Let A , ••• ,A be the matrices of the point graph of H. Define

0 3

2
E := A - (s - 1)A 1 + (s - s + 1) I - _!._l J
2 s+

Then we have the following lemma.

s.2.2. LEMMA. The eigenvalues of E al"e

2 2
0 and s + st + t

of multiplicity
2 2
1 + st(s+1)(t+1) s2t+st2-st+s+t and 3 s t +st+1
s
2
s + st +t
2 s 2 + st + t 2 '

respectively.

~· It is clear that EJ = O, and that

(E - (s2 -s+l)I + s!1 J)2 = (s-1) 2 At- (s-l)(A1A2+A~1> +A~.

A~
A A =
1 2 1 = st A1 + (s-1)A 2 + (t+1)A3 ,
Ai= s(t+l)I + (s-1)Al + A2 .

By use of A + A + A
1 2 3
=J - I and EJ = 0, this leads to

2 2
E(E - (s +st+t )I) = 0 •

Hence 0 and s
2 + st + t 2 are the eigenvalues of E. Finally,

yields the multiplicities. D

~· In the terminology of DELSARTE [D1], the matrix (s 2 + st + t 2 ) -l E

is a minimal idempotent in the Bose-Mesner algebra of the association
scheme on the points of H. The underlying theory provides a more elegant
 55

way to prove the previous lemma.

Let L0 be a line of H, let Li (i = 1,2) denote the set of points at

distance 2i + 1 from L0 • Partition E according to L , L1 and L :
0 2

EOl
E = roo ElO Ell
•oj
E12

E20 E21 E22

so Eij *
= Eji for i,j = 0,1,2.

5. 2.3. LEMMA. The eigenvalues of E 11 a:r>e o., s 2 and st of nruZtipliaity

st - s + t, t(s 2 -1) and s, respeatively.

~· Let A1 , 11 and A be the submatrices of A1 and A2 , respectively,

2111
corresponding to L1 • We easily see that without loss of generality

holds, hence

From the eigenvalues of Is+l ® Jst' Ist+t ® Js' I and J, and the fact that
these four matrices have a common basis of eigenvectors, the eigenvalues of
E11 and their multiplicities follow. D

5.2.4. THEOREM. A generalized he:cagon 'With s + 1 points on a line and t + 1

lines through a point satisfies
3
i. t s s or s = 1 ..

ii. s s t
3 or t =1 •

~· From 5.2.2 and 5.2.3 it follows that

rank E = s 3 (s 2t 2 +st+1) I (s 2 +st+t2 l

Since rank E s rank E, we have

11
56

2 2 2 3 2 2
(s t+s-t) (s +st+t) :$; s (s t +st+l) •

This yields

3
Thus s =1 or t ~ s • Applying this result to the dual of H yields (ii) • D

For another proof of this inequality see [E3]. Next we shall derive
some additional regularity for hexagons meeting the above bound. To achieve
this we need some properties of the matrices Eij. First we observe that

1-1/(s+l) if A(x,y) =4 I

(E02) xy =
t -1/(s+l) otherwise,

as follows directly from the definition of E and E • This implies

02

*
CE02 Eo2> xy =
l 1-1/(s+l)
-1/(s+l)
if A(x,z)

otherwise.
= A(y,z) for some z E L0 ,

Hence, without loss of generality

E02 = j
* ®
1 s+1 - _l_J
s+l r(s+l) '

where r s 2t 2 • Now the positions of the points of L relative to the

:=
2
points of L0 give rise to a partition of E into (s + 1) 2 square block
22
2
matrices F .. of size s t 2 :
l.J

:r·l .
ssJ

It is a matter of straightforward counting to see that F.j has constant row

2 2 2 l.
sums equal to t (s-1+i5ij) - s t /(s+l} for i,j = O, ••• ,s. This, and
the structure of E imply that
02
 57

'l'he following identities are now quickly seen to be true:

EOO = s2 (I - s ! 1 J) EijJ = 0 for i,j = 0,1,2,

5.2.5. LEMMA. The matT'i:x;

hae eiuenvaiues
2 2
0 , st + t and s + st + t 2 ;

2
the muitipUaity of s + st + t 2 equa"ls
2 2 3 2 2
t (s - 1) (s - t) / (s + st + t ) •

PROOF. Define

Using the above formulas we obtain

(*) UU U
* = s 2 (s 2 + t 2)U ,

Since rank u = s, the last formula reflects that the columns of U span an
2
s-dimensional eigenspace of E' corresponding to the eigenvalue s + t 2
Thanks to 5.2.2 and 5.2.3 the eigenvalues of E and E are known. By use of
11
1.3.3 we obtain that the non-zero eigenvalues of E' are

2
s2 +t2 , st + t. and s 2 + st + t 2 ,

of multiplicity

S I t(s 2 -1)

respectively. Now from (*) it follows that

E' - s -2 U * U

2 2 2
has just two distinct non-zero eigenvalues st + t and s + st + t , with the
SB

same multiplicities as before. On the other hand one easily verifies that

2
E' - ,- 0 •u. ~ .:,] ,

which proves the lemma. 0

2
The important thing in the last lemma is that the eigenvalue s 2 + st + t
disappears if t = s 3 • In order to give a combinatorial interpretation of
this phenomenon, we need two definitions. For a line L and. points x and y
of a generalized hexagon we define:

p .. k(L,x,y) :=
l.J
I{z I A(z,L) = 2i + 1, A(z,x} = 2j, A(z,y) = 2k} I ,
for i = 0,1,2, j,k = 0,1,2,3; the configuration induaed by L, x and y is the
configuration formed by the points and the lines, which are on a shortest
path between L and x, L and y, or x and y. For example, Figure 3 gives all
possible configurations induced by L, x and y if A(L,x) = A(L,y) =5 and
A(X,y) = 4.

7 4 3 2 7 4 3 7 4 3 7 4 3
s +s -s -s -1 s +s -s -1 s +s -s -s s +s -s -s

FIGURE 3

5.2.6. THEOREM. If a generu'lized he:TJagon has ordsra (s,s 3 ), then pijk(L,x,y)

onty depends on i, j, k and the configuraation induaed by L, x and y.

PROOF. First observe that

 59

2 3 3
l Pi'k(L,x,y) , l pijk(L,x,y) and l pijk(L,x,y)
i=O J j=O k=O

only depend on i, j, k and the configuration induced by L, x and y (in fact,

2 .
lp~jk(L,x,y) = ppj~x,yl). Subsequently, we verify (this is an easy but
i=O
tiresome job), that the theorem is true if i = O, or j ~ 1, or k ~ 1, and
also if A(L,x) < 5 or A(L,y) < 5. Thus it suffices to prove the theorem for
i = j = k = 2, A{L,x) = A(L,y) = 5. From the definitions of E, E and
22
Pijk(L,x,y) it follows that

( (E 1 J) 2)
22 - s+T xy

Now take t = s 3•
Lemma 5.2.5 implies that i 22 has just two distinct eigen-
4
values O and s (1 + s 2) • Bence

-2 4 2 -
E = s (1 +s )E
22 22

Using the formulas for the matrices Eij this yields

1
(E 22 - S'+T J)
2
= s 4 (1 + s 2 )E 22 * E + s 8 (s + 1) -1 J .
- s 2 E02 02

-1 2
This implies that {(E 22 - (s + 1) J) ) xy only depends on x, y and the

configuration induced by L0 , x and y. Combination with the previous steps

yields that for x,y L , p 222 (L ,x,y) only depends on the configuration
E
2 0
induced by L0 , x and y. This ·completes the proof. 0

With the available formulas the values of pijk(L,x,y) are readily

computed. For example, in Figure 3 we give p 222 CL,x,y) for the given con-
figurations.
RONAN [Rl], [R2] and THAS [T2], give sufficient conditions for a
generalized hexagon to be one of the known ones. One hopes of course that a
result like the one above will imply such a sufficient condition. Unfor-
60

tunately, the gap between the condition we have and the condition we need,
still seems to be too large to close up. It is worthwhile to remark that the
known hexagons of order (s,t) with t ~ s 3 do not satisfy the condition of
the above theorem.
Finally we remark that similar techniques yield the inequality {see
MATHON [M6])

3 2
t s s + t 2 (s -s+1)

for a regular near hexagon with parameters (s,t,t 2 ), as introduced by

= O, then a near hexagon is
SHULT & YANUSBKA [S7] (see also [Y2]). If t 2
the same as a generalized hexagon.

5.3. GEOMETRIC AND PSEUDO-GEOMETRIC GRAPHS FOR GENERALIZED POLYGONS

In this section we deal with the question whether a pseudo-geometric

graph is geometric for a generalized n-gon. It is clear that for n € {2,3}
the point graph of an n-gon is the complete graph. Assume G is the point
graph of an n-gon D with n > 3. Then three points of D which form a triangle
in G, must lie on one line of D. This implies that the graph ~ cannot
be an induced subgraph of G. The next result states that the converse is
also true.

5. 3 .1. LEMMA. For a generalized n-gon with n > 3,, a paeudo-geometl'ic graph
G ia geometric iff ~ ia not an induced aubgraph of G.

~· Only the "if" part remains to be proved. Take n even {the case n odd
is not difficult, but superfluous because of the Feit-Higman theorem). Let
D be the incidence structure whose points are the vertices of G, and whose
lines are the cliques complete subgraphs) of G of size p 1 + 2. For two
{=
11
adjacent vertices of G, there are p 1 vertices adjacent to both, but all
11
these vertices are mutually adjacent since otherwise ~ occurs. This
means that eVf!JIY edge of G determines a unique line of D. This proves that
n
D is a 1 - (v,s+1,t+1) design, where s = p 111 +1 and t = Pi,n-i -1.
k
Let G.' denote the incidence graph of D. Then, because p. k i = 1 for
J., -
k = O, ••• ,~n-1, i = O, ••• ,k, the girth of G' is at least 2n. Now, since
each point of D is incident with pn lines, it follows that the distance
1 ,n-1
between a point and a line of D (regarded as vertices of G') is at most
 61

n - 1. Hence G' has diameter n and girth 2n. This proves that D is an n-qon,
whose point graph is G. D

A direct consequence of this lemma (which was pointed out to me by

D.E. Taylor) is the following.

S. 3. 2. COROLLARY. For> a genero.lized n-gon UJith n > 4, a pseudo-geomet'l'ia

gr>aph is geomet'l'ia.

PROOF. If n > 4, then p~ 1 .. 1. Hence ~ does not occur in a pseudo-

geometric graph. Now S.3.1 gives the result. D

What remains to be studied are generalized quadrangles. A very easy

counting argument shows that the point graph G of a quadrangle of order
1 2
(s, t) is strongly regular with intersection numbers p = s - 1, p = t + 1,
11 11
0
p 11 = d 1 = s (t + 1) (this proves that the point graph of an n-gon is dis-
tance regular, in the case n .. 4). This implies (see [CS], [Tl] or Appendix
I) that the eigenvalues of G, and hence the eigenvalues of any pseudo-geo-
metric graph for a quadrangle of order (s,t), are

s(t+1) , s-1 and -t-1

of multiplicity

1 , s 2 (st+1) I (s+t) and st(s+l)(t+l} I (s+t} ,

respectively.
There exist many pseudo-geometric graphs for quadrangles, which are
not geometric. The Shrikhande graph (see cover} is one of them. More
examples (including an infinite family} are given in Section 6.2. The
following theorem, which is due to CAMERON, GOETHALS & SEIDEL [CS], gives
a sufficient condition for a pseudo-geometric graph to be geometric, as
well as the extension of Higtllan's inequality to pseudo-geometric graphs
for generalized quadrangles.

S.3.3. 'l.'HEOREM. Let G be a pseudo-geomet'l'ic gmph for> a thiak genero.lized

quad.Pangle of order> (s,t). Then

i. t ::;; s 2 ..

ii. if equality holds, then G is geomet'l'ia,

62

iii. equality impl.ies that aU eubaonstituents of G are strort{Jl.y

1'6(/U l.al'.

!!!QQ!.· Let A be the adjacency matrix of G. Define

2
E :== - (s+1}A + (s -1}I + J .

Then from the eigenvalues of A it follows that E has just one non-zero
2
eigenvalue (s + 1) (s + t} of multiplicity s (st + 1) I (s + t). Hence

(*) rank E = s 2 (st + l) I (s + t) •

Lat x be a vertex of G. Partition A and E according to x, the vertices ad-
jacent to x, and the vertices not adjacent to x:

-sj *

A•~
j*

A11

A21
.:'.]
A22
E =
f:j E11

E21
J*]
E12

E22
I

where A = A; and E = E; • For i .. 1,2, let Gibe the graph with ad-
12 1 12 1
jacency matrix Aii (so Gi is a subconstituent of .G). Then G has s(t + 1)
1
1
vertices, and is regular of degree p = s - 1. Bence s -1 is an eigenvalue
11
of A of multiplicity c, say. It is known (see [BS], [C12] or Appendix I}
11
that c equals the number of components of G1 • Clearly each component has
at least s vertices. Hence

(**) CS S S(t+l) •

The matrices A , I and J have a common basis of eigenvectors. Using this

11
it follows that E has an eigenvalue 0 of multiplicity c - 1 (one of the
11
eigenvalues s - 1 of A corresponding to the eigenvector j leads to the
11
eigenvalue s(t + 1) of Eli). Hence

rank E = s (t + 1) - (c - 1) •
11

Now using (*) and (**) we have

s(t+l) - t s rank E
11
s rank E = s 2 (st+1) I (s+t) •

This yields
 63

2
t(s-l)(t-s) s 0,

proving (i).Suppose equality holds. Then we must have equality in(**),

which means that G1 is a disjoint union of complete graphs on s vertices.
Since, x is arbitrary, this yields that ~ does not occur in G. Now
(ii} follows on applying 5.3.1.
Since the disjoint union of complete graphs of the same size is strongly
regular, it only remains to be proved that G2 is strongly regular. This we
shall prove analogously to the proof of 5.2.5. We know that the eigenvalues
of E are O, s (s + 1) and s (s 2 + 1) of multiplicity s 2 , (s - 1) (s 2 + 1) and 1,
11
respectively. Now using 1.3.3 and the eigenvalues of E, we obtain that the
matrix

2
has eigenvalues O, s 2 (s + 1) and 2s 2 , where 2s is a simple eigenvalue with
eigenvector [s 2 j *] * • Bence E , and also A22 , has just two distinct eigen-
22
values not belonging to the eigenvector j. This proves that G2 is strongly
regular. 0

From (iii) of the above theorem it follows that the number of points
adjacent to three mutually non-adjacent points of a quadrangle of order
(s,s 2 ) is constant. This result was first proved by BOSE [B7].
A quadrangle of order (s,t) is the same as a partial geometry with
parameters (s,t,1) (see [H11], [Tl] or Appendix I). For pseudo-geometric
graphs for a partial geometry with parameters (s,t,a), where a> 1, a result
like Lemma 5.3.1 does not hold anymore. Therefore the question in the be-
ginning of this section is much more difficult to answer for these geo-
metries.
64

CHAPTER 6

CONSTRUCTIONS

6.1. SOME 2 - (71,15,3) DESIGNS

In this section we shall construct eight non-isomorphic 2 - (71, 15,3)

designs. First we shall construct a 2 - (56,12,3) design D, which satisfies
the hypothesis of Theorem 3.2.4. Next we show that D is embeddable in a
2- (71,15,3) design. A less extensive treatment of this construction appear-
ed in [B2]. Designs with these parameters seem to be new (see [C4] p.104,
or [H5] p.297).

The most important ingredient for our construction is Fa, the field
with eight elements. Let G be the full automorphism group of Fa, that is,
i
the group of order 16a defined by x.+ ax 2 +b, a,b E lFa,a;. O, i €72:. We
shall identify F with AG(3,2), the 3-dimensional affine space over JF •
8 2
Although G is the full automorphism group of JF , G is not the full auto-
8
morphism group of AG(3,2). Yet G acts transitively on the elements (we
reserve the word points for points of a design), the lines (i.e. unordered
pairs of elements} , the planes (sets of four linearly dependent ele-
ments), and the sets of four linearly independent elements. Moreover, the
stabilizer of a line L has four orbits on lines: L itself, the lines inter-
secting L, the lines parallel to L, and the lines skew to L.
Now we shall define the incidence structure D. The points of D are the
fiftysix ordered pairs of distinct elements of lFa. The blocks of D are the
3
seventy 4-subsets of Fa· Let a, satisfying a. = a+ 1, be a primitive element
of JFa· The point (0,1) is defined to be incident with the following blocks:

{0,1,a,a 3 }, {0,1,a,a 2}, {0,1,a 2 ,a3 }, {a,a 2 ,a 3 ,a4 }, {a 2 ,a 3 ,a4 ,a 5 },

{O,l,a 2 ,a 6 }, {0,1,a 2 ,a 4 }, {0,1,a4 ,a6 }, {a,a 2 ,a4 ,a6 }, {a,a 3 ,a4 ,a6 },

{0,1,a4,a5 }, {0,1,a,a4 }, {O,l,a,a 5 }, {a,a 2,a4,a

5
}, {a,a 2,a5,a
6 }.

Now we let G act on D. This defines D, because G acts transitively on the

points of D, and because the map x I+ x 2 , which fixes the point (0, 1), also
fixes the set of blocks incident with (0,1). A point (x,y) is called
 65

equivatent to a point (w,z) i f {x,y} = {w,z}. 'l'wo blocks b c Fa and

b' c Fa are called equivatent i f b = b', orb n b' = 11 (i.e. bub' = F ).
8
We know that G has two orbits on the blocks of D. The first orbit contains
fifty~ix blocks, they are the linearly independent 4-subsets of JF ; these
8
blocks will be called blocks of type I. The second orbit consists of the
fourteen linearly dependent 4-subsets of JF (Le. planes), that is, they
8
have the form {x,y,z,x+y+z}; these blocks are of type II. It is clear that
equivalent blocks are of the same type.

6.1.l. LEMMA. Let (x,y) be a point of D. Let b c F and b' c:: JF be dis-
8 8
tinct equivatent btoaks of D.
i. If (x,y) is inaident with b, then {x,y} c b OP {x,y} c b'.

ii. If {x,y} c:: b, bof type I, then e:caatiy one of the foitowing two
statements is tPUe.
1. (x,y) is incident with b and not with b', and (y,x) is inai-
dent with b' and not with b;
2. (x,y) is incident with b' and not with b and (y,z) is incident
with b and not with b'.
iii. If b is of type II, then (x,y) is inaident with b iff {x,y} c:: b.

iv. b and b' have no points of D in aonm:m.

!:!QQ!.· Withoutlossof generality take (x,y) = (0,1). We may do so, because

G is transitive on the points of D. Blocks incident with (0,1) are given in
(*). on applying the map x 1+ x + 1 to (*) we find that the blocks incident
with (1,0) are the following ones:

{0,1,a,a 3 }, {0,1,a 3 ,a6 }, {0,1,a,a6 }, {a,a 3 ,a 5 ,a6 }, {a,a 4 ,a 5 ,a 6 },

{0,1,a 2 ,a6 }, {0,1,a.5 ,a6 }, {0,1,a 2 ,a 5 }, {a 2 ,a3 ,a 5 ,a6 }, {a,a 2 ,a 3 ,a 5},

{0,1,a 4 ,a s }, {0,1,a 3 ,a s}, }0,1,a3 ,a 4 }, {a 3 ,a4 ,a s ,a6 }, {a 2 ,a 3 ,a4 ,a6 }.

The first column of (*) and (**) consists of blocks of type II, all other
blocks are of type I. Now (i), (ii) and {iii) are just a matter of verifica-
tion. (iv) follows immediately from (ii) for blocks of type I, and from
(iii) for blocks of type II. D
66

From (ii) and (iii} of this lemma we conclude the following. If each
point of D is replaced by its equivalent partner and each block of D of
type I by its equivalent partner, then incidence is not changed. Hence the
peJ:IDutation of the points of D, which interchanges the equivalent pairs of
points, is an automorphism of D. This automorphism is different from, and
commutes with, the automorphisms provided by the group G.

6.1.2. THEOREM. D is a 2 - (56,12,3) design.

~· First we prove that D is a 1-design. Since every point is incident

with fifteen blocks, the average number of points incident with a block
equals twelve. By 6.1.1.iii a block of type II is incident with exactly
twelve points. Now, since G acts transitively on blocks of the same type,
also the blocks of type I are incident with exactly twelve points. We have
seen that G has three orbits on the (unordered) pairs of lines of JF
8
(intersecting, parallel, skew}, for which the following pairs are repre-
sentatives:

{{0,1},{0,a}} , {{O,l},{a,a 3 }} , {{0,1},{a 2 ,a 4 }} •

From this it follows that the group 2 x G, which is an automorphism group of

D, has seven orbits on the (unordered) pairs of points of D, for which the
following ones are representatives:

{(0,1),(0,a)}, {(0,1),(a,a 3 )} , {(0,1),(a 2 ,a)},

4

{(1,0),(0,a)}, {(1,0),(a,a 3 )} , { (1, 0) , (a 2 ,a4 ) } ,

{ (1,0), (0, 1)}

Blocks incident with (0,1) and (1,0) are given in(*) and (**),respective-
ly. Using the maps x r+ ax, x i+ x +a and x i+ a (x +a) , we obtain the blocks
incident with (0,a), (a,a 3) and (a 2 ,a 4 }. The blocks incident with (0,a) are

{O,a,a 2 ,a4 }, {O,a,a 2 ,a 3 }, {O,a,a 3 ,a4 }, <a 2 ,a 3 ,a 4 ,a 5 }, {a 3 ,a 4 ,a 5 ,a6 ),

~0,1,a,a
3 1, {O,a,a 35
,a }, \0,1,a,a 5 }, {1,a 2
,a3,a
5 }, {1,a 2,a4,a
5} ,

{O,a,a 5 ,a6 }, {O,a,a 2 ,a 5 }, {O,a,a 2 ,a6 }, {a 2 ,a 3 ,a 5 ,a 6 ), {1,a 2 ,a 3 ,a 6 }.

The blocks incident with (a,a 3 ) are

 67

(0,1,a,a 3 ), {O,a,a 3 ,a4 }, {1,a,a 3 ,a4 }, (0,1,a 2 ,a4 }, {1,a 2 ,a 4 ,a 6 },

{a,a 3 ,a4 ,a 5 }, "(a,a 2 ,a 3 ,a 4 }, {a,a 2 ,a 3 ,a 5 ), {O,a 2 ,a4 ,a 5 }, {0,1,a 2 ,a 5 ),

{a,a 2,a 3 ,a6 }, {O,a,a 2 ,a 3 }, {O,a,a 3 ,a6 }, {O,a 2 ,a4 ,a 6 }, {O,a 4 ,a 5 ,a 6 }.

The blocks incident with (a 2 ,a 4 ) are

{O,a,a 2 ,a4 }, {O,a 2 ,a4 ,a 5 }, (a,a 2 ,a4 ,a 5 }, {O,a,a 3 ,a 5 }, {1,a,a 3 ,a 5 },

{a 2 ,a4 ,a 5 ,a6 }, (a 2 ,a 3 ,a4 ,a 5 }, {a 2 ,a 3 ,a4 ,a6 ), {O,a 3 ,a 5 ,a6 }, {O,a,a 3 ,a 6 },

{1,a 2 ,a 3 ,a4 }, {O,a 2 ,a 3 ,a4 }, f{0,1,a 2,a4 }, {0,1,a 3 ,a 5 ), {0,1,a 5 ,a 6 ).

Blocks incident with (0,1) are marked by(. Blocks incident with (1,0) are
marked by ). We see that for each of the seven pairs of points there are
exactly three blocks incident with both points. Hence D is a 2 - (56,12,3)
design. 0

Next we shall see that D satisfies the hypothesis of 3.2.4. 'l'he "line
graph of a geometry is the graph whose vertices are the lines, two vertices
being adjacent iff the lines intersect.

6.1.3. 'I'HEOREM. D has just three interieeation numbers 3, 2 and 0 (== k-r+>..).
The alaee graph of Dis the comptement of the tine graph of PG(3,2).

~· Let b 1 c JF and b c JF be non-equivalent blocks of D. Let bi and

8 2 8
b2 be the equivalent partners of b 1 and b 2 , respectively. Let B1 , B2 , Bi,
B2 be the sets of points incident with b 1 , b 2 , bi, b2, respectively. Then
by 6.1.1

Hence by 3.2.1.ii
IB .n B 1
1 2

This implies

From (ii) and (iii) of 6.1.1 it follows that

68

I ca 1 u B2) n <B 2 u B2l I =

2 I x _,, y,
.. I{<x,y> E F
8
{x,y} c b1 or {x,y} C bi I

{x,y} c b2 or {x,y} c b2}I

if jbl n b2I .. 2 I

L: if lb1 n b2I E {1,3}

Hence 3, 2 and 0 are the intersection numbers of D. It is clear that the

thirtyfive equivalence classes of blocks of D can be represented by the
4-subsets of JF containing O. Therefore, they can also be represented by
8
all 3-subsets of :m- - {O}. Now using Result 4.2.3 we see that the class
8
graph of Dis indeed the complement of the line graph of PG(3,2). 0

By the above theorem, there exists a 2-1 correspondence between the

blocks of D and the lines of PG(3,2), such that two blocks have no point in
common if£ they correspond to the same line of PG(3,2}, two blocks have two
points in common if£ they correspond to intersecting lines and two blocks
have three points in common if£ they correspond to skew lines.

6.1.4. THEOREM. Dis embeddabZe in a eymmetl"ia 2- (71,15,3) design.

PROOF. We extend D to o 1 with fifteen points (called net.i points), being the
points of PG(3,2) and one block (nav bZoak). The points incident with the
new block are precisely the new points. We define a new point to be inci-
dent with an old block (block of D) iff the line of PG(3,2) corresponding
to that block contains that point. Now it is easily seen that o1 is a
1 - (71, 15, 15) design, and that any two distinct blocks of o have three
1
points in common. This proves that the dual of o , and therefore o 1 itself,
1
is a symmetric 2 - (71,15,3) d~sign. 0

From 6.1;1.iii it follows that the seven equivalence classes of blocks

of type II of D correspond to seven mutually intersecting lines of PG(3,2).
For these lines we may take seven lines through one point or seven lines in
one plane. Let o 1 be the embedding of D in which blocks of type II corre-
spond to linea through one point, and J.et o be the other embedding of D.
2
Define o*1 and o*2 to be the dual of o and o 2 , respectively. We shall show
1
 69

that these four 2- (71,15,3) designs are non-isomorhic. To achieve this we

define a block B of a 2 - (71,15,3) design to be special if the derived
design with respect to B (i.e. the 2 - (15,3,2) subdesign of the 2 - (71,15,3)
. design, formed by the points of B and the blocks distinct from B) consists
of twO identical copies of a 2 - (15,3,1) design. It is not difficult to see
that the residual design with respect to a special block B (i.e. the
2- (56,12,3) subdesi9n of the 2 - (71,15,3) design, formed by the points off
B and the blocks distinct from B) satisfies the hypothesis of Theorem 3.2.4
and that its class graph is isomorphic to the complement of the block graph
of the 2 - (15,3,1) design associated to B (which clearly is quasi-symmetric,
since A= 1). From the proof of 6.1.4 it follows that the new blocks of o
1
and D2 are special. Moreover, 6.1.1.iii implies that the new point of o1
which is incident with all old blocks of type II, is a special block of D~.
By verification it turns out that these are the only special blocks of o 1 ,
o*1 a nd o • However, o*h .
2 as seven special blocks. They are the seven new
2
points of o 2 , which lie in the plane of PG(3,2) correspondinq to the old
blocks of type II. This already shows that

we know that any 2- (15,3,1) design associated to a special block of o 1 or

o 2 is the design formed by the points and lines of PG(3,2). By use of
6.1.1.iii it follows that also the 2 - (15,3,1) design associated to the
special block of D~ is the desi9n which comes from PG(3,2). Leto* be the
residual design of D~ with respect to the special block. Then the class
graph of o* is again the complement of the line graph of PG(3,2). This
means that, similarly as for the design o, interchanging points and planes
of PG(3,2) yields a second embedding of o* into a 2- (71,15,3) design. Let
o*3 be this 2 - (71,15,3) design and leto3 be the dual of o *3 • By verifica-
*
tion it follows that o 3 has just one special block (the one we started
with), but that o has seven·special blocks. This already shows that
3
o * tA o *3 • By further investigation it turns out that the 2- (15,3,1) design
1
associated to any of the seven special blocks of o 3 is again the design
obtained from PG(3,2), however, the seven special blocks of o~ give other
2- (15,3,1) designs (in fact all seven of them give the second design in
the list of WHITE, COLE & CUMMINGS [W3]). This proves that o *2 </.< o 3 , and
therefore o .:;. o * • Bence o , o * , o , o *2 , o and o *3 are all non-isomorphic.
1 1 1 1 2 3
But there is still more. As remarked before, the 2 - ( 15, 3, 1 ) design associ-
70

ated to one of the seven special blocks of n (these seven special blocks
3
form an orbit under the automorphism group of o3 ) is the design formed by
the points and lines of PG(3,2). This implies that once again we can make
another 2- (71,15,3) design by taking away the (two identical) 2 - (15,3,1)
designs and putting .them back again after having interchanged points and
* .
planes. Call this new design n4 , and let n4 be its dual. It turns out that
D4 has one special block, and that o* has seven special blocks. Bence n is
4 4
.
not isomorphic to o1 , o*1 , n* , n3 and o* • In .
addition, o is not isomorphic
2 4 4
to o 2 and o*3 , since otherwise n3 would have been isomorphic to o or n*1 •
1
Because of time considerations we did not check whether it is possible
to produce still more 2 - (71,15,3) designs by playing once again the same
game with respect to a special block of o*4 • Thus we have the following
result.

6.1.5. THEOREM. Thel'e e:cist at least eight 2 - (71,15,3) cle.signa.

The designs D*1 , n*2 , o*3 and n*

4 are given explicitly in Appendix II. By
taking residual designs with respect to various special blocks we obtain
(at least) four non-isomorphic 2 - (56,12,3) designs which satisfy the hypo-
thesis of Theorem 3.2.4. One of these designs has a class graph which is
non-isomorphic to the class graph of the other ones.

An ova 7, in a 2 - ( 71, 15, 3) design is a set S of six points such that

any block has two or no points with S in common, see [A4]. Let S be an oval.
It is clear that exactly twentysix blocks do not meet s. Therefore, by
3 • 1. 1 an oval of a 2 - ( 71 , 15, 3) design is equivalent to an empty sub-
design with six points and twentysix blocks. From 3.1.1.ii (see also [A4])
it follows that the subdesign of the 2- (71,15,3) design formed by the
points off sand the blocks not meetings is the dual of a 2- (26,6,3)
design. By verification it follows that the following blocks of D provide
. n* :
an oval in
1

{1,a,a 2 ,a 3 }, {1,a,a 2 ,a 5 }, {0,1,a,a 5 }

{1,a 2 ,a4 ,a 5 }, {a,a 2 ,a4 ,a 5 }, {O,a 3 ,a4 ,a6 }.

We conclude this section with a remark about automorphism groups. The

group 2 x G of order 336 is an automorphism group of D1 , n2 , o3 and their
 71

duals. The designs o4 and o*4 have an automorphism group of order 48, viz.
the stabilizer of a special block of o3 • F.C. Bussemaker has verified by use
of a computer that the groups mentioned above are the full automorphism
groups.

ACKNOWLEDGEMENT. I thank H.J. Beker, F.C 1 Bussemaker, R.H.P. Denniston and

M. Ball Jr. for various contributions to this section.

6.2. SafE STRONGLY REGULAR GRAPHS

Suppose A is the adjacency matrix of a strongly regular graph G on n

vertices of degree d. Furthermore, assume that A admits the following
structure:

where Aij is a square matrix of size c := n/m having constant row sums
equal to bij' say, for i,j = 1, ••• ,m. Prom 1.2.3.iii it follows that the
eigenvalues of the matrix B := (bij) satisfy

Hence (*) yields directives for the construction of strongly regular graphs,
whose adjacency matrix admits this block structure. Let us consider two
special cases of this structure.

CASE 1: All diagonal entries of B are equal to r , say, and all off-diagonal
0
entries of Bare equal to r 1 , say, that is,

Hence by (*)

This implies

o. 2 (A) (m -1) +d) /m

or
72

6.2.1. EXAMPLE. We wish to construct a strongly regular graph G with para-

3 2 2
meter set (q +q +q+l,q +q,q-1,q+l), admitting the block structure of case 1
with m = q 2 + 1 and c = q + 1. Then (see Appendix I) A2 (G) = q - 1,
An (G) ,= - q - 1 and the formulas above yield r 0 = q, r l = 1. It is indeed
possible to construct G, by use of this framework. To do so we define the
permutation matrices P and Q of size q + 1 by

if i + j = q + 2'
and (Q) ij := { :
otherwise.

It easily follows that

Pk Q = (Pk Q) * = QP-k ' r

i=O
pi =J •

Fork= l, ••• ,q, define 1\. := PkQ = QP-k. Then

= p k-t , J I (q+l)J I

for k,t = 1, ••• ,q.Let p 1 , ••• ,pv (v := q 2) be the points, and let
c1 , ••. ,cq+l be the parallel classes of an affine plane of order q. For
i,j = 1, ... ,q2+1, define the (q+1) x (q+l) matrices

J-Iifi=j,

I if i .J.
r j , i = q2 + 1 or j = q 2 + 1. ,
2
1\. if i :F j, i s q 2 , j s q , anq: ck contains the
line through pi and pj.

By use of the formulas above it is straightforward to verify that the square

matrix A of size q 3 +q2 +q+ 1, built up with these block matrices Aij'
satisfies

(A+I) 2 •(q+l)J+qI,
2

Therefore, A is the adjacency matrix of the desired strongly regular graph.

 73

A strongly regular graph with the same parameters is provided by the

point graph of a generalized quadrangle of order (q,q) (see Section 5.1). However,
for q > 4 the graphs constructed above need not be geometric. Indeed, we can
order the points and the parallel classes of an affine plane of order q in
such a way that the graph we obtained by our construction has ~ as an
induced su.bgraph. Then by 5.3.1, this graph is not geometric, and therefore
non-isomorphic to any of those which come from generalized quadrangles.

As a second example of graphs admitting the block structure of case 1

we mention the eleven incidence graphs of three linked 2 - (16,6,2) designs
(see Section 4.2).

CASE 2; All diagonal entries of B are equal to r , say. The off-diagonal

0
entries of B take exactly two values r and r 2 , say Cr 1 > r 2J. Then by(*)
1
the (0,1) matrix

1
B ;= - - - (B - r (J - I) - r I)
r -r 2 O
1 2

has just three distinct eigenvalues, one of which is simple and belongs to
the eigenvector j. Hence B is the adjacency matrix of a strongly regular
graph G' with eigenvalues

6.2.2. EXAMPLE. We wish to construct a strongly regular graph G with para-

meters (40,12,2,4), admitting the block structure of case 2 with m = 10,
c = 4,
r 0 = r 2 = O, r 1 = 2. From t.. 2 (G) = 2, t.. 40 CG) = -4 and the above for-
mulas it foilows that d' = 6,· t.. 2 (G') • 1, t.. CG') = - 2. Hence G' is the
10
complement of .the Petersen graph. For h = 0,1 and i,j = 2,3,4, define the
square (0,1) matrices Thij of size four, by

h i f (k,.I!.) € {(1,1),(1,.1!.),(k,1),(k,.I!.)},

1 - h otherwise.
74

Then we have for h,h' 0,1; i,j,i' ,j' .. 2,3,4,

Thij *
Thji Thij J .. 2J, Thij + T(l-h)ij .. J '

*
Thij Th'i'j' J if j "' j' , *
Thij Thi'j 2Tlii 1 ,

Thij T(l-h)i'j = 2TOii' •

With the help of these properties it is relatively easy to check that the
following two matrices are adjacency matrices of strongly regular graphs
with parameters (40,12,2,4).

0 0 '1'022 T032 '!'042 T022 T032 T042

'1'044 T034 '1'044 Ton Tl22 T024
0 0 T122 T034 0 T1:22 T022 0 To33 "'o•3 T022 T032 T042
T044 T044 Tl34
0 0 T133 0 T022 T032 0 T042
"043 T133 T043 T024 T024 T023 T033 T044
0 T024 T034 T044 0 T022 T023 T024
T044 T044 T133 T123 T033 T024

0 0 0 T033 T043 T033 T033

T033 Tl43 T034 Tl23 '!'144 T122 T022 T022

0 0 0 T044 Tl34 0 '1'022 TQ33 0 T044 T033 T043

Tl22 T042 Tl32 T124 T023
0
T042 Tl22 Tl32 T142 T032 T022 T024 T022 T034 T044 T033 T043

0 T123 T123 T023 T023 0 T033 T033 T044 T044

T043 T 133 T033 T144
0 0 0 0 T033 T044 T044
T042 T122 T044 T023 '1'132 T033 T024 T023 T033

0 0 0 0 T024 T024 0 T0.34 T034 '1'044 T044

T122 T042 T143 T022 T132 T033

MATHON [MS] used the block structure of case 2 for the construction of
2 2
strongly regular graphs with parameters {pq , ~(pq 2 -1) , l.o(pq 2-s) , l.o(pq -1))
for prime powers p and q, p = 1, q: -1 (mod 4). The strongly regular graph
G', which provides the framework for Mathon's construction has parameters
(p , ~ (p-1) , l.o (p-5) , l.o (p-1)) •

Any graph constructed in one of the examples above has the property
1 2
that p 11 + 2 p • This implies that its adjacency matrix A satisfies
11

where d denotes the degree of the graph. This yields the well-known fact
that A+ I is the incidence matrix of a symmetric 2 - (n,d+l ,pi ) design.
1
Similarly, if a strongly regular graph satisfies pi 1 = pi 1 , then its ad-
1
jacency matrix itself is the incidence matrix of a symmetric 2 - (n,d,p 11 )
design. This phenomenon is behind the next example, where we derive a
1 2 1 2
strongly regular graph with p
11
+2 = p 11 from one with p
11
p
11
•
 75

6.2.3. EXAMPLE. We start with a description of the generalized quadrangle Q

1 1
of order {2 -1,2 +1), 1 e :N, due to BALL [H6]. Consider AG(3,q), the three
1
dimensional affine geometry over JF , with q = 2 • Let S be a set of
q
m := q + 2 lines from AG(3,q) passing through one point, such that no three
lines lie in one plane. Such a set exists, because it corresponds to a com-
plete oval in the projective plane PG(2,q), which exists iff q is even, see
[D3]. It is easy to prove that each plane of AG(3,q) contains two or no
lines from s. The points of Qare the points of AG(3,q); the lines of Qare
the lines of S, and the lines of AG(3,q) which are parallel to a line of S;
a point and a line are incident in Q, iff they are incident in AG(3,q). Now
it is easy to prove that Q is a generalized quadrangle of order (q-1,q+l).
Let us partition the adjacency matrix A of the line graph of Q {i.e. the
point graph of the dual of Q) into m2 square block matrices of size q 2 ,
according to the m parallel classes in Q:

From the structure of Q it follows that we may arrange the lines such that

Aij 0 if i • j 1

Aij = Iq $ Jq if i odd and j 1 + i, or j odd and i = j + l ;

for all other values of i,j, where PkR. is a permutation matrix of size q,
for k,i =1 1 ••• ,q. Now we derive a new matrix Afrom A by replacing

for i 1,3, ••• ,m-1. Then it follows that

~2 2 2
A =A =qJ+qI,

Since Ais symmetric with all diagonal entries equal to one, the matrix
 76

A- I is the adjacency matrix of a strongly regular graph with parameters

(q (q+2),q 2+q-1,q-2,q). For q = 2 this graph is the Clebsch graph (see
2

Section 4.1, Figure 2), but for all other values of q = 2.t these strongly
. regular graphs seem to be new.

We remark that in the above example A has the·block structure of case

1, whilst Ahas the block structure of case 2 with the cocktailparty graph
on m vertices (complete ~-partite graph) as the underlying strongly regular
graph.

The remainder of this section will be devoted to strongly regular

graphs with parameters (40,12,2,4). For convenience we call such graphs
40-graphs. Examples 6.2.1 (q = 3) and 6.2.2 provide 40-graphs. The point
graph of a generalized quadrangle of order {3,3) is a 40-graph. PAYNE [PS]
proved that there are exactly two generalized quadrangles of order (3,3)
{one being the dual of the other). In fact, these two geometric 40-graphs
are the graph of Example 6.2.1 with q = 3, and the second graph of Example
1
6.2.2. From p 11 = 2 it follows that a subgraph of a 40-graph induced by all
vertices adjacent to a given vertex, is regular of degree two, so a dis-
joint union of cycles. But we can say more.

6.2.4. LEMMA. Let G be a 40-graph. Let x be a verte:x: of G and Zet Gx be the

subgraph of G induced by the vertiaes adjaaent to x. Then Gx is one of the
following graphs:
i. a 12-ayale;
ii. the disjoint union of a 9-ayale and a triangle;
iii. the disjoint union of wo 6-ayales;
iv. the disjoint union of a 6-ayale and wo triangles;
v. the disjoint union of four triangles.

~·We only have to prove.that the number of vertices of any component

of Gx is divisible .by three. If Gx is connected, there is nothing to prove.
Suppose Gx has a component c of size c < 12. We partition A into sixteen
block matrices according to: the vertex x, the vertices of c, the remaining
vertices of Gx, and the vertices not adjacent to x. Then the entries of
 77

0 c 12-c 0

1 2 0 9
B :=
1 0 2 9

0 c/3 4-c/3 a
are the average row sums of the block matrices of A. It is easy to see that

On the other hand we know

So the eigenvalues of B interlace the eigenvalues of A tightly. Bence, by

1.2.3.ii the row sums of the block matrices are constant, so c/3 is an
integer. D

We associate with a 40-graph a 5-tuple (a 1 , ••• ,a ), where a , ••• ,a

5 1 5
denote the number of vertices x for which Gx has the form (i), ••• ,(v),
respectively, of the above lemma. Using 5.3.1 we observe that a 40-graph is
the point graph of a generalized quadrangle iff its 5-tuple is (0,0,0,0,40).
The first graph of Example 6.2.2 has 5-tuple (0,0,4,24,12). R. Mathon
(private communication) constructed a 40-graph with 5-tuple (0,0,0,36,4).

WEISFEILER [Wl] describes an algorithm for generating strongly regular

graphs with a given parameter set, based on the principle of backtracking.
By use of this algorithm we wrote a computer program (in Algol 60) for the
construction of 40-graphs. Weisfeiler's algorithm rejects isomorphism only
partially. This means that some of the produced 40-graphs may be isomorphic.
We had our program run for about ten minutes. It turned out that, although
we obtained about twohundred (not necessarily non-isomorphic) 40-graphs,
the process of finding all 40'-graphs still was in the beginning phase. For
this reason there was no hope for completing the whole search. It seems
that there are thousands of 40-graphs. Still we wanted to test the few
hundred 40-graphs we found on isomorphism. A complete test on isomorphisms
would have been.too expensive. Therefore we just computed the 5-tuple of
each 40-graph. It turned out that twentyone of these 40-graphs had different
5-tuples. So we found at least twentyone non-isomorphic 40-graphs. one of
these is the first graph of Example 6.2.2. But none of these graphs has the
 78

5-tuple of a geometric 40-graph or Mathon's 40-graph. So we have the follow-

ing result.

· 6.2.5. THEOREM. ThePe exist at Zeast twentyfoUI' strongZy !'egu"la!' gruphs

i»ith pa!'ametePs (40,12,2,4).

These 40-graphs are given in Appendix II, except for the three 40-
graphs that are already exhibited in the Examples 6.2.1 and 6~2.2. The
first graph in the list is Mathon's 40-graph.

We noticed already that a 40-graph gives a 2 - (40,13,4) design. There

is no reason why non-isomorphic graphs should lead tot non-isomorphic
designs. However, it has been checked that our twentyfour 40-graphs do
produce twentyfour non-isomorphic 2 - (40,13,4) designs.
An oval in a 2 - (40,13,4) design is a set S of four points, such that
any block has at most two pofnts in common with S, see [A4]. Easy counting
arguments give that twelve blocks are disjoint from s and four blocks have
exactly one point in common with S. Suppose we have a 40-graph with a co-
clique of size four, such that any vertex is adjacent to two or to no
vertices of that coclique. Then this coclique of the 40-graph produces an
oval in the corresponding 2 - (40, 13,4) design. Conversely, it can be proved
that any oval in a 2 - (40, 13, 4) design, obtained from a 40-graph, corre-
sponds to such a coclique. We see that the two 40-graphs of Example 6.2.2
produce 2 - (40, 13, 4) designs with ten disjoint ovals. Also the last six
40-graphs of Appendix II supply designs with ovals. The remaining sixteen
40-graphs have no ovals.

In 4.2.2 we saw that the chromatic number of any 40-graph is at least

five. Since the complement of the Petersen graph is 5-colourable, it follows
that the two 40-graphs of Example 6.2.2 are 5-colourable as well.
 79

APPENDIX I

GRAPHS AND DESIGNS

This appendix contains the basic concepts and results from the theory
of graphs and designs, which are used in the present monograph. Some general
references are [BS], [C12], [H8], [W6] for graphs, [D3], [HS], [H17], [R3]
for designs, and [B7], [C6] for both. We shall assume knowledge of Section
1.1.
A gpaph consists of a finite non-empty set of vertiaes together with a
set of edges, where each edge is an unordered pair of vertices (so our
graphs are finite, undirected and without loops or multiple edges) • The two
vertices of an edge are called adjaaent (or joint). A graph is aomplete if
every pair of vertices is an edge. The complete graph on n vertices is
denoted by Kn. A graph without edges is called void (or null) • The aomple-
ment of a graph G is the graph G on the same vertex-set as G, where any two
vertices are adjacent whenever they are not adjacent in G. The disjoint
union of a collection of graphs G1 , ••• ,Gm on disjoint vertex sets is the
graph whose vertex-set is the union of all vertex-sets, and whose edge-set
is the union of all edge-sets of G1 , ••• ,Gm. A graph is disaonnected if it
is the disjoint union of two or more graphs. Any graph G is the disjoint
union of one or more connected (= not disconnected) graphs, called the
aomponente of G.
Let G be a graph on n vertices. A sequence of distinct vertices
x0 , ••• ,xt of G is a path of length£ between x 0 and x£ if {xi_ 1 ,xi} is an
edge for i = 1, ••• ,£. The distance p(x,y) between two vertices x and y is
the length of the shortest path between x and y (p(x,y) =•if x and y are
in distinct components of G) •· The diameter of G is the largest distance in
G. A sequence.of vertices x 0 , ••• ,xt is a circuit of length t if x 1 , ••• ,xt
are distinct, x 0 = xt' t > 2 and {x _ 1 ,xi} is an edge for i = 1, ••• ,t. The
1
gizoth of G is the length of the shortest circuit in G. The adjaaenay ma.t:ri:r;
of G is the n x n matrix A, indexed by the vertices of G, defined by

1 if {x,y} is an edge,
(A) =
xy {0 otherwise.
80

Obviously, G has adjacency matrix J - A - I. The eigenvalues of G are the

eigenvalues of A; they are denoted by A1 (G) ~ ••• ~ An(G) (the eigenvalues
are real, because A is symmetric). We easily have that

The inaidence matri:x: N of G, whose rows are indexed by the vertices and
whose columns are indexed by the edges, is defined by

if x E EI

otherwise.

The graph with adjacency matrix N*N - 2I is called the line graph of G,
denoted by L (G) • The aubgrcr:ph of G induced by a set S of vertices of G is
the graph with vertex-set s, where two vertices are adjacent whenever they
are adjacent in G (a subgraph is always an induced subgraph) • Note that the
adjacency matrix of a subgraph of G is a principal submatrix of A. A alique
is a complete subgraph; a aoalique (or independent set of vertiaea) is a
void subgraph. The size of the largest clique and coclique is denoted by
w(G) and a(G), respectively. A aolou:t.>ing of G is a colouring of the verti-
ces, such that adjacent vertices have different colours (i.e. a partition
of the vertices into cocliques) • Vertices which are coloured with the same
colour form a aoloUP alass. G is k-aolo'UI'able if G admits a colouring with
k colours; the smallest possible value of k is the ahz>omatia numbeP of G,
denoted by y (G) • It easily follows that

y (G) ~ w (G) , y (G) a (G) ~ n •

If y(G) = 2, then G is bipa.I'tite. By use of 1.1.2 it follows that if G is

bipartite, then

Ai (G) = -1.n+l-i (G) for i = 1, ••• ,n.

Conversely, A1 {G} = -An{G) in!Plies that G is void or bipartite; this follows
from the Perron-Frobenius theorem on non-negative matrices (see [C12],
[M3]). If G can be coloured with y colours, such that all pairs of differ-
ently coloured vertices are edges, then G is aomplete y-paPtite (i.e. the
complement of the disjoint union of complete graphs). The complete bipartite
graph (Y = 2) is denoted by K.,,m
0 , where R. and m are the sizes of the two
colour classes. By use of 1.1.2 it follows that
 81

The de(JNJe (or valency) of a vertex is the number of vertices adjacent to

that vertex. A graph is l'egu'La:P (of degree d) if all its vertices have the
same degree (equal to d).
Let G be regular of degree d. Then Aj = dj, hence d is an eigenvalue
of G with eigenvector j. Moreover, the matrices A, J and I have a common
basis of eigenvectors. By use of this we obtain

If G is connected, then the Perron-Frobenius theorem yields that d is the

largest eigenvalue of G with multiplicity one. Hence, d = Al(G) and its
multiplicity equals the number of components of G (see (BS] for an element-
ary proof) • If G is connected and d = 2, then G is called an n-ayate (or
ail'auit}. G is strongty regula't' i f G is not void or complete and the ad-
jacency matrix A satisfies

(1) AJ=dJ,

2
for some number p 11 • This is equivalent to requiring that A has precisely
two distinct eigenvalues not belonging to the eigenvector j.
Now let G be strongly regular. Then also G is strongly regular. Com-
putation of the diagonal entries and the row sums of both sides of the
second equality of (1) yields

(2)

2 1 2
AJ=dJ, A = dI + p 11 A + p 11 CJ-A-I) •

This reflects that G satisfies the following three properties: G is regular

1
of degree d; for any pair of adjacent vertices there are exactly p 11 ver-
tices adjacent to both; for any pair of non-adjacent vertices there are
2 1 2
exactly p 11 vertices adjacent to both. The integers n, d, p 11 , p are the
11
panzmeters of G. Let f be the multiplicity of the eigenvalue A2 (G). Then
82

whence

d+ (n-1).\.n(G)
f= .\.h(G)-.\.2(G) ,

So, if n Y, 2f + 1, then .\.n (G} is rational, and therefore A-

2
(G) and .\.n (G) are
integers. It is an easy exercise (see [C6]) to show that n = 2f + 1 implies

By use of the above results and trace A= O, it follows that 2 (G) ~ 0 and
A-

.\.n(G) ~ -1. From (2) we have that A- (G) = 0 iff

2
p~l =d. It is easily seen
that p = d reflects that G is a complete y-partite graph. If .\.n(G) = -1,
11
then A- 2 (G) = O, hence G is the disjoint union of complete graphs and
2 2
p = O. conversely, by (1) p 11 = 0 implies d = >.. 2 (G} and .\.n{G) = -1. These
11
two families of strongly regular graphs are called imprimitive. Let x be a
vertex of G. The two subaonstituents of G with respect to x are the sub-
graphs of G induced by the vertices adjacent to x and by the vertices non-
1 1
adjacent to x. The subconstituents are regular of degree p 11 and d - p 11 ,
respectively. Examples of primitive (= not imprimitive) strongly regular
graphs are: the pentagon (5-cycle), L(Km) form~ 5 (= tPiangulaP gruph},
L{Km,m ) for m ~ 3 (= Zattiae graph) and their complements. The Petersen
gruph is the complement of L(K } and has parameters (10,3,0,1).
5

An inaidenae struature consists of a finite non-empty set v1 of points

and a finite non-empty set v2 of blocks, together with a subset of v1 x v2
of [lags. A point and a block are incident i f they form a flag. Often blocks
are identified with the sets of points with which they are incident; if so,
blocks are denoted by capitals, otherwise we use small types. An incidence
structure without flags is called empty.
Let D be an incidence structure with v points and b blocks. An inci-
dence structure D' formed by points and blocks of D is a substruatu:re of D
whenever a point and a block are incident in D' iff they are incident in D.
The inaidenae·mat:r>i:x: N of D, whose rows are indexed by the points and whose
columns are indexed by the blocks, is defined by

1 ifXEB,
(N) =
x,B {
0 otherwise.
 83

The .incidence structures with incidence matrices N* and J - N are called the
dual. and the aompl.ement of D, respectively. The graph with adjacency matrix
[~* ~] is the inaidence graph of D. Clearly D and its dual have the same
incidence graph. We call D a t - (v,k,A.) design (or t-design with paI'alfleters
(v,k,i)) if all blocks have size k, and if any set oft points is contained
in exactly }. blocks. A t - (v,k,>.) design with k < t or v-k < t is degenerate.
A design is a t-design for some t. A eubdesign is a substructure which is a
design. Note that we allow repeated blocks (two or more blocks incident
with exactly the same points).
Let D be a non-degenerate t - (v,k,>.) design. The complement of D is
also a non-degenerate t-design. By elementary counting we see that for t ~ 1
D is also a (t-1) - (v,k,}. (v-t+l) I (k-t+l)) design. In particular, D is a
1 - (v,k,r) design, where

r equals the number of blocks incident with any point. counting flags yields
bk = vr. If D is a 1 - (v,k,r) design, then rk is the largest eigenvalue of
NN* and hence by 1.1. 2 rrk is the largest singular value of N. A 2-design
is also called (bal.anaed incompZ.ete) bwak design. Now let D be a non-
degenerate 2 - (v,k,A.) design. In terms of the incidence matrix N this means

(3) N*J=kJ, NJ=rJ, NN*=A.J+(r-A)I.

NN* has eigenvalues A.v + r - ). = rk and r - ). of multiplicity 1 and v - 1,

respectively. By 1.1.2 these eigenvalues are the squares of the singular
value!il of N. From rk '# O, r - ). ttj. O, it follows that v = rank NN* = rank N,
hence b ~ v (Fisher's inequality). If b =v (i.e. r = k), Dis called
symmetria. Formula (3) yields N(N* - (A./r)J) = (r - >.)I. If D is symmetric, N
is square, hence (N* - (A./r)J)N = (r-A.)I, i.e. N*N = A.J + (r-A.)I, and
therefore the dual of D is a symmetric 2 - (v,k,A.) design as well. Let D be
symmetric, and let B be a block of D. The subdesign formed by the points
incident with.Band the blocks distinct from B is a 2 - (k,A.,A.-1) design
(possibly degenerate), called the derived design of D with respect to B.
Similarly, the subdesign formed by the points not incident with B and the
blocks distinct from Bis a 2- (v-k,k-A.,>.) design (possibly degenerate),
called the residual. design of D with respect to B. A 2- (v-k,k->.,A.) design
D' is embeddabZ.e in D if D' is a residual design of D. For example, the
84

points and lines of PG(2,q), the projective plane of order q, form a sym-
metric 2- (q2+q+1,q+1,1) design; the degenerate 2- (q+l,1,0) design is a
derived design, and the affine plane of order q, which is a 2 - (q 2 ,q,1)
design, is a residual design.
A pai>tiaZ. geometry with pal'al11eters (s,t,a), s,t,a E JN, is a
1 - (v,s+l,t+l) design satisfying the following two conditions:

any two blocks have at most one point in common;

(4) for any non-incident point-block pair (x,B) the number of
blocks incident with x and intersecting B equals a.

For a partial geometry we speak of lines rather than blocks. Let D denote a
partial geometry with parameters (s,t,a) and incidence matrix N. The graph
with adjacency matrix NN* - (t + 1) I is the point (J!'aph of D. The Une graph
of D is the point graph of the dual of D. From the definition it readily
follows that the point graph G of D is strongly regular with parameters

(v, s(t+l) ,t(a-1) +s-1,a(t+l)) •

By use of our identities for strongly regular graphs we obtain

t. 1 CG) = s(t+l), t. 2 (G) = s-a, /.n(G) = -t-1,

v = (s + 1) (st +a) /a , b = (t + 1) (st +a) /a •

Hence NN* has the eigenvalues (s + 1) (t + 1) , 0 and s + t + 1 - a of multiplicity

1, s(s+1-a)(st+a) /a(s+t+l-a) and st(s+1)(t+1) /a(s+t+1-a),
respectively. By 1.1.2 the square roots of these eigenvalues are the singular
values of N. A partial- subgeometry is a substructure, which itself is a
partial qeometry. Let D' be a partial subgeometry of D with parameters
(s',t',a). Then the point graph of D' is an induced subgraph of the point
graph of D. This can be seen as follows. Let x and y be two points of D' and
suppose there exists a line L of o, which is not a line of D', incident with
x and y. Let M be a line of D.' incident with x. By (4) there are a lines of
D' incident with x and intersecting M. Hence there are at least a+ 1 such
lines in D. This is a contradiction. so two points are on a line of D' iff
they are on a line of o, which proves the claim. A partial geometry with
parameters (s,t,1) is the same as a generoUzed quad:riangZ-e of order (s,t).
 85

APPENDIX II

TABLES

First we list the twentyone strongly regular graphs with parameters

(40,12,2,4), as promised in Section 6.2. Together with the ones of Examples
6.2.1 and 6.2.2 they form the twentyfour 40-graphs of Theorem 6.2.5. The
respective 5-tuples of the 40-graphs listed below are:

(0,0,0,36,4) (8, 18,2,9,3) (16,14,4,5,1)

(12, 18,6,3, 1) (4,20,4,10,2) (18,18,0,3,1)
(12, 12,0, 14, 2) (9,18,0,9,4) (18 I 20 I 0 I 0 r 2)
(6,22,6,3,3) (18, 12,9,0, 1) (27, 12,0,0, 1)
(0,32,0,0,8) (0, 12,0, 18, 10) (0,36,0,0,4)
(0,24,0, 12,4) (8,8,0,20,4) (6,12,12,10,0)
(2, 8, 12, 16, 2) (10,8, 18,4,0) (0, 16,6, 12,6) •

2 3 4 5 6 7 8 9 10 11 12 13 10!S Ul 17 21 22 24 2B 32 35 38 9 10 15 16 17 21 22 24 28 32 35 38
1 3 4 14 15 16 17 18 19 20 21 22 B 9 10 14 19 23 24 32 33 34 39 6 8 9 10 14 19 23 24 32 33 34 39
1 2 4 23 24 25 26 27 28 29 30 31 712 15 16 19 23 24 30 31 35 39 4 7 12 15 16 19 23 24 30 31 35 39
1 2 3 32 33 34 35 36 37 38 39 40 810 13 17 18 20 21 30 32 37 39 3 0 10 13 11 1e 20 21 30 32 31 39
t 6 7 14 17 20 23 27 31 32 37 39 912 14 18 20 21 22 28 31 34 39 7 9 12 14 10 20 21 22 28 31 34 39
I 5 7 15 18 21 24 2B 29 33 35 40 9 12 13 15 17 18 23 31 36 37 38 2 9 11 15 17 18 20 23 ll 35 36 37
I 5 6 16 19 22 25 26 30 34 36 38 5 10 11 14 16 18 24 33 36 37 38 3 5 10 11 14 16 18 24 33 36 37 38
1 9 10 14 17 20 25 26 30 33 35 40 4 II 15 IS 20 22 24 29 34 35 36 2 4 12 13 15 18 22 24 is 3:436 38
1 B 10 15 18 21 23 27 31 34 36 38 2 5 6 11 16 20 25 30 35 37 39 I 2 5 6 12 13 16 25 30 37 38 39
1 a 9 16 19 22 24 20 29 32 37 39 2 4 7 12 13 22 25 29 33 38 39 I 2 4 7 11 20 22 25 29 33 35 39
1 12 13 14 17 20 24 28 29 34 36 38 8 9 13 15 19 21 25 ;;a 30 33 36 6 7 10 13 15 19 21 25 28 30 34 36
I 11 13 IS IS 21 25 26 30 32 37 39 5 6 10 17 19 21 25 29 34 35 36 3 5 8 9 17 19 21 25 29 33 35 36
1 11 12 16 19 22 23 27 31 33 35 40 6 10 11 16 19 22 26 28 31 34 37 4 9 11 ~16 19 22 26 28 31 33 37
2 5 8 II IS 16 23 28 30 32 36 40 5 7 15 17 19 22 26 29 30 32 37 2 7 15 17 19 22 26 29 30 32 37
2 6 9 12 14 16 24 26 31 33 37 38 3 6 8 11 14 21 26 29 31 38 39 1 6 B 11 14 21 26 29 31 38 39
2 7 10 13 14 15 25 27 29 34 35 39 3 7 9 13 20 23 26 29 32 34 36 1 7 9 13 20 23 26 29 32 34 36
2 5 6 11 19IS 24 26 31 34 35 39 4 6 12 14 20 24 26 28 30 33 36 l 6 12 14 20 24 26 28 30 33 36
2 6 9 12 1917 25 27 29 32 36 40 s s 1 a 23 25 26 20 32 35 38 4 5 8 7 8 23 25 26 28 32 35 38
2 7 10 13 1817 23 28 30 33 37 38 3 11 12 13 14 20 27 28 32 35 38 2 3 11 12 13 14 20 27 28 32 35 38
2 5 s 11 2221 25 27 29 33 37 38 5 8 9 16 17 19 27 29 31 33 38 4 5 6 10 16 17 19 27 29 31 34 38
2 6 9 12 2220 23 28 30 34 35 39 4 5 11 12 15 23 27 32 33 34 37 4 5 11 12 15 23 27 32 33 34 37
2 7 10 13 2120 24 26 31 32 36 40 5 a 10 u 1• 23 21 30 31 35 36 5 8 10 13 14 23 27 30 31 35 36
3 5 9 13 1914 21 24 25 32 35 38 3 6 16 IS 21 22 27 28 29 30 33 3 6 16 1S 21 22 27 28 29 30 33
3 6 10 11 1715 22 39
23 25 33 36 l 2 3 1 a 11 25 21 28 31 34 31 1 2 3 7 8 17 25 27 28 31 34 37
3 1 a 12 16 18 20 23 24 34 37 40 9 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 1e 24 26 21 29 30 31 32
3 1 a 12 1s 11 22 21 2s 32 35 38 13 14 15 16 17 16 25 27 33 34 35 39 u 14 1s 16 11 1e 25 21 33 34 Js 39
3 s 9 13 16 18 20 26 28 33 36 39 19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39
3 6 10 11 14 19 21 26 27 34 37 40 1 s 11 13 11 ia t9 23 24 29 39 40 1 5 11 13 17 18 19 23 24 29 39 40
3 6 10 11 16 18 20 30 31 32 35 38 B 10 12 14 15 16 20 23 25 28 37 40 e 10 12 14 1s 16 20 23 25 .za31 40
3 7 8 12 14 19 21 29 31 33 36 39 3 4 9 11 14 17 22 23 25 34 38 40 3 4 9 11 14 17 22 23 25 34 38 40
3 5 9 13 15 17 22 29 30 34 37 40 3 5 6 13 15 20 22 24 25 32 33 40 3 5 6 13 15 20 22 24 25 32 ll 40
4 5 10 12 14 18 22 23 26 29 33 34 1 2 4 14 16 10 19 21 25 31 36 40 I 2 4 14 16 18
19 21 25 31 36 40
4 6 8 13 15 19 20 24 27 30 32 34 2 7 10 11 17 20 21 23 26 31 35 40 2 7 10 12 13 17
21 23 26 31 38 40
4 7 9111617212528313233 2 5 8 12 13 16 21 24 26 30 30 40 2 5 8 11 16 20
21 24 26 30 35 4()
4 6 8 13 16 17 21 23 26 29 36 37 1 3 8 9 12 10 19 22 26 33 37 40 1 3 6 10 12 18
19 22 26 34 37 4()
4 7 9 11 14 18 22 24 27 30 35 37 6 8 II 12 16 17 22 27 32 39 40 6 7 8 11 12 16
17 22 27 32 39 40
4 5 10 12 15 19 20 25 28 31 35 36 4 7 9 13 14 21 24 27 29 35 40 6 7 9 13 14
21 24 27 29 35 40
4 7 9 11 1S 19 20 23 26 29 39 40 1 7 10 15 18 19 20 27 30 34 40 7 a 9 15 rn 19 20 21 30 n 40
4 5 10 12 16 17 21 24 27 30 38 4() 2 3 4 5 9 10 15 26 27 28 36 40 2 3 4 5 9 10 15 26 27 28 36 40
4 6 S 13 14 18 22 25 2B 31 38 39 20 29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39
86

9 10 15 16 17 21 22 24 28 32 35 3B 9 10 15 16 17 21 22 24 28 32 35 38 9 10 15 16 17 21 22 24 20 32 35 38
7 8 9 14 15 19 23 24 30 32 33 39 7 8 9 12 IS 24 30 33 35 39 6 7 10 12 15 19 23 24 30 34 35 39
4 6 10 12 16 19 23 24 31 34 35 39 6 10 14 16 24 31 32 34 39 8 9 14 16 19 23 24 31 32 33 39
3 8 10 13 17 18 20 21 30 32 37 39 8 10 II 17 18 20 21 30 35 37 39 8 10 11 17 18 20 21 30 35 37 39
7 9 12 14 18 20 21 22 28 31 34 39 9 12 14 18 20 21 22 28 31 34 39 7 9 12 14 18 20 21 22 28 31 34 39
3 9 !1 14 15 17 18 23 31 36 37 38 9 12 13 15 17 18 23 31 36 37 38 2 9 13 15 17 18 20 23 31 32 36 37
2 5 10 12 13 16 18 24 33 36 37 38 5 10 11 14 16 IS 24 33 36 37 38 2 s 10 11 14 16 18 24 33 36 37 38
2 -.4 11 15 18 20 22 24 29 34 35 36 4 13 15 18 20 22 24 29 32 34 36 3 • 12 13 15 18 22 24 29 34 36 38
1 2 5 6 11 16 20 25 30 35 37 39 2 5 6 13 16 2() 25 30 32 37 39 1 3 5 6 12 13 16 25 30 37 38 39
l 3 4 7 11 14 22 25 29 33 38 39 3 4 7 12 13 22 25 29 33 38 39 1 . 2 4 7 13 20 22 25 29 32 33 39
6 a a 10 13 19 21 25 28 33 34 36 7 13 15 16 19 21 25 28 30 31 36 4 7 13 15 16 19 21 25 28 30 31 36
3 5 7 15 17 19 21 25 29 30 35 36 5 6 10 17 19 2l 25 29 34 35 36 2 8 9 17 19 21 25 29 33 35 36
• 7 11 15 16 19 22 26 28 30 31 37 g 9 10 11 19 22 26 28 33 34 37 6 9 10 H 19 22 26 28 33 34 37
2 5 6 10 17 19 22 26 29 32 34 37 5 7 15 17 19 22 26 29 JO 32 37 3 7 15 17 19 22 26 29 30 32 37
1 2 6 8 12 13 21 26 29 31 38 39 2 6 8 11 14 21 26 29 31 38 39 l 6 8 11 14 21 26 29 31 38 39
1 3 7 9 13 20 23 26 29 32 34 36 3 7 9 11 20 23 26 29 34 35 36 1 7 9 11 20 23 26 29 34 35 36
6 12 14 20 24 26 28 30 33 36 4 6 12 14 20 24 26 28 30 33 36 l 4 6 12 14 20 24 26 28 30 33 36
6 7 8 23 25 26 28 32 35 38 • 5 6 1 8 23 2s 26 2a 32 35 38 • 5 6 1 a 23 2s 26 28 32 35 38
1l 12 13 14 20 27 26 32 35 38 2 3 11 12 13 14 20 27 26 32 35 38 2 3 11 12 13 14 20 27 28 32 35 38
a a 16 17 19 21 29 31 33 38 • 5 8 9 16 17 19 27 29 31 33 30 4 5 6 10 16 17 19 27 29 31 34 38
4 s 11 12 15 23 27 32 33 34 37 1 4 s 11 12 15 23 21 32 33 34 37 1 4 5 11 12 15 :13 27 32 33 34 37
s a 10 13 1• 23 21 30 31 35 36 1 s 8 10 13 14 n 21 30 31 35 36 I 5 8 10 13 14 23 27 30 31 35 36
3 6 16 !B 21 22 27 28 29 30 33 2 3 6 16 18 21 22 27 28 29 30 33 2 3 6 16 1e 21 22 21 20 29 30 33
1 2 3 7 8 17 25 27 28 31 34 37 1 2 3 7 8 17 25 27 28 31 34 37 1 2 3 7 8 17 25 27 28 31 34 37
9 JO 11 12 IS 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32
13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 39
19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39
1 5 11 13 17 18 19 23 24 29 39 40 1 5 11 13 17 18 1'9 23 24 29 39 40 1 5 11 13 17 18 19 23 24 29 39 40
8 10 12 14 15 16 20 23 25 28 37 40 8 10 12 14 15 16 20 23 25 2S 37 40 a 10 12 14 15 10 20 23 2s 2s 37 40
2 4 g 12 13 17 22 23 25 34 38 40 2 4 9 11 14 17 22 23 25 34 38 40 2 4 9 11 14 17 22 23 25 34 38 40
3 5 6 13 15 20 22 24 25 32 33 40 3 5 6 11 15 20 22 24 25 33 35 40 3 s 6 11 15 20 22 24 25 33 35 40
1 2 4 14 16 18 19 21 25 31 36 40 1 3 8 9 14 18 19 21 25 33 36 40 l 3 6 10 14 18 19 21 25 34 36 40
2 7 10 11 17 20 21 23 26 31 35 40 2 7 10 13 17 20 21 23 26 31 32 40 3 7 10 12 13 17 21 23 26 31 38 40
3 s a 11 14 16 21 24 26 30 3e 40 3 s a 12 13 16 21 24 26 30 Ja 40 2 s 8 13 16 20 21 24 26 30 32 40
1 3 8 9 12 18 19 22 26 33 37 40 4 12 16 1B 19 22 26 31 37 40 l 4 12 16 18 19 22 26 31 37 40
6 1 8 11 12 16 17 22 27 32 J9 40 a 11 12 16 11 22 21 32 39 40 6 a 11 12 16 11 22 27 32 39 40
4 6 7 9 13 14 21 24 27 29 35 40 7 9 13 14 21 24 27 2:9 35 40 4 7 9 13 14 21 24 27 29 35 40
1 6 7 10 15 18 19 20 27 30 34 40 7 10 15 18 19 20 27 30 34 40 1 8 9 15 18 19 20 27 30 33 40
2 3 4 5 9 10 15 26 27 28 36 40 2 3 4 5 9 10 1 s 26 27 28 36 40 2 3 4 5 9 10 15 26 27 28 36 40
28 29 30 31 32 33 34 35 36 37 38 39 26 29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39

9 10 15 16 17 21 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 35 38
6 8 9 10 12 19 23 24 33 34 35 39 4 7 9 10 15 16 23 24 32 33 37 39 s 7 9 10 15 22 23 24 32 33 34 39
7 14 15 16 19 23 24 30 31 32 39 4 8 11 12 14 19 23 24 30 34 35 39 4 a 11 12 14 19 23 24 Jo 34 35 39
a 10 12 13 11 10 21 30 31 3S 39 2 3 11 13 17 18 21 22 31 32 38 39 3 e 10 13 11 rn
21 22 31 32 36 39
9 12 14 18 20 21 22 28 31 34 39 1 a 10 12 13 11 10 20 30 36 37 39 2 7 11 12 13 17 18 20 30 37 38 39
9 13 15 17 18 20 23 31 32 36 37 10 11 14 15 18 20 21 23 28 31 34 37 10 11 14 15 18 20 21 23 28 31 34 37
5 10 11 16 18 20 24 33 35 36 37 2 5 14 16 18 19 21 24 29 31 35 36 . 2 5 14 16 18 19 21 24 29 31 35 36
4 11 14 15 18 22 24 29 34 36 38 3 5 9 15 18 20 22 24 29 33 34 38 3 4 9 15 16 18 20 24 29 33 37 38
2 5 6 11 14 16 25 30 37 38 39 2 8 10 11 13 14 19 25 28 33 36 38 2 8 10 11 13 14 19 25 28 33 36 38
2 4 7 13 20 22 25 29 32 33 39 2 5 6 9 21 22 25 30 34 35 39 2 4 6 9 16 21 25 30 35 37 39
8 9 13 15 19 21 25 28 JO 33 36 4 6 9 16 17 20 25 29 35 36 37 5 6 9 17 20 22 25 29 34 35 36
4 5 16 17 19 21 25 29 31 35 36 3 5 13 14 21 23 25 29 32 33 37 3 s 13 14 21 23 25 29 32 33 37
6 10 11 16 19 22 26 28 31 34 37 5 9 12 16 19 22 26 28 31 34 37 5 9 12 16 19 22 26 28 31 34 37
5 8 9 17 19 22 26 29 32 33 37 3 6 7 9 12 15 26 31 36 38 39 3 6 7 9 12 IS 26 31 36 38 39
3 6 8 l1 20 21 26 29 31 35 39 6 a 14 11 19 22 26 29 30 32 37 6 8 14 17 19 22 26 29 30 32 37
3 7 9 12 13 23 26 29 34 36 38 2 7 11 13 20 23 26 29 30 34 38 1 a 10 13 20 23 26 29 30 34 36
• 6 12 14 20 24 26 28 30 33 36 4 5 11 15 21 24 26 28 30 33 36 4 5 11 15 21 24 26 28 30 33 36
5 6 7 a 23 25 26 28 32 3S 3& 5 6 7 a 23 2S 26 28 32 35 38 5 6 7 B 23 25 26 28 32 35 38
3 1l 12 13 14 20 27 28 32 JS 38 l 7 9 13 15 20 21 27 2S 30 32 35 7 9 13 15 20 21 27 28 30 32 35
6 7 10 15 17 19 27 29 30 34 38 1 s 6 8 !I 16 19 27 31 32 33 39 5 6 8 11 16 19 27 31 32 33 39
4 5 11 12 15 23 27 32 33 34 37 4 6 7 10 12 17 19 27 29 33 34 38 6 7 10 12 17 19 27 29 33 34 38
5 8 10 13 14 23 27 30 31 35 36 1 4 a 10 u 15 23 21 29 31 Js 36 2 4 11 13 15 23 27 29 31 35 38
3 6 16 18 21 22 27 28 29 30 33 2 3 6 12 16 18 22 27 28 30 33 36 3 6 12 16 18 22 27 28 30 33 36
1 2 3 1 a 11 25 21 2a 31 34 31 1 2 3 7 8 17 25 27 28 31 34 37 1 2 3 / a 17 2s 21 2e 31 34 37
9 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32
13 14 15 16 17 !B 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 -39
19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39
1 s 11 13 17 18 19 23 24 29 39 40 1 6 9 13 17 18 19 23 24 29 39 40 1 6 9 13 17 18 19 23 24 29 39 40
a 10 12 14 1s 16 20 23 2s 2s 37 40 7 8 II 12 15 16 21 22 25 28 39 40 7 8 11 12 15 16 21 22 25 28 39 40
3 4 9 11 17 20 22 23 25 34 35 40 3 s 10 15 16 17 19 23 25 31 38 40 S 10 IS 16 17 19 23 25 31 38 40
3 5 6 12 13 15 22 24 25 33 38 40 6 7 13 14 20 22 24 25 30 33 40 6 7 13 14 20 22 24 25 30 33 40
1 3 6 10 14 18 19 21 25 34 36 40 2 4 12 15 18 19 20 25 34 36 40 2 4 12 15 18 19 20 25 34 36 40
2 7 10 11 14 17 21 23 26 31 38 40 s 9 12 11 20 21 23 26 :n JS 40 a 9 12 11 20 21 23 26 31 35 40
2 s a D 16 20 21 24 26 Jo 32 40 6 8 10 13 16 21 24 26 32 36 40 6 11 13 16 21 24 26 32 38 40
2 7 12 15,18. 19 22 26 30 37 40 3 7 10 11 18 19 22 26 33 37 40 7 10 11 lS 19 22 26 33 37 40
7 8 11 12 16 17 22 27 32 39 40 7 9 11 14 17 22 23 27 32 34 40 4 9 11 14 16 17 23 27 32 37 40
6 7 9 13 14 21 24 21 29 35 40 S 6 11 12 13 1S 24 27 35 38 40 5 8101213152427353640
4 a 9 16 1a 19 20 21 31 33 40 4 8 9 14 16 1B 21 27 30 37 40 '1 8 9 14 18 21 22 27 30 34 40
2 3 4 5 9 10 15 26 27 28 36 40 3 4 5 10 14 20 26 27 26 2:9 40 2 4 5 10 14 20 26 27 28 29 40
26 .29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39
 87

10 12 14 16 17 20 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 35 38
5 7 10 15 16 19 23 24 30 32 34 39 5 7 10 15 16 19 23 .2430 31 35 39 s 7 10 15 16 19 23 24 31 32 33 39
4 8 11 12 14 19 23 24 31 33 35 39 4 8 11 12 14 19 23 24 32 33 34 39 4 8 11 12 14 19 2330 24 34 35
39
l 0 10 13 17 18 21 22 30 32 37 39 3 8 10 1.3 17 18 21 22 30 31 3$ 39 3 a 10 13 11 rn 2131 22 32 36
39
2 7 II 12 13 17 18 20 31 36 38 39 2 7 11 12 13 17 18 20 32 36 37 39 7 11 12 13 17 1830 20 37 38
39
10 11 14 15 18 20 21 23 28 31 34 37 10 11 14 15 18 20 21 23 28 31 34 37 w 11 14 15 18 20 2128 23 31 34
37
2 5 9 14 18 21 22 24 29 33 35 37 2 5 9 14 18 21 22 24 29 33 34 38 2 5 9 14 18 21 2229 24 34 35
36
3 '4 9 15 16 18 20 24 29 34 36 38 3 4 9 15 16 18 20 24 29 35 36 37 3 4 9 15 16 1S 2029 24 33 37
38
7 8 10 13 14 19 20 25 28 30 33 36 7 8 10 13 14 19 20 25 28 30 33 36 7 8 10 1J 14 19 20 25 2$ 30 33 36
I 2 4 6 9 12 15 25 35 '36 37 39 4 6 9 11 16 25 34 36 38 39 l' 4 6 9 12 15 25 35 36 37 39
3 5 6 14 16 1'i 19 25 29 32 36 37 6 10 17 19 22 25 29 33 35 37 3 6 14 16 17 19 25 29 32 36 37
I 3 5 10 13 21 23 25 29 33 34 JS 5 13 14 15 23 25 29 31 36 38 1 s 10 13 21 23- 25 29 33 34 38
4 5 9 12 16 19 22 26 2B 31 34 37 9 12 16 19 22 26 2a 31 34 37 4 12 16 19 22 26 28 31 34 37
I 3 6 7 9 11 22 26 30 34 38 39 3 6 7 9 12 21 26 30 35 37 39 7 9 11 22 26 31 33 38 39
2 6 8 10 17 19 22 26 29 31 33 3B 6 8 12 17 19 20 26 29 30 34 3B a 10 11 19 22 26 29 30 34 Ja
1 2 8 11 13 20 23 26 29 30 35 37 2 8 10 13 21 23 26 29 32 33 37 B 11 13 20 23 26 29 31 35 36
I 4 5 11 1S 21 24 26 28 30 33 3£ 4 5 II 15 21 24 26 28 30 33 36 5 11 15 21 24 26 28 3(} 33 36
4 5 6 7 a 23 25 26 20 32 35 3B 5 6 7 a 23 25 26 28 32 35 38 6 7 8 23 25 26 28 32 35 38
2 3 9 II 13 15 21 27 28 32 35 38 3 9 11 13 15 21 27 28 32 35 38 9 11 13 15 21 27 28 32 JS 38
1 5 6 a 9 16 21 27 31 32 33 39 5 6 8 9 15 22 27 31 32 33 39 6 a9 16 21 27 30 32 34 39
4 6 7 12 17 19 20 27 29 30 34 35 6 7 14 16 17 19 27 29 31 32 36 7 12 17 19 20 27 29 31 .)3. 35
l 4 7 13 14 15 23 27 29 31 32 36 4 7 11 13 20 23 27 29 30 34 35 7 13 14 15 23 27 29 30 32 37
2 3 6 12 16 18 22 27 28 30 33 36 2 3 6 12 16 18 22 27 28 30 33 36 6 12 16 18 22 27 28 30 33 36
1 2 3 7 s 11 2s z1 za 31 34 31 I 2 3 7 8 17 25 27 28 31 34 37 1 2 3 7 8 17 25 27 28 31 34 37
9 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 IB 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32
13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 JS 39
19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 2636 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39
1 6 9 13 17 10 19 23 24 29 39 40 l 6 9 13 17 16 19 23 24 29 39 40 I 6 9 13 17 18 15} 23 24 29 39 40
7 a 11 12 1s 16 21 22 2s w 39 •o 7 s 11 12 1$ 16 2:1 42 25 28 39 40 7 S 11 12 1S 16 21 22 25 28 39 40
2 4 9 14 16 17 21 23 25 31 38 40 2 4 9 14 15 17 22 23 25 32 37 40 5 9 15 17 20 22 23 25 31 35 40
3 5 6 13 15 20 22 24 25 30 35 40 2 4 6 12 13 20 21 24 25 33 35 40 4 6 13 14 16 21 24 25 30 38 40
1 2 4 II IS 19 20 22 25 33 34 40 1 3 5 t6 ta 19 20 21 25 3o 34 40 2 4 11 18 19 20 22 25 33 34 40
3 7 9 12 15 17 20 23 26 32 37 40 3 7 9 11 16 17 20 23 26 31 38 40 a 9 12 14 17 21 23 26 32 37 40
2 6 0 12 13 u 21 24 26 32 36 40
l 3 1 10 16 18 19 21 26 31 36 40
5 a 9 10 ll 17 22 23 27 34 35 40
6 7
2 8
8 9
10
11
10
13
14
12
15
18
17
22
19
21
24 26
22 26
23 27
32
31
34
36
36
35
40
40
40
1 3
6
; :~ :~ :~ 17~ 23;~ ;}
7 9 10 11 16
i~ ~ :~
27 34 38 40
4 6 7 10 11 13 16 24 27 33 38 40 6 8 11 13 14 16 24 27 30 30 40 6 8 10 11 13 22 24 27 33 35 40
1 5 8 12 14 15 18 19 27 30 37 40 4 7 10 12 15 18 19 27 33 37 40 5 a 12 14 1s rn 19 21 31 36 40
2 3 4 s 10 14 20 26 27 28 29 40 2 3 4 5 10 14 20 26 27 28 29 40 2 3 4 5 10 14 20 26 27 28 29 40
28 29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39 2B 29 30 31 32 33 34 35 36 37 38 39

10 12 14 16 17 20 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 35 38 10 12 14 16 17 20 22 24 28 32 :JS 30
4 5 12 15 16 19 23 24 31 32 34 39 4 7 10 11 12 19 23 24 33 34 35 39 5 7 10 15 16 19 23 .24 31 32 33 39
7 8 10 11 14 19 23 24 30 33 35 39 5 8 14 15 16 19 23 24 30 31 32 39 4 8 11 12 14 19 23 24 30 34 35 39
2 S 10 11 13 17 IS 22 31 36 38 39 2 8 10 13 17 18 21 22 31 32 36 39 3 8 10 13 17 1B 21 22 31 32 36 39
2 7 9 12 17 18 20 21 30 35 37 39 3 7 11 12 13 17 18 20 30 37 38 39 2 7 11 12 13 17 1B 20 30 37 38 39
10 11 14 15 18 20 21 23 28 31 34 37 10 II 14 15 18 20 21 23 28 31 34 37 10 11 14 16 18 20 21 23 28 31 34 37
3 5 9 14 1S 1e 22 24 29 34 36 38 2 5 9 14 15 18 22 24 29 34 36 38 2 5 9 14 18 21 22 24 29·34 35 36
3 4 1l 16 18 20 21 24 29 32 33 37 3 4 9 16 18 20 21 24 29 33 35 37 4 9 15 16 18 20 24 29 33 37 38
5 7 10 13 16 19 20 25 28 31 33 36 7 a 10 u t4 19 20 25 2s 3o 33 36 8 10 13 14 19 20 25 28 30 33 36
1 3 4 6 9 15 16 25 30 37 38 39 1 2 4 6 9 15 16 25 30 31 38 39 2 4 6 9 16 21 25 30 35 37 39
3 4 6 12 17 19 20 25 29 34 35 36 2 5 6 16 17 19 2() 25 29 31 35 36 5 6 16 17 19 20 25 29 31 35 36
1 2 5 11 13 14 23 25 29 33 37 38 1 2 5 13 14 21 23 25 29 32 33 37 3 5 13 14 21 23 25 29 32 33 37
4 a 9 12 14 19 22 26 28 30 34 37 5 9 12 16 19 22 26 28 31 34 37 5 9 12 16 19 22 26 28 31 34 37
1 3 6 7 12 13 21 26 31 32 36 39 3 6 7 9 12 21 26 31 35 36 39 3 6 7 9 12 15 26 31 36 38 39
2 6 7 10 17 19 22 26 29 32 33 37 6 7 10 17 19 22 26 29 32 33 37 6 a 14 11 19 22 26 29 30 32 37
1 2 8 9 10 21 23 26 29 34 35 36 3 0 10 11 13 23 26 29 3• 36 3B 0 10 11 13 23 26 29 34 36 38
1 4 5 11 15 21 24 26 28 30 33 36 4 5 11 15 21 24 26 28 30 33 36 5 11 15 21 24 26 28 30 33 36
4 5 6 7 a 23 25 26 2s 32 35 Ja s 6 7 823252628323538 6 7 s 23 25 26 26 32 35 38
2 3 9 11 13 15 21 27 28 32 35 38 3 9 11 13 15 21 27 28 32 35 38 9 11 13 15 21 27 28 32 35 38
1 5 6 8 9 11 22 27 32 33 34 39 6 a 9 11 22 27 32 33 34 39 6 s 9 11 22 27 32 33 34 39
5 6 a 14 16 11 19 21 29 30 31 38 8 12 14 17 19 27 29 30 34 38 7 10 12 17 19 21 29 33 34 38
1 7 13 15 20 23 27 29 30 31 35 7 13 15 20 23 27 29 30 31 35 7 13 15 20 23 27 29 30 31 35
2 6 12 16 18 22 27 28 30 33 36 6 12 16 18 22 27 28 3<l 33 36 6 12 16 18 22 27 2S 30 33 36
1 3 7 8 17 25 27 28 31 34 37 2 3 7 a 11 2s 21 28 31 34 37 1 2 3 7 8 17 25 27 28 31 34 37
9 10 11 12 18 24 26 27 29 30 31 32 10 11 12 18 24 26 27 29 30 31 32 9 10 11 12 18 24 26 27 29 30 31 32
13 14 15 16 17 18 25 27 33 34 35 39 13 14 15 16 17 18 25 27 33 34 35 39 u 14 15 16 11 1a 25 21 33 34 35 39
19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39 19 20 21 22 23 24 25 26 36 37 38 39
I 6 9 13 17 18 19 23 24 29 39 40 1 6 9 13 17 18 19 23 24 29 39 4Q I 6 9 13 17 18 19 23 24 29 39 40
7 B 11 12 15 16 21 22 25 28 39 40 7 8 11 12 15 16 21 22 25 28 39 40 7 8 11 12 IS 16 21 22 25 28 39 40
3 5 10 13 17 21 22 23 25 32 34 40 3 5 9 10 17 21 22 23 25 34 35 40 3 9 10 15 17 22 23 25 34 3$ 40
2 4 8 9 14 21 22 24 25 33 35 40 3 ~ 6 11 13 14 22 24 25 33 38 40 2 6 11 13 14 22 24 25 33 38 40
I 2 8 14 15 18 19 20 25 30 36 40 1 3 4 12 15 18 19 20 25 34 36 40 I 4 12 15 18 19 20 25 34 36 40
3 a 9 12 15 17 20' 23 26 31 38 40 2 8 9 12 15 17 20 23 26 31 38 40 2 9 12 17 20 21 23 26 31 35 40
2 6 7 11 13 16 20 24 26 30 38 40 6 7 13 16 20 21 24 26 30 32 40 3 7 13 16 20 21 24 26 30 32 40
I 3 s 11 16 IS 19 22 26 31 37 40 2 8 11 14 18 19 22 26 30 37 40 7 10 11 18 19 22 26 33 37 40
7 9 11 14 16 17 23 27 32 37 40 7 9 11 14 16 17 23 27 32 37 40 9 11 14 16 17 23 27 32 37 40
6 8 10 12 13 15 24 27 35 36 40 6 8 10 12 13 15 24 27 35 36 40 6 8 10 12 13 15 24 27 35 36 40
4 7 10 12 18 19 21 27 33 34 40 5 7 10 16 18 19 21 27 31 33 40 1 5 8 14 16 18 19 21 27 30 31 40
2 3 4 5 10 14 20 26 27 28 29 40 2 3 4 5 10 14 20 26 27 28 29 40 2 3 4 5 10 14 20 26 27 28 29 40
28 29 JO 31 32 33 34 35 36 37 38 39 28 29 30 31 32 H 34 35 36 37 JS 39 28 29 30 31 32 33 34 35 36 37 39 39
88

9 10 12 17 20 21 23 .26 27 30 31 9 10 12 17 20 21 23 26 27 30 31 s 9 10 12 17 20 22 23 25 27 30 31
" 6 B 12 lS 20 22 24 25 27 29 32 6 11 12 18 2:0 22 24 25 27 29 31 5 7 '8 10 18 20 21 24 26 27 29 32
9 10 11 18 19 22 23 25 28 29 31 8 9 10 18 19 22 23 25 28 29 32 6 9 11 12 18 19 21 23- 26 28 29 31
6 7 11 17 19 21 24 26 28 30 32 6 7 11 17 19 21 24 26 28 30 32 5 6 7 11 17 19 22 24 25 28 30 32
4 10 12 14 1 s 22 23 26 28 33 35 4 10 12 14 15 22 23 26 28 33 35 2 4 10 12 14 15 22 23 26 28 33 35
4 9 11 13 16 18 20 30 31 33 35 4 8 9 13 16 18 20 30 32 33 35 3 4 8 9 13 15 18 20 30 32 33 35
4 8 9 14 15 17 20 29 32 34 35 4 9 11 14 lS 17 20 29 31 34 35 2 4 9 11 14 16 17 20 29 31 34 35
-;, 7 10 13 16 21 24 25 28 34 35 3 6 12 13 15 21 24 25 28 34 35 2 6 12 13 16 21 24 25 28 34 35
3 6 7 14 16 22 24 26 27 33 36 3 6 7 14 16 22 24 26- 27 33 36 3 6 7 14 16 22 24 26 27 ..33 J6
3 5 a 13 1s 1a 19 30 J2 33 36 3 5 11 13 15 18 19 30 31 33 36 . 2 5 11 13 16 18 19 30 31 33 36
4 6 12: u 15 21 23 25 27 34 36 4 7 10 13 16 21 23 25 27 34 36 4 7 10 13 15 21 23 25 27 34 36
1 2 5 11 14 16 17 19 29 31 34 36 l 2 5 8 14 16 17 19 29 32 34 36 1 3 5 8 14 15 17 19 29 32 34 36
6 8 10 11 17 22 26 29 35 36 37 38 6 a 10 11 11 22 26 29 Js 36 31 30 6 a io 11 11 u 26 29 35 36 37 3s
5 7 9 12 18 21 25 30 35 36 37 38 s 7 9 12 lS 21 25 30 35 36 37 38 5 7 9 12 18 21 25 30 35 36 37 38
5 7 10 11 20 24 27 31 33 34 37 38 5 7 B 10 20 24 27 32 33 34 37 38 5 6 11 12 20 24 27 31 33 34 37 38
6 8 9 12 19 23 28 32 33 34 37 38 6 9 11 12 19 23 28 31 33 34 37 38 7 8 9 10 19 23 28 32 33 34 37 38
l 4 7 12 13 22 25 31 32 33 37 39 1 4 7 12 13 22 25 31 32 33 37 39 1 4 7 12 13 21 26 31 32 33 37 39
2 3 6 10 14 21 26 31 32 34 37 39 2 3 6 10 14 21 26 31 32 34 37 39 2 3 6 10 14 22 25 31 32 34 37 39
3 4 10 12 16 24 27 29 30 35 37 39 3 4 10 12 16 24 27 29 30 35 37 39 3 4 10 12 16 24 27 29 30 35 37 39
1 2 6 7 15 23 2B 29 30 36 37 39 2 6 1 15 23 28 29 30 36 37 39 l 2 6 7 1s 23 28 29 30 36 37 39
l 4 8 11 14 18 27 28 29 33 37 40 4 a 11 14 10 21 20 29 JJ 37 40 2 3 a 11 u 11 21 20 JO 33 31 40
2 3 5 9 13 17 27 28 30 34 37 40 3 5 9 13 17 27 28 30 34 37 40 4 5 9 13 18 27 28 29 34 37 40
1 3 5 11 16 20 25 26 32 35 37 40 3 5 11 16 20 25 26 32 35 37 40 3 5 11 16 20 25 26 32 35 37 40
2 4 8 9 15 19 25 26 31 36 37 40 4 8 9 15 19 25 26 31 36 37 40 8 9 15 19 25 26 31 36 37 40
2 3 8 11 14 17 23 24 30 33 38 39 3 8 11 14 17 23 24 30 33 38 39 1 8 11 14 18 23 24 29 33 38 39
1 4 5 9 13 18 23 24 29 34 38 39 4 5 9 13 18 23 24 29 34 38 39 2 5 9 13 17 23 24 30 34 38 39
1 2 9 11 15 19 21 22 32 35 38 39 2 9 11 15 19 21 22 32 35 38 39 I 9 11 15 19 21 22 32 35 38 39
3 4 5 8 16 20 21 22 31 36 38 39 4 5 8 16 20 21 22 31 36 38 39 3 5 e 16 20 21 22 31 36 38 39
2 3 7 12 13 19 20 21 26 33 38 40 3 7 12 13 19 20 21 26 33 38 40 2 7 12 13 19 20 22 25 33 38 40
1 4 6 10 14 19 20 22 25 34 38 40 6 10 14 19 20 22 25 34 38 40 I 4 6 10 14 19 20 21 26 34 38 40
1 3 6 12 15 17 18 24 28 35 38 40 7 10 16 17 18 24 28 35 38 40 1 3 7 10 15 17 18 24 28 35 38 40
2 4 7 10 16 17 18 23 27 36 38 40 6 12 15 17 18 23 27 36 38 40 2 4 6 12 16 17 18 23 27 36 3B 40
5 6 9 10 15 16 17 21 25 29 39 40 6 9 10 15 16 17 21 25 29 39 40 5 6 9 10 15 16 17 21 25 29 39 40
7 8 11 12 15 16 19 22 26 30 39 40 8 11 12 15 16 18 22 26 30 39 40 7 8 11 12 15 16 19 22 26 30 39 40
5 6 7 s 13 14 19 23 27 31 39 40 5 6 7 3 13 14 19 23 27 31 39 40 5 6 7 e u 14 19 23 21 31 39 40
9 10 11 12 13 14 20 24 28 32 39 40 9 10 11 12 13 14 20 24 28 32 39 40 9 10 u 12 13 14 20 24 w 32 39 40
13 14 15 16 l7 18 19 20 21 22 23 24 13 14 15 16 17 18 19 20 21 22 23 24 13 14 15 16 17 l8 19 20 21 22 23 24
13 14 15 16 25 26 27 28 29 30 31 32 13 14 15 16 25 26 27 28 29 30 31 32 13 14 15 16 25 26 27 28 29 30 " 32
17 18 19 20 25 26 27 28 33 34 35 36 17 18 19 20 25 26 27 28 33 34 35 36 17 18 19 20 25 26 27 28 33 34 35 36
21 22 23 24 29 30 31 32 33 34 35 36 21 22 23 24 29 30 31 32 33 34 35 36 21 22 23 24 29 30 31 32 33 34 35 "'

a 9 10 12 17 20 22 23 25 27 JO 31 8 9 10 12 18 20 21 23 26 27 29 31 a 9 10 12 lB 20 22 23 25 27 29 31
6 11 12 '16 20 21 24 26 27 29 31 5 7 10 11 17 20 22 24 25 27 30 31 5 7 10 11 17 20 21 24 26 27 30 31
a 9 10 16 19 21 23 26 28 29 32 6 8 9 12 17 19 22 23 25 28 30 32 6 6 9 12 17 19 21 23 26 28 30 32
6 7 11 17 19 22 24 25 28 30 32 5 6 1 11 10 19 21 24 26 29 29 32 5 6 7 11 18 19 22 24 25 28 29 32
4 10 12 14 15 22 23 26 28 33 35 2 4 10 12 14 15 22 23 26 26 33 35 2 4 10 12 14 15 22 23 20 28 33 35
2 4 a 9 13 16 18 20 30 32 33 35 3 4 9 11 13 15 1B 20 30 31 33 JS 3 4 9 11 u 15 18 20 30 31 33 35
3 4 9 11 14 15 17 20 29 31 34 35 2 4 a 9 14 16 17 20 29 32 34 35 2 4 8 9 14 16 17 20 29 32 34 35
1 3 6 12 13 15 21 24 25 28 34 3S 1 3 7 10 13 15 21 24 25 28 34 35 1 3 7 10 13 15 21 24 25 28 34 35
1 3 6 7 14 16 22 24 26 27 33 36 1 3 6 7 14 16 22 24 26 27 33 36 3 6 7 14 16 22 24 26 27 33 36
l 3 5 11 l3 15 18 19 30 31 33 36 l 2 5 8 13 16 18 19 30 32 33 36 2 5 8 13 16 18 19 30 32 33 36
2 4 7 10 l3 16 21 23 25 27 34 36 2 4 6 12 13 16 21 23 25 27 34 36 4 6 12 13 16 21 23 25 27 34 36
I 2 5 8 14 16 17 19 29 32 34 36 l 3 5 11 14 15 17 19 29 31 34 36 3 5 11 14 15 17 19 29 31 34 36
6 8 10 11 17 22 26 29 35 36 37 38 6 a 10 11 17 22 26 29 35 36 37 38 8 10 11 17 22 26 29 35 36 37 38
5 7 9 12 18 21 25 30 35 36 37 38 5 7 9 12 18 21 25 30 35 36 37 38 7 9 12 18 21 25 30 35 36 37 38
5 7 8 10 20 24 21 32 33 34 37 38 5 6 8 12 20 24 27 32 33 34 37 38 6 8 12 20 24 27 32 33 34 37 38
6 9 11 12 19 23 2S 31 34 37 38 7 9 10 11 19 23 28 31 33 34 37 38 9 10 11 19 23 28 31 33 34 37 3B
1
• 7 12
3 s 10
13
14
21
22
26
25 31
32
32
33
34
37
37
39
39
2
1
3 7 12 13 21 26
4 6 10 14 22 25
31
31
32
32
33
34
37
37
39
39
3 7 12
4 6 10
13 22
14 21
25
26
31
31
32 33
32 34
37
37
39
39
4 10 12 16 24 27 29 30 35 37 ~9 3 4 10 12 16 24 27 29 30 35 37 39 4 10 12 16 24 27 29 30 35 37 39
2 6 7 15 23 28 29 30 36 37 9 2 6 7 15 23 2l! 29 30 36 37 39 2 6 7 15 23 28 29 30 36 37 39
3 8 11 14 17 27 28 30 33 37 40 4 8 11 14 17 27 28 30 33 37 40 8 11 14 lB 27 28 29 33 37 40
4 5 9 13 18 27 28 29 34 37 40 3 5 9 13 18 27 28 29 34 37 40 5 9 13 17 27 28 30 34 37 40
3 5 11 16 20 25 26 32 35 37 40 1 3 5 11 16 20 25 26 32 35 37 40 s 11 16 20 25 26 32 35 37 40
4 8 9 15 19 25 26 31 36 37 40 2 8 9 15 19 25 26 31 36 37 40 8 9 15 19 25 26 31 36 37 40
4 8 11 14 IS 23 24 29 33 38 39 2 8 11 14 IS 23 24 29 33 38 39 1 8 11 14 17 23 24 JO 33 38 .39
3 5 9 13 17 23 24 30 34 3B 39 l 5 9 13 17 23 24 30 34 3S 39 2 5 9 13 18 23 24 29 34 38 39
2 9 11 15 19 21 22 32 35 38 39 1 9 11 15 19 21 22 32 35 38 39 1 9 11 15 19 21 22 32 35 JS 39
4 5 8 16 20 21 22 31 36 36 39 3 5 8 16 20 21 22 31 36 38 39 3 5 8 16 20 21 22 31 36 38 39
3 7 12 13 19 20 22 25 33 38 40 1 7 12 13 19 20 22 25 33 38 40 1 7 12 13 19 20 21 26 33 38 40
6 10 14 19 20 21 26 34 38 40 2 6 10 14 19 20 21 26 34 38 40 2 6 10 14 19 20 22 25 34 38 40
7 10 16 17 18 24 28 35 38 40 1 8 12 16 17 18 24 28 35 38 40 l 6 12 16 17 18 24 28 35 3S 40
6 12 IS 17 18 23 27 36 38 40 3 4 1 10 15 11 18 23 21 36 Ja 40 3 7 10 15 17 18 23 27 36 38 40
9 10 15 16 17 21 25 29 39 40 6 9 10 15 16 17 21 25 29 39 40 "5 6 9 10 15 16 17 21 25 29 39 40
11 12 15 16 18 22 26 30 39 40 8 11 12 15 16 18 22 26 30 39 40 7 8 11 12 15 16 18 22 26 30 39 40
5 6 7 8 13 14 19 23 27 31 39 40 6 7 B 13 14 19 23 27 31 39 40 5 6 7 8 13 14 19 23 27 31 39 40
9 10 11 12 13 14 20 24 28 32 39 40 9 10 11 12 13 14 20 24 28 32 39 40 9 10 11 12 13 14 20 24 28 32 39 40
13 14 15 16 17 18 19 20 21 22 23 24 13 14 15 16 17 18 19 20 21 22 23 24 13 14 15 16 17 18 19 20 21 22 23 24
13 14 15 16 25 26 27 28 29 30 31 32 13 14 15 16 25 26 27 28 29 JO 31 32 13 14 15 16 25 26 27 28 29 30 31 32
17 18 19 20 25 26 27 28 33 34 35 36 11 18 19 20 25 26 27 28 33 34 35 36 17 18 19 20 25 26 27 28 33 34 35 36
21 22 23 24 29 30 31 32 33 34 35 36 21 22 23 24 29 30 31 32 33 34 35 36 21 22 23 24 29 30 31 32 33· 34 35 36
 89

Here are listed the ten systems of three linked 2 - (16,6,2) designs,
as promised in Section 4.2. The incidence graphs of these systems, together
with the point graph of the generalized quadrangle of order (3,5) form the
eleven 4-colourable strongly regular graphs with parameters (64,18,2,6) of
Theorem 4 • 3 • 1.

3 4 5 9 13 2 5 9 13 3 5 9 13 3 7 !S 16 7 15 16
3 4 6 10 14 l 6 10 14 3 6 10 14 5 9 10 11 9 10 11
I 2 4 7 l1 15 l 2 7 !l 15 l 2 4 7 11 15 3 5 6 8 14 5 6 B 14
1 2 3 8 12 16 2 8 12 16 1 2 J 8 12 16 4 6 7 9 13 4 6 7 9 13
1 6 7 8 9 13 6 8 9 13 l 6 7 8 9 13 4 10 12 14 16 4 10 12 14 16
2 5 7 B 10 14 s 8 10 14 2 s 7 8 10 14 l 8 11 12 !3 15 B ll 12 13 15
3 5 6 B 11 15 5 6 8 11 15 3 5 6 B 11 15 2 3 4 5 12 13 2 3 4 5 12 13
4 5 6 7 12 16 5 6 7 12 16 4 5 6 7 12 16 2 4 6 11 14 15 2 4 6 11 14 15
1 5 10 11 12 13 10 ll 12 13 l 5 10 11 12 13 2 6 8 10 13 16 2 6 8 10 13 16
2 6 9 11 12 14 6 9 11 12 14 2 6 9 11 12 14. 2 7 a 9 12 14 2 7 8 9 12 14
3 7 9 10 12 15 3 7 9 10 12 15 3 7 9 10 12 15 3 4 8 9 10 15 3 4 e 9 10 15
4 8 9 10 11 16 4 8 9 10 11 16 4 8 9 10 11 16 3 6 9 11 12 16 3 6 9 11 12 16
1 5 9 14 15 16 1 5 9 14 15 16 1 5 9 14 15 16 3 7 10 11 13 14 3 7 10 ll 13 14
2 6 10 13 15 16 2 6 10 13 15 16 2 6 10 13 15 16 4 5 7 8 11 16 4 5 7 8 11 16
3 7 ll 13 14 16 7 11 13 14 16 3 7 11 13 14 16 5 6 7 10 12 15 5 6 7 10 12 15
4 8 12 13 14 15 4 8 12 13 14 15 4 8 12 13 14 15 5 9 13 14 15 16 5 9 13 14 15 16

2 3 4 611 16 3 4 6 11 16 4 6 11 16 1 2 3 6 11 13 3 6 11 13
1 3 4 512 15 3 4 8 9 15 4 8 9 15 I 2 5 12 14 15 5 12 14 15
1 2 4 8 9 14 2 4 5 12 14 2 4 5 12 14 1 3 4 5 7 10 4 5 7 10
1 2 3 710 13 2 3 7 10 13 2 3 7 10 13 1 4 9 11 15 16 4 9 11 15 16
1 5 7 811 16 5 6 e 10 16 5 6 8 10 16 I 6 1 8 12 16 6 7 8 12 16
2 6 7 8 12 15 5 9 10 11 15 2 5 10 11 15 1 8 9 10 13 14 8 9 10 13 14
3 5 6 7. 9 14 8 10 11 12 14 3 8 10 11 12 14 2 3 5 B 9 16 3 5 8 9 16
4 5 6 8 10 13 4 5 7 B 11 13 4 7 8 11 13 2 4 6 9 10 12 4 6 9 10 12
1 6 9 10 12 16 l 7 9 11 12 16 1 9 11 12 16 2 4 7 13 14 16 4 7 !3 14 16
2 5 9 10 11 15 2 6 7 8 12 15 2 6 7 8 12 15 2 7 8 10 11 15 8 10 11 15
3 8 10111214 3 5 6 7 9 14 3 5 6 7 9 14 3 4 8 11 12 14 4 8 ll 1412
4 7 9 11 12 13 4 6 9 10 12 13 4 6 9 10 12 13 3 6 10 14 15 16 6 10 14 1615
l 6 11131415 l 6 11 13 14 15 1 6 11 13 14 15 3 7 9 12 13 15 7 9 12 1513
2 s 12 13 14 16 2 8 9 13 14 16 2 8 9 13 14 16 4 5 6 e 13 15 4 s 6 8 13 15
3 8 9 13 15 16 3 5 12 13 15 16 3 5 12 13 15 16 5 6 7 9 11 14 5 6 9 11 14
4 7 10 14 15 16 4 7 10 14 15 16 4 7 10 14 15 16 5 10 11 12 13 16 s 10 11 12 13 16

2 3 4 7 12 14 6 10 12 5 11 12 1 2 6 7 10 14 6 9 14 16
I 3 4 8 ll 13 4 6 B 11 12 1 4 5 6 9 11 1 2 9 12 13 16 2 7 10 12 13
l 2 4 5 10 16 2 9 12 13 15 l 2 10 11 14 16 4 6 8 9 14 16 6 7 8 10 14
l 2 3 6 9 15 2 6 7 14 16 1 2 s 8 13 15 4 8 10 12 13 4 8 9 12 13 16
l 5 6 8 12 14 1 4 5 7 9 10 1 4 7 8 10 12 I 4 6 12 15 l 3 4 13 14 15
2 5 6 7 11 13 2 3 7 8 9 !l 2 3 6 6 9 10 1 4 5 11 13 14 4 5 6 11 12
3 6 7 8 10 16 3 4 6 7 13 15 3 4 5 8 14 16 2 3 8 13 14 15 3 6 8 12 15
4 5 7 8 9 15 3 4 9 12 14 16 3 4 10 11 13 15 2 5 6 8 11 12 5 8 l 1 13 14
1 7 9 10 11 14 1 3 10 11 15 16 1 3 6 7 15 16 3 7 8 9 11 3 8 10 11 16
2 8 9 10 12 13 2 4 5 8 15 16 2 4 9 12 15 16 5 6 10 15 16 1 5 7 8 9 15
3 5 9 11 12 16 5 6 9 11 13 16 5 6 10 12 14 15 2 3 4 10 !l 16 2 3 4 7 9 11
4 6 10 11 12 15 5 7 11 12 14 15 6 8 11 12 13 16 2 4 5 7 9 15 2 4 s 10 15 16
l 7 12 13 15 16 1 3 5 B 13 14 1 3 9 12 13 14 3 5 6 9 10 13 3 5 6 7 13 16
2 8 11 14 15 16 2 4 10 11 13 14' 2 4 6 7 13 14 3 5 7 12 14 16 3 5 9 10 12 14
3 5 10 13 14 15 6 8 9 10 14 15 5 7 9 10 13 16 6 7 11 13 15 16 6 9 10 11 13 15
4 6 9 13 14 16 7 8 10 12 13 16 7 8 9 11 14 15 9 10 11 12 14 15 7 11 12 14 15 16
90

2 3 15 16 l 3 7 15 16 1 3 7 15 16 2 3 7 15 16 3 4 5 9 13
2 5 9 10 1l 1 5 9 10 l1 1 5 9 10 11 5 9 10 11 3 4 6 10 14
3 5 6 a 14 l 5 6 8 14 l 5 6 8 14 3 5 6 8 14 2 4 11 15
4 6 7 9 13 l 4 6 7 9 13 1 4 6 7 9 13 4 6 1 9 13 2 3 12 16
4 10 12 14 16 l 4 10 12 14 16 1 4 10 12 14 16 4 10 12 14 16 l 6 7 8 9 13
1 8 11 12 13 15 l 8 11 12 13 15 1 8 11 12 13 15 8 11 12 13 15 2 5 1 8 10 14
2 '3 4 5 12 13 2 3 4 s 12 13 2 3 4 s 12 13 3 4 s 12 13 3 s 6 8 ll 15
2 4 6 11 14 15 2 .4 6 11 14 15 2 4 6 11 14 15 4 6 11 14 15 4 5 6 7 12 16
2 6 8 10 13 16 6 8 10 13 16 2 6 8 10 13 16 6 8 10 13 16 l 5 10 11 12 13
2 7 8 9 12 14 7 8 9 12 14 2 7 8 9 12 14 7 8 9 12 14 2 6 9 11 12 14
3 4 8 9 10 15 4 8 9 10 15 3 4 a 9 10 15 4 8 9 10 15 3 7 9 10 12 15
3 6 9 11 12 16 3 6 9 11 12 16 3 6 9 ll 12 16 6 9 11 12 16 4 8 9 10 11 16
3 7 10 11 13 14 3 7 10 II 13 14 3 7 10 11 13 14 7 10 11 13 14 1 5 9 14 15 16
4 5 7 8 11 16 4 5 7 8 11 16 4 5 7 a 11 16 5 7 8 11 16 2 6 10 13 15 16
5 6 7 10 12 15 5 6 7 10 12 15 5 6 7101215 6 7 10 12 15 3 7 11 13 14 16
5 9 13 14 15 16 5 9 13 14 15 16 5 9 13 14 15 16 9 13 14 15 16 4 8 12 13 14 15

2 6 ll 13 3 6 11 13 2 6 ll 13 2 3 6 II 13 3 4 6 11 16
2 12 14 15 5 12 14 15 2 5 12 14 15 2 5 12 14 15 l 3 4 8 9 15
3 4 5 7 10 4 5 7 10 3 4 5 7 10 3 4 5 7 10 2 4 5 12 14
I 4 9 !! 15 16 I 4 9 11 15 16 l 4 9 ll 15 16 4 9 !! 15 16 2 3 7 10 13
1 6 7 B 12 16 l 6 7 8 12 16 l 6 7 8 12 16 6 7 8 12 16 5 6 8 10 16
1 8 9 10 13 14 1 8 9 10 13 14 1 8 9 10 13 14 B 9 10 13 14 2 5 9 10 11 15
2 3 5 8 9 16 2 3 5 8 9 16 2 5 8 9 16 2 3 5 8 9 16 3 8 10 11 12 14
2 6 9 10 12 2 4 6 9 10 12 2 6 9 10 12 2 4 6 9 10 12 4 5 7 8 II 13
2 4 7 13 14 16 2 4 7 13 14 16 2 7 13 14 16 2 4 7 13 14 16 l 7 9 11 12 16
2 7 8 10 II 15 2 7 8 10 l l 15 2 8 10 !l 15 2 7 8 10 ll IS 2 6 7 a 12 15
3 4 8 !! 12 14 3 4 e 11 12 14 3 4 6 11 12 14 3 4 8 11 12 14 3 s 6 7 9 14
3 6 10 14 15 16 3 10 14 15 16 3 6 10 14 15 16 3 6 10 14 IS 16 4 6 9 10 !2 13
3 7 9 12 13 IS 3 9 12 !3 15 3 7 9 12 13 15 3 7 9 12 13 15 1 6 11 13 14 IS
4 s 6 8 13 IS 4 5 6 8 13 IS 4 s 6 8 13 15 4 5 6 8 13 IS 2 8 9 13 14 16
s 6 7 9 11 14 5 6 7 9 !l 14 5 6 7 9 II 14 5 6 7 9 11 14 3 5 12 13 lS 16
5 10 !l 12 13 16 s 10 11 12 13 16 s 10 11 12 13 16 5 10 11 12 13 16 4 7 10 14 IS 16

6 r 10 14 1 2 6 9 14 16 10 14 l 2 6 9 14 16 8 9 11
2 9 12 13 16 I 2 7 10 12 13 2 9 12 13 16 2 7 10 12 13 4 6 9 11
4 6 8 9 14 16 4 6 7 8 10 14 6 8 9 14 16 6 7 8 10 14 2 10 11 14 16
4 7 8 10 12 13 4 8 9 12 13 16 4 7 a lO 12 13 8 9 12 13 16 2 9 12 13 15
1 3 4 6 12 15 I 3 4 13 14 15 l 3 4 13 14 15 3 4 6 12 15 4 7 8 10 12
I 4 s 11 13 14 1 4 5 6 ll 12 1 4 5 6 11 12 I 4 5 11 13 14 3 5 6 10 12
2 3 8 13 14 15 2 3 6 8 12 15 2 3 6 8 12 IS 2 3 8 13 14 15 4 9 12 14 16
2 5 6 8 11 12 2 5 8 11 13 14 2 5 8 11 13 14 2 s 6 8 11 12 4 10 11 13 15
l 3 8 10 11 16 1 3 7 8 9 11 l 3 8 10 11 16 1 3 7 B 9 II 3 5 8 13 14
1 5 7 8 9 15 l 5 8 10 15 16 l s 7 8 9 15 1 5 8 10 15 16 4 6 7 13 14
2 3 4 7 9 11 2 3 4 10 II 16 2 3 4 7 9 11 2 3 4 10 II 16 5 7 9 10 13 16
2 4 5 10 15 16 2 4 5 7 9 15 2 4 5 10 lS 16 2 4 5 7 9 15 5 7 11 12 14 lS
3 5 6 9 10 13 3 5 6 7 13 16 3 5 6 9 11 13 3 5 6 7 13 16 l 3 6 7 15 16
3 s 7 12 14 16 3 5 9 10 12 14 3 5 7 12 14 16 3 5 9 10 12 14 2 4 5 8 15 16
6 7 11 13 15 16 6 9 10 11 13 15 6 7 11 13 15 16 6 9 10 11 13 15 6 8 9 10 14 IS
9 10 I! 12 14 15 7 !ll 12 14 15 16 9 10 ll 12 14 15 7 11 12 14 15 16 6 8 ! I 12 13 16
 91

Here we give the 2- (71,15,3) designs 07, o;, o; and D*4 , which are
constructed in Section 6.1. Together with their duals they form the eight
2 - (71,15,3) designs of Theorem 6.1.5.
2 3 4 5 6 7 8 9 10 11 12 13 14 15 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 2 3 16 17 18 19 20 21 22 23 24 25 26 27 2 3 16 17 18 19 20 21 22 23 24 25 26 27
'2 3 28 29 30 31 32 33 34 35 36 37 38 39 1 2 3 28 29 30 31 32 33 34 35 36 37 38 39
4 s 16 17 28 29 40 41 42 43 44 45 46 47 4 5 16 17 28 29 40 41 42 43 44 45 46 47
4 5 18 19 30 31 48 49 50 51 52 53 54 55 4 18 19 30 31 48 49 so 51 52 53 54 55
6 7 16 17 30 31 56 57 58 59 60 61 62 63 6 16 17 30 31 S6 57 58 59 60 61 62 63
6 7 18 19 28 29 64 65 66 67 68 69 70 71 6 7 18 19 28 29 64 65 66 67 68 69 70 71
8 9 20 21 32 33 40 41 48 49 56 57 64 65 8 9 20 21 32 33 40 41 48 49 56 57 64 65
8 9 22 23 34 35 42 43 so 51 58 59 66 67 1 8 9 22 23 34 3S 42 43 50 51 58 59 66 67
10 11 20 21 34 35 44 45 52 53 60 61 68 69 10 11 20 21 34 35 44 45 52 53 60 61 68 69
10 11 22 23 32 33 46 47 54 55 62 63 70 71 10 11 22 23 32 33 46 47 54 55 62 63 70 71
12 13 24 25 36 37 40 41 50 51 60 61 70 71 12 13 24 25 36 37 40 41 50 51 60 61 70 71
12 13 26 27 38 39 42 43 48 49 62 63 68 69 12 13 26 27 38 39 42 43 48 49 62 63 68 69
1 14 15 24 25 38 39 44 45 54 55 S6 57 66 67 14 15 24 2S 38 39 44 45 54 55 56 57 66 67
1 14 15 26 27 36 37 46 47 52 S3 58 59 64 6S 1 14 15 26 27 36 37 46 47 52 53 58 59 64 65
2 4 6 20 22 36 38 40 42 52 54 56 58 68 70 2 4 6 20 22 36 38 40 42 52 54 56 58 68 70
2 4 6 21 23 37 39 41 43 53 55 57 59 69 71 2 4 6 21 23 37 39 41 43 53 55 57 59 69 71
2 5 7 20 22 37 39 44 46 48 50 60 62 64 66 2 5 7 20 22 37 39 44 46 48 so 60 62 64 66
2 5 7 21 23 36 38 45 47 49 51 61 63 65 67 2 5 7 21 23 36 38 45 47 49 51 61 63 65 67
2 a 10 24 26 2s 30 40 46 48 54 59 61 67 69 2 8 10 24 26 28 30 40 47 48 SS 59 60 67 68
2 a 10 25 21 29 31 41 47 49 55 58 Go 66 68 2 8 10 25 27 29 31 42 45 50 53 57 62 65 70
2 9 11 24 26 29 31 43 45 51 53 56 62 64 70 2 9 11 24 26 29 31 43 44 51 52 56 63 64 71
2 9 11 2S 27 28 30 42 44 50 52 57 63 65 71 2 9 11 25 27 28 30 41 46 49 54 58 61 66 69
2 12 14 16 18 32 34 40 44 51 55 59 63 64 68 2 12 14 16 18 32 34 40 44 50 54 59 63 6S 69
2 12 14 17 19 33 35 41 45 50 54 58 62 65 69 2 12 14 17 19 33 35 42 46 48 52 57 61 67 71
2 13 15 16 18 33 35 42 46 49 53 57 61 66 70 2 13 15 16 18 33 35 41 45 51 55 58 62 64 68
2 13 15 17 19 32 34 43 47 48 52 56 60 67 71 2 13 lS 17 19 32 34 43 47 49 53 56 60 66 70
3 4 7 24 27 32 35 40 42 53 55 60 62 65 67 3 4 7 24 27 32 35 40 42 53 55 61 63 64 66
3 4 7 25 26 33 34 41 43 52 S4 61 63 64 66 3 4 7 25 26 33 34 41 43 52 54 60 62 65 67
3 5 6 24 27 33 34 44 46 49 51 56 58 69 71 3 5 6 24 27 33 34 44 46 49 51 57 59 68 70
3 5 6 25 26 32 35 45 47 48 so S7 59 68 70 3 5 6 25 26 32 35 45 47 48 so 56 58 69 71
3 8 11 16 19 37 38 41 44 49 52 59 62 67 70 3 8 11 16 19 37 38 41 45 48 52 59 63 66 70
3 8 11 17 18 36 39 40 45 48 53 58 63 66 71 3 6 11 17 18 36 39 43 47 so 54 57 61 64 68
3 9 10 16 19 36 39 43 46 51 54 57 60 65 68 3 9 10 16 19 36 39 40 44 49 53 58 62 67 71
3 9 10 17 18 37 38 42 47 50 55 56 61 64 69 3 9 10 17 18 37 3B 42 46 51 55 56 60 65 69
3 12 lS 20 23 29 30 40 47 51 S2 57 62 66 69 3 12 15 20 23 29 30 43 44 48 55 58 61 6S 70
3 12 15 21 22 28 31 41 46 50 53 56 63 67 68 3 12 15 21 22 28 31 41 46 50 53 56 63 67 68
3 13 14 20 23 28 31 43 44 48 55 S8 61 65 70 3 13 14 20 23 28 31 40 47 51 52 57 62 66 69
3 13 14 21 22 29 30 42 4S 49 54 59 60 64 71 13 14 21 22 29 30 42 45 49 54 59 60 64 71
4 8 13 18 23 2S 30 34 37 45 46 56 62 65 68 8 12 18 21 27 31 33 38 44 47 58 60 66 71
4 8 13 19 22 24 31 35 36 44 47 S7 63 64 69 4 8 12 19 23 24 30 34 36 45 46 56 62 64 69
4 9 12 18 21 26 30 33 39 44 47 58 60 67 70 4 9 13 18 20 26 31 32 39 45 46 59 61 67 70
4 9 12 19 20 27 31 32 38 45 46 59 61 66 71 4 9 13 19 22 25 30 35 37 44 47 57 63 65 68
4 10 15 16 21 25 28 3S 38 48 51 58 62 64 71 4 10 14 16 23 26 29 33 37 49 50 56 61 66 68
4 10 15 17 20 24 29 34 39 49 50 59 63 65 70 4 10 14 17 21 25 28 34 39 48 51 58 63 64 70
4 !! 14 16 23 26 28 32 36 49 50 56 60 66 69 11 15 16 22 27 29 32 36 48 51 57 60 67 69
4 11 14 17 22 27 29 33 37 48 51 57 61 67 68 11 15 17 20 24 28 35 38 49 50 59 62 65 71
s 8 12 16 22 26 29 35 39 52 55 56 61 6S 71 5 8 13 16 22 27 28 34 39 52 55 56 61 65 71
5 B 12 17 23 27 28 34 38 53 54 57 60 64 70 5 8 13 17 20 24 29 33 37 53 54 58 63 67 69
s 9 13 16 20 25 29 32 37 53 54 58 63 67 69 5 9 12 16 23 26 28 35 38 53 54 57 60 64 10
5 9 13 17 21 24 28 33 36 52 55 59 62 66 68 5 9 12 17 21 25 29 32 36 52 55 59 62 66 68
s 10 14 18 20 26 31 34 36 41 42 57 62 67 71 5 10 15 18 20 26 30 34 36 41 42 57 63 66 71
5 10 14 19 21 27 30 35 37 40 43 56 63 66 70 5 10 15 19 22 25 31 33 38 40 43 59 61 64 69
5 11 15 18 22 25 31 33 38 40 43 59 60 65 69 s 11 14 18 21 27 30 35 37 40 43 56 62 67 70
5 11 15 19 23 24 30 32 39 41 42 58 61 64 68 5 11 14 19 23 24 31 32 39 41 42 58 60 65 68
6 8 15 18 21 27 29 32 36 43 44 50 54 61 62 6 8 lS 18 23 25 29 35 39 40 46 49 52 60 63
6 8 15 19 20 26 28 33 37 42 45 51 55 60 63 6 8 15 19 21 26 28 32 37 42 44 51 54 61 62
6 9 14 18 23 24 29 35 38 41 46 48 S2 60 63 6 9 14 18 22 24 29 34 38 41 47 48 53 61 62
6 9 14 19 22 25 28 34 39 40 47 49 53 61 62 6 9 14 19 20 27 28 33 36 43 45 50 55 60 63
6 10 13 16 23 27 31 33 39 40 4S so S2 64 67 6 10 13 16 2.1 24 31 35 36 43 46 48 54 65 66
6 10 13 17 22 .26 30 32 38 41 44 51 S3 65 66 6 10 13 17 23 27 30 32 38 41 44 50 52 64 67
6 11 12 16 21 24 31 34 37 42 47 48 54 6S 66 6 11 12 16 20 25 31 34 37 42 47 49 55 64 67
6 11 12 17 20 25 30 35 36 43 46 49 55 64 67 6 11 12 17 22 26 30 33 39 40 45 51 53 65 66
7 8 14 16 20 24 30 33 36 43 47 so 53 68 71 7 8 14 16 20 25 30 32 38 43 46 51 53 68 71
1 e 14 11 21 25 31 32 39 42 46 s1 52 69 10 7 8 14 17 22 26 31 35 36 41 44 49 55 69 70
7 9 15 16 22 27 30 34 36 41 45 48 5S 69 70 7 9 15 16 21 24 30 33 39 42 47 50 52 69 70
7 9 15 17 23 26 31 35 37 40 44 49 54 68 71 9 1 S 17 23 27 31 34 37 40 45 48 54 68 71
7 10 12 18 22 24 28 32 37 43 45 49 52 57 58 10 12 18 22 24 28 32 37 43 4S 49 52 S7 58
7 10 12 19 23 25 29 33 36 42 44 .48 53 S6.,59 10 12 19 20 27 29 35 39 41 47 51 54 56 59
7 11 13 18 20 27 28 35 39 41 47 51 54 sfi 59 11 13 18 23 25 28 33 36 42 44 48 53 56 59
7 11 13 19 21 26 29 34 38 40 46 50 55 57 58 11 13 19 21 26 29 34 38 40 46 50 55 57 58
92

3 4 5 6 7 8 9 10 !I 12 13 14 15 4 5 6 7 8 9 10 11 12 13 14 15
3 16 17 18 19 20 21 22 23 24 25 26 27 16 17 18 19 20 21 22 23 24 25 26 27
1 2 3 28 29 30 31 32 33 34 35 36 37 38 39 28 29 30 31 32 33 34 35 36 37 38 39
4 5 16 17 28 29 40 41 42 43 44 45 46 47 4 5 16 17 28 29 40 41 42 43 44 45 46 4 7
4 5 18 19 30 31 48 49 50 51 52 53 54 55 4 5 18 19 30 31 48 49 50 51 52 53 54 55
6 7 16 17 30 31 56 57 58 59 60 61 62 63 6 7 16 17 30 31 56 57 58 59 60 61 62 63
6 7 18 19 28 29 64 65 66 67 68 69 70 71 6 7 18 19 28 29 64 65 66 67 68 69 70 71
B 9 20 21 32 33 40 41 48 49 56 57 64 65 1 8 9 20 21 32 33 40 41 48 49 56 57 64 65
8 9 22 23 34 35 42 43 50 51 58 59 66 67 1 a 9 22 23 34 35 42 43 so s1 s0 59 66 67
10 11 20 21 34 35 44 45 52 53 60 61 6s 69 10 II 20 21 34 35 44 45 52 53 60 61 68 69
10 11 22 23 32 33 46 47 54 55 62 63 70 71 10 11 22 23 32 33 46 47 54 55 62 63 70 71
12 13 24 25 36 37 42 43 48 49 62 63 68 69 12 13 24 25 36 37 42 43 48 49 62 63 68 69
12 13 26 27 3B 39 40 41 50 51 60 61 70 71 1 12 13 26 27 38 39 40 41 50 51 60 61 70 71
I 14 15 24 25 38 39 46 47 52 53 58 59 64 65 1 14 15 24 25 38 39 46 47 52 53 58 59 64 65
1 14 15 26 27 36 37 44 45 54 55 56 57 66 67 14 15 26 27 36 37 44 45 54 55 56 57 66 67
2 4 6 24 26 32 34 44 46 48 50 56 58 68 70 4 6 24 26 32 34 44 46 48 50 60 62 64 66
2 4 6 25 27 33 35 45 47 49 51 57 59 69 71 2 4 6 25 27 33 35 45 47 49 51 61 63 65 67
2 5 7 24 26 33 35 40 42 52 54 60 62 64 66 2 5 7 20 24 34 36 40 47 51 54 56 58 68 70
2 5 7 25 27 32 34 41 43 53 55 61 63 65 67 5 7 21 25 35 37 41 46 50 55 57 59 69 71
2 8 10 16 18 36 38 41 45 51 55 58 62 64 68 8 10 16 18 37 38 42 45 51 52 56 62 64 71
8 10 17 !9 37 39 43 47 49 53 56 60 66 70 2 8 10 17 19 36 39 43 44 50 53 57 63 65 70
9 11 16 18 37 39 40 44 50 54 59 63 65 69 2 9 11 16 27 28 38 40 48 53 55 58 63 66 69
9 11 17 19 36 38 42 46 48 52 57 61 67 71 2 9 11 17 26 29 39 41 49 52 54 59 62 67 68
12 14 20 22 28 30 42 45 51 52 56 63 65 70 2 12 14 18 22 30 32 40 42 44 54 59 61 65 69
2 12 14 21 23 29 31 40 47 49 54 58 61 67 68 2 12 14 19 23 31 33 41 43 45 55 58 60 64 68
2 13 15 20 22 29 31 41 46 48 55 59 60 66 69 2 13 15 20 22 28 30 43 47 48 52 57 60 67 7!
2 13 15 21 23 28 30 43 44 50 53 57 62 64 71 2 13 15 21 23 29 31 42 46 49 53 56 61 66 70
3 4 20 23 36 39 44 46 49 51 60 62 65 67 3 4 7 22 26 32 39 43 45 49 53 56 58 69 71
3 4 7 21 22 37 38 45 47 48 so 61 63 64 66 3 4 7 23 27 33 38 42 44 48 52 57 59 68 70
3 5 6 20 23 37 38 40 42 53 55 56 58 69 71 3 5 6 20 22 36 39 41 42 52 55 61 63 64 66
3 5 6 21 22 36 39 41 43 52 54 57 59 68 70 3 5 6 21 23 37 38 40 43 53 54 60 62 65 67
3 8 11 24 27 28 31 42 47 50 55 57 60 65 68 3 a 11 24 26 29 31 42 47 s1 ss 57 60 65 69
3 8 11 25 26 29 30 40 45 48 53 59 62 67 70 3 8 11 25 27 28 30 43 46 50 54 56 61 64 68
3 9 10 24 27 29 30 41 44 49 52 58 63 66 71 3 9 10 18 25 30 37 41 44 47 49 58 60 66 70
3 9 10 25 26 28 31 43 46 51 54 56 61 64 69 3 9 10 19 24 31 36 40 45 46 48 59 61 67 71
3 12 15 16 19 32 35 41 47 50 52 56 62 67 69 3 12 15 16 18 32 34 41 46 51 53 57 63 67 68
3 12 15 17 18 33 34 43 45 48 54 58 60 65 71 3 12 15 17 19 33 35 40 47 50 52 56 62 66 69
3 13 14 16 19 33 34 40 46 51 53 57 63 66 68 3 13 14 16 20 28 34 45. 49 50 55 59 62 65 70
3 13 14 17. 18 32 35 42 44 49 55 59 61 64 70 3 13 14 17 21 29 35 44 48 51 54 58 63 64 71
4 8 12 16 20 24 28 32 36 53 54 59 61 66 71 4 8 12 16 20 25 29 33 36 53 54 59 60 66 71
4 8 12 17 23 26 29 34 39 52 55 57 63 64 69 4 8 12 17 21 24 28 32 37 52 55 58 61 67 70
4 9 13 16 21 25 28 33 37 52 55 58 60 67 70 4 9 13 16 19 23 34 37 39 47 54 57 61 64 69
4 9 13 17 22 27 29 35 38 53 54 56 62 65 68 4 9 13 17 18 22 35 36 38 46 55 56 60 65 68
4 10 14 18 22 24 30 32 38 40 43 57 60 67 69 4 10 14 20 23 24 29 30 38 41 50 56 63 67 69
4 10 14 19 21 26 31 34 37 41 42 59 62 65 71 4 10 14 21 22 25 28 31 39 40 51 57 62 66 68
4 11 15 18 23 25 30 33 39 41 42 56 61 66 68 4 11 15 18 20 26 31 35 37 40 43 59 63 64 70
11 15 19 20 27 31 35 36 40 43 58 63 64 70 4 11 15 19 21 27 30 34 36 41 42 58 62 65 71
8 13 18 21 24 31 35 39 45 46 56 63 67 71 5 8 13 18 20 27 31 33 39 44 46 58 62 67 69
8 13 19 22 26 30 33 36 44 47 58 61 65 69 5 8 13 19 21 26 30 32 38 45 47 59 63 66 68
5 9 12 18 20 25 31 34 38 44 47 57 62 66 70 5 9 12 22 24 27 29 30 35 45 53 57 62 64 70
5 9 12 19 23 27 30 32 37 45 46 59 60 64 68 s 9 12 23 25 26 28 31 34 44 52 56 63 65 71
5 10 15 16 23 24 29 35 37 48 51 57 61 65 70 5 10 15 16 22 26 29 33 37 48 50 58 61 65 68
5 10 15 17 20 26 28 33 38 49 50 59 63 67 68 5 10 15 17 23 27 28 32 36 49 51 59 60 64 69
5 11 14 16 22 25 29 34 36 49 50 56 60 64 71 5 11 14 16 19 25 32 35 39 42 48 56 60 67 70
5 11 14 17 21 27 28 32 39 48 51 58 62 66 69 5 11 14 17 18 24 33 34 38 43 49 57 61 66 71
6 8 14 18 20 27 29 33 37 43 46 50 52 61 62 6 8 15 16 23 24 30 35 39 40 44 49 55 68 71
6 8 14 19 23 25 28 35 38 41 44 48 54 60 63 6 8 15 17 22 25 31 34 38 41 45 48 54 69 70
6 9 15 18 21 26 29 32 36 42 47 51 53 60 63 6 9 14 16 21 26 30 33 36 43 46 51 52 69 70
6 9 15 19 22 24 28 34 39 40 45 49 55 61 62 6 9 14 17 20 27 31 32 37 42 47 50 53 68. 71
6 10 12 16 22 27 31 33 39 42 44 48 53 64 67 6 10 12 18 21 27 29 34 39 43 47 48 55 56 59
6 10 12 17 21 25 30 35 36 40 46 50 55 65 66 6 10 12 19 20 26 28 35 38 42 46 49 54 57 58
6 11 13 16 23 26 31 32 38 43 45 49 52 65 66 6 11 13 18 23 25 29 32 36 40 45 50 52 57 58
6 11 13 17 20 24 30 34 37 41 47 51 54 64 67 6 11 13 19 22 24 28 33 37 41 44 51 53 56 59
7 8 15 16 21 27 30 34 38 42 46 49 54 69 70 7 8 14 18 23 26 28 35 36 41 47 48 53 61 62
7 8 15 17 22 25 31 32 37 40 44 51 52 68 71 7 8 14 19 22 27 29 34 37 40 46 49 52 60 63
7 9 14 16 20 26 30 35 39 43 47 48 55 68 71 7 9 15 18 21 24 28 33 39 42 45 50 54 60 63
7 9 14 17 23 24 31 33 36 41 45 50 53 69 70 9 15 19 20 25 29 32 38 43 44 51 55 61 62
7 10 13 18 23 27 28 34 36 !10 47 48 52 56 59 10 13 16 24 27 31 32 35 41 43 52 54 65 66
7 10 13 19 20 25 29 32 39 42 45 so 54 57 58 7 10 13 17 25 26 30 33 34 40 42 53 55 64 67
7 11 12 18 22 26 28 35 37 41 46 49 53 57 58 7 11 12 16 21 22 31 36 38 44 47 49 50 64 67
7 11 12 19 21 24 29 33 38 43 44 51 55 56 59 7 11 12 17 20 23 30 37 39 45 46 48 51 65 66
 93

REFERENCES

[Al] H. AGRAWAL, On the bounds of the nwribez> of aommon t:r>eatments between

bloaks of aez>tain two assoaiate PBIB designs, Calcutta Statist. Assoc.
Bull • .!,l (1964), 76-79.

[A2] R.w. AHRENS & G. SZEKERES, On a aombinatonal genemlization of the

27 lines assoaiated with a aubic sUPfaae, J. Austral. Math. Soc. 10
(1969), 485-492.

[A3] N. ARONSZAJN, Rayleigh-Ritz and A. Weinstein method foz> appro:x:imation

of eigenvalues. I. Opemtors in a Hilbez>t spaae, Proc. American Nat.
Acad. Sci. 34 (1948), 474-480.

[A4] E.F. ASSMUS, Jr. & J.H. van LINT, OVals in projeative designs, J.
Combinatorial Theory A (to appear).

[Bl] H.J. BEKER, Construations and deaorrrpositions of designs, Ph.D. Thesis,

University of London, 1976.

[B2] H.J. BEKER & w. HAEMERS, 2-designs having an interseation nwriber k-n,
J, Combinatorial Theory A (to appear).

[B3] H.J. BEKER & c.J. MITCHEL, A aonstl'Uation method for point divisible

designs, J. Statist. Planning and Inference 3, (1978), 293-306.

[B4] N. BIGGS, Finite groups of automorphisms, London Math. Soc. Lecture
Notes No. 6, Cambridge University Press, London, 1971.

[B5] N. BIGGS, Algebraia Graph Theory, Cambridge tracts in math. No. 67,
Cambridge University Press, Cambridge, 1974.

[B6] a.c. BOSE, Strongly l'e(J}A-lar graphs, partial geometries and partia7.Zy
balanaed designs, Pacific J. Math • .!,l (1963), 389-419.
[B7] R.C. BOSE, Graphs and designs, in "Finite geometric structures and
their applications", C.I.M.E., Bressanone, 1972, pp. 3-104.

[BB] R.c. BOSE & D.M. MESNER, On linear assoaiative algebras aornsponding
to association sahemes of partiaily balanaed designs, Ann. Math.
Statist. lQ_ (1959), 21-38.
94

[B9] c. BUMILLER, Pointwise products of inaid&nae matriaes of subgraphs

(preprint} •

[B10] K.A. BUSH, Bounds for the nwriber of aommon syrriJoZs in balanaed and
aertain partiaiLy balanaed d&signs, J. Combinatorial Theory A~

(1977) I 46-51,

[B11] F .c. BUSSEMAKER & J .J. SEIDEL, Symmetria Hadarna:.Nl matriaes of order
36, Ann. N.Y. Acad. Sci. 175 (1970), 66-79; also: Report Techn. Univ.
Eindhoven 70-WSK-02.

[cl] P.J. CAMERON, On (f!'OupS iuith several doubZy-transitive pe!'rrtUtation

representations, Math. z. ~ {1972), 1-14.

[C2] P.J. CAMERON, E:ctending symmetria d£si{Jn8, J. Combinatori~ Theory A

.!! (1973), 215-220.

[c3] P.J. CAMERON, Partial quadrangles, Quart. J. Math. Oxford (3) 25

(1974), 1-13.

[C4] P.J. CAMERON, ParaiLeUsms of CompZete Designs, London Math. Soc.

Lecture Notes 23, Cambridge University Press, Cambridge, 1976.

[CS] P.J. CAMERON, J.M. GOETHALS & J.J. SEIDEL, Strongly reguZar graphs
having strongZy regular subaonstitusnts, J. Algebra 55 (1978), 257-280.
[c6] P.J. CAMERON & J.B. van LINT, Graph theory, aoding theory and bZoak
d£si{Jn8, London Math. Soc. Lecture Notes No. 19, Cambridge University
Press, Cambridge, 1975.

[c7] F. de CLERCK, Partial,. geomet'I'iea: a aorriJinatoX'iaZ survey, Bull. ·

Belgisch Wiskundig Genootschap B ~ (1979) {to appear) •

[CB] w.s. CONNOR, On the structure of balanaed inaomplete btoak desi{JnB,

Ann. Math. Statist. 23 {1952), 57-71.

[C9] w.s. CONNOR & W.H. CLATWORI'HY, Some theorems for partia'lZy batanaed
d.esi{JnB, Ann. Math. Statist. ~ (1954), 100-112.

[C10] G.M. CONwELL, The 3-spaae PG(3,2) and its (f!'Oup, Ann. Math. ll (1910),
60-76.

[c11] D.M. CVETKOVIC, Graphs and their speatra, Thesis, Univ. Beograd.
Publ. Elektrotehn. Fak. s,er. Mat. Fiz. IB-356 (1971) I 1-50.

[C12] o.M. CVETKOVIC, M. DOOB & H. SACHS, Speatra of Grpahs, a Monograph,

V.E.B. Deutscher Verlag der Wissenschaften, Berlin, 1979.
 95

[Dl] P. DELSARTE, An a'Lgebruia approach to the assoaiation schemes of

coding theory, Philips Res. Rep. Suppl. lQ. (1973}.
[D2] P. DELSARTE, J.M. GOETHALS & J.J. SEIDEL, Spheriaa'L codes and designs,
Geometriae Dedicata .§.. (1977), 363-388.

[03] P. DEMBOWSKI, Finite Geometries, Springer Verlag, Berlin-Heidelberg-

New York, 1968.

[Fl] w. FEIT & G. HIGMAN, The non-e:i:ristenae of certain genero.Uzed po'Ly-

gons, J. Algebra.!. (1964), 114-131.

[G1] A. GEWIR'l'Z, The uniqueness of g (2, 2, 10, 56) , Trans. New York Acad.
Sci. l!_ (1969), 656-675.

[G2] J.M. GOETHALS & J .J. SEIDEL, Strong'Ly regu'Lar graphs derived from
aord>ina:toria'L designs, canad. J. Math. B. (1970), 597-614.

[H1] w. HAEMERS, A generuUzation of the Higman-Sims teahnique, Proc. Kon.

Nederl. Akad. Wetensch. A!!.!_ (= Indag. Math. 40) (1978), 445-447.

[H2] w. HAEMERS, Eigenva'Lue methods, in: Packing and covering in Combina-

torics" (A. Schrijver ed.), Tract 106, Mathematical centre, Amsterdam,
1979, pp. 15-38.

[H3] w. HAEMERS & c. ROOS, An inequaUty for genero'Lized he:1:agons,

Geometriae Dedicata (to appear) •

[H4] w. HAEMERS & M.S. SHRU'.HANDE, Some remarks on subdesigns of symmetria

designs, J. Statist. Planning and Inference (to appear).
[HS] M. HALL, Jr., CordJinatoriaZ Theory, Blaisdell, Weltham-Toronto-London,
1967.

[H6] M. HALL, Jr., Affine genera'Lized quadriZateru'Ls, in: "Studies in pure

mathematics" (L. Mirsky ed.), Academic Press, 1971, pp. 113-116.

[H7] H.L. HAMBURGER & M.E. GRIMSHAW, Linear transformations in n-dimension-

a'L veate>r spaae, cambridge University Press, London-New York, 1951.
[HS] F. HARARY, Graph Theory, Addison-Wesley, Reading, Mass., 1969.

[H9] E. V. HAYNSWOR'l'H, App'Liaations of a theorem on partitioned matrices,

J. Res. Nat. Bur. Standards B 63 (1959), 73-78.
96

[HlO] M.D. HESTENES & D.G. HIGMAN, Rank 3 groups and strongly regular graphs,
in: "Computers in Algebra and Number Theory" (G. Birkhoff & M. Hall,
Jr., eds.), SIAM-AMS Proc. IV, Amer. Math. Soc., Providence, R.I.,
1971, pp. 141-159.

[B11] D.G. HIGMAN, Partial geometries, generalized qu.aih•angles and strongly

regular graphs, in: Atti Convegno di Geometria e sue Applicazioni",
Perugia, 1971, pp. 263-293.

[H12] D.G. HIGMAN, Invariant relations, aoherent aonfigurations and

generalized polygons, in: "combinatorics; part 3: combinatorial group
theory" (M. Hall, Jr. & J.H. van Lint, eds.), Tract 57, Mathematical
Centre, Amsterdam, 1974, pp. 27-43

[Hl3] A.J. HOFFMAN, On eigenvalues and aotourings of graphs, in: "Group

Theory and its Applications" (B. Harris, ed.), Acad. Press, New York,
1970, pp. 79-91.

[H14] A.J. HOFFMAN, Eigenvalues of graphs, in: "Studies in Graph Theory;

part II" (D.R. Fulkerson, ed.), M.A.A. Studies in Math., 1975, PP·
225-245.

[B15] A.J. HOFFMAN & R.R. SINGLETON, On Moon graphs of diamete-r-s 2 and 3,
I.B.M. J. Res. Develop.,! (1960), 497-504.

[H16] D.R. HUGHES & F.C. PIPER, Projeative Planes, Springer Verlag, Berlin-
Heidelberg-New York, 1973.

[H17] D.R. HUGHES & F.c. PIPER, Block Designs (in preparation}.

[Kl] W.M. KANTOR, Generalized quad:r>angles associated liJith G2 (q) (preprint).

[1<2] R. I<ILMOYER & L. SOLOMON, On the theorem of Feit-Higman, J. Combina-

torial Theory A.!2, (1973), 310-322.

[Ll] J.B. van LINT & H.J. RYpER, Block designs with repeated bloaks,
Discrete Math.~ (1972}, 381-396.

[L2] L. LOVAsZ, On the Shannon capacity of a graph, IEEE Trans. Information

Theory 25 (1979), 1-7.

[Ml] K.N. MAJUMDAR, On some theo-r-ems in combinatorics relating to incom-

plete bloak designs, Ann. Math. Statist. ~ (1953), 379-389.
 97

[M2] H.B. MANN, A note on balanced incomplete block designs, Ann. Math.
Statist. !Q. ( 1969) , 679-680.
[M3] M. MARCUS & H. MINC, A sUPVey of matrix theory and matrix inequalities,
,Allyn & Bacon, Boston, 1964.

[M4] R. MATHON, 3-dlase association schemes, in: "Proc. Conf. on Algebraic

Aspects of Combinatorics", (D.G. Corneil & E. Mendelsohn, eds.),
Congressus NUmerantium XIII, Utilitas, Winnipeg, 1975, pp. 123-1SS.

[MS] R. MA'I'HON, Syrmnetric confe?'ence matrices of Ol'der> pq2 + 1, canad. J.

Math. 30 (1978), 321-331.

[M6] R. MATHON, On p'l'imitive association schemes urith th?'ee classes

(in preparation) •

[M7] R. MATHON, The systems of Unked {16,6,2) designs (submitted to Ars

Combinatoria),

[MS] c.J. MITCHELL, An infinite family of symmetna designs, Discrete Math.

26 (1979), 247-2SO.

[N1] B. NOBLE & J.W. DANIEL, Applied Linear- Algebl'a (second edition),
Prentice-Hall, Englewood Cliffs, 1977.

[P1] S.E. PAYNE, Nonisomorrphia generolized quad?'anglee, J. Algebra 18

(1971), 201-212.

[P2] S.E. PAYNE, Genemlized quad?'angles of o?'der (s-1,s+l), J. Algebra 22

(1972) I 97-119.

[P3] S.E. PAYNE, A roeetroiation on the parameter-a of a aubquad?'angle, Bull.

Amer. Math. Soc. 79 (1973), 747-748.

[P4] s.E. PAYNE, Finite generulized lijuad:Panglea: a eu:rvey, in: "Proc. Int.
Conf. on Projective Planes", Wash. State Univ. Press, 1973, pp. 291-
261.

[PS] S.E. PA¥NE, All genemlized quadrangles of oz>der> 3 az>e knoli!n, J.

Combinatorial Theory A !!!_ (197S), 203-206.

[P6] S.E. PAYNE, An inequaUty foz> genemlized quad:Panglea, Proc. Amer.

Math. Soc. Z!. (1978), 147-1S2.
98

[Rl] M.A. RONAN, A geometric charaaterization of Moufang he:x:agons, Ph. D.

Thesis, University of Illinois, 1978.

[R2] M.A. RONAN, A corribinatorial characterization of the dual Moufang

_he:x:agon (preprint) •

[R3] H.J. RYSER, Corribinatorial Mathematics, Carus Mathematical Monograph

14, Wiley, New York, 1963.

[Sl] G.J. SCBELLEKENS, On a he:x:ayonia structure I & II, Proc. Kon. Nederl.
Akad. Wetensch. A 65 (= Indag. Math. 24) (1962}, 201-217, 218-234.

[S2] A.J. SCHWENK & R.J. WILSON, Eigenvalues of gr>aphs, in: "Selected
Topics in Graph Theory" (L.W. Beineke & R.J. Wilson, eds.), Academic
Press, London-New York-San Francisco, 1978, pp. 307-336.

[S3] J.J. SEIDEL, Strongly regular graphs with C-1,1,0) ad;jacenay matri:c
having eigenvalue 3, Linear Algebra and Appl. ! (1968), 281-298.

[S4] J .J. SEIDEL, Strongly regular gr>aphs, in: "Surveys in Combinatorics"

(B. Bollobas, ed.), London Math. Soc. Lecture Notes No. 38, cambridge
University Press, Cambridge, 1979, pp. 157-180.

[SS] M.S. SHRIKID\NDE, Strongly reguZar graphs and quasi-syrrmetric d.esigns,

Utilitas Mathematica 2_ (1973), 297-309.

[S6] s.s. SHRIKHANDE, The uniqueness of the L2 association saheme, Ann.

Math. Statist. 30 (1959), 781-798.

[S7] E.E. SHULT & A. YANUSHKA, Near> n.,-gons and Une systems, Geometriae
Dedicata (to appear) •

[Tl] J .A. THAS, Combinatorias of partial geometries and gener>a.Uzed

quadrangles, in: "Higher Combin'ltorics", (M. Aigner, ed.), Reidel,
DOrdrecht, 1977, pp. 183-199.

[T2] J.A. THAS, A Pemark on -the theoPem of Yanushka-Ronan aharaaterizing

the genex-alized he:x:agon H(q) arising from the (ll'Oup G2 (q) (preprint} •

[T3] J .A. THAS & S.E .• PAYNE, Classiaal finite genex-aZized quad:r>anglee: a
aorribinatoPial study, Ars Combinatoria ~ (1976) , 57-110.

[T4] J. TITS, SUP 7,e triaUU et cel'tains groupes qui s 'en Mduisent,
Publ. Math. I.H.E.S. ~ (1959), 14-60.
 99

[W1] B. WEISFEILER, On Con.atruction and Identification of Grp:pha, Lecture

Notes in Math. No. 558, Springer Verlag, Berlin-Heidelberg-New York,
1976.

[W2] H. WEYL, Das asymptotiache Verteilungagesetz der Eigenwerte Unearer

partieller Differentialgleichungen (mit einer Anwendung auf die
Theorie der Hohlraumstrahlung), Math. Ann. 21.. (1912), 441-479.
[W3] H.S. WHITE, F .N. COLE & L .D. CUMMINGS, Complete classification of the
triad systems on fifteen elements, Memoir Nat. Acad. Sci. _!!, second
memoir (1925), 1-89.

[W4] H.A. WILBRINK & A.E. BROUWER, A (57,14,1) strongZy reguZar graph does
not exist, Report ZW 121/78, Math. Centre Alllsterdam, 1978.
[W5] J.H. WILKINSON, The AZgebraic Eigenvalue Problem, Clarendon, Oxford,
1965.

[W6] R.J. WILSON, Introduction to graph Theory (second edition), Longman,

London, 1979.

[Yl] A. YANUSHKA, GeneraZized he:cagons of order (t,t}, Israel. J. Math. 23

(1976), 309-324.

[Y2] A. YANUSHKA, A sufficient condition for the existence of an associa-

tion scheme (preprint).
100

NOTATIONS

hermitian transpose of matrix A.

(A}ij ij-th entry of matrix A
(bij} matrix with entries bij"
!.. l (A) ~ ••• ~ An (A) eigenvalues (if real) of matrix A.
A!&>B Kronecker product of mat.rices A and B.
diag Ca , •.• , an) diagonal matrix with diagonal entries a 1 , ••. ,an.
1
I or In n x n identity matrix.
J all-one matrix.
n x n all-one matrix.
all-one vector.
length of vector u.
< u , ... ,u > linear span of vectors u 1 , ••• ,un.
1 n .L
< u , ... ,un > orthogonal complement of< u 1 , ••• ,un >.
1
G complement of graph G.
a(G) size of largest clique of G.
y(G) chromatic number of G.
w(G) size of largest coclique of G.
t.. 1 (G) ~ ••• ~ Xn (G) eigenvalues of G.
fi(G) multiplicity of eigenvalue >..i(G).
L(G) line graph of G.
D(G,G )
1
incidence structure formed by G and G1 , seep. 17.
Kn complete graph on n vertices.
KR.,m complete bipartite graph on R. + m vertices.
p(x,y) distance between vertices x and y, see p. 79.
k
pij(x,y) or pij intersection numbers of a graph, seep. 51 and p. 81.
/..(e ,e l
2
distance between elements of an n-gon, see p. 52.
1
pijk(L,x,y) see p. 58.
lZ integers.
lN positive ·integers.
lF field with q elements.
q
PG(n,q) n-dimensional projective geometry over lFq.
AG(n,q) n-dimensional affine geometry over JFq.
1 if i = j; 0 if i ~ j.
lower integer part of real number x.
upper integer part of real number x.
 101

INDEX

adjacency matrix 79 distance regular graph 51

adjacent vertices 79 dual of an incidence structure 83
balanced incomplete block design 83 edge 79
bipartite graph 80 eigenvalues of a graph 15,80
block 82 element of a generalized n-gon 52
block design 83 embeddable 83
block graph 35 empty incidence structure 82
cauchy inequalities 8 equivalent blocks 34,65
chromatic number 80 equivalent points 65
circuit 79,81 flag 82
class graph 35 (generalized) n-gon 50
Clebsch graph 39 (generalized) hexagon 53
clique 80 (generalized) polygon 50
coclique 80 (generalized) quadrangle 84
colour class 80 geometric graph 52
colourable, t- 80 Gewirtz graph 42
colouring 80 girth 79
complement of a graph 79 graph 79
complement of an inc. structure 83 graph, 40- 76
complete graph 79 Higman-Sims technique 10
complete y-partite graph 80 Hoffman-Singleton graph 42
component 79 imprimitive str. reg. graph 37,82
connected graph 79 incidence 82
cycle, n- 81 incidence graph 46,83
degenerate design 83 incidence matrix 80,82
degree 81 incidence structure 82
derived design 83 induced configuration 58
design 83 induced subqraph 80
design, t- 83 independent set of vertices 80
design, t - (v·,k,A) 83 interlacing eigenvalues 7
diameter 79 intersection numbers of a design 31
disconnected graph 79 intersection numbers of a graph 51
disjoint union of graphs 79 joint vertices 79
distance in a graph 79 lattice graph 82
distance in a generalized n-gon 52 length of a circuit 79
102

length of a path in a graph 79 quasi-symmetric 2-design 35

length of a path in a gen. n-gon 52 regular graph 81
line graph 67,84 residual design 83
linked 2 - (16,6,2) designs 45 Schlllfli graph 26
new block 68 Shrikhande graph cover
new point 68 singular values 5
null graph 79 special block 69
order of a generalized n-gon 50 strongly regular graph 81
oval 70,78 subconstituent 82
parameters of a t-design 83 subdesign 83
parameters of a partial geometry 84 subgraph 80
parameters of a str. reg. graph 81 substructure 82
partial geometry 84 symmetric 2-design 83
partial subgeometry 84 thick generalized n-gon 50
path in a generalized n-gon 52 tight interlacing 7
path in a graph 79 triangular graph 82
Petersen graph 82 type I, II block 65
point 82 valency 81
point graph 52,84 vertex 79
primitive str. reg. graph 37,82 void graph 79
pseudo-geometric graph 52
 103

SAMENVA'l'TING

Matrices en eigenwaarden zijn al vaak benut bij het bestuderen van

eindige grafen en incidentiestructuren (designs). Dit gebeurt ook in dit
proefschrift. Het uitgangspunt is een stelling over de eigenwaarden van ge-
partitioneerde matrices. Het toepassen van deze stelling op matrices geas-
socieerd met graf en of designs levert grenzen voor de grootte van deelgra-
f en, de afmetingen van subdesigns en de doorsnijdingsgetallen van designs.
Bet geval dat deze grenzen worden bereikt wordt ook behandeld. De genoemde
stelling geeft bovendien aanleiding tot een nieuw bewijs voor de ongelijk-
heden van A.J. Hoffman, betreffende de eigenwaarden van gepartitioneerde
matrices. Deze ongelijkheden leveren grenzen voor het kleuringsgetal van
een graaf. Mede dankzij die grenzen is het mogelijk alle 4-kleurbare sterk
reguliere grafen te bepalen.
De behandeling van het bovenstaande gebeurt in de eerste vier hoofd-
3
stukken. Hoofdstuk 5 behandelt de ongelijkheid t s s , alsook het geval
vangelijkheid,voor veralgemeende zeshoeken van de orde (s,t), s ~ 1. Dit
gebeurt met behulp van eigenwaardetechnieken. Dezelfde methode toegepast op
veralgemeende vierhoeken levert nieuwe bewijzen voor een aantal bekende
stellingen.
Met de hulp van eigenwaardemethoden kan men richtlijnen opstellen voor
het construeren van designs en grafen met speciale eigenschappen. Langs
deze weg worden er in Hoofdstuk 6 nieuwe 2-designs en sterk reguliere gra-
fen geconstrueerd.
104

CURRICULUM VITAE

De schrijver van dit proefschrift werd op 29 december 1948 te Eindhoven

geboren. In 1967 behaalde hij zijn H.B.S.-b diploma aan het Van der Putt-
lyceum te Eindhoven. Daarna studeerde hij aanvankelijk natuurkunde en later
wiskunde aan de 'l'echnische Bogeschool te Eindhoven. In oktober 1975 behaal-
de hij het diploma wiskundig ingenieur. Gedurende de laatste 2 jaren van
zijn studie was hij student-assistent bij het rekencentrum van de hoge-
school. Zijn afstudeerwerk, dat werd verricht onder leiding van Prof .dr.
J.J. Seidel, bestond uit het onderzoeken van relaties tussen sterke grafen
en 2-designs. Sinds zijn afstuderen werkt hij aks wetenschappelijk ambte-
naar aan de Technische Hogeschool te Eindhoven bij Prof.dr. J.J. Seidel.
 STELLINGEN

behorende bij het proefschrift

Eigenvalue techniques in design and

graph theory

door
W.H. Haemers
 STELLINGEN

Stel men wil boodschappen overseinen die gevormd warden met letters uit een
alfabet A van n letters genummerd van 1 t/m n, waarvan sommige paren letters
met elkaar verwarbaar zijn. Beschouw een nxn matrix M over een willekeurig
lichaam, die voldoet aan (M) i, i 'f 0 voor i = 1, .•• , n en (M)i ,j = 0 als de
ie en je letter van A niet verwarbaar zijn. Dan is de Shannon capaciteit
van A, die een maat is voor de hoeveelheid foutloos overseinbare informatie,
ten hoogste gelijk aan de rang van M.
Ref.: Willem Haemers, An upperbou'nd for the Shannon capacity of a graph,
Proc. Conf. Algebraic Methods in Graph Theory, Szeged, 1978 (te
verschijnenl.

Het antwoord op de vragen van Probleem I, 2 en 3 uit [2) is ontkennend.

Ref.: [1) Willem Haemers, On some problems of Lovasz concerning the
Shannon capacity of a graph, IEEE Trans. Information Theory ~

(1979) 231 - 232.

[2] Laszlo Lovasz, On the Shannon capacity of a graph, IEEE Trans.
Information Theory 25 (1979) 1 - 7.

Zeals bekend verscheen een speciaal geval van de stalling van Turan reeds
als opgave 28 in het tiende deel van "Wiskundige Opgaven met de Oplossingen"
(1910). De aldaar afgedrukte oplossing is echter onjuist.

Er bestaat geen stark reguliere graaf met 49 hoekpunten en graad 16.

Ref.: F.C. Bussemaker, w. Haemers, R. Mathon & H.A. Wilbrink, The non-
existence of a strongly regular graph with parameters (49, 16, 3, 6l
(in voorbereiding).
 5

Het is bekend dat de verbindingsmatrix van een sterk reguliere graaf soms
tevens de incidentiematrix van een symmetrisch 2-design is. Het is mogelijk
dat op deze manier twee niet isomorfe grafen isomorfe 2-designs opleveren.
Dit is echter niet mogelijk als een van de grafen een automorfismegroep
heeft van oneven orde. Met de resultaten van het onderstaande rapport vo:gt
hieruit dater tenminste 15531 niet isomorfe 2-(36, 15, 6) designs zijn.
Ref.: F.C. Bussemaker, R. Mathen & J.J. Seidel, Tables of ·two-graphs,
T.H.-rapport (in voorbereiding).

6
2
Voor het bestaan van een 3-(,k -,k+1,k,2)design is nodig dat k-3 een twaalvoud
is.

7
2
Als er voor n > 1 een projectief vlak van de orde 2 n-l bestaat, dan bes~aat
4 2 4 3
er een "near-square A-linked design" met 2 n-2 " punten en A • 2 "- 2 - 3.2 n- 2 +
- 2n-l+l, en is dus het antwoord op de eerste vraag van het Probleem uit §9
van onderstaand artikel, bevestigend.
Ref.: D.R. Woodall, Square A-linked designs, Proc. London Math. Soc. 20
(1970) 669 - 687.

La at M~ 0
-- ( s* Be) (B* ~<s der hermi·tl.'sch geconjungeerde van B) een hermitische

matrix zijn met grootste eigenwaarde A+ en kleinste eigenwaarde A-.

Veronderstel B heeft afmetingen mxn en gemiddelde rijsom r. Dan geldt
-A+A- <: r 2m/n.

Beschouw een rechthoekig veld, betegeld met een eindig aantal rechthoekige
tegels. Een tegel heet zuiderbuur van een andere tegel als de noordkant van de
eerste op dezelfde lijn ligt als de zuidkant van de andere tegel. Een noord-
zuid pad is een rijtje tegels waarbij elke tegel (behalve de eerste) zuiderbuur
is van zijn voorganger. Analoog is een oost-west pad gedefinieerd. Er geldt nu
dat elk paar tegels op een noord-zuid pad of op een cost-west pad ligt.
 10

De enige samenhangende planaire ster~ reguliere grafen zijn:

DO 11

Meetkundig inzicht is een van de meest toepasbare dingen die men zich
binnen de wiskunde kan verwerven. Daarom is het treurig dat het gewijzigde
onderwijsprogramma het ruimtelijk voorstellingsvermogen van de doorsnee
middelbare-school-verlater in tien jaar tijd met ongeveer een dimensie heeft
doen dalen.

Eindhoven, 30 oktober 1979 W.H. Haemers.