Classification of divisible design graphs with at most 39 vertices
Dmitry Panasenkoa,b, Leonid Shalaginova
a Chelyabinsk State University, Brat’ev Kashirinyh st. 129, Chelyabinsk, 454021, Russia
b Krasovskii Institute of Mathematics and Mechanics, S. Kovalevskaja st. 16, Yekaterinburg, 620990, Russia
arXiv:2109.12805v1 [math.CO] 27 Sep 2021
Abstract
A k-regular graph is called a divisible design graph (DDG for short) if its vertex set can be
partitioned into m classes of size n, such that two distinct vertices from the same class have
exactly λ1 common neighbors, and two vertices from different classes have exactly λ2 common
neighbors. A DDG with m = 1, n = 1, or λ1 = λ2 is called improper, otherwise it is called
proper. We present new constructions of DDGs and, using computer enumeration algorithm, we
find all proper connected DDGs with at most 39 vertices, except for three tuples of parameters:
(32, 15, 6, 7, 4, 8), (32, 17, 8, 9, 4, 8), (36, 24, 15, 16, 4, 9).
Keywords: divisible design graph, divisible design, walk-regular graph
2010 MSC: 05C50, 05E10, 15A18
1. Introduction
An incidence structure with constant block size k is a (group) divisible design whenever the
point set can be partitioned into m classes of size n, such that two points from one class occur
together in exactly λ1 blocks, and two points from different classes occur together in exactly λ2
blocks. A divisible design D is called symmetric or to have the dual property (SDD for short) if
the dual of D (that is, the design with the transposed incidence matrix) is again a divisible design
with the same parameters as D. A divisible design graph is a graph whose adjacency matrix is
the incidence matrix of a symmetric divisible design. A DDG with m = 1, n = 1, or λ1 = λ2 is
called improper (these DDGs are (v, k, λ)-graphs), otherwise it is called proper.
At first divisible design graphs were studied in master’s thesis by M.A. Meulenberg [8] and
then studied in more details in 2011 in two following papers by D. Crnkovic, W.H. Haemers,
H. Kharaghani and M.A. Meulenberg [2, 6] and in 2013 by D. Crnkovic and W.H. Haemers [3].
In 2008 M.A. Meulenberg presented the list of feasible parameters of proper DDGs up to
50 vertices. In 2011 and in 2013 feasible parameters of proper DDGs up to 27 vertices were
studied and the existence of graphs was resolved in all but one cases, however, the exact number
of graphs corresponding to these tuples of parameters remained unknown.
In this paper we present new constructions of DDGs and we find all proper connected
DDGs with at most 39 vertices, except for three tuples of parameters: (32, 15, 6, 7, 4, 8),
(32, 17, 8, 9, 4, 8), (36, 24, 15, 16, 4, 9).
The paper is organised as follows. In Section 2 we give some definitions, notations and
preliminaries about DDGs. In Section 3 we give an overview of some known constructions of
DDGs and in Section 4 we describe some new constructions. In Section 5 we describe some new
sporadic constructions of DDGs. In Section 6 we describe the algorithm used for enumerating
DDGs. In Section 7 we present table with enumeration results with constructions, which can be
used to obtain these graphs.
2. Preliminaries
Let It and Jt be an identity t × t and all-ones t × t matrix, respectively, and K(m,n) be Im ⊗ Jn =
diag(Jn , . . . , Jn ). Then a graph Γ is a DDG with parameters (v, k, λ1 , λ2 , m, n) if and only if Γ has
an adjacency matrix A that satisfies
A2 = kIv + λ1 (K(m,n) − Iv ) + λ2 (Jv − K(m,n) ).
In DDGs v = mn, and taking row sums on both sides of equation above gives
k2 = k + λ1 (n − 1) + λ2 n(m − 1).
A2 equation also gives us strong information about the eigenvalues of A and their multiplici-
ties.
√ √
Lemma
p 1 ([6,pLemma 2.1]). A has at most five distinct eigenvalues k, k − λ1 , − k − λ1 ,
k2 − λ2 v, − k2 − λ2 v with corresponding multiplicities 1, f1 , f2 , g1 , g2 , where f1 + f2 = m(n−1)
and g1 + g2 = m − 1.
Some of the multiplicities may be 0, and some values may coincide. In general, the multiplic-
ities f1 , f2 , g1 and g2 are not determined by the parameters, but if we know one, we can compute
the rest because f 1 + f 2 = m(n − 1), g1 + g2 = m − 1 and
p p
trace(A) = 0 = k + ( f1 − f2 ) k − λ1 + (g1 − g2 ) k2 − λ2 v.
This equation leads to the following result.
Lemma 2 ([6, Theorem 2.2]). Consider a proper DDG with parameters (v, k, λ1 , λ2 , m, n) and
eigenvalue multiplicities f1 , f2 , g1 , g2 . Then
(1) k − λ1 or k2 − λ2 v is a nonzero square.
(2) If k − λ1 is not a square, then f1 = f2 = m(n − 1)/2.
(3) If k2 − λ2 v is not a square, then g1 = g2 = (m − 1)/2.
Let V1 ∪ V2 ∪ . . . ∪ Vt be the partition of the vertex set of a graph Γ with the property that
every vertex of Vi has exactly ri j neighbours in V j . Then V1 ∪ V2 ∪ . . . ∪ Vt will be an equitable
t-partition of Γ. Matrix R = (ri j )t×t is called the quotient matrix of equitable partition.
Lemma 3 ([6, Theorem 3.1]). The vertex partition from the definition of a DDG (the canonical
partition) is equitable and the quotient matrix R satisfies
R2 = RRT = (k2 − λ2 v)Im + λ2 nJm .
p p
The eigenvalues of R are k, k2 − λ2 v, − k2 − λ2 v with corresponding multiplicities 1, g1 ,
g2 .
2
Lemma 4 ([6, Proposition 3.2]). The quotient matrix R of a DDG satisfies
P
ri j = k for j = 1, . . . , m,
i
ri2j = trace(R2 ) = mk2 − (m − 1)λ2 v,
P
i, j
p
0 6 trace(R) = k + (g1 − g2 ) k2 − λ2 v 6 m(n − 1).
A graph is walk-regular, whenever for every l > 2 the number of closed walks of length l at
a vertex x is independent of the choice of x. Note that walk-regularity implies regularity (take
l = 2).
A DDG with four distinct eigenvalues is walk-regular, provided it is connected ([2, Corollary
4.2]). A DDG with five distinct eigenvalues can also be walk-regular. To decide on this the
following lemma can be used.
Lemma 5 ([2, Theorem 4.3]). A proper DDG is walk-regular if and only if the quotient matrix
R has constant diagonal.
3. Known constructions
3.1. (v, k, λ)-graphs and designs
The incidence graph of a design with incidence matrix N is the bipartite graph with adjacency
matrix
" #
O N
N ⊤
O.
Construction 1 ([6, Construction 4.1]). The incidence graph of an (n, k, λ1 )-design with
1 < k 6 n is a proper DDG with λ2 = 0.
Proposition 1 ([6, Proposition 4.3]). For a proper connected DDG Γ the following are equiva-
lent.
(1) λ2 = 0.
(2) Γ comes from Construction 1.
Construction 2 ([6, Construction 4.4]). If A′ is the adjacency matrix of an (m, k′ , λ′ )-graph
(1 6 k′ < n), then A = A′ ⊗ Jn is the adjacency matrix of a proper DDG with k = λ1 = nk′ ,
λ1 = nλ′ .
Proposition 2 ([6, Proposition 4.5]). For a proper DDG Γ the following are equivalent.
(1) λ1 = k.
(2) Γ comes from Construction 2.
Construction 3 ([6, Construction 4.6]). Let A1 , . . . , Am (m > 2) be the adjacency matrices of
m (n, k′ , λ′ )-graphs with 0 6 k 6 n − 2. Then A = Jv − K(m,n) + diag(A1, . . . , Am ) is the adjacency
matrix of a proper DDG with k = k′ + n(m − 1), λ1 = λ′ + n(m − 1), λ2 = 2k − v.
Proposition 3 ([6, Proposition 4.7]). For a proper DDG Γ the following are equivalent.
(1) λ2 = 2k − v.
(2) Γ comes from Construction 3.
3
3.2. DDGs with λ1 = k − 1
The lexicographic product or graph composition G[H] of graphs G and H is a graph such
that the vertex set of G[H] is V(G) × V(H) and adjacency defined by
(u1 , u2 ) ∼ (v1 , v2 ) if and only if u1 ∼ v1 or (u1 = v1 and u2 ∼ v2 ).
Construction 4 ([6, Theorem 4.11]). Let Γ′ be a strongly regular graph with parameters
(m, k′ , λ, λ + 1). Then Γ′ [K2 ] is a DDG with parameters (2m, 2k′ + 1, 2k′ , 2λ + 2, m, 2). Or
let Γ′ be K x×y , the complete multipartite graph containing x parts of y vertices. Then Γ′ [K2 ] is a
DDG with parameters (2xy, 2y(x − 1) + 1, 2y(x − 1), 2y(x − 2) + 2, x, 2y).
An involutive automorphism of a graph is called Seidel automorphism if it interchanges only
non-adjacent vertices.
Construction 5 (Dual Seidel switching; [4, Construction 2]). Let Γ be a DDG obtained with
the first case of Construction 4. Let M be the adjacency matrix of Γ, and P be a non-identity
permutation matrix of the same size. Then PM is the adjacency matrix of a DDG if and only if P
represents a Seidel automorphism.
Proposition 4 ([4, Theorem 2]). For a proper DDG Γ with λ2 , 0 the following are equivalent.
(1) λ1 = k − 1.
(2) Γ comes from Construction 4 or 5.
3.3. Distance-regular graphs
Suppose that G is a graph with diameter d. For any vertex u and for any integer i with 0 6
i 6 d, let Gi (u) denote the set of vertices at distance i from u. If u′ ∈ Gi (u) and w is a neighbour
of u′ , then w must be at distance i − 1, i or i + 1 from u. Let ci , ai and bi denote the number of
such vertices w. G is a distance-regular graph if and only if parameters ci , ai , bi depends only
on the distance i, and not on the choice of u and u′ . An array {k, b1 , . . . , bd−1 ; 1, c2 , . . . , cd } is
called an intersection array of a distance-regular graph. A distance-regular graph of diameter d
is called antipodal if being at distance d or 0 defines an equivalence relation on the vertices. For
a distance-regular graph the parameters λ and µ give the number of common neighbors of a pair
of vertices at distance 1 and 2, respectively.
Distance regular graphs of diameter 2 are known as strongly regular graphs with parameters
(v, k, λ, µ).
Construction 6 ([6, Theorem 4.13]). Suppose Γ is an antipodal distance-regular graph of di-
ameter 3. If λ = µ, then Γ is a proper DDG with parameters (n(µn + 2), µn + 1, 0, µ, µn + 2, n).
If λ = µ − 2, then the complement of Γ is a proper DDG with parameters (µn2 , µn(n − 1),
µn(n − 2), µ(n − 1)2 , µn, n).
3.4. Partial complements
The partial complement of a DDG is a graph whose adjacency matrix can be obtained as the
complement of all blocks of the canonical partition except the diagonal blocks.
Proposition 5 ([6, Proposition 4.15]). The partial complement of a proper DDG Γ is again a
DDG if one of the following holds:
(1) The quotient matrix R equals t(Jm − Im ) for some t ∈ {1, . . . , n − 1}.
(2) m = 2.
4
An equitable partition of a (v, k, λ)-graph that satisfies (1) from Proposition 5 is a so-called
Hoffman coloring. Note that the diagonal blocks are zeros, so the partition corresponds to a
vertex coloring.
Construction 7 ([6, Construction 4.16]). Let Γ be a (v, k, λ)-graph. If Γ has a Hoffman coloring
or an equitable partition into two parts of equal size, then the partial complement is a DDG.
3.5. Hadamard matrices
An m × m matrix H is a Hadamard matrix if every entry is 1 or −1 and HH ⊤ = mIm . A
Hadamard matrix H is called graphical if H is symmetric with constant diagonal, and regular if
all row and column sums are equal.
Construction 8 ([6, Construction √ 4.8]). Consider a regular graphical Hadamard matrix H of
order m > 4 and row sum l = ± m. Let n > 2. Replace each entry with value −1 by Jn − In ,
and each +1 by In , then we obtain the adjacency matrix of a DDG with parameters (mn, n(m −
l)/2 + l, (n − 2)(m − l)/2, n(m − 2l)/4 + l, m, n).
Construction 9 ([6, Construction 4.9]). Consider a regular graphical Hadamard matrix H of
order l2 > 4 with diagonal entries −1 and row sum l. The graph with adjacency matrix
M N O
A = N O M ,
O M N
where " # " #
1 Jl2 + H Jl2 + H 1 Jl2 + H Jl2 − H
M= and N = ,
2 Jl2 + H Jl2 + H 2 Jl2 − H Jl2 + H
is a DDG with parameters (6l2 , 2l2 + l, l2 + l, (l2 + l)/2, 3, 2l2 ).
Construction 10 ([2, Theorem 3.2]). If there exist a regular graphical Hadamard matrix of or-
der 4u2 with row sum 2u and a Hadamard matrix of order 2u2 , then there exists a DDG with
parameters (24u2 , 12u2 − 2u, 4u2 − 2u, 6u2 − 2u, 12u2, 2).
3.6. DDGs with parameters (4n, n + 2, n − 2, 2, 4, n) and (4n, 3n − 2, 3n − 6, 2n − 2, 4, n)
An m × n-lattice graph is a line graph of complete bipartite graph Km,n .
Construction 11 ([9, Theorem 1]). An m × n-lattice graph is a DDG if and only if n = 4 (or
m = 4). These graphs have parameters (4n, n + 2, n − 2, 2, 4, n).
Construction 12 ([7, 9, Generalised dual Seidel
! switching]). Let M be the adjacency matrix
M11 M12
of a 4 × n-lattice graph. M = for some submatrices M11 , M12 , M21 , M22 of M,
M21 M22
! H, which isomorphic to 2 × n-lattice
where M11 is the adjacency matrix of an induced subgraph
P11 0
graph. Consider the permutation matrix P = 0 I , where P11 is the permutation matrix of
!
P11 M11 M12
Seidel automorphism ϕ corresponding to central symmetry of H. Then matrix
M21 M22
is adjacency matrix of DDG with parameters (4n, n + 2, n − 2, 2, 4, n).
5
Construction 13 ([9, Construction 5]). Consider C s [K2 ], where s = 4t, t > 1 and 4-cube.
Each of these graphs has equitable partition with quotient matrix J4 . Consider some copies of
C s [K2 ] and some copies of 4-cube with a given fixed equitable partition. Denote by V1 , V2 , V3 , V4
the classes of this partition. The switching of edges between V1 and V2 and also between V3 and
V4 gives DDG with parameters (4n, n + 2, n − 2, 2, 4, n).
Note that graphs obtained from Construction 11, Construction 12 and Construction 13 have
different spectra.
Proposition 6 ([9, Theorem 1]). Let Γ be a DDG with parameters (4n, n + 2, n − 2, 2, 4, n). Then
one of the following cases hold.
(1) If n is odd, then Γ is isomorphic to 4 × n-lattice graph.
(2) If n is even, then Γ comes from Construction 11, 12 or 13.
Construction 14 ([9, Theorem 2]). Let Γ be a DDG with parameters (4n, 3n − 2, 3n − 6, 2n −
2, 4, n). Then Γ can be obtained from DDG with parameters (4n, n + 2, n − 2, 2, 4, n) by switching
of edges between two classes and other two classes of canonical partition.
4. New constructions
Construction 15. Suppose Γ is an antipodal distance-regular graph of diameter 3 with antipo-
dal classes of size r. Denote by Ai the matrix of a relation ‘to be at distance i’ on the vertices
of Γ. If λ = µ + 2, then matrix A = A1 + A3 is a matrix of a DDG with parameters (r(2µ + 4),
2µ + r + 2, r − 2, µ + 2, 2µ + 4, r).
Proof. The intersection array of an antipodal distance-regular graph of diameter 3 is {k, µ(r − 1),
1; 1, µ, k} [1, p. 431]. From this the statement of the construction follows straightforwardly.
Let G be a k-regular graph on v vertices with the smallest eigenvalue λmin . A Hoffman col-
oring in G is a partition of the vertices into Hoffman-cocliques, that is, cocliques meeting the
Hoffman upper bound c = vλmin /(λmin − k).
Construction 16. Suppose Γ is a strongly regular graph with parameters (v, k, µ + 2, µ) and Γ
has Hoffman coloring with Hoffman-cocliques of size n (v = mn). Let A be an adjacency matrix
of Γ, in which Hoffman-cocliques are located on the main diagonal, K = K(m,n) , I = Iv . Then
matrix A + K − I is the adjacency matrix of a DDG with parameters (mn, k + n − 1, n + µ − 2,
2k
µ + m−1 , m, n).
Proof. Let’s calculate (A + K − I)2 (we denote J = Jv ).
(A + K − I)2 = A2 + K 2 + I 2 + AK + KA − 2A − 2K.
k
Since A2 = kI + λA + µ(J − I − A), K 2 = nK and AK = KA = m−1 (J − K),
2k
(A + K − I)2 = kI + λA + µ(J − I − A) + nK + I + (J − K) − 2A − 2K.
m−1
2k 2k
(A + K − I)2 = (k − µ + 1)I + (µ + )J + (n − − 2)K + (λ − µ − 2)A.
m−1 m−1
6
Since λ = µ + 2,
2k 2k
(A + K − I)2 = (k − µ + 1)I + (µ + )J + (n − − 2)K,
m−1 m−1
which can rewritten as
2k
(A + K − I)2 = (k + n − 1)I + (n + µ − 2)(K − I) + (µ + )(J − K).
m−1
So, A + K − I is an adjacency matrix of a DDG with parameters (mn, k + n − 1, n + µ − 2,
2k
µ + m−1 , m, n).
Construction 17. Let Γ be a DDG with parameters (v, k, λ1 , λ2 , m, n) and quotient matrix
R = aIm + b(Jm − Im ). Take s copies of Γ and label all blocks of canonical partition in each
copy with numbers 1, . . . , m. Then connect all vertices from blocks with the same label (ad-
jacency inside the block does not change). The resulting graph is a DDG with parameters
(vs, k + (s − 1)n, λ1 + (s − 1)n, λ2, ms, n), if λ2 = 2b = 2a + (s − 2)n.
Proof. Consider two vertices from the same block of the same copy of Γ. Then they have λ1
common neighbours in Γ plus n common neighbours in each other copy of Γ. Thus, the number
of common neighbours equals λ1 + (s − 1)n.
Consider two vertices from different blocks of the same copy of Γ. Then they have λ2 com-
mon neighbours in Γ and no new common neighbours in other copies of Γ. Thus, the number of
common neighbours equals λ2 .
Consider two vertices from the same block of different copies of Γ. Then they have a common
neighbours in their blocks (2a in total) plus n common neighbours in each other copy of Γ. Thus,
the number of common neighbours equals 2a + (s − 2)n.
Consider two vertices from different blocks of different copies of Γ. Then they have b com-
mon neighbours in their copies of Γ (2b in total) and no new common neighbours in other copies
of Γ. Thus, the number of common neighbours equals 2b.
So, the resulting graph is a DDG with parameters (vs, k + (s − 1)n, λ1 + (s − 1)n, λ2 , ms, n), if
λ2 = 2b = 2a + (s − 2)n.
Construction 18. Let D be a symmetric divisible design with parameters (v, k, λ1 , λ2 , m, n), such
that every block contains the same number of points mk from each class. Γ is the incidence graphs
of D. Construct new graph Γ∗ with the same vertex set as Γ and vertices x, y are adjacent in Γ∗
when x, y are adjacent in Γ, or x, y are points from different classes of D, or x, y are blocks from
different classes of dual design of D. Γ∗ is a DDG with parameters (2v, k +(m−1)n, λ1 +(m−1)n,
λ2 + (m − 2)n, 2m, n), if 2(m − 1) mk = λ2 + (m − 2)n.
Proof. Consider two vertices, corresponding to two points from the same class of D. They have
λ1 common neighbours in G plus n common neighbours in each other class of points. Thus, we
have λ1 + (m − 1)n common neighbours. Similarly for two vertices, corresponding to two blocks
from the same class.
Consider two vertices, corresponding to two points from different classes of D. They have
λ2 common neighbours in G plus n common neighbours in each other class of points. Thus, we
have λ2 + (m − 2)n common neighbours. Similarly for two vertices, corresponding to two blocks
from different classes.
7
Consider two vertices, the first corresponding to a point of D and the second corresponding
to a block of D. They have no common neighbours in G plus mk common neighbours in each
class of points except for the class the first vertex belongs to and plus mk common neighbours in
each class of blocks except for the class the second vertex belongs to. Thus, we have 2(m − 1) mk
common neighbours.
So, Γ∗ is a DDG with parameters (2v, k + (m − 1)n, λ1 + (m − 1)n, λ2 + (m − 2)n, 2m, n), if
2(m − 1) mk = λ2 + (m − 2)n.
Construction 19. Let D be a symmetric divisible design with parameters (v, k, λ1 , λ2 , m, n) with
even m, such that every block contains the same number of points mk from each class. Γ is the
incidence graphs of D. Split the set of classes of blocks and the set of classes of points into pairs.
Construct a new graph Γ∗ with the same vertex set as Γ and vertices x, y are adjacent in Γ∗ when
x, y are adjacent in Γ or x, y are from different classes of the same pair of classes partition. Γ∗ is
a DDG with parameters (2v, k + n, λ1 + n, λ2 , 2m, n), if 2k m = λ2 .
Proof. The proof is similar to Construction 18.
A weighing matrix W(n, k) of order n and weight k is an n × n (0, 1, −1)-matrix, such that
WW T = kIn .
Construction 20. Let W be a (4t, 4(t − 1))-weighing matrix, such that the main diagonal of W
contains blocks of zeros of size 4. Construct matrix A′ by replacing each 0 with O2 , each 1 with
I2 and each −1 with J2 − I2 . Then matrix A = A′ + It ⊗ ((J4 − I4 ) ⊗ J2 ) is the adjacency matrix of
a DDG Γ with parameters (8t, 4t + 2, 6, 2t + 2, 4t, 2).
Proof. Consider two vertices corresponding to two rows of A, which were obtained from one row
of W after replacing. These vertices form a block of canonical partition of Γ. They have only 6
common neighbours in Γ, which come from addition of It ⊗ ((J4 − I4 ) ⊗ J2 ).
Consider two vertices corresponding to two rows of A, which were obtained from different
rows of W after replacing. We have two cases: these vertices correspond to the same block of
zeros from W or they correspond to different blocks of zeros from W.
Consider two vertices corresponding to the first case. They have 4 common neighbours from
addition of It ⊗ ((J4 − I4 ) ⊗ J2 ) plus 4(t − 1)/2 common neighbours from replacing. Thus, these
vertices have 2t + 2 common neighbours.
Consider two vertices corresponding to the second case. They have 3 common neighbours
in each block of Γ, which correspond to block of zeros from W. Thus, they have 6 common
neighbours. Plus they have 4(t − 2)/2 common neighbours from replacing. Thus, these vertices
have 2t + 2 common neighbours total.
So, Γ is a DDG with parameters (8t, 4t + 2, 6, 2t + 2, 4t, 2).
Construction 21. Let Γ be a DDG obtained from Construction 20 with adjacency matrix A. The
main diagonal of the A consists of It ⊗ ((J4 − I4 ) ⊗ J2 ), which gives a partition of Γ into complete
multipartite graphs with 4 parts of size 2. Construct new graph Γ′ by removing the edges of
the complete bipartite subgraph K4,4 from each part of this partition. Then Γ′ is a DDG with
parameters (8t, 4t − 2, 2, 2t − 2, 4t, 2).
Proof. The proof is similar to Construction 20.
8
5. Sporadic constructions
Construction 22. The following matrix M is the adjacency matrix of a DDG with parameters
(27, 8, 4, 2, 9, 3):
O J−I J−I J−I J−I O O O O
J − I O T1 T2 T3 J O O O
J − I T 1 O T 3 T 2 O J O O
J − I T2 T3 O T1 O O J O
M = J − I T3 T2 T1 O O O O J ,
O J O O O J−I I I I
O O J O O I J−I I I
O O O J O I I J−I I
O O O O J I I I J−I
1 0 0 0 1 0 0 0 1
where J = J3 , I = I3 , T 1 = 0 0 1 , T 2 = 1 0 0 , T 3 = 0 1 0 .
0 1 0 0 0 1 1 0 0
Note that blocks 2, 3, 4, 5 induce Deza graph with parameters (12, 3, 1, 0) and blocks 6, 7, 8,
9 induce 4 × 3-lattice graph.
Construction 23. The following matrix M is the adjacency matrix of a DDG with parameters
(28, 6, 2, 1, 7, 4):
O A B C O O O
AT D O O E O O
T
B O D O O E O
M = C T O O D O O E , where
O E O O O A C
O O E O AT O F
O E CT FT
O O O
1 1 0 0 1 1 0 0 1 1 0 0
1 1 0 0
0 0 1 1
0 0 1 1
A = 0 0 1 1
, B = , C =
1 ,
1 1 0 0 0 0 1
0 0 1 1 0 0 1 1 1 1 0 0
0 1 1 0 1 0 1 0 1 0 0 1
1 0 0 1 0 1 0 1 0 1 1 0
D = , E = , F = .
1 0 0 1 1 0 1 0 0 1 1 0
0 1 1 0 0 1 0 1 1 0 0 1
Remark 1. The following matrix M is the adjacency matrix of a DDG with parameters
(32, 10, 2, 3," 8, 4): #
A B
M = , where A and C are not necessarily the same adjacency matrices of
BT C
Shrikhande graph and B is an incidence matrix of some symmetric divisible design with pa-
rameters (16, 4, 0, 1, 4, 4).
Note that there are two non-isomorphic DDGs with parameters (32, 10, 2, 3, 8, 4) and both of
them have the structure described above.
9
6. Enumeration algorithm
6.1. Search for feasible parameters
At the first step, for a fixed number of vertices v, we calculate all feasible parameters
(v, k, λ1 , λ2 , m, n) of DDGs with the approach based on M.A. Meulenberg’s work [8]:
(1) All possible number of classes m and class sizes n are calculated, m ∗ n must be equal to
v.
(2) k runs from 3 to v − 3, both λ1 and λ2 run from max(0, 2k − v) to k, λ1 , λ2 .
(3) For the remaining possibilities it is checked whether k2 − λ2 v or k − λ1 are squares. If one
of them is not a square, then the corresponding multiplicities are equal. If both of them are not
squares, then the corresponding parameters are excluded.
(4) The following equalities checked:
• k2 = k + λ1 (n − 1) + λ2 n(m − 1);
• λ1 (n − 1) + λ2 n(m − 1) ≡ 0 mod k − 1 [8, Corollary 3.2.1];
• v(λ1 (n − 1) + λ2 n(m − 1)) ≡ 0 mod k(k − 1) [8, Corollary 3.2.1].
(5) The following cases are excluded [8, Chapter 5.1]:
• λ1 = k − 1 and n is odd.
• k2 − λ2 v = k − λ1 .
• f1 = f2 and g1 = g2 .
• f1 = f2 or g1 = g2 are not integers.
• If only one of k2 − λ2 v or k − λ1 is not a square, the yet undetermined multiplicities can be
computed. The case when they are not integers is excluded.
6.2. Enumeration of quotient matrices
Given feasible parameters (v, k, λ1 , λ2 , m, n) of a DDG, we initially construct all possible
quotient matrices R using the following method:
(1) We generate all possible r11 , 0 6 r11 6 n − 1.
(2) For all obtained r11 we generate all unordered partitions of k−r11 as sum of m−1 numbers.
We use unordered partitions since the permutation of r1 j , j > 2 does not change the final result.
For example, these two rows will give the same quotient matrix up to classes renumbering:
(1 5 5 7), (1 7 5 5).
(3) We check if generated row satisfies the equality for R2 = (k2 − λ2 v)Im + λ2 nJm . If not, we
reject this row. Also in case of odd n all rii must be even numbers.
(4) We repeat steps (1)–(3) for the next rows, but we generate ordered partitions instead.
After we generate all possible i > 2 rows of the quotient matrix, we leave only non-equal up to
classes renumbering options.
After we generate all non-equal quotient matrices for given parameters, in the case where
both k2 − λ2 v and k − λ1 are squares, we also compute all possible values of g1 using the equation
p p
trace(A) = 0 = k + ( f1 − f2 ) k − λ1 + (g1 − g2 ) k2 − λ2 v.
Then we find the corresponding multiplicity for the generated matrix and check if the found
value is among the calculated possible values of g1 . If not, we reject this quotient matrix.
10
6.3. Constructing adjacency matrices
For given quotient matrix R and tuple of parameters (v, k, λ1 , λ2 , m, n) we generate all possible
rows of the adjacency matrix, relying on known partition into classes.
We say that two partially filled matrices are equivalent if the graphs determined by them are
isomorphic with respect to partition into classes.
After adding a new row of the adjacency matrix, all partial matrices were checked for equiv-
alence and only nonequivalent were considered for adding the next row.
7. Enumeration results
In the table below, # gives a number of non-isomorphic DDGs with specified parameters;
f f g g
v, k, λ1 , λ2 , m, n are the parameters of a DDG; θ11 , θ22 , θ31 , θ42 are non-principal eigenvalues with
corresponding multiplicities; ‘WR’ denotes whether a graph is walk-regular. The column ‘con-
structions’ describes constructions from Section 4, which can be used to obtain specified graphs.
Tuples of parameter with λ1 ∈ {k − 1, k} and λ2 ∈ {0, 2k − v}, which have been characterized in
Sections 4.1 and 4.2, are omitted.
Note that for # ‘!’ denotes the exact number of non-isomorphic graphs, ‘+’ denotes the
number of known non-isomorphic graphs (there can be more DDGs with given parameters), ‘?’
denotes that the question of the existence of DDGs with given parameters remains unresolved.
For the ‘constructions’ column, ‘c’ is short for ‘construction‘, ‘p’ is short for ‘proposition‘,
‘r’ is short for ‘remark‘, ‘DSS’ means ‘dual Seidel switching’, ‘PC’ means ‘partial complement’,
‘L(G)’ denotes the line graph of G, ‘J(m,n)’ denotes Johnson graph, ‘T(n)’ denotes triangular
graph. For Construction 14, we note with ‘.’ the number of the construction for which the
switching of edges was implemented.
Table 1: Proper DDGs with v 6 39, 0 6 λ1 < k − 1, λ2 < {0, 2k − v}
f f g g
# v k λ1 λ1 m n θ11 θ22 θ3 1 θ4 2 WR constructions
1 3 3
1! 8 4 0 2 4 2 2 −2 0 – + 2 × 4-lattice [c8, c11]
√ 3 √
1! 12 5 0 2 6 2 5 − 53 – −15 + icosahedron [c6]
1! 12 5 1 2 4 3 22 −26 13 – + 3 × 4-lattice [c8, c11]
1! 12 6 2 3 3 4 23 −26 02 – + L(octahedron)
1! 12 7 3 4 4 3 22 −26 12 −11 + c8, c14.11
1! 15 4 0 1 5 3 25 −25 – −14 + L(Petersen graph) [c6]
√ 6 √
1! 18 9 6 4 6 3 3 − 36 31 −34 + c18 from (9, 3, 0, 1)-SDD
1! 20 7 3 2 4 5 24 −212 33 – + 5 × 4-lattice [c8, c11]
1! 20 9 0 4 10 2 34 −36 13 −16 – DSS(J(6,3))
1! 20 9 0 4 10 2 35 −35 – −19 + J(6,3) [c6]
1! 20 13 9 8 4 5 24 −212 32 −31 + c8, c14.11
√ 1 √
1! 24 6 2 1 3 8 29 −212 12 − 121 – c9
√ 8 √
1! 24 7 0 2 8 3 7 − 78 – −17 + Klein graph [c6]
1! 24 8 4 2 4 6 25 −215 43 – + 6 × 4-lattice [c8, c11]
1! 24 8 4 2 4 6 27 −213 42 −41 – c12
6! 24 8 4 2 4 6 29 −211 41 −42 + c13
√ 6 √
5! 24 10 2 4 12 2 8 − 86 23 −28 + c21
√ 8 √
2! 24 10 3 4 8 3 7 − 78 21
−26 + one graph is Cayley (see [5])
√ 1 √
1! 24 10 6 3 3 8 28 −213 28 − 281 – c9
√ 6 √
1! 24 14 6 8 12 2 8 − 86 22 −29 + c10, c20
√ 8 √
1! 24 14 7 8 8 3 7 − 78 – −27 + PC(Klein graph) [p5]
11
f f g g
# v k λ1 λ1 m n θ11 θ22 θ3 1 θ4 2 WR constructions
5 15 2
1! 24 16 12 10 4 6 2 −2 4 −41 + c8, c14.11
1! 24 16 12 10 4 6 27 −213 41 −42 – c14.12
4! 24 16 12 10 4 6 29 −211 – −43 + c14.13
√ 4 √
1! 27 8 4 2 9 3 27 −211 10 − 104 – c22
2! 27 18 9 12 9 3 36 −312 08 – + c16 from Schläfli graph
√ 3 √
1! 28 6 2 1 7 4 29 −212 8 − 83 – c23
1! 28 9 5 2 4 7 26 −218 53 – + 7 × 4-lattice [c8, c11]
√ 7 √
1! 28 13 0 6 14 2 13 − 137 – −113 + Taylor graph [c6]
16! 28 13 4 6 7 4 39 −312 11 −15 – –
56! 28 15 6 8 7 4 37 −314 16 – + c16 from T(8), Chang graphs
4! 28 15 6 8 7 4 38 −313 13 −13 – DSS of the entry above
1! 28 19 15 12 4 7 26 −218 52 −51 + c8, c14.11
√ 12 √
2! 32 10 2 3 8 4 8 − 812 21 −26 + r1
1! 32 10 6 2 4 8 27 −221 63 – + 8 × 4-lattice [c8, c11]
1! 32 10 6 2 4 8 210 −218 62 −61 – c12
15! 32 10 6 2 4 8 213 −215 61 −62 + c13
√ 8 √
15! 32 14 2 6 16 2 12 − 128 24 −211 + c21
2+ 32 15 6 7 4 8 312 −316 – −13 + Cayley graphs (see [5])
1! 32 16 0 8 16 2 46 −410 015 – + c15 from halved 6-cube
? 32 17 8 9 4 8 317 −311 12 −11 – –
√ 8 √
1! 32 18 6 10 16 2 12 − 128 23 −212 + c20
1! 32 22 18 14 4 8 27 −221 62 −61 + c8, c14.11
1! 32 22 18 14 4 8 210 −218 61 −62 – c14.12
9! 32 22 18 14 4 8 213 −215 – −63 + c14.13
2! 35 12 3 4 7 5 312 −316 23 −23 – –
3854! 35 12 3 4 7 5 314 −314 – −26 + c7 from (35, 18, 9)-graphs
√ 12 √
3! 36 9 3 2 12 3 6 − 612 34 −37 + two graphs from c19
√ 9 √
7! 36 9 4 2 18 2 5 − 59 37 −310 + c17 from icosahedron
1! 36 11 7 2 4 9 28 −224 73 – + 8 × 4-lattice [c8, c11]
√ 9 √
1! 36 17 0 8 18 2 17 − 179 – −117 + Taylor graph [c6]
3+ 36 24 15 16 4 9 312 −320 03 – + Cayley graphs (see [5])
1! 36 25 21 16 4 9 28 −224 72 −71 + c8, c14.11
√ 12 √
1! 36 27 21 20 12 3 6 − 612 31 −310 + c18 from (18, 12, 6, 8)-SDD
√ 9 √
2! 38 9 0 2 19 2 38 −311 5 − 59 – –
8. Conclusion
The list of proper divisible design graphs up to 39 vertices is available by
http://alg.imm.uran.ru/dezagraphs/ddgtab.html. This web page provides access to
adjacency matrices, quotient matrices and WL-closures of the graphs we found.
Acknowledgments
The reported study was funded by RFBR according to the research project 20-51-53023.
References
[1] A.E. Brouwer, A. Cohen, A. Neumaier, Distance-Regular Graphs, Springer, 1989.
[2] D. Crnkovic, W.H. Haemers, More about Divisible Design Graphs. CentER Discussion Paper, Vol. 2011-140
(2011).
12
[3] D. Crnkovic, W.H. Haemers, Walk-regular divisible design graphs. Designs, Codes and Cryptography, 72 (2014)
165–175.
[4] S. Goryainov, W.H. Haemers, V.V. Kabanov, L. Shalaginov, Deza graphs with parameters (n, k, k − 1, a) and β = 1,
Journal of Combinatorial Designs, 17(3) (2019) 188–202.
[5] S. Goryainov, L. Shalaginov, Cayley-Deza graphs, on less than 60 vertices, Siberian Electronic Mathematical
Reports, 11 (2014) 268–310 (in Russian).
[6] W.H Haemers, H. Kharaghani, M.A. Meulenberg, Divisible Design Graphs, Journal of Combinatorial Theory,
Series A, 118 (2011) 978–992
[7] V.V. Kabanov, E.V. Konstantinova, L. Shalaginov, Generalised dual Seidel switching and Deza graphs with strongly
regular children, Discrete Mathematics, 344(3) (2021).
[8] M.A. Meulenberg, Divisible Design Graphs, Master’s thesis, Tilburg University, 2008.
[9] L. Shalaginov, Divisible design graphs with parameters (4n, n + 2, n − 2, 2, 4, n) and (4n, 3n − 2, 3n − 6, 2n − 2, 4, n)
https://arxiv.org/abs/2106.08677
13
