Distance matrices on the H − join of graphs: a

general result and applications

Domingos M. Cardoso∗
Roberto C. Díaz and Oscar Rojo†

Abstract

Given a graph H with vertices 1, . . . , s and a set of pairwise vertex disjoint graphs
G1 , . . . , Gs , the vertex i of H is assigned to Gi . Let G be the graph obtained from the graphs
G1 , . . . , Gs and the edges connecting each vertex of Gi with all the vertices of Gj for all edge
ij of H. The graph G is called the H − join of G1 , . . . , Gs . Let M ( G ) be a matrix on a graph G.
A general result on the eigenvalues of M ( G ), when the all ones vector is an eigenvector of
M ( Gi ) for i = 1, 2, . . . , s, is given. This result is applied to obtain the distance eigenvalues,
the distance Laplacian eigenvalues and as well as the distance signless Laplacian eigenvalues
of G when G1 , . . . , Gs are regular graphs. Finally, we introduce the notions of the distance
incidence energy and distance Laplacian-energy like of a graph and we derive sharp lower
bounds on these two distance energies among all the connected graphs of prescribed order
in terms of the vertex connectivity. The graphs for which those bounds are attained are
characterized.

MSC 2010: 05C50, 05C76, 05C35, 15A18.

Keywords: Graph operations, vertex connectivity, distance matrix, eigenvalues, distance inci-
dence energy, distance Laplacian-energy like.

1 Introduction


The distance matrix of a graph G of order n is the n × n matrix D( G ) = di,j , indexed by the
vertices of G, where di,j is the distance (number of edges of a shortest path) between vertices
vi and v j . The very beginning of the distance matrix research goes back to the thirties of
the twentieth century [22] and [33]. The main motivation from graph theory point of view
was the problem of realizability of distance matrices, first presented in [8] and investigated in
[23, 24, 25, 29], [5], [30] and [4], among many other papers. However, it was proven in [1] and
∗ Department of Mathematics, Universidade de Aveiro, Aveiro, Portugal.
† Department of Mathematics, Universidad Católica del Norte, Antofagasta, Chile.

1
[31] that the problem of determining the minimum weight graph realization of a distance matrix
with integer entries is NP-complete. The distance matrices have also deserved the attention of
the spectral graph theory community since the paper published by Graham and Pollak in 1971
[14], where a relationship between the number of negative eigenvalues of the distance matrix
and the addressing problem in data communication systems is established. In the same paper
the authors also proved the following very impressive result, which drew the attention of many
researchers, transforming the spectral properties of distance matrices in a hot topic of spectral
graph theory. Assuming that T is a tree of order n ≥ 2 with distance matrix D( T ), then

detD( T ) = (−1)n−1 (n − 1) 2n−2 . (1)

Thus, the determinant of the distance matrix of a tree depends only from its number of vertices.
The concepts of distance Laplacian and distance signless Laplacian were introduced in [3], where
the authors proven the equivalence between the distance signless Laplacian, distance Laplacian
and the distance spectra for the class of transmission regular graphs.
A very complete survey of the state of the art on distance matrices up to 2014 appears in [2]
(see also [27]). More recently, extremal graphs were characterized in terms of the eigenvalues
of distance signed Laplacian and distance Laplacian matrices. Namely, the graphs of order n
with least distance signed Laplacian eigenvalue equal to n − 2 and the graphs of order n ≥ 11
with second least distance signed Laplacian eigenvalue in the interval [n − 2, n] [19]; the graphs
of order n with largest distance Laplacian eigenvalue equal to n − 2 [18] (in this paper the
authors proven the conjecture proposed in [2] that the multiplicity of the largest eigenvalue of
the distance Laplacian matrix of a non complete graph is not greater than n − 2 with equality
if and only if either the graph is isomorphic to a star or to a regular complete bipartite graph).

This paper is devoted to the determination of distance, distance Laplacian and distance
signless Laplacian eigenvalues of graphs obtained by the H-joint operation [7] over a family of
regular graphs. Furthermore, sharp lower bounds on distance incidence energy and distance
Laplacian-energy like of a graph are deduced for graphs of prescribed order in terms of the
vertex connectivity and the graphs for which these bounds are attained are characterized.
In the remaining part of this section, we introduce the notation and basic definitions of the
concepts used throughout the paper.
Let G = (V ( G ) , E ( G )) be a simple undirected graph of order n, that is, on n vertices,
with vertex set V ( G ) and edge set E ( G ) . The cardinality of V ( G ) is called the order of G.
If e ∈ E ( G ) has end vertices u and v, then we say that u and v are adjacent and this edge is
denoted by uv. If u ∈ V ( G ), then NG (u) is the set of neighbors of u in G, that is, NG (u) = {v ∈
V ( G ) : uv ∈ E ( G )}. The cardinality of NG (u) is said to be the degree of u. A graph G is called
d-degree regular when every vertex has the same degree equal to d.
The vertex connectivity (or just connectivity) of a graph G, denoted by κ ( G ) , is the minimum
number of vertices of G whose deletion disconnects G. It is conventional to define κ (Kn ) =
n − 1.
The adjacency matrix of a graph G of order n, A( G ), is a 0 − 1-matrix of order n with entries

2
aij such that aij = 1 if ij ∈ E( G ) and aij = 0 otherwise. Other matrices on G are the Laplacian
matrix L ( G ) = D ( G ) − A ( G ) and the signless Laplacian matrix Q ( G ) = D ( G ) + L ( G ) ,
where D ( G ) is the diagonal matrix of vertex degrees. It is well known that L ( G ) and Q ( G )
are positive semidefinite matrices and that (0, 1) is an eigenpair of L ( G ) where 1 is the all
ones vector. Let us denote by σ( M) the spectrum (the multiset of eigenvalues) of a square
matrix M. An eigenvalue of a matrix M will be denoted by λ( M ) and throughout the paper,
assuming that M has order n, the eigenvalues M are indexed in non increasing order, that is,
λ1 ( M) ≥ · · · ≥ λn ( M). Given an eigenvalue λi ( M) its eigenspace will be denoted by Λλi ( M ).
If M is a nonnegative matrix then, by the Perron-Frobenius Theorem, M has an eigenvalue
equal to its spectral radius, called the Perron root of M. In addition, if M is irreducible then the
Perron root of M is a simple eigenvalue with a corresponding positive eigenvector, called the
Perron vector of M.
Given two vertex disjoint graphs G1 and G2 , the join of G1 and G2 is the graph G = G1 ∨ G2
such that V ( G ) = V ( G1 ) ∪ V ( G2 ) and E ( G ) = E( G1 ) ∪ E( G2 ) ∪ { xy : x ∈ V ( G1 ) , y ∈ V ( G2 )} .
This join operation can be generalized as follows [6, 7]: Let H be a graph of order s. Let
V ( H ) = {1, . . . , s}. Let { G1 , . . . , Gs } be a set of pairwise vertex disjoint graphs. For 1 ≤ i ≤ s,
the vertex i ∈ V ( H ) is assigned to the graph Gi . Let G be the graph obtained from the graphs
G1 , . . . , Gs and the edges connecting each vertex of Gi with all the vertices of Gj if and only if
ij ∈ E ( H ) . That is, G is the graph with vertex set V ( G ) = is=1 V ( Gi ) and edge set
S

!  
s
[ [
E( G ) = E( Gi ) ∪ {uv : u ∈ V ( Gi ), v ∈ V ( Gj )} .
i =1 ij∈ E( H )

This graph operation introduced in [6] under the designation of H − join of the graphs G1 , . . . , Gs
it is denoted by _
G = { Gi : 1 ≤ i ≤ s}.
H
The same graph operation was introduced in [11] as a generalization of the lexicographic prod-
uct [9, 10]. In [11] this graph operation was designated generalized composition and denoted by
H [ G1 , G2 , . . . , Gs ]. Clearly if each Gi is a graph of order ni , then the H − join of G1 , . . . , Gs (gen-
eralized composition H [ G1 , G2 , . . . , Gs ]) is a graph of order n = n1 + n2 + . . . + ns . The same
operation appear in [26] under the designation of joined union. In [26] the distance spectrum
of the joined union of regular graphs is determined and this technique is applied to the con-
struction of distance equienergetic graphs with diameter greater than two.
As usual, let Pn , Cn , Sn and Kn be the path, cycle, star and the complete graph on n vertices,
respectively.

H { G1 , G2 , G3 }
W
Example 1 Let H = P3 , G1 = P2 , G2 = K4 and G3 = P3 . Then the graph is depicted
in Figure 1.

3
 P3 { P2 , K4 , P3 }.
W
Figure 1. The graph

1
The Wiener index W ( G ) of a connected graph G is W ( G ) = 2 ∑u,v∈V (G) d(u, v),

1
2 u,v∈∑
W (G) = d(u, v),
V (G)

where d(u, v) is the distance between u, v ∈ V ( G ), that is, the length of the shortest path
connecting u and v. The transmission Tr (v) of a vertex v ∈ V ( G ) is the sum of the distances
from v to all other vertices of G, that is,

Tr (v) = ∑ d(v, u).

u ∈V ( G )

A graph H is said to be r − transmission regular if Tr (v) = r for each vertex v ∈ V ( H ). The

eigenvalues of the distance matrix D( G ) of the graph G are called the distance eigenvalues of G
and they are denoted by
∂1 ( G ) ≥ ∂2 ( G ) ≥ . . . ≥ ∂ n ( G ).
In [3] Aouchiche and Hansen introduce, for a connected graph G, the distance Laplacian
matrix L( G ) and the signless Laplacian matrix Q( G ), respectively, as follows

L( G ) = Tr( G ) − D( G ) and Q( G ) = Tr( G ) + D( G ),

where Tr( G ) = diag[Tr(v1 ), Tr(v2 ), . . . , Tr(vn )] is the diagonal matrix of the vertex transmissions
in G. The eigenvalues of L( G ) and Q( G ) are called the distance Laplacian eigenvalues and the
distance signless Laplacian eigenvalues of G and they are denoted by

∂1L ( G ) ≥ ∂2L ( G ) ≥ . . . ≥ ∂nL ( G ) and ∂1Q ( G ) ≥ ∂2Q ( G ) ≥ . . . ≥ ∂nQ ( G ),

respectively. Notice that L( G ) and Q( G ) are both real symmetric matrices and then, from
Geršgorin’s Theorem, it follows that their eigenvalues are nonnegative real numbers. Let 1 be

4
the all one vector. Clearly each row sum of L( G ) is 0. Then (0, 1) is an eigenpair of L( G ) and,
when G is connected graph, 0 is a simple eigenvalue.
It is clear that if G is a r − transmission regular graph, then

L( G ) = rIn − D( G ) and Q( G ) = rIn + D( G ),

where In is the identity matrix of order n and, for i = 1, . . . , n, ∂iL ( G ) = r − ∂n−i+1 ( G ) and
∂iQ ( G ) = r + ∂i ( G ).
A basic result on ∂1L ( G ) is the following:

Theorem 1 [3, Cor. 3.6] Let G be a connected graph of order n ≥ 3. Then

∂iL ( G ) ≥ ∂iL (Kn ) = n, for i = 1, 2, . . . , n − 1,

and ∂nL ( G ) = ∂nL (Kn ) = 0.

Throughout this paper, G = H { Gi : 1 ≤ i ≤ s} where, for 1 ≤ i ≤ s, it is assumed that Gi

W

is a graph order ni such that the all one vector 1ni is an eigenvector for the eigenvalue µi of Gi .
Moreover, I and 0 are the identity and the zero matrices of the appropriate order, respectively.
Let M( G ) be a matrix on a graph G. In this paper, a general result on the eigenvalues of
M ( G ), when the all one vector is an eigenvector of M ( Gi ) for i = 1, 2, . . . , s, is given. This
result is applied to obtain the distance eigenvalues, the distance Laplacian eigenvalues and as
well as the distance signless Laplacian eigenvalues of G when G1 , . . . , Gs are regular graphs.
Finally, we introduce the notions of the distance incidence energy and distance Laplacian-
energy like of a graph and derive sharp lower bounds on these two distance energies among
all the connected graphs G of order n, with m ≥ n edges and κ ( G ) ≤ k. The graphs for which
those bounds are attained are characterized.

2 A general result on the H-join of graphs

Consider the vertices of G with the labels 1, . . . , ∑is=1 ni starting with the vertices of G1 , contin-
uing with the vertices of G2 , G3 , . . . , Gs−1 and finally with the vertices of Gs .

5
Example 2 For the graph in Example 1, our labeling is

5
7

1
3
8
4
2

9
6
Figure 2. The graph of Fig. 1 with vertex labels as described.

With the above mentioned labeling, we get

δ12 1n1 1nT2 δ1s 1n1 1nTs

 
M1 ...
... ..
 δ12 1n2 1nT
 
M2 . 
M (G) = 

..
1
.. ..
 (2)
. . . δ(s−1)s 1ns−1 1nTs

 
δ1s 1ns 1nT1 ... δ(s−1)s 1ns 1nTs−1 Ms

where the diagonal blocks Mi are symmetric matrices such that

Mi 1 n i = µ i 1 n i (3)

for i = 1, . . . , s and δi,j are scalars for 1 ≤ i < j ≤ s.

Using a strategy similar to the one used in [6], the following lemmas are proven and the
spectrum of the matrix M( G ) is deduced as it is stated by Theorem 2.

Lemma 1 Consider the block matrices Mi which appear in the expression of M( G ) in (2) and the
eigenvalues µi in (3), both for i = 1, . . . , s. Then

∪is=1 σ( Mi ) − {µi } ⊆ σ( M ( G )).

Proof Let λi 6= µi be an eigenvalue of Mi with multiplicity mi and let (λi , ui ) be an eigenpair of

Mi , where ui is an arbitrary vector in Λλi ( Mi ) \ {0}, for i = 1, . . . , s. Then, for i = 1, we have
       
u1 M1 u1 λ1 u1 u1
 0   0   0   0 
M (G)  = =  = λ1 
       
.. .. .. .. 
 .   .   .   . 
0 0 0 0

6
and thus λ1 ∈ σ( M ( G )) with multiplicity m1 . Similarly,
       
0 0 0 0
 u2   u2   ..   .. 
M (G)   = λ2  , ··· M ( G )  .  = λs  .
       
.. .. 
 .   .   0   0 
0 0 us us

and thus λ2 , . . . , λs ∈ σ ( M ( G )), with multiplicity m2 , . . . , ms , respectively. Furthermore, for

each i ∈ {1, . . . , s}, if µi has multiplicity pi > 1, we can consider pi − 1 linear independent
vectors from Λµi ( Mi ) \ {0} orthogonal to 1i and using each of them in a similar way as above,
we obtain pi − 1 linear independent eigenvectors of M( G ) associated to µi . Therefore, µi ∈
σ ( M ( G )) with multiplicity pi − 1, for i = 1, . . . , s. 2

Lemma 2 Considering the matrix M( G ) in (2) and the s × s symmetric matrix

 √ √ √ 
µ1 δ12 n1 n2 . . . δ1(s−1) n1 ns−1 δ1s n1 ns
√ √ √

 δ12 n1 n2 µ2 . . . δ2(s−1) n2 ns−1 δ2s n2 ns 


Fs =  .. .. .. .. .. 
, (4)
. . . . .
√ √ √
 
δ1(s−1) n2 ns−1 δ2(s−1) n2 ns−1 ... δ(s−1)s ns−1 ns
 
µ s −1
√ √

√

δ1s n1 ns δ2s n2 ns ... δ(s−1)s ns−1 ns µs

it follows that
σ ( Fs ) ⊆ σ( M( G )).
 T  T
Proof Let λ ∈ σ( Fs ). There exists x1 x2 . . . x s 6= 0 0 ... 0 such that
 √ √ 
δ12 n1 n2 ... δ1s n1 ns
  
µ1 x1 x1
√ .. ..
x2 x2
 
 δ12 n1 n2 µ2 . .    
 = λ . (5)
  

.. .. .. √
 .. ..
. . . δ(s−1)s ns−1 ns . .
    
√ √
 
δ1s n1 ns ... δ(s−1)s ns−1 ns µs xs xs

We claim that λ ∈ σ ( M ( G )) with an associated eigenvector

h q q q i
T T n1 T n1 T n1 T
y = x 1 1 n 1 x 2 n 2 1 n 2 . . . x s −1 n s −1 1 n s −1 x s n s 1 n s .

7
Indeed
 
δ12 1n1 1nT2 δ1s 1n1 1nTs x 1 1 n1

M1 ... q
... ..  x2 nn12 1n2

 δ12 1n2 1nT

M2 .  
M (G) y =  1  
.. .. ..
 .
  q ..

. . . δ(s−1)s 1ns−1 1nTs
  
 
δ1s 1ns 1nT1 ... δ(s−1)s 1ns 1nTs−1 Ms xs nn1s 1ns

 √ √ 
( µ 1 x 1 + δ12 n 1 n 2 x 2 + · · · + δ1s n 1 n s x s ) 1 n
 1 
 q √
δ12 n1 x1 + µ2 x2 nn12 + · · · + δ1s n2 ns xs 1n2 


=
 
.. 
 . 
  √ q  
δ1s n1 x1 + δ2s n1 n2 x2 + · · · + nn1s µs xs 1ns
 √ √   
q ( µ1 x1 + δ12 n1 n2 x2 + · · · + δ1s n1 ns xs ) 1n1 q x 1 1 n1
n1 √ √ n1
n2 ( δ12 n1 n2 x1 + µ2 x2 + · · · + δ1s n2 ns xs ) 1n2 n2 x 2 1 n2
   
   
=  = λ .
   
.. ..
 . 
 q .
 
 q
n1 √ √ 
n1

ns ( δ1s n1 ns x1 + δ2s x2 n2 ns + · · · + µs xs ) 1ns ns x s 1 ns

The last equality is a consequence of (5). Hence M ( G ) y = λy and then σ ( Fs ) ⊆ σ( M( G )). 2

From Lemma 1 and Lemma 2, we get the following more strong result on the spectrum of
M ( G ), where G = H { Gi : 1 ≤ i ≤ s}.
W

Theorem 2 Consider the matrix M ( G ) (2) and the matrix Fs (4). Then the spectrum of M ( G ) is

σ( M( G )) = ∪is=1 σ ( Mi ) − {µi } ∪ σ ( Fs ),

3 Determination of the eigenvalues of H-join distance matrices

In this section, we assume that H is a connected graph of order s and di,j denotes the distance
between i, j ∈ V ( H ). Here, we apply Theorem 2 to determine the eigenvalues of the distance
matrix, distance Laplacian matrix and distance signless Laplacian matrix of the H-join G =
H { Gi : 1 ≤ i ≤ s }, when G1 , . . . , Gs are regular graphs. Therefore, throughout this section we
W

deal with the H-join of a family of regular graphs G1 , . . . , Gs .

8
3.1 Distance eigenvalues
Taking into account (2), we get that the distance matrix of the H-join G = H { Gi : 1 ≤ i ≤ s},
W

where Gi is a di -degree regular graph of order ni , for each i = 1, . . . , s, it follows that

d1,2 1n1 1nT2 d1,s 1n1 1nTs

 
M1 ...
 d1,2 1n2 1nT
 ... .. 
M 2 . 
D( G ) =  ..
1
.. ..
, (6)

 . . . T
ds−1,s 1ns−1 1ns 

d1,s 1ns 1nT1 ... ds−1,s 1ns 1nTs−1 Ms

and from (4)

 √ √ 
λ1 ( M1 ) d1,2 n1 n2 ... d1,s n1 ns
 √ .. .. 
 d1,2 n1 n2 λ1 ( M2 ) . . 
Fs = 
 .. ... ... √
.
 (7)
. ds−1,s ns−1 ns
√ √
 
d1,s n1 ns ... ds−1,s ns−1 ns λ 1 ( Ms )

Remark 1 Since H is connected, then Mi = 2( Jni − Ini ) − A( Gi ) (in (6)), where Jni is the all one
square matrix of order ni , Ini is the identity matrix of order ni and A( Gi ) is the adjacency matrix of Gi ,
for each i = 1, . . . , n. Further, as Gi is a di -degree regular graph it follows that Mi 1ni = λ1 ( Mi )1ni ,
where the eigenvalues λ1 ( Mi ) (in (7)) are given by

λ 1 ( Mi ) = 2 ( n i − 1 ) − d i ,

for i = 1, . . . , s. Moreover, since for each i ∈ {1, . . . , s}, the matrices Jni − Ini and A( Gi ) commute,
then the spectrum of Mi is completely determined by the spectrum of A( Gi ). That is,

σ( Mi ) = {2(ni − 1) − di , −2 − λni ( A( Gi )) , . . . , −2 − λ2 ( A( Gi ))}, (8)

for i = 1, . . . , s.

Taking into account the Remark 1, applying Theorem 2, we obtain the following corollary.

Corollary 1 Let H a connected graph of order s. If for each i ∈ {1, . . . , s}, Gi is a di -degree regular
graph, then the spectrum of D( G ), where G = H { Gi : 1 ≤ i ≤ s}, is
W

 
s
σ(D( G )) = ∪i=1 σ( Mi ) − {2(ni − 1) − di } ∪ σ( Fs )

where Fs is the s × s matrix given in (7).

Notice that the matrices Mi in the Theorem 2 are not necessarily the distance matrices D ( Gi ).
This can be observed with the following example.

9
 { G1 , G2 , G3 }.
W
Example 3 Let H = P3 , G1 = C6 , G2 = K2 , G3 = K3 and consider the graph G = H

{C6 , K2 , K3 }.
W
Figure 3. The graph G = P3

In this example, we have σ ( M1 ) = {8, 0, −1[2] , −3[2] }, σ( M2 ) = {1, −1} and σ( M3 ) = {2, −1[2] }.
Notice that M1 6= D(C6 ). Further,
 √ √ 
8 12 2 √18
√
F3 =  12 6 .
√ √1
2 18 6 2

Therefore, to four decimal places σ( F3 ) = {15.2621, −0.2621, −4} and from Corollary 1, we get

σ(D( G )) = {15.2621, 0, −0.2621, −1[5] , −3[2] , −4}.

3.2 Distance Laplacian eigenvalues

For i = 1, . . . , s, let us consider the matrices Li = k i Ini − Mi , where k i = λ1 ( Mi ) + ∑ j6=i di,j n j ,
with Mi , λ1 ( Mi ) as in the Remark 1. Notice that in this case Li 1ni = λni ( Li )1ni , where

λ ni ( L i ) = ∑ di,j n j , (9)
j 6 =i

for each i = 1, . . . , s and therefore, we can write

k i = λ 1 ( Mi ) + λ n i ( L i ) , (10)

for each i = 1, . . . , s. Taking into account (2), we get that the distance Laplacian matrix of the
H-Join G = H { Gi : 1 ≤ i ≤ s} is
W

−d1,2 1n1 1nT2 −d1,s 1n1 1nTs

 
L1 ...
 −d12 1n2 1nT
 ... ..  s
L2 .  M
L( G ) =  ..
1
.. ..
 = k i Ini − D( G ),
. . . −ds−1,s 1ns−1 1nTs
 
  i =1
−d1,s 1ns 1nT1 ... −ds−1,s 1ns 1nTs−1 Ls

10
where D( G ) is as in (6). Notice also that L( G )1n = 0n .
On the other hand, from (4) we obtain
 √ √ 
λ n1 ( L 1 ) −d1,2 n1 n2 ... −d1,s n1 ns
 −d1,2 √n1 n2 .. ..
 
λ n2 ( L 2 ) . . 
Fs = 
.. ... ... √
, (11)
−ds−1,s ns−1 ns

.
√ √
 
−d1,s n1 ns ... −ds−1,s ns−1 ns λ ns ( L s )
where for each i = 1, . . . , s, λni ( Li ) is as in (9). Therefore, applying Theorem 2, we obtain the
following corollary.
Corollary 2 Let H a connected graph of order s. If for each i = 1, . . . , s, Gi is a regular graph, the
spectrum of L( G ) where G = H { Gi : 1 ≤ i ≤ s} is
W

σ(L( G )) = ∪is=1 σ ( Li ) − {λni ( Li )} ∪ σ ( Fs )

where Fs is the s × s matrix in (11) and λni ( Li ) is as in (9), for each i = 1, . . . , s.

Example 4 Consider the same H-join graph of Example 3. Making some computation we have that
k1 = 16, k2 = 10 and k3 = 16. Thus, L1 = 16I6 − M1 , L2 = 10I2 − M2 , L3 = 16I3 − M3 ,
σ ( L1 ) = {19[2] , 17[2] , 16, 8}, σ ( L2 ) = {11, 9}, σ( L3 ) = {17[2] , 14}, and
 √ √ 
8 − 12 − 2 √18
 √
F3 =  − 12 9 − 6 .
√ √
−2 18 − 6 14
Therefore, σ( F3 ) = {20, 11, 0} and, from Corollary 2, we get
σ(L( G )) = {20, 19[2] , 17[4] , 16, 11[2] , 0}.

3.3 Distance signless Laplacian eigenvalues

Finally, we consider the distance signless Laplacian matrix of the H-join of regular graphs. For
i = 1, . . . , s, let us consider the matrices Qi = k i Ini + Mi , where k i is again as in (10) and Mi ,
λ1 ( Mi ) are as in the Remark 1, for each i = 1, . . . , s. Moreover, notice that Qi 1ni = λ1 ( Qi )1ni ,
where
λ 1 ( Q i ) = k i + λ 1 ( Mi ) = 2 ( 2 ( n i − 1 ) − d i ) + λ n i ( L i ) , (12)
for each i = 1, . . . , s. Taking into account (2), we get that the distance signless Laplacian matrix
of G = H { Gi : 1 ≤ i ≤ s} is
W

d1,2 1n1 1nT2 d1,s 1n1 1nTs

 
Q1 ...
 d12 1n2 1nT
 .. ..  s
Q2 . .  M
Q( G ) =  ..
1
.. ..
 = k i Ini + D( G ),
. . . ds−1,s 1ns−1 1nTs
 
  i =1
d1,s 1ns 1nT1 ... ds−1,s 1ns 1nTs−1 Qs

11
where D( G ) is as in (6). Also, from (4) we obtain
 √ √ 
λ1 ( Q1 ) d1,2 n1 n2 ... d1,s n1 ns
 d1,2 √n1 n2 .. ..
 
λ1 ( Q2 ) . . 
Fs = 
.. ... ... √
,
 (13)
. ds−1,s ns−1 ns
√ √
 
d1,s n1 ns ... ds−1,s ns−1 ns λ1 ( Q s )
where for each i ∈ {1, . . . , s}, λ1 ( Qi ) is as in (12). Therefore, applying Theorem 2, we obtain
the following corollary.
Corollary 3 Let H be a connected graph of order s. If for each i ∈ {1, . . . , s}, Gi is a regular graph, the
spectrum of Q( G ), with G = H { Gi : 1 ≤ i ≤ s}, is
W

σ (Q( G )) = ∪is=1 σ ( Qi ) − {λ1 ( Qi )} ∪ σ( Fs )

where Fs is the s × s matrix in (13) and λ1 ( Qi ) is as in (12) for each i = 1, . . . , s.

Example 5 Consider the same H-join as in Example 3. Similarly to Example 4, we get that Q1 =
16I6 + M1 , Q2 = 10I2 + M2 , and Q3 = 16I3 + M3 , σ ( Q1 ) = {24, 16, 15[2] , 13[2] }, σ( Q2 ) = {11, 9}
and σ( Q3 ) = {18, 15[2] } and
 √ √ 
24 12 2 √18
√
F3 =  12 11 6 .
√ √
2 18 6 18

Therefore, to four decimal places σ( F3 ) = {30.9043, 12, 10.0957} and thus, from Corollary 3, we get

σ(Q( G )) = {30.9043, 16, 15[4] , 13[2] , 12, 10.0957, 9}.

4 Sharp lower bounds on the distance incidence energy and

distance Laplacian-energy like
In this section, we define the distance incidence energy and the distance Laplacian-energy like
of connected graphs. Then we derive sharp lower bounds on these two energies among the
graphs G on n vertices in terms of connectivity. These lower bounds are attained if and only if
G = K k ∨ ( K1 ∪ K n − k −1 ) .
Three well known energies on a graph are the (adjacency) energy, introduced by Gutman in
1978 as the sum of the absolute values of the eigenvalues of the adjacency matrix, the Laplacian
energy introduced by Gutman and Zhou in [15] and the signless Laplacian energy which was
defined in analogy with the Laplacian energy. The distance energy [28], the distance Laplacian
energy [32] and the distance signless Laplacian energy [13] have been introduced as follows
n n   n  
2W ( G )  2W ( G ) 
E ( G ) = ∑ |∂ j ( G )|, E ( G ) = ∑ ∂ j ( G ) −
D L  L
and E ( G ) = ∑ ∂ j ( G ) −
Q  Q
,
j =1 j =1
n  j =1
n 

12
respectively.
Other energies on a graph are the incidence energy [17] and the Laplacian-energy like [34].
For a graph G of order n, its incidence energy, denoted by IE( G ), becomes
n q
IE ( G ) = ∑ qi ( G )
i =1

where q1 ( G ) ≥ q2 ( G ) ≥ . . . ≥ qn ( G ), are the signless Laplacian eigenvalues of G; and its

Laplacian-energy like energy, denoted by LEL( G ), is
n
√
LEL ( G ) = ∑ µi
i =1

where µ1 ( G ) ≥ µ2 ( G ) ≥ . . . ≥ µn ( G ) = 0, are the Laplacian eigenvalues of G.

From now on, V (n, k) is the family of connected graphs G of order n such that κ ( G ) ≤ k.

In [21] and in [34], sharp upper bounds on the incidence energy and on the Laplacian-energy
like, respectively, for the graphs in V (n, k ) are obtained. These upper bounds are attained if
and only if G = Kk ∨ (K1 ∪ Kn−k−1 ).
In this section we extend the above concepts of incidence energy and Laplacian-energy
like to the distance matrix context, introducing the notions of distance Laplacian energy like,
denoted by DLEL( G ), and distance incidence energy, denoted by DIE( G ), as follows
n q n q
DLEL( G ) = ∑ ∂iL ( G ) and DIE( G ) = ∑ ∂iQ ( G ).
i =1 i =1

In this study the following theorem plays a crucial role.

Theorem 3 ([3], Theorem 3.5) Let G be a connected graph on n vertices and m ≥ n edges. Consider
the connected graph G
e obtained from G by the deletion of an edge.

• Let ∂1L ( G ), . . . , ∂nL ( G ) and ∂1L ( G

e), . . . , ∂nL ( G
e) be the distance Laplacian eigenvalues of G and G,
e
respectively. Then ∂i ( G L L
e) ≥ ∂ ( G ) for i = 1, . . . , n.
i

• Let ∂1Q ( G ), . . . , ∂nQ ( G ) and ∂1Q ( G

e), . . . , ∂nQ ( G
e) be the distance signless Laplacian eigenvalues of G
e respectively. Then ∂Q ( G
and G, e) ≥ ∂Q ( G ) for i = 1, . . . , n.
i i

Clearly trace(L( G
e)) > trace(L( G )) and trace(Q( G
e)) > trace(Q( G )). Therefore, from Theo-
rem 3, the following corollaries are immediate.

Corollary 4 If G and G
e are connected graphs such that G
e is obtained from G by the deletion of an edge,
then DLEL( G
e) > DLEL( G ) and DIE( G e) > DIE( G ).

13
Corollary 5 Among the all connected graphs on n vertices and m ≥ n edges, the complete graph Kn
has the smallest distance Laplacian-energy like and the smallest distance incidence energy.

For i = 1, 2, 3, let Gi be a di −regular graph of order ni . Then G = G1 ∨ ( G2 ∪ G3 ) is a graph

of order n = n1 + n2 + n3 . Observe that G = G1 ∨ ( G2 ∪ G3 ) is a P3 − join graph in which the
central vertex of P3 is assigned to G1 , one pendent vertex of P3 is assigned to G2 and the other
to G3 .
Labelling the vertices of G = G1 ∨ ( G2 ∪ G3 ) starting with the vertices of G1 , continuing with
the vertices of G2 and finishing with the vertices of G3 , and using the results obtained in the
Subsection 3.3, the distance signless Laplacian matrix Q ( G ) becomes

1n1 1nT2 1n1 1nT3

 
Q1
Q ( G ) =  1n2 1nT1 Q2 21n2 1nT3  (14)
T
1n3 1n1 21n3 1n2T Q3

where, for i = 1, 2, 3,
Qi = k i Ini + 2( Jni − Ini ) − A( Gi ) (15)
and
k 1 = 2( n1 − 1) − d1 + n2 + n3
k2 = 2(n2 − 1) − d2 + n1 + 2n3 (16)
k3 = 2(n3 − 1) − d3 + n1 + 2n2
Clearly, the largest eigenvalues of Q1 , Q2 and Q3 are

λ1 ( Q1 ) = 4(n1 − 1) − 2d1 + n2 + n3
λ1 ( Q2 ) = 4(n2 − 1) − 2d2 + n1 + 2n3 (17)
λ1 ( Q3 ) = 4(n3 − 1) − 2d3 + n1 + 2n2

with eigenvectors 1n1 , 1n2 and 1n3 , respectively.

Applying Corollary 3, we get the following result.

Theorem 4 If G = G1 ∨ ( G2 ∪ G3 ) and, for i = 1, 2, 3, Gi is a di −regular graph then

σ(Q( G )) = (σ ( Q1 ) ∪ σ( Q2 ) ∪ σ( Q3 ) − {λ1 ( Q1 ), λ1 ( Q2 ), λ1 ( Q3 )}) ∪ σ( F3 )

where Q1 , Q2 , Q3 are as in (15), λ1 ( Q1 ), λ1 ( Q2 ), λ1 ( Q3 ) are as in (17) and

 √ √ 
λ1 ( Q1 ) n1 n2 n1 n3
√ √
F3 ( G ) =  n1 n2 λ1 ( Q2 ) 2 n2 n3  . (18)
√ √
n1 n3 2 n2 n3 λ1 ( Q3 )

Let n and k be positive integers, with k ≤ n − 1 and consider the graph

G ( i ) = K k ∨ ( Ki ∪ K n − k −i ), (19)

14
 j k
n−k
where, without loss of generality, we assume 1 ≤ i ≤ 2 . Then, for the graph G (i ), the
matrices Q1 , Q2 and Q3 in (15) are
Q1 = (n − 1) Ik + A(Kk )
Q2 (i ) = (2n − k − i − 1) Ii + A(Ki ) (20)
Q3 (i ) = (n + i − 1) In−k−i + A(Kn−k−i ),
respectively, and the matrix F3 ( G (i )) in (18) becomes
 √ p 
n +√k − 2 ki pk ( n − k − i )
F3 ( G (i )) =  ki 2n − k − 2 2 i (n − k − i )  .
 
p p
k (n − k − i ) 2 i (n − k − i ) 2n − k − 2
Taking into account that the adjacency eigenvalues of Ks are s − 1 and −1 with multiplicity
s − 1, the spectra of Q1 , Q2 (i ) and Q3 (i ) in (20) are

n o
σ ( Q1 ) = n + k − 2, (n − 2)[k−1] ,
n o
[ i −1]
σ ( Q2 (i )) = 2n − k − 2, (2n − k − i − 2) and
n o
σ ( Q3 (i )) = 2n − k − 2, (n + i − 2)[n−k−i−1] ,

where λ[t] denotes that λ is an eigenvalue with multiplicity t. One can obtain that the spectrum
of F3 ( G (i )) which is
σ ( F3 ( G (i ))) = { f 1 (i ) , f 2 , f 3 (i )} ,
q
where f 1,3 (i ) = 2(n − 1) − 2k ± ( 2k )2 + 4i (n − k − i ) and f 2 = n − 2.
Now, applying Theorem 4 to Q( G (i )) the next corollary follows.

Corollary 6 Let G (i ) be the graph defined in (19). Then

n o
σ (Q ( G (i ))) = f 1 (i ) , f 2 , f 3 (i ) , (n − 2)[k−1] , (2n − k − i − 2)[i−1] , (n + i − 2)[n−k−i−1] .

Let |S| be the cardinality of a finite set S and let us define W (n, k) as follows:

W (n, k) = { G ∈ V (n, k) : | E( G )| ≥ n}.

Furthermore, consider
s
√ √
r
k k
b(n, k ) = k n − 2 + (n − k − 2) n − 1 + 2( n − 1) − + ( )2 + 4( n − k − 1)
2 2
s r
k k
+ 2( n − 1) − − ( )2 + 4( n − k − 1).
2 2
Then, denoting by K0 the empty graph (that is, the graph without edges and without ver-
tices) we have the following theorem.

15
Theorem 5 If G ∈ W (n, k), then

DIE ( G ) ≥ b (n, k) , for k = 1, . . . , n − 1. (21)

Additionally, the inequalities (21) hold as equalities if and only if G = Kk ∨ (K1 ∪ Kn−k−1 ) .

Proof Let G ∈ W (n, k). We first consider k = n − 1. From Corollary 5, DIE ( G ) ≥ DIE (Kn )
with equality if and only if G = Kn . Moreover
q √
b (n, n − 1) = 2(n − 1) + (n − 1) n − 2 = DIE (Kn ) .

Then the result is true for k = n − 1. Now let 1 ≤ k ≤ n − 2 and let G ∈ W (n, k ) such
that DIE ( G ) is a minimum. Let S ⊆ V ( G ) such that G − S is a disconnected graph. Let
C1 , C2 , . . . , Cr be the connected components of G − S. We claim that r = 2. If r > 2 then we can
construct a new graph H = G ∪ {e} where e is an edge connecting a vertex in C1 with a vertex
in C2 . Clearly, H ∈ W (n, k ) and G = H − e. By Corollary 4, DIE ( G ) > DIE ( H ) , which is a
contradiction. Therefore r = 2, that is, G − S = C1 ∪ C2 . By hypothesis |S| ≤ k. Now, we claim
that |S| = k. Suppose |S| < k. Since G − S = C1 ∪ C2 , we may construct a graph H = G + e
where e is an edge joining a vertex u ∈ V (C1 ) with a vertex v ∈ V (C2 ) . We see that H − S is a
connected graph and the deletion of the vertex u disconnected it. This tell us that H ∈ W (n, k ).
By Corollary 4, DIE ( G ) > DIE ( H ) , which is also a contradiction. Hence G − S = C1 ∪ C2
and |S| = k. Let |C1 | = i. Then |C2 | = n − k − i. Repeated application of Corollary 4 enables to
conclude that
G = K k ∨ ( Ki ∪ K n − k −i ) = G ( i )
j k
for some 1 ≤ i ≤ n− 2
k
. We have proved DIE ( G ) ≥ DIE ( G (i )) for all G ∈ Wnk . We now
search for the value of i for which DIE ( G (i )) is minimum. From Corollary 6,
√ √ √
DIE ( G (i )) = k n − 2 + (i − 1) 2n − k − i − 2 + (n − k − i − 1) n + i − 2
s r s r
k k 2 k k
+ 2n − 2 − + ( ) + 4i (n − k − i ) + 2n − 2 − − ( )2 + 4i (n − k − i ).
2 2 2 2
Defining the function
√ √
f ( x ) = ( x − 1) 2n − k − x − 2 + (n − k − x − 1) n + x − 2
s r s r
k k 2 k k
+ 2n − 2 − + ( ) + 4x (n − k − x ) + 2n − 2 − − ( )2 + 4x (n − k − x )
2 2 2 2

and observing that f ( x ) = f (n − k − x ), for x ∈ [0, n − k ], after some algebraic manipula-

tion, we may conclude that f is a strictly increasing function in the interval [0, n− k
2 ]. Hence
DIE ( G ) ≥ DIE ( G (1)) for all G ∈ W (n, k ). Moreover, since G (1) = Kk ∨ (K1 ∪ Kn−k−1 )
and DIE ( G (1)) = b (n, k ), the inequality (21) holds as equality if and only if G = Kk ∨
( K1 ∪ K n − k −1 ) . 2

16
 To find a sharp lower bound on the distance Laplacian-energy like among the graphs in
W (n, k) is easier. Using the above mentioned labelling for the vertices of G = G1 ∨ ( G2 ∪ G3 )
and the results obtained in the subsection 3.2, we obtain
−1n1 1nT2 −1n1 1nT3
 
L1
L ( G ) =  −1n2 1nT1 L2 −21n2 1nT3  (22)
T
−1n3 1n1 −21n3 1n2T L3

where, for i = 1, 2, 3,
Li = k i Ini − 2( Jni − Ini ) + A( Gi ) (23)
and k i are as in (16).
The smallest eigenvalues of L1 , L2 and L3 are

λn1 ( L1 ) = n2 + n3 , λn2 ( L2 ) = n1 + 2n3 and λn3 ( L3 ) = n1 + 2n2 , (24)

with eigenvectors 1n1 , 1n2 and 1n3 , respectively.

Applying Corollary 2, the following theorem is obtained.

Theorem 6 If G = G1 ∨ ( G2 ∪ G3 ) and, for i = 1, 2, 3, Gi is a di −regular graph then

σ (L( G )) = (σ( L1 ) ∪ σ( L2 ) ∪ σ ( L3 ) − {λn1 ( L1 ), λn2 ( L2 ), λn3 ( L3 )}) ∪ σ( F3 )

where L1 , L2 , L3 are as in (23), λn1 ( L1 ), λn2 ( L2 ), λn3 ( L3 ) are as in (24) and

√ √
λ n1 ( L 1 ) − n 1 n 2 − n 1 n 3
 
√ √
F3 ( G ) =  − n1 n2 λn2 ( L2 ) −2 n2 n3  . (25)
√ √
− n 1 n 3 −2 n 2 n 3 λ n3 ( L 3 )
For the graph G (i ), the matrices L1 , L2 and L3 in (23) are

L1 = (n − 1) Ik − A(Kk ),
L2 (i ) = (2n − k − i − 1) Ii − A(Ki ), (26)
L3 (i ) = (n + i − 1) In−k−i − A(Kn−k−i ),

respectively, and the matrix F3 ( G (i )) in (25) becomes

 √ p 
n√ −k − ki − pk (n − k − i )
F3 ( G (i )) =  − ki 2n − k − 2i − 2 i ( n − k − i ) .
 
p p
− k ( n − k − i ) −2 i ( n − k − i ) k + 2i

The spectra of L1 , L2 (i ) and L3 (i ) in (26) are

n o
[ k −1]
σ ( L1 ) = n ,n−k ,
n o
σ ( L2 (i )) = (2n − k − i )[i−1] , 2n − k − 2i and
n o
σ ( L3 (i )) = (n + i )[n−k−i−1] , k + 2i .

17
In this case, the spectrum of F3 ( G (i )) is

σ ( F3 ( G (i ))) = {2n − k, n, 0}.

Applying Theorem 6 to L( G (i )), we obtain the next corollary.

Corollary 7 Let G (i ) be the graph defined in (19). Then

n o
σ (L ( G (i ))) = 2n − k, n[k] , (2n − k − i )[i−1] , (n + i )[n−k−i−1] , 0 .

By similar arguments to the ones used in the proof of Theorem 5, we get the following
result.

Theorem 7 If G ∈ W (n, k), then

DLEL ( G ) ≥ c(n, k ) (27)
√ √ √
where c(n, k ) = 2n − k + k n + (n − k − 2) n + 1, for k = 1, . . . , n − 1. The inequality (27) holds
as equality if and only if G = Kk ∨ (K1 ∪ Kn−k−1 ) .

Acknowledgements. The research of D.M. Cardoso was partially supported by the Por-
tuguese Foundation for Science and Technology (“FCT-Fundação para a Ciência e a Tecnolo-
gia”), through the CIDMA - Center for Research and Development in Mathematics and Ap-
plications, within project UID/MAT/04106/2013. The research of R. C. Díaz was supported
by Conicyt-Fondecyt de Postdoctorado 2017 N o 3170065, Chile. The research of O. Rojo was
supported by Project Fondecyt Regular 1170313, Chile.

References
[1] I. Altöfer, On optimal realizations of finite metric spaces by graphs, Discrete Comput.
Geom. 3 (1988) 103–122.

[2] M. Aouchiche, P. Hansen, Distance spectra of graphs: A survey, Linear Algebra Appl. 458
(2014) 301–386.

[3] M. Aouchiche, P. Hansen, Two Laplacians for the distance matrix of a graph, Linear Alge-
bra Appl., 439(2013) 21-33.

[4] F.T. Boesch, Properties of the distance matrix of a tree, Quart. Appl. Math. 26 (1968-1969)
607–609.

[5] P. Buneman, A note on matric properties of trees, J. Comb. Theory Ser. B 17 (1974) 48–50.

[6] D. M. Cardoso, M. A. A. de Freitas, E. Martins., M. Robbiano, Spectra of graphs obtained

by a generalization of the join graph operation, Discrete Mathematics 313 (2013) 733-741.

18
 [7] D. M. Cardoso, E. Martins., M. Robbiano, O. Rojo, Eigenvalues of a H-generalized opera-
tion constrained by vertex subsets, Linear Algebra Appl. 438 (2013) 3278 - 3290.

[8] S.L. Hakimi, S.S. Yau, Distance matrix of a graph and its realizability, Quart. Appl. Math.
22 (1964) 305–3017.

[9] F. Harary, On the group of compositions of two graphs. Duke Math. J. 26 (1959) 29–34.

[10] G. Sabiduss, The composition of graphs. Duke Math. J. 26 (1959) 693–696.

[11] A. J. Schwenk, Computing the characteristic polynomial of a graph, Graphs and Combi-
natorics (Lecture notes in Mathematics 406, eds. R. Bary and F. Harary), Springer-Verlag,
Berlin-Heidelberg-New York, 1974, 153–172.

[12] D. M. Cardoso, D. Cvetković, P. Rowlinson, S. K. Simić

[13] K. Ch. Das, M. Aouchiche, P. Hansen, On (distance) Laplacian energy and (distance) sign-
less laplacian energy of graphs, Les Cahiers du GERAD, G-2017-96, Nov. 2017.

[14] R.L. Graham, H.O. Pollak, On the addressing problem for loop switching, Bell Syst. Tech.
J. 50 (1971) 2495–2519.

[15] I. Gutman, B. Zhou, Laplacian energy of a graph, Linear Algebra Appl. 414 (2006) 29-37.

[16] I. Gutman, D. Kiani, M. Mirzkhah, B. Zhou, On incidence energy of a graph, Linear Alge-
bra Appl. 431 (2009) 1223-123.

[17] M.R. Jooyandeh, D. Kiani, M. Mirzakhah, Incidence energy of a graph, MATCH Commun.
Math. Comput. Chem. 62 (2009) 561–572.

[18] H. Lin, B. Wu, Y. Chen, J. Shu, On the distance and istance Laplacian eigenvalues of graphs,
Linear Algera Appl. 492 (2016) 128–135.

[19] H. Lin, K. Ch. Das, Characterization of extremal graphs from distance signless Laplacian
eigenvalues, Linear Algebra Appl. 500 (2016) 77–87.

[20] J. Liu, B. Liu, A Laplacian-energy like invariant of a graph, MATCH Commun. Math.
Comput. Chem. 59 (2008) 397–419.

[21] O. Rojo, E. Lenes, A sharp upper bound on the incidence energy of graphs in terms of
connectivity, Linear Algebra Appl. 438 (2013) 1485 âĂŞ 1493.

[22] I. Schonberg, Remarks to Maurice Fréchet’s article "Sur la définition axiomatique d’une
classe d’espacedstanciés vectoriellement applicable sur léspace de Hilbert," Ann. of Math.
36 (1935) 724–732.

19
[23] J.M.S. Simões-Pereira, An algorithm and its role in the sudy of optimal graph realizations
of distance matrices, Discrete Math. 79 (1990) 299-312.

[24] J.M.S. Simões-Pereira, A note on distance matrices with unicyclic graph realizations, Dis-
crete Math. 65 (1987) 277–287.

[25] M.S. Simões-Pereira, A note on a tree realizability of a distance matrix, J. Combin. Theory
B 6 (1969) 303–310.

[26] D. Stevanović, Large sets of long distance equienergetic graphs, Ars Math. Contemp. 2
(2009) 35–40.

[27] D. Stevanović, A. Ilić, Spectral properties of distance matrix of graphs, in I. Gutman, B. Fur-
tula (Eds), Distance in Molecular Graph Theory, Math. Chem. Monogr., vol. 12, University
of Kragujevac, Kragujevac, 2010, pp. 139–176.

[28] H.S. Ramane, Distance Energy of Graphs, in I. Gutman, X. Li (Eds.), Energies of Graphs -
Theory and Applications, Mathematical Chemistry Monographs 17, University of Kragu-
jevac, Faculty of Sciences, Kragujevac, 2016, pp. 123–144.

[29] M.S. Simões-Pereira, Some results on the tree realization of a distance matrix, in Théory
des Graphs, J. Int. Étude, Dunod, Rome, 1966, pp. 383–388.

[30] S.C. Varone, Trees related realizations of distance matrices, Discrete Math. 192 (1998) 337–
346.

[31] P. Winkler, The complexity of metric realization, SIAM J. Discrete Math. 1 (1988) 552–559.

[32] J. Yang, L. You, I. Gutman, Bound on the distance Laplacian energy of graphs, Kragujevac
Jour. Math. 37(2) (2013) 245–255.

[33] G. Young, A. Householder, Discussion of a set of points in terms of their mutul distances,
Psychomatika 3 (1938) 19–22.

[34] B-X Zhu, The Laplacian-energy like of graphs, Applied Mathematical Letters 24 (2011)
1604-1607.

20