Journal of Interdisciplinary Mathematics

ISSN: 0972-0502 (Print) 2169-012X (Online) Journal homepage: https://tandfonline.com/loi/tjim20

Bounds for eigenvalues of the adjacency matrix of

a graph

Pintu Bhunia, Santanu Bag & Kallol Paul

To cite this article: Pintu Bhunia, Santanu Bag & Kallol Paul (2019): Bounds for eigenvalues
of the adjacency matrix of a graph, Journal of Interdisciplinary Mathematics, DOI:
10.1080/09720502.2019.1630938

To link to this article: https://doi.org/10.1080/09720502.2019.1630938

Published online: 03 Sep 2019.

Submit your article to this journal

Article views: 4

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://tandfonline.com/action/journalInformation?journalCode=tjim20
Journal of Interdisciplinary Mathematics
ISSN: 0972-0502 (Print), ISSN: 2169-012X (Online)
DOI : 10.1080/09720502.2019.1630938

Bounds for eigenvalues of the adjacency matrix of a graph

Pintu Bhunia§
Department of Mathematics
Jadavpur University
Kolkata 700032
West Bengal
India

Santanu Bag †
Department of Mathematics
Vivekananda College For Women
Barisha
Kolkata 700008
West Bengal
India

Kallol Paul *
Department of Mathematics
Jadavpur University
Kolkata 700032
West Bengal
India

Abstract
We obtain bounds for the largest and least eigenvalues of the adjacency matrix of a
simple undirected graph. We find upper bound for the second largest eigenvalue of the

The first author would like to thank UGC, Govt. of India for the financial support in the form
of Junior Research Fellowship. The third author would like to thank RUSA 2.0, Jadavpur
University for the partial support.

§
E-mail: [email protected]
†
E-mail: [email protected]
*E-mail: [email protected] (Corresponding Author)

©
2 P. BHUNIA, S. BAG AND K. PAUL

adjacency matrix. We prove that the bounds obtained here improve on the existing bounds
and also illustrate them with examples.

Subject Classification: (2010) Primary 05C50, Secondary 15A42.

Keywords: Adjacency matrix, Eigenvalues, Spectral radius.

1. Introduction
Suppose G = (V, E) be a simple undirected graph with n vertices
and e edges. Let V = {v1, v2, , vn} and di = deg(vi) denotes the degree of
the vertex vi for i = 1, 2,, n. The adjacency matrix associated with the
graph G is defined by A = (aij)n×n, where aij = 1, if vi ~ vj i.e., the vertex vi
is adjacent to vertex vj and aij = 0 otherwise. It is evident that A(G) is a
symmetric matrix with entries 0, 1 and entires in the main diagonal are
zero. For an n×n matrix A, the specrtum, s (A) is the set of all eigenvalues
of A and spectral radius, rs(A), is the largest absolute value of the
eigenvalues i.e., largest of |λ1 |,|λ2 |, … ,|λn |. Being a real symmetric
matrix A(G) has n real eigenvalues, say, λ1 (G), λ2 (G), … , λn (G), with
λi (G) ≥ λi +1 (G), i = 1, 2, … , n − 1. It is well known [1] that l1(G) > 0 is the
spectral radius of the adjacency matrix A(G) i.e., l1(G) = rs(A(G)). The
numerical range of an n × n real matrix A, written as, W(A), is defined as
 t n

W ( A) = x Ax : x= (x1 , x2 , … , xn ) ∈  , ∑|xi |= 1 .
t n 2

 i =1 
It is easy to verify that the numerical range contains the spectrum and
for a symmetric matrix A(G), λn (G) ≤ x t A(G)x ≤ λ1 (G) for all x ŒRn with
∑ in=1|xi | = 1. Also λ1 (G)= sup{x A(G)x : x= (x1 , x2 , … , xn ) ∈  , ∑ i =1|xi |= 1}
2 t t n n 2

and λn (G)= inf{x t A(G)x : x= (x1 , x2 , … , xn )t ∈  n , ∑ in=1|xi |2= 1}. Many

mathematicians over the years have obtained upper bounds for the largest
and the smallest eigenvalue of the adjacency matrix A(G) associated with
the graph G. We here note some of the bounds obtained by mathematicians
over the years.

(1) Nikiforov [14] proved that

δ −1 1
λ1 (G) ≤ + 2e − nδ + (1 + δ )2
2 4

and
EIGENVALUES OF THE ADJACENCY MATRIX 3

cl(G) − 1
λ1 (G) ≤ 2e ,
cl(G)

where δ = min 1≤ i ≤ n{di } and cl(G) is the maximum size of the complete
subgraph of the graph G.
(2) Wilf [17] proved that
n −1
λ1 (G) ≤ 2e .
n

(3) Berman and Zhang [3] proved that

λ1 (G) ≤ max { }
di d j : 1 ≤ i , j ≤ n, vi ∼ v j .

(4) Favaron et al.[7] proved that if mi be the average degree of the vertex
vi, then
λ1 (G) ≤ max{di mi : vi ∈ V }.

(5) Stanley [16] proved that

λ1 (G) ≤ (−1 + 1 + 8e )/ 2.

(6) Hong [10] proved that if G is a connected graph, then

λ1 (G) ≤ 2e − n + 1.

(7) Brighman and Dutton [2] proved that

λn (G) ≥ − 2e(n − 1)/ 2).

(8) Constantine [5], Powers [15], Hong [9] proved that

λn (G) ≥ − (n / 2)[(n + 1)/ 2].

where [x] denotes the largest integer not greater than x.

(9) Das and Kumar [6] proved that
λ1 (G) ≤ 2e − (n − 1)dn + (dn − 1)d1

and
λn (G) ≥ − 2e − (n − 1)dn + (dn − 1)d1 ,
4 P. BHUNIA, S. BAG AND K. PAUL

where d1 and dn are maximum and minimum degree of the vertices

of the graph respectively.
(10) Collatz and Sinogowitz [4] proved that if G is a connected graph of
order n, then
π
λ1 (G) ≥ 2 cos .
n+1

(11) Hong [8] proved that if G is a connected unicyclic graph

λ1 (G) ≥ 2.

(12) Favaron et al. [7] proved that

λ1 (G) ≥ d1 .

(13) Das and Kumar [6] proved that if G be a simple graph with at least
one edge and d1 be the highest degree of G, then

(d1 + d j − 1) + (d1 + d j − 1)2 − 4(d1 − 1)(d j − 1) + 4c12j + 8c1 j d1

λ1 (G) ≥ ,
2

where
= d j max{dk : v1 ∼ vk } and c1j is the cardinality of the common
neighbor between v1 and vj.
(14) Kumar [12] proved that if G be a simple graph of order n ≥ 2, then
1
λ1 (G) ≥ max di + d j + (di − d j )2 + 4cij2 .
j <i
2
In this paper we improve on the upper and lower bounds of the
largest eigenvalue. We also improve on the bounds of least eigenvalue.
We use the relation λn (G) ≤ x A(G)x ≤ λ1 (G) to improve on the bounds.
t

Finally, we consider examples of two graphs and their Huckel matrices

[11] associated with the molecular structure of the carbon skeleton of
1,2-divinylcyclobutadiene and 1,4 - divinylbenzene respectively to show
that the bounds obtained by us improve on the previous bounds.

2. On upper bounds of largest eigenvalues of a graph

We first obtain an upper bound of largest eigenvalue l1(G). For this
we need the following lemma:
EIGENVALUES OF THE ADJACENCY MATRIX 5

Lemma 2.1 : Let l be any eigenvalue of the adjacency matrix A(G) of a simple
graph G with n vertices and e edges. If x = (x1, x2, , xn)t is an eigenvector
corresponding to the eigenvalue l then
=λ xi ∑ x j for i 1, 2,..., n.
=
j , vi ∼ v j

Proof : Since l is an eigenvalue of A(G) and x = (x1, x2, , xn)t is an associated

eigenvector, so we have
A(G)x = λ x
⇒ x t A(G)x =
λ xt x
n
⇒ 2 ∑
vi ∼ v j , i < j
λ ∑xi2 .
xi x j =
1
i=

Differentiating partially with respect to xi on both sides, we have for

each i = 1, 2,, n,
2 ∑x
j , vi ∼ v j
j
= 2λ x i

⇒ λ xi =∑ xj . j , vi ∼ v j

This completes the proof of the lemma. 

Theorem 2.2 : Suppose G be a simple graph with n vertices and e edges. Let l1(G)
be the largest eigenvalue of the adjacency matrix A(G) and di denotes the degree
of the vertex vi. Then
λ1 (G) ≤ max
1≤ i ≤ n ∑d.
j , vi ∼ v j
j

Proof : Let x = (x1, x2, , xn)t ŒRn be an unit eigenvector corresponding to

the eigenvalue l. Then from Lemma 2.1 it follows that
n
λ xi
= ∑ xj
=
j , vi ∼ v j
∑a x .
1
j=
ij j

So,
2
 n n 
λ2 = ∑  ∑aij x j 
=i 1=j 1 
6 P. BHUNIA, S. BAG AND K. PAUL

n n
λ2
⇒ = ∑(a
=i 1
2
1i
+ a22i +…+ ani2 )xi2 + 2 ∑
i ,=j 1, i < j
xi x j (a1i a1 j + a2 i a2 j +…+ ani anj )
n n
⇒ λ2 ≤ ∑d x
=i 1
i
2
i
+ ∑
i ,=j 1, i < j
(xi2 + x 2j )(a1i a1 j + a2 i a2 j +…+ ani anj ).

Now the coefficient of xi2 in the 2nd summation of the above

expression is

a11a1i + a21a2i + ... + an1ani

+ a12a1i + a22a2i + ... + an1ani
+ ... + ... + ... + ...
+ a1, i–1a1i + a2, i–1a2i + ... + an, i–1ani
+ a1, i–1a1i + a2, i–1a2i + ... + an, i–1ani
+ ... + ... + ... + ...
+ a1na1i + a2na2i + ... + annani

which is equal to a1i (d1 − a1i ) + a2 i (d2 − a2 i ) +…+ ani (dn − ani ). From
the definition of adjacency matrix it follows that aij = 1 if vi ~ vj, aij = 0
otherwise, degree of the vertex vi = di = ∑ nj=1 aij and so coefficient of xi2 is
∑ j , v ∼v (d j − 1). Hence
i j
n n
λ2 ≤
=i 1
∑d x + ∑ ∑
i
2
i
=i 1 j , v ∼ v
(d j − 1)xi2
i j

n  
⇒ λ2 ≤
=i 1
∑  d
i
j , v ∼v
(
+ d j ∑
− 1)  x2
 i
 i j 
n  
⇒ λ 2 ≤ ∑  di + ∑ d j − di  xi2
=i 1  j , v ∼v 
 i j 
n  
⇒ λ 2 ≤ ∑  ∑ d j  xi2
=i 1  j , v ∼ v 
 i j 
n
⇒ λ 2 ≤ max
1≤ i ≤ n ∑ d ∑x
j , v ∼v
j
1
i=
2
i
i j

⇒ λ ≤ max
1≤ i ≤ n
∑d.
j , v ∼v
j
i j

As l is an arbitrary eigenvalue so we get,

EIGENVALUES OF THE ADJACENCY MATRIX 7

λ1 (G) ≤ max
1≤ i ≤ n ∑d.
j , v ∼v
j 
i j

Corollary 2.3 : If G is a regular graph with n vertices and degree of each vertex
is m then l1(G) = m. In particular, if G = Kn is a complete graph with n vertices
then l1(Kn) = n – 1.

Proof : Here di = m for each i = 1, 2,, n and so from Theorem 2.2, we get
1
l1(G) £ m. Choose x1= x2= ...= xn= so that xt A(G) x = m. Therefore
n
l1(G) = m. For a complete graph Kn, di = n – 1 and so l1(Kn) = n – 1. 

Corollary 2.4 : If G = Km,n is a complete bipartite graph with the set of vertices V
and U consisting of m and n vertices respectively then λ1 (K m , n ) = mn .

Proof: Let V = {v1, v2, , vn} and U = {vm+1, vm+2, , vm+n} Then d1 = d2 =
... = dm = n and dm+1 = dm+2 = ... = dm+m = m. Then from Theorem 2.2, we
cos θ
get λ1 (G) ≤ mn . Choose xi = for each i = 1, 2,, m and xm + i = sin θ
m n
x t A(G)x
for each i = 1, 2,, n. Then= mn sin 2θ ≤ mn and for θ = π4 ,
x t A(G)x = mn . Hence λ1 (K m , n ) = mn . 

We next obtain an upper bound for a simple circuitless connected

graph G with n vertices and e edges.

Theorem 2.5 : Let T be a simple circuitless connected graph with n(≥ 2) vertices
and e edges. Then
max
1≤ i ≤ n
∑d
j , v ∼v
j
≤ e.
i j

Proof : We prove the theorem by the method of induction on the numer

of vertices n. Let T be a simple circutless connected graph with 2 vertices.
Then T has exactly 1 edge and so
max
1≤ i ≤ 2
∑d
j , v ∼v
j
≤e=1.
i j

Next assume the result to be true for any simple circuitless connected
graph with k vertices and e edges. Consider a simple circuitless connected
graph T¢ with k + 1 vertices, e¢ = e + 1 edges. Since T¢ is simple circuitless
8 P. BHUNIA, S. BAG AND K. PAUL

connected, there exists at least one vertex v1 (say) of degree 1. Let the
vertex v1 be adjacent to v2 by the edge e1. Let T″ be the graph obtained
by deleting the vertex v1 and the edge e1 from T¢. Therefore T″ is a simple
circuitless connected graph with k vertices and e¢ – 1 edges. Hence by
induction hypothesis we must have
max
2≤ i ≤ k +1 ∑d
j , v ∼v
j
≤ e ′ − 1.
i j

Let v be any vertex in T¢. If v is not adjacent to v2 in T¢ then

∑d
j , v ∼v
j
≤ e ′ − 1 < e ′.
j

If v is adjacent to v2 in T¢ then
∑d
j , v ∼v
j
≤ e +1 =e ′.
j

Therefore,
max
1≤ i ≤ k + 1
∑d
j , v ∼v
j
≤ e ′.
i j

This completes the proof.

Remark 2.6 : For a simple circuitless connected graph G with at least one
edge, cl(G)= 2, n − 1= e , dn= 1 and hence
cl(G) − 1
λ1 (G) ≤ max
1≤ i ≤ n
∑d
j , v ∼v
j
≤ 2
cl(G)
e= 2e − (n − 1)dn + (dn − 1)d1 .
i j

This shows that the upper bounds mentioned in [14, Th. 1] and [6,
Th. 2.7] are weaker than the bound obtained by us for simple circuitless
connected graph with at least one edge.

Remark 2.7 : Suppose G be a simple circuitless graph. Then the edges of

G which are used to calculate ∑ j , v ∼v d j ,(for i = 1, 2,..., n) are considered
i j
atmost one time. For, if e1 be an edge of G which is used two times to
calculate ∑ j , v ∼v d j then two end vertices v and u of e1 must be adjacent to
k j
vk which forms a circuit of triangle form. For any triangle free connected
graph G,
cl(G) − 1
max ∑ d j ≤ 2 e.
1≤ i ≤ nj , v ∼v
i j cl(G)
 EIGENVALUES OF THE ADJACENCY MATRIX 9

This shows that the upper bound mentioned in [14, Th. 1] is weaker
than the bound obtained by us for a simple triangle free connected graph.

3. On lower bounds of largest eigenvalues of a graph

Das and Kumar [6] proved that if G is a simple graph with at least one
edge and d1 be the highest degree of G, then

(d1 + d j − 1) + (d1 + d j − 1)2 − 4(d1 − 1)(d j − 1) + 4c12j + 8c1 j d1

λ1 (G) ≥ ,
2
where d j max{dk : v1vk ∈ E} and c1j is the cardinality of common
=
neighbour between v1 and vj. Kumar [12] proved that for a simple graph
G of vertices n ≥ 2
1
λ1 (G) > max di + d j + (di − d j )2 + 4cij2 ,
j <i
2
where cij is the cardinality of the common neighbour between vi and vj.
Here we obtain some lower bounds for the largest eigenvalue l1(G) for a
simple graph G and prove that bound is better than that obtained by Das
and Kumar [6] and Kumar [12] for certain classes of graphs.

Theorem 3.1 : Suppose G(n, e) be a simple graph with n vertices v1, v2, , vn
and e(≥ 1) edges. Then
2e
(i ) λ1 (G) ≥ ,
n
2
(ii ) λ1 (G) ≥ n ∑ di dj ,
∑j d 2 v ∼v , i < j

j =1
i j

where di = degree of vi, for each i = 1, 2,, n.

Proof : (i) We know that

 n

λ1 (G)= sup x t A(G)x : x= (x1 , x2 , … , xn )t ∈  n , ∑|xi |2= 1 .
 i =1 
1
= Let x (x1 , x2 ,..., xn )t ∈  n where x1= x2= ...= xn= . Then
n
∑ i =1|xi | = 1 and
n 2

1 1 2e
= ∑ xi x j 2
x t A(G)x 2= ∑= n
.
v ∼v , i < j
i j
v ∼v , i < j
i j n n
10 P. BHUNIA, S. BAG AND K. PAUL

2e
Therefore, λ1 (G) ≥ n
.

(ii) Choose x = (x1, x2, , xn)t ŒRn with

di n
xi = 1, 2, … , n, so that ∑|xi |2 =
,i = 1.
n
i =1
∑d j =1
2
j

Then
x t A(G)x = 2 ∑
v ∼v , i < j
xi x j
i j

2
⇒ x t A(G)x =
n ∑ di dj
∑dj j
2 v ∼v , i < j
i
j =1

2
⇒ λ1 (G) ≥ n ∑ di d j .
∑d
j =1
2 v ∼v , i < j
j
i j

This completes the proof of the theorem. 

Theorem 3.2 : Suppose G(n, e) be a simple graph with n vertices and e(≥ 1)
edges. Then
1
λ1 (G) ≥ ∑ dd ,
e vi ∼ v j , i < j i j

where di = degree of vi, for each i = 1, 2,, n. Moreover the equality holds
if G be a regular or complete bipartite graph.

Proof : We know that

 n

λ1 (G)= sup x t A(G)x : x= (x1 , x2 , … , xn )t ∈  n , ∑|xi |2= 1 .
 i =1 
di
Let x = (x1, x2, , xn)t ŒRn where x1= x2= ...= xn= 2e
.

Then ∑ in=1|xi | = 1 and

2

x t A(G)x = 2 ∑ xi x j
v v , i < j
i j
EIGENVALUES OF THE ADJACENCY MATRIX 11

di d j
⇒ x t A(G)x = 2 ∑
v v , i < j 4e 2
i j

1
⇒ x t A(G)x = ∑ dd
e vi  v j , i < j i j
1
⇒ λ1 (G) ≥ ∑ dd .
e vi  v j , i < j i j

If G be a regular graph and m be the degree of each vertex, then

1 me
∑ d= d
e vi ∼ v j , i < j i j
= m
e

and so l1(G) = m. Finally, if G be a complete bipartite graph with two set

of vertices containing m and n number of vertices then
1 e mn
e
∑ di d j
=
v ∼v , i < j
=
e
mn
i j

and so λ1 (G) = mn . 

Remark 3.3 : The bound obtained by Das and Kumar [6, Th. 3.2] is weaker
than our bound for a complete graph with n > 2 vertices. For a complete
graph with n vertices we have, di = n – 1, for each i = 1, 2,, n. Then the
bound obtained by us in Theorem 3.2 is
1 1 e(n − 1)
e
∑
v ∼v , i < j
di d j =
e
∑
v ∼v , i < j
(n − 1) =
e
= n − 1.
i j i j

Again d1 ==d j n − 1, c1 j =
n − 2 and so the bound obtained by Das
and Kumar [6, Th. 3.2] is

(d1 + d j − 1) + (d1 + d j − 1)2 − 4(d1 − 1)(d j − 1) + 4c12j + 8c1 j d1

2
2(n − 1) − 1 + (2(n − 1) − 1)2 − 4(n − 2)2 + 4(n − 2)2 + 8(n − 2) n − 1
=
2
2n − 3 + (2n − 3)2 + 8(n − 2) n − 1
= .
2
12 P. BHUNIA, S. BAG AND K. PAUL

Now,

2n − 3 + (2n − 3)2 + 8(n − 2) n − 1

(n − 1) − >0
2
if,[2(n − 1)2 − {2n − 3}]2 − {2n − 3}2 + 8(n − 2) n − 1 > 0
i.e. if, m 4 − m2 {2m − 1} − 2(m − 1) m > 0,[put m = n − 1 > 1]
i.e. if, m2 (m − 1)2 − 2(m − 1) m > 0
i.e. if, m3 (m − 1)3 − 4 > 0

which is always true for each m ≥ 2 i.e., n > 2.

Similarly we can verify that the bound obtained by Kumar [12, Th.
1] is weaker than that obatined in Theorem 3.2 for a complete graph with
n > 2 vertices.

Remark 3.4 : Similarly we can show that for a complete bipartite graph
G = Km,n with vertices m, n (m > 1, n > 1, m £ n the bound obtained in
Theorem 3.2 is better than that obtained by Das and Kumar [6, Th. 3.2],
Kumar [12, Th. 1].

4. On bounds of least eigenvalue of adjacency matrix of a graph

In this section we give a lower bound for the least eigenvalue of the
adjacency matrix associated with a simple graph and an upper bound
for the least eigenvalue of the adjacency matrix associated with a simple
graph having chromatic number 2.

Theorem 4.1 : Suppose G be a simple graph of order n with chromatic number

2. Then
1
λn (G) ≤ −
e
∑
v ∼v , i < j
di d j ,
i j

where e is the number of edges, v1, v2, , vn are the vertices and di = degree
of the vertex vi.

Proof : Let x = (x1, x2, , xn)t ŒRn. Since chromatic number of the graph is 2
so it is possible to choose xi and xj so that they are of opposite signs when
vi ~ vj for all 1 ≤ i , j ≤ n, i ≠ j. Let

di dj
xi =− , x j = , if vi ∼ v j , 1 ≤ i , j ≤ n, i ≠ j.
2e 2e
EIGENVALUES OF THE ADJACENCY MATRIX 13

Then ∑ in=1|xi |2 = 1 and

di d j
x t A(G)x = −2 ∑
v ∼v , i < j 4e 2
i j

1
⇒ x t A(G)x =
− ∑ dd
e vi ∼ v j , i < j i j
1
⇒ λn (G) ≤ − ∑ dd .
e vi ∼ v j , i < j i j 

Theorem 4.2 : Suppose G be a simple graph with n vertices and e edges. Then
 
λn (G) ≥ − max 
1≤ i ≤ n
∑d
 j , v ∼v j
,

 i j 
where v1, v2, , vn are the vertices and di = degree of the vertex vi.

Proof : We note that l is an eigenvalue of A(G) if and only if –l is an

eigenvalue of – A(G). Now proceeding as in Lemma 2.1 and Theorem 2.2
we get for any eigenvalue l of A(G),
 
−λ ≤ max  ∑ d j  .
1≤ i ≤ n  
 j , vi ∼ v j 

So we get λn (G) ≥ − max 1≤ i ≤ n(∑ j , v ∼v d j ). 

i j

Remark 4.3 : For any tree T with n vertices and e(≥ 1) edges the bound
obtained by us in Theorem 4.2 is better than the bound obtained in [6, Th.
4.1]. This follows from the fact that
 
− max  ∑ d  ≥ − e = − 2e − (n − 1)d + (d − 1)d .
1≤ i ≤ n  j n n 1

 j , vi ∼ v j 

5. An upper bound of second largest eigenvalue of adjacency matrix of

a simple graph
Suppose G be a simple graph with n vertices v1, v2, , vn with deg(vi)
= di for each i = 1, 2,, n and e edges. Let G¢ = G – {vj} be the simple
graph obtained from G by deleting the vertex vj Œ{v1, v2, , vn}. Then for
each i ∈ {1, 2, … j − 1, j + 1, … , n} degree d¢i of the vertex vi in the graph G¢ is
defined as follows:
14 P. BHUNIA, S. BAG AND K. PAUL

di − 1, vi ∼ v j
di′ = 
d′i , otherwise

Using the above notations we now obtain an upper bound for the
second largest eigenvalue l2(G).

Theorem 5.1 : Suppose G be a simple graph with n vertices v1, v2, , vn with
deg(vi) = di, for each i = 1, 2,, n. Also assume that dj = max {d1, d2, , dn}. Then
λ2 (G) ≤ max
1≤ k ≤ n , k ≠ j
∑ d′ .
i , v ∼v
i
k i

Proof : Marcus and Mink [13] proved that if G be any simple graph and v
be any vertex in G, then
λi (G) ≥ λi (G − v) ≥ λi +1 (G) for i = 1, 2, … , n − 1.

Suppose G¢ = G – {vj}. Then λ1 (G ′) ≤ max 1≤ k ≤ n& k ≠ j(∑ i , vk ∼vi d′i ). Also

λ2 (G) ≤ λ1 (G − v j ) = λ1 (G′). Therefore,
 
λ2 (G) ≤ max  ∑ d′i .
1≤ k ≤ n & k ≠ j  
 i , v k ∼ vi  

Example 5.2 : Here we present two graphs G1 and G2 and their Huckel
matrices [11] associated with the molecular structure of the carbon skeleton
of 1,2-divinylcyclobutadiene and 1,4 - divinylbenzene respectively.

Figure 1
Graph G1 and the adjacency matrix A(G1)
EIGENVALUES OF THE ADJACENCY MATRIX 15

Figure 2
Graph G2 and the adjacency matrix A(G2)

Comparison of bounds.
In the following table we compare our upper bounds of largest
eigenvalue with a few known results for the Huckel matrices [11] A(G1)
and A(G2) representing the carbon skeleton of 1,2-divinylcyclobutadiene
and 1,4 - divinylbenzene respectively.

Bounds
Berman
Nikiforov Stanley obtained
Graphs and l1(G)
[14] [16] by us in
Zhang [3] Th. 2.2
G1 2.828 3.000 3.531 2.645 2.303
G2 3.162 2.449 4.000 2.449 2.214

The next table shows the comparison between our lower bounds of
largest eigenvalue with a few known results for the same Huckel matrices
A(G1) and A(G2).
16 P. BHUNIA, S. BAG AND K. PAUL

Bounds
Das and
Favaron Kumar obtained
Graphs Hong [8] Kumar l1(G)
et al. [7] [12] by us in
[6]
Th. 3.2
G1 2.000 1.732 2.000 2.135 2.203 2.303
G2 2.000 1.732 1.847 1.902 2.152 2.214

Looking at the two tables we conclude that the bounds obtained by

us both in case of lower and upper bounds are better than the existing
bounds as referred above.

References

[1] R.B. Bapat, Graphs and Matrices, Hindustan Book Agency, Springer.
[2] R.C. Brigham and R.D. Dutton, Bounds on the graph spectra, J. Com-
bin. Theory Ser. B 37 (1984) 228-234.
[3] A. Berman and X.D. Zhang, On the spectral radius of graphs with cut
vertices, J. Combin. Theory Ser. B 83 (2001) 233-240.
[4] L. Collatz and U. Sinogowitz, Spektren Endlicher Grafen, Abh. Math.
Sem. Univ. Hamburg 21 (1957) 63-77.
[5] G. Constantine, Lower bounds on the spectra of symmetric matrices
with non negative entries, Linear Algebra Appl. 65 (1985) 171-178.
[6] K.C. Das and P. Kumar, Some new bounds on the spectral radius of
graphs, Discrete Math. 281 (2004) 149-161.
[7] O. Favaron, M. Maheo and J.F. Sacle, Some eigenvalue properties in
graphs ( conjecture of Graffitic-II), Discrete Math. 111 (1993) 197-220.
[8] Y. Hong, On the spectra of unicycle graph, J. East China Norm. Univ.
Natur. Sci. Ed. 1 (1986) 31-34.
[9] Y. Hong, Bounds of eigenvalues of graphs, Acta Math. Appl. Sinica 2
(1988) 165-168.
[10] Y. Hong, A bound on the spectral radius of graphs, Linear Algebra
Appl. 108 (1988) 135-140.
[11] D. Janežić, A. MiliČević, S. Nikolić, and N. Trinajstić, Graph-theoret-
ical matrices in chemistry, CRC Press, Taylor & Francis Group, 2015.
[12] R. Kumar, Bounds for eigenvalues of a graph, J. Math. Inequal. 4(3)
(2010) 399-404.
EIGENVALUES OF THE ADJACENCY MATRIX 17

[13] M. Marcus and H. Mink, A Survey of Matrix Theory and Matrix In-
equalities, Allyn and Bacon, Boston. 1964.
[14] V. Nikiforov, Some inequalities for the largest eigenvalue of a graph,
Combin. Probab. Comput. 11 (2002) 179-189.
[15] D.L. Powers, Graph partitioning by eigenvectors, Linear Algebra
Appl. 101 (1988) 121-133.
[16] R.P. Stanley, A bound on the spectral radius of graphs with e edges,
Linear Algebra Appl. 67 (1987) 267-269.
[17] H. Wilf, The eigenvalues of a graph and its chromatic number, J. Lon-
don Math. Soc., 42(1967), 330-332.

Received October, 2018