2018 Czechoslovak Mathematical Journal 10 pp

Online first

NOTE ON A CONJECTURE FOR THE SUM OF SIGNLESS

LAPLACIAN EIGENVALUES

Xiaodan Chen, Nanning, Guoliang Hao, Nanchang,

Dequan Jin, Nanning, Jingjian Li, Nanning

Received October 18, 2016. Published online June 7, 2018.

Abstract. For a simple graph G on n vertices and an integer k with 1 6 k 6 n, denote

by Sk+ (G) the sum of k largest signless Laplacian eigenvalues of G. It was conjectured that


Sk+ (G) 6 e(G) + k+12 , where e(G) is the number of edges of G. This conjecture has been
proved to be true for all graphs when k ∈ {1, 2, n − 1, n}, and for trees, unicyclic graphs,
bicyclic graphs and regular graphs (for all k). In this note, this conjecture is proved to be
true for all graphs when k = n − 2, and for some new classes of graphs.
Keywords: sum of signless Laplacian eigenvalues; upper bound; clique number; girth
MSC 2010 : 05C50, 15A18

1. Introduction

All graphs considered in this note are finite, undirected and simple. Let G
be a graph with vertex set V (G) = {v1 , v2 , . . . , vn } and edge set E(G), and let
e(G) = |E(G)|. Denote by dG (vi ) the degree of the vertex vi in G. The adja-
cency matrix of G is A(G) = [aij ]n×n , where aij = 1 if the vertices vi and vj
in G are adjacent, and aij = 0 otherwise. The Laplacian matrix and the signless
Laplacian matrix of G are, respectively, defined to be L(G) = D(G) − A(G) and
Q(G) = D(G) + A(G), where D(G) is the diagonal matrix of vertex degrees of G.
The eigenvalues of L(G) and Q(G), usually called the Laplacian eigenvalues and the
signless Laplacian eigenvalues of the graph G, are arranged (in non-increasing order)

This work was supported by National Natural Science Foundation of China (Nos.
11501133, 11661010, 11461004), Natural Science Foundation of Guangxi Province
(Nos. 2016GXNSFAA380293, 2014GXNSFBA118008) and Natural Science Foundation
of Fujian Province (No. 2015J05017).

DOI: 10.21136/CMJ.2018.0548-16 1
as µ1 (G) > µ2 (G) > . . . > µn (G) = 0 and q1 (G) > q2 (G) > . . . > qn (G) > 0,
respectively. It is known that if G is bipartite, then L(G) and Q(G) are similar, and
hence their eigenvalues are identical (see, e.g., [4], p. 217). For more details on the
Laplacian eigenvalues and the signless Laplacian eigenvalues of graphs one may refer
to [3], [4].
For an integer k with 1 6 k 6 n, let Sk (G) be the sum of k largest Laplacian eigen-
k
P
values of a graph G, that is, Sk (G) = µi (G). Grone and Merris [9] conjectured
i=1
that for any graph G with n vertices and each k ∈ {1, 2, . . . , n},
k
X
Sk (G) 6 |{v ∈ V (G) : dG (v) >}|.
i=1

This conjecture has been proved to be true by Bai [2] recently and now is called
the Grone-Merris theorem. As a variation of the Grone-Merris theorem, Brouwer [3]
conjectured that for any graph G with n vertices and each k ∈ {1, 2, . . . , n},
 
k+1
Sk (G) 6 e(G) + .
2

Brouwer’s conjecture has attracted the attention of many researchers, but has not
been settled yet. For the progress on this conjecture one can see [5], [7], [8], [10],
[11], [12].
Analogously to the definition of Sk (G), let Sk+ (G) be the sum of k largest signless
k
Laplacian eigenvalues of the graph G, that is, Sk+ (G) =
P
qi (G), k ∈ {1, 2, . . . , n}.
i=1
Motivated by Brouwer’s conjecture, Ashraf et al. [1] posed the following conjecture:
Conjecture 1.1. For any graph G with n vertices and each k ∈ {1, 2, . . . , n},
 
k+1
Sk+ (G) 6 e(G) + .
2

By a computer search, Ashraf et al. [1] confirmed Conjecture 1.1 for all graphs
with at most 10 vertices. They also proved that Conjecture 1.1 is true for all graphs
when k ∈ {1, 2, n − 1, n}, and for regular graphs (for all k). In addition, they pointed
out that Conjecture 1.1 holds for trees (for all k), since Sk+ (G) = Sk (G) holds when G
is bipartite. Recently, Yang and You [13] further showed that Conjecture 1.1 is true
for unicyclic and bicyclic graphs (for all k).
In this note, we continue to explore Conjecture 1.1. We will show that Conjec-
ture 1.1 holds for all graphs when k = n − 2, and for some new classes of graphs.

2 Online first
 2. Lemmas and results

As usual, we denote by Kn and K1, n−1 the complete graph and the star with n ver-
tices, respectively. Let G1 ∪ G2 denote the vertex-disjoint union of two graphs G1
and G2 , and let kG denote the vertex-disjoint union of k copies of the graph G. For
a subgraph H of G, write G − E(H) for the spanning subgraph of G whose edge set
is E(G) \ E(H). The complement of G is denoted by G.

Lemma 2.1 ([4]). Let G be a graph with n > 2 vertices and let G′ be an edge-
deleted subgraph of G, that is, G′ = G − E(K2 ). Then q1 (G) > q1 (G′ ) > q2 (G) >
q2 (G′ ) > . . . > qn (G) > qn (G′ ).
It is known that q1 (Kn ) = 2n−2 and q2 (Kn ) = . . . = qn (Kn ) = n−2 (see, e.g., [1]).
This, together with Lemma 2.1, yields the next lemma.

Lemma 2.2. If G is a graph of order n > 2, then q1 (G) 6 2n−2 and qi (G) 6 n−2
for i = 2, . . . , n.
For an n × n Hermitian matrix M , we arrange its eigenvalues (in non-increasing
order) as λ1 (M ) > λ2 (M ) > . . . > λn (M ). The following result is the well-known
Courant-Weyl inequality (see, e.g., [4], p. 19).

Lemma 2.3 ([4]). If A and B are n×n Hermitian matrices, then for n > i > j > 1,
λi (A + B) 6 λj (A) + λi−j+1 (B).
The next result gives a relation between the signless Laplacian eigenvalues of G
and those of G, which can be deduced from Lemma 2.3 by bearing in mind that
Q(Kn ) = Q(G) + Q(G) and qn (Kn ) = n − 2.

Lemma 2.4. If G is a graph with n > 2 vertices and G is its complement, then
for i = 1, 2, . . . , n, qi (G) > n − 2 − qn−i+1 (G).

Lemma 2.5 ([13]). If G is a graph with n vertices and G1 , G2 , . . . , Gt are its edge-
St
disjoint subgraphs with E(G) = E(Gi ), then for any integer k with 1 6 k 6 n,
i=1
t
Sk+ (G) Sk+ (Gi ), Sk+ (Gi ) = Sn+i (Gi ) if k > |V (Gi )| = ni .
P
6 where
i=1

Lemma 2.6 ([13]). If G is a connected graph with n vertices, then for any integer k
with
p
3n − 4 + 8n2 (e(G) − n + 1) + (n − 4)2
 
k+1
6 k 6 n, Sk+ (G) 6 e(G) + .
2n 2

Online first 3
 It is shown [10], [12] that for any acyclic graph (i.e., tree or forest) F of order n,
Sk (F ) 6 e(F )+ k+1


2 holds for all k ∈ {1, 2, . . . , n}. This, together with the fact that
Sk+ (G) = Sk (G) holds for any bipartite graph G, yields the following result directly.

Lemma 2.7. If F is an acyclic graph (i.e., tree or forest) with n > 2 vertices,
then for any integer k with 1 6 k 6 n, Sk+ (F ) 6 e(F ) + k+1


2 .

Lemma 2.8 ([6]). If G is a connected graph with n > 2 vertices, then q1 (G) 6
2e(G)/(n − 1) + n − 2, with equality if and only if G ∼
= K1,n−1 , or G ∼
= Kn .
Now, we are in position to present the main results of this note.

Theorem 2.9. For n > 3, let p be an integer with 1 6 p 6 n/3. If Conjecture 1.1
holds for all graphs when k = p, then Conjecture 1.1 holds for all graphs when
k = n − p as well.
P r o o f. Suppose that G is any graph with n > 3 vertices and G is its comple-
ment. The hypothesis of the theorem implies that
   
p+1 p+1
Sp+ (G) 6 e(G) + and Sp+ (G) 6 e(G) + .
2 2
We now just need to show that
 
+ n−p+1
Sn−p (G) 6 e(G) + .
2
n
P
Indeed, bearing in mind the well-known fact that qi (G) = 2e(G), we have
i=1
n
X
+
Sn−p (G) = 2e(G) − qi (G)
i=n−p+1
Xp
6 2e(G) − (n − 2 − qi (G)) (by Lemma 2.4)
i=1
= 2e(G) − p(n − 2) + Sp+ (G)
 
p+1
6 2e(G) − p(n − 2) + e(G) +
2
   
n p+1 n


= e(G) + − p(n − 2) + (as e(G) + e(G) = 2 )
2 2
n2 − (2p + 1)n + p2 + 5p
= e(G) +
 2
n−p+1
6 e(G) + (as n > 3p)
2
as desired. 

4 Online first
 It is known [1] that Conjecture 1.1 holds for all graphs when n 6 10 or k = 2.
This, together with Theorem 2.9, yields the following corollary, which asserts that
Conjecture 1.1 holds for all graphs when k = n − 2.

+ n−1


Corollary 2.10. If G is a graph with n > 3 vertices, then Sn−2 (G) 6 e(G)+ 2 .
It is also worth pointing out that Theorem 2.9 suggests that to prove Conjec-
ture 1.1, it is sufficient to prove Conjecture 1.1 for all graphs when 1 6 k 6 2n/3.
As an application of this idea, we may derive the following result, which, in some
sense, can be regarded as a partial solution to Conjecture 1.1.

Theorem 2.11. Let G be a graph with n > 3 vertices. If

(n − 1)(4n2 − 15n)
e(G) > ,
9(n − 3)

then Sk+ (G) 6 e(G) + k+1

2 holds for 1 6 k 6 2n/3.
P r o o f. For 1 6 k 6 2n/3, by Lemmas 2.2 and 2.8, we have
 
+ 2e(G) k+1
Sk (G) 6 + n − 2 + (k − 1)(n − 2) 6 e(G) + ,
n−1 2

provided that

2(n − 3)
(2.1) k 2 − (2n − 5)k + e(G) > 0.
n−1

To complete this proof, we just need to prove that (2.1) holds for 1 6 k 6 2n/3 when

(n − 1)(4n2 − 15n)
e(G) > .
9(n − 3)

Now, consider the quadratic equation

2(n − 3)
f (x) = x2 − (2n − 5)x + e(G) = 0,
n−1
with the discriminant being

8(n − 3)
∆ = (2n − 5)2 − e(G).
n−1
It is easy to see that if ∆ 6 0, that is,

(n − 1)(2n − 5)2  (n − 1)(4n2 − 15n) 

e(G) > > ,
8(n − 3) 9(n − 3)

Online first 5
then f (x) > 0 holds for any real number x and hence, (2.1) follows. Otherwise, the
least root of f (x) = 0 is
q
2n − 5 − (2n − 5)2 − 8(n−3)
n−1 e(G)
α= ,
2
which shows that f (x) > 0 holds for x 6 α. Thus, (2.1) holds for 1 6 k 6 2n/3 if
q
2n − 5 − (2n − 5)2 − 8(n−3)
n−1 e(G) 2n
> ,
2 3
that is,
h 8(n − 3) i
(2n − 15)2 > 9 (2n − 5)2 − e(G) ,
n−1
which follows when
(n − 1)(4n2 − 15n)
e(G) > .
9(n − 3)
This completes the proof. 
Remark. By the above proof, one may draw a stronger conclusion (from a some-
what stronger assumption): for any graph G with n > 3 vertices, if

(n − 1)(2n − 5)2
e(G) > ,
8(n − 3)

then Sk+ (G) 6 e(G) + k+1

2 holds for 1 6 k 6 n; in other words, Conjecture 1.1
holds for the graphs with at least

(n − 1)(2n − 5)2
8(n − 3)

edges. In particular, a direct calculation shows that

(n − 1)(2n − 5)2
 
n
6 − (n − 2),
8(n − 3) 2

which implies that Conjecture 1.1 holds for the graphs obtained from Kn by deleting
at most n − 2 edges.
Recall that the clique number of a graph G, denoted by ω(G), is the number of
vertices of a maximum complete subgraph contained in G. By Theorem 2.11, we
may obtain the next corollary.

Corollary 2.12. Let G be a connected graph with n > 3 vertices. If ω(G) >
√
(2 2n − 1)/3, then Sk+ (G) 6 e(G) + k+1


2 holds for 1 6 k 6 2n/3.

6 Online first
 P r o o f. Note that for a connected graph G we have e(G) > 21 ω(G)(ω(G) − 1) +
n − ω(G). Thus, by Theorem 2.11, it suffices to show that

1 (n − 1)(4n2 − 15n)
ω(G)(ω(G) − 1) + n − ω(G) > ,
2 9(n − 3)

that is,
9(n − 3)ω(G)2 − 27(n − 3)ω(G) − 4n(2n2 − 14n + 21) > 0.
Consider the following quadratic equation:

h(x) = 9(n − 3)x2 − 27(n − 3)x − 4n(2n2 − 14n + 21) = 0.

It is easy to check that the largest root of h(x) = 0 is

p
9(n − 3) + 81(n − 3)2 + 16n(n − 3)(2n2 − 14n + 21)
β= ,
6(n − 3)

which shows that h(x) > 0 holds for x > β. Thus, to complete this proof, we just
need to prove that
p √
9(n − 3) + 81(n − 3)2 + 16n(n − 3)(2n2 − 14n + 21) 2 2n − 1
6 ,
6(n − 3) 3

that is,
√
2
81(n − 3) + 16n(2n2 − 14n + 21) 6 (n − 3) 4 2n − 11 ,
that is,
√
√
16 − 11 2 n2 + 33 2 − 37 n − 15 > 0,

which holds for n > 3 (since its largest root is 1.4543), completing the proof. 
Recall that the girth of a graph G (of order n), denoted by g(G), is the length
(i.e., the number of edges) of a shortest cycle contained in G. Clearly, 3 6 g(G) 6 n.
We here make a convention that a graph G is acyclic if and only if g(G) > n.

Theorem 2.13. If G is a graph with n vertices and n > g(G) > g > 4, then
Sk+ (G)
6 e(G) + k+1


2 holds for 1 6 k 6 ⌊g/4⌋.
P r o o f. Some idea of the proof comes from Lemma 15 in [1]. Let G be a coun-
terexample for the theorem having a minimum number of edges. Then we have
that
 
+ k+1
(2.2) Sk (G) > e(G) +
2

holds for some k with 1 6 k 6 ⌊g/4⌋.

Online first 7
 Note that G contains a cycle of length at least g and hence, it contains a subgraph
⌊g/2⌋K2 (belonging to the cycle) as well. It is easy to check that for 1 6 k 6 ⌊g/4⌋,

(2.3) Sk+ (⌊g/2⌋K2) = 2k 6 ⌊g/2⌋ = e(⌊g/2⌋K2 ).

Now, let G′ = G − E(⌊g/2⌋K2 ). By Lemma 2.5, together with (2.2) and (2.3), we
obtain
 
k+1
e(G) + < Sk+ (G) 6 Sk+ (⌊g/2⌋K2) + Sk+ (G′ ) 6 e(⌊g/2⌋K2) + Sk+ (G′ ),
2

from which, as well as the fact that e(G′ ) = e(G) − e(⌊g/2⌋K2), we conclude that
 
k+1
(2.4) Sk+ (G′ ) ′
> e(G ) +
2

holds for some k with 1 6 k 6 ⌊g/4⌋.

To complete the proof, it remains to be shown that n > g(G′ ) > g, which, together
with (2.4), would yield that G′ is also a counterexample for the theorem but has
fewer edges than G, contradicting the minimality of G. Indeed, on one hand, by the
definition of girth, we have g(G′ ) > g(G) > g. On the other hand, if g(G′ ) > n,
then G′ is acyclic and consequently, by Lemma 2.7, we have that Sk+ (G′ ) 6 e(G′ ) +
k+1


2 holds for all k, contradicting (2.4).
This completes the proof. 

Theorem 2.14. Let G be a connected graph with n > 4 vertices. If

p
(2.5) g(G) > 6 + 2 8(e(G) − n + 1) + 1,

then Sk+ (G) 6 e(G) + k+1

2 holds for 1 6 k 6 n.

P r o o f. Since G is connected, we have e(G) − n + 1 > 0 and hence, by (2.5), we

obtain g(G) > 6. Consequently, for any integer k with 1 6 k 6 g(G)/4, Theorem 2.13
yields directly that Sk+ (G) 6 e(G) + k+1


2 . On the other hand, for g(G)/4 < k 6 n,
again by (2.5) we get
p
g(G) 3n − 4 + 8n2 (e(G) − n + 1) + (n − 4)2
n>k> > ,
4 2n

which, together with Lemma 2.6, also yields that Sk+ (G) 6 e(G) + k+1


2 . Combining
these two cases, we obtain the desired result, completing the proof. 

8 Online first
 Remark. Theorem 2.14 asserts that Conjecture 1.1 holds for the connected
graphs having sufficiently large girth relative to the number of edges. Moreover,
recall that a connected graph with n vertices and c cycles, usually called a c-cyclic
graph, has n−1+c edges. We can now restate Theorem 2.14 as follows: if G a c-cyclic
√
graph with g(G) > 6 + 2 8c + 1, then Sk+ (G) 6 e(G) + k+1


2 holds for 1 6 k 6 n,
from which one can easily conclude that Conjecture 1.1 is true for the 3-cyclic graphs
with girth at least 16, and for the 4-cyclic graphs with girth at least 18, and so on.
It should be mentioned that for the cases of c = 0, 1 and 2, the theorem still holds
√
when the restriction g(G) > 6 + 2 8c + 1 is removed (see [1], [13] for details).

Acknowledgements. The authors are very grateful to the anonymous referees

for their valuable comments and suggestions which have considerably improved the
presentation of this paper.

References

[1] F. Ashraf, G. R. Omidi, B. Tayfeh-Rezaie: On the sum of signless Laplacian eigenvalues

of a graph. Linear Algebra Appl. 438 (2013), 4539–4546. zbl MR doi
[2] H. Bai: The Grone-Merris conjecture. Trans. Am. Math. Soc. 363 (2011), 4463–4474. zbl MR doi
[3] A. E. Brouwer, W. H. Haemers: Spectra of Graphs. Universitext, Springer, Berlin, 2012. zbl MR doi
[4] D. Cvetković, P. Rowlinson, S. Simić: An Introduction to the Theory of Graph Spectra.
London Mathematical Society Student Texts 75, Cambridge University Press, Cam-
bridge, 2010. zbl MR doi
[5] Z. Du, B. Zhou: Upper bounds for the sum of Laplacian eigenvalues of graphs. Linear
Algebra Appl. 436 (2012), 3672–3683. zbl MR doi
[6] L. Feng, G. Yu: On three conjectures involving the signless Laplacian spectral radius of
graphs. Publ. Inst. Math., Nouv. Sér. 85 (2009), 35–38. zbl MR doi
[7] E. Fritscher, C. Hoppen, I. Rocha, V. Trevisan: On the sum of the Laplacian eigenvalues
of a tree. Linear Algebra Appl. 435 (2011), 371–399. zbl MR doi
[8] H. A. Ganie, A. M. Alghamdi, S. Pirzada: On the sum of the Laplacian eigenvalues of
a graph and Brouwer’s conjecture. Linear Algebra Appl. 501 (2016), 376–389. zbl MR doi
[9] R. Grone, R. Merris: The Laplacian spectrum of a graph II. SIAM J. Discrete Math. 7
(1994), 221–229. zbl MR doi
[10] W. H. Haemers, A. Mohammadian, B. Tayfeh-Rezaie: On the sum of Laplacian eigenval-
ues of graphs. Linear Algebra Appl. 432 (2010), 2214–2221. zbl MR doi
[11] I. Rocha, V. Trevisan: Bounding the sum of the largest Laplacian eigenvalues of graphs.
Discrete Appl. Math. 170 (2014), 95–103. zbl MR doi
[12] S. Wang, Y. Huang, B. Liu: On a conjecture for the sum of Laplacian eigenvalues. Math.
Comput. Modelling 56 (2012), 60–68. zbl MR doi

Online first 9
[13] J. Yang, L. You: On a conjecture for the signless Laplacian eigenvalues. Linear Algebra
Appl. 446 (2014), 115–132. zbl MR doi

Authors’ addresses: X i a o d a n C h e n, College of Mathematics and Information Sci-

ence, Guangxi University, No. 100, Daxue Road, Nanning, Guangxi, P. R. China, e-mail:
[email protected]; G u o l i a n g H a o (corresponding author), College of Science, East
China University of Technology, No. 418, Guanglan Road, Nanchang 330013, Jiangxi,
P. R. China, e-mail: [email protected]; D e q u a n J i n, J i n g j i a n L i, College of
Mathematics and Information Science, Guangxi University, No. 100, Daxue Road, Nan-
ning, Guangxi, P. R. China, e-mail: [email protected], [email protected].

10 Online first