0% found this document useful (0 votes)
33 views19 pagesOn A Conjecture For The Signless Laplacian Eigenvalues
Uploaded by
Guilherme PortoCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
33 views19 pagesOn A Conjecture For The Signless Laplacian Eigenvalues
Uploaded by
Guilherme PortoCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 19
Linear Algebra and its Applications 446 (2014) 115-132
Contents lists available at ScienceDirect
ie
Linear Algebra and its Applications Applications
www elsevier.com/locate/laa
On a conjecture for the signless Laplacian Dew
cigenvalues
Jieshan Yang, Lihua You *
School of Mathematical Sciences, South China Normal University, Guangzhou,
510631, China
ARTICLE INFO ABSTRACT
Article history: Let G be a simple graph with n vertices and e(G) edges,
Received 23 May 2018 and q@(G) > q2(@) > > qn(G) > 0 be the signless
Rear nc od hemnarce 20L4 Laplacian eigenvalues of @, Let $}(G) = DE, 4i(G), where
Submitted by R. Brualdi k = 1,2,...,n. F, Ashraf et al, conjectured that ${(G@) <
6(G) + (*5') for
1,2,...,m. In this paper, we give various
MSC upper bounds for S{(G), and prove that this conjecture is
05C50 i
ee true for the following cases: connected graph with sufficiently
® large k, unicyclic graphs and bicyclic graphs for all k, and
tricyclic graphs when & # 3. Finally, we discuss whether the
Keywords. y y
Sighleee Laplacian eigenvalues upper bound given in this conjecture is tight or not for c-cyclic
Conjecture graphs and propose some problems for future research.
Connected graph © 2013 Elsevier Inc. All rights reserved.
Unieyelic graph
Bicyclic graph
‘Tricyclic graph
1. Introduction
Let @ be a simple graph with vertex set V(@) and edge set E(G). The Laplacian
matrix and the signless Laplacian matrix of @ are defined as L(G) = D(G) — A(@) and
* Research supported by National Natural Science Foundation of China (No. 10901061), the Zhujlang
‘Technology New Star Foundation of Guangzhou (No. 201132200090), and Program on International
Cooperation and Innovation, Department of Education, Guangdong Province (No. 2012g}hz0007)
* Corresponding author.
E-mail address: ylhuaMscnu.educn (L. You)
(0024-8705/8 — see front matter © 2013 Elsevier Inc. All rights reserved.
http:/ dx dol.org/ 10.1016 /}.1aa.2013.12.032 16 J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-132
Q(G) = D(G) + A(G) respectively, where A(G) is the adjacent matrix and D(G) is the
diagonal matrix of vertex degrees of @. It is well known that both L(G) and Q(@) are
symmetric and positive semidefinite, then we can denote the eigenvalues of L(G) and
Q(C), called respectively the Laplacian eigenvalues and the signless Laplacian eigenvalues
of G, by pui(G) 2 o(G) > --- > in(@) = 0 and qi(G) 2 q2(G) > --- 2 an(G@) 2 0.
Let |U| be the cardinality of a finite set U, and |E(@)| = e(G). If (G) =n+e-1,
then G is called a c-cyclic graph. Especially, if ¢ = 0 (¢ = 1, ¢ = 2, c= 3), then G is
called an acyclic (a tmicyclic, a bicyclic, a tricyclic) graph.
Grone and Merris [10] conjectured that for a graph with n vertices and degree sequence
{d, | v € V(G)}, the following holds:
k k
SG) = l@) < VI \{v EV(G) | dy >HI, for K=1,2,.
a a
Recently, it was proved by Bai [2]. As a variation of the Grone-Merris conjecture,
Brouwer [4] conjectured that for a graph G with n vertices,
k+1
S4(@) As(M) > + > A,(M) be the eigenvalues of any matrix M, and o(M)
be the spectrum of M.
Lemma 2.1. (See [9).) Let A and B be two real symmetric matrices of s
any Sk n, we denote $$(G) = St (G) since SP(GU (k — n)Ky) =
Sn (G).
Lemma 2.2. Let G@ be a graph with n vertices, G1,G2,...,Gr be the edge disjoint sub-
graphs of G such that E(@) = Uf; B(G:), where t > 1. Then for any integer k with
l e(H) for every nonempty subgraph H of @
Proof, Let @ be a counterexample of Conjecture 1.1 for some integer k, having the
minimum number of edges. If G has a nonempty subgraph H with $¢(G) < e(H), then
by Lemma 2.2,
k+L ¢-
eG) + ( ; ) < SE(@) < S¢(H) + S¢(G— EH) < e(H) + S$ (@ — EU).
This implies that Sf (G@ — E(H)) > e(@) — e(H) + (*$!) = e(@ — E(H)) + (52), which
is a contradiction to the choice of G. 0
Remark 2.4, Corollary 2.3 is a natural extension of Lemma 15 in [1]
Lemma 2.5. (See [1).) If for some k, Conjecture 1.1 holds for graphs G and H, then it
holds for GU H.
Therefore, in order to prove Conjecture 1.1, it suffices to do so for connected graphs.
Thus we only need to consider the following conjecture,
Conjecture 2.6. For any connected graph G with n vertices and any k = 1,2.....7,
SH(G) < (CG) + ("3"):
2
3. Upper bounds for Sz (@)
In this section, we give various upper bounds for $f (@) in terms of the clique number
w and e(@), the maximum degree A and e(@), the matching number m and e(G),
respectively. Furthermore, we also show Conjecture 2.6 is true for connected graph with
sufficiently large k, which plays an important role in the proofs of our main results.
Recall that the clique number of G is the number of vertices of a maximum complete
subgraph of @. A matching M of G is a subset of E(G) such that no two edges in M share
a common vertex. A maximum matching is a matching which covers as many vertices
as possible. The matching number of G is the number of edges in a maximum matching
of G. J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-192 119
Theorem 3.1. Let G be a graph with clique number w. Then for any k = 1,2....,0,
Sit (G) < 2e(G) — w? + (k + 2)w — 2k
Proof. Obviously, Kus is a subgraph of G. Note that o(Q(K.)) = (20 — 2,0 — 2-H},
where \!J means that \ is an eigenvalue with multiplicity t. Thus, by Lemma 2.2,
SE(G) < Sf(Ku) + [e(@) — e(Ku)] Sf (Ka)
= 20-24 (k= —2)+2[e(6) - (3)|
=2(G)—w+(k+2w—-2k 0
‘Theorem 3.2. Let G@ be a graph with maximum degree A. Then for any k= 1.2..
Si(G) < 2(G)—A+k.
Proof. Obviously, Ki,4 is a subgraph of @. Note that 7(Q(Ki,a)) = {4 + 1,1!4-4, 0}.
Thus, by Lemma 2.2,
SE(C) < S$ (Kia) + [e(G) — e(Ki,a)| Sf (Ka)
=At+k+2[e(G)- A] =2(G)-A+k. Oo
Theorem 3.3, Let G be a graph with matching number m. Then for any k = 1,2,....m,
Si(G) < 2e(G) — 2m + 2k.
Proof. Obviously, mK is a subgraph of G. Note that o(Q(mKy)) = {20,0}. Thus,
by Lemma 2.2, 5(G) < $}(mB2) + [e(@) — e(mK2)]$} (Ke) = 2e(@) — 2m +2k. 0
Lemma 3.4, (See [5).) If @ is bipartite, then Q(@) and L(G) share the same eigenvalues.
By Lemma 3.4, the correctness for Conjecture 2.6 for trees follows from that of Brou-
wer's conjecture, that is, for any tree T with n vertices and any
2 cel) + (3) a)
Theorem 3.5. Let G be a connected graph with n vertices. Then
SE(L) O and bpp = kypgg = ++ = ky = O where 1 Sz (Gi) < Yo [2e(Gi) + 2ks — mj] = 2e(H) + 2k — |V(H)|.
ia i J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-192 121
Case 1: r = t, then the result is obvious.
Case 2:1 < t.
Note that e(G;) > 1 for r+1 < @ < t since G has no isolated vertices. Then for
r+1 e(Gi) + (mi = 1) — mi = e(Gi) -1 > 0,
which implies that 2e(G) — n > 2e(H) — |V(H)|. Thus, Sjf(@) <2e(@)+2k—n. a
4. Conjecture 2.6 for unicyclic, bieyelic and tricyclic graphs
In this section, we prove that Conjecture 2.6 is true for unicyclic and bicyclic graphs
with n vertices for all integer k, and tricyclic graphs with n vertices when k #3, where
i 3 holds.
Since @ is a unicyclic graph, e(@) =n. Then
Bn 4+ YSPC) SF FONE FTG _ Bn — A + VON BNF TS
2n ~ Qn
B 4 holds.
Since G is a bicyclic (respectively tricyclic) graph, e(@) = n +1 (respectively e(G) =
n+2), Then
3 vitn? =8n+16 | 3n
ona se vie sents
In
Then by Corollary 3.6, we have Sf (G) < e(G) + ("£1) for any integer 1 < k 11. The following
notations, put forward in [14], and lemmas are needed in the proofs of our main results.
Let G; ~ Go denote the graph obtained from G: and G2 by connecting a vertex of G1
with a vertex of Go. Let G1 & Go denote the graph obtained from Gy and G2 by
two edges between V(G1) and V(G2). The following two lemmas show that if Conjec-
ture 2.6 holds for G, and G2, then Conjecture 2.6 is also true for G; ~ Gy and G; = Ga.
serting
Lemma 4.3. Let G, be a nonempty graph with nj vertices, where i = 1,2. If Sf (Gi) <
(Gi) + ($1) for ky = 1, ny and i= 1,2, then for k =1,2,...,m, +n,
k+l
SCG ~ G2) < (Gi ~ Gs) + (51).
Proof. Assume that k; of the k largest siguless Laplacian eigenvalues of Gy U G2 come
from o((Q(Gi)), where i = 1,2 and ky +k = k. Since G; is nonempty, e(G;) > 1 and
e(G1 ~ Ga) = e(G1) + e(G2) +1 2 e(Gi) +2 fori = 1,2.
Case 1: kik = 0.
Without loss of generality, suppose that ky = 0, then ky = k. By Lemma 2.2,
Sy(G1 ~ Ga) < Sf(G1 UG2) + Sf (Ka)
= Si (G1) + Sf (Ka)
< (Gi) + ("3 ‘) +2
1. Then by Lemma 2.2,
Sif(Gi ~ Ga) < Sf (Gi UG2) + Sf (Ke)
= SSE (GC) + Sf (2)
2
2
Shor]
AR +A +h the $2
2
thi tke
=e(Gi~ G2) +
4) vertices such that there is a hanging tree
T of G with \V(D)| > 2. Suppose |V(G —T)| = s, where2 2, where i = 1,2. If S(Gi) <
(Gi) + ("E1) for ki = 1,2,...,m: and i= 1,2, then for k= 1,2,...,m1 + ne,
k+.
SE(Gi = Ga) < (Gi ~G) + ( ; ):
Proof. The cases of k= Lis trivial, and k = 2 have been confirmed in {1}. We assume that
k > 3 in the following, and k, of the k largest signless Laplacian eigenvalues of Gy U Ga
come from o(Q(G,)), where i = 1,2 and ky + ky = k. Since o(G,) > 2, e(Gy ~ Ge) =
e(G1) + eG) +2 > eG.) +4 for #=1,2.
Case 1: kiky = 0.
Without loss of generality, suppose kz = 0, then ky =k. Hence, by Lemma 2.2,
Si(G1 © Ga) < Sf (Gy UG2) + 287 (Ke)
= Sf (G1) + 28f (Ke)
=aeny+(" "ites
3, kykz > 2. Then by Lemma 2.2,
SE(G1 © G2) < Sf (G1 UG2) + 28f (Ka)
2
= 2 SEG) + 2575)
p > 3. Let 00(p,q,t) denote the bicyclic
graph obtained from a path P; and two cycles C,, C, by identifying a vertex of C, with
one end of P; and a vertex of Cy with the other end vertex of Pr, where q > p > 3 and
t > 2 (see Fig. 1).
A bicyclic graph is called oo-type if it is oo(p,q,t) or it can be obtained by attach-
ing some hanging trees to co(p,q,t), where g > p > 3 and t > 1, We will show that
Conjecture 2.6 is true for oo-type bicyclic graphs when k
Theorem 4.7. Let n,k be integers with n > 11 and 1 < k p > 3 and t > 1. By Corollary 4.4, it will
suffice to consider the oo-type bicyclic graph G which is obtained by attaching some
pendent vertices to 00(p,q,t) or G & 00(p,q,t)- J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-192 125
- >
v v
Un (a,b) Un(a,b)
Fig. 2. The graphs U(a,b) and U2(a, 6)
Case 1: t > 2
Let e be an edge of P; in c0(p,q,t), then G —e is the union of two unicyclic graphs.
By Theorem 4.1 and Lemma 4.3, we get the desired result.
Case 2: 1.
Let o0'(3,3,1) be the graph obtained from 00(3.3, 1) by attaching n — 5 pendent
vertices to the common vertex of two eyeles. If G © o0/(3,3.1), then by Lemma 4.6 and
direct calculation, ¢(co!(3,3, 1), ©) = (e — 1)" “4(w — 3)[e — (n + 3)? + 3na — 8). Thus
S¥ (0o!(3,3,1)) < (n +3) +341=(n +1) + ($') = e(00!(3,3, 1) + C3’).
Otherwise, there exits two edges e1 and ¢2 of a cycle such that @—{e1, €2} is the union
of a unicyelic graph and a tree with at least two edges. Combining the Inequality (1),
‘Theorem 4.1 and Lemma 4.5, the result is obtained. 0
Let p.q,t be integers with p > 3, q > 3, and 2 11. The following lemmas are essential in our proof.
Let U}(a,b) be the graph obtained by attaching a and b pendent vertices to two
vertices of a triangle, respectively, where a +b =n —3,n 2 4 and a > b > 0. Let
U2(a,b) be the graph obtained by attaching a and b pendent vertices to two nonadjacent
vertices of a quadrangle, respectively, where a +b =n —4,n > 5 and a > b > 0 (sce
Fig. 2).
Lemma 4.8. For n > 5, a> b> 0 and a > 2, q3(U}(a,b)) <2.
Proof. By Lemma 4.6 and direct calculation, we have
(Uj (a,b), ) = (x — 1)? F(x),
where 126, J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-132
(a) = 2° —(n+5)a4 + [(a+5)n—a? — 3047)
+(3n +8)
— [2a +7)n ~ 2a? — 6a+7
Let ry > ay > rg > x4 > ats be the roots of f(«) = 0. Note that
f0)=-4<0, f)=-ab<0, — f(2)=2a+2-2>0,
f(a +2) = -a4 +05 + 3ab — 305 — 5a? + Bab — 5a + 2b-2
<-at +a‘ +3a° — 3a — 5a? + 3a? — 5a + 2a-2
= -20° —3a-2<0
1() = 2p — (4a + 6)n# + (40? + 160 + 6)n4 — (4a? + 12a — 20)n? — 40n + 16]
= Flnin — (2a +3)]? + (4a —3)n® — (4a + 120 — 20)n? — 40n + 16}
= Sw’ [(n — 2a = 3)"n + (4a — 3)n — 4a? — 120 + 9] + Ln? = 40n + 16}
2 {2 [(0 = a)?(a+b-+3) + dab — 3a ~ 3] + 11n? = 40n + 16}
m
Jee [(a = b)?(a + b) + 30? + 36? — 2ab — 3a — 3b] + 11n? — 40n + 16}.
Case 1: a
(2) = Fn ? [(2 —b)?(2 +b) + 30? — 7b + 6] + 11n? — 40n + 16}
= fe [e-»? ery ss(s—Z) +8] suu(n— 2)" 2h
> a {w leo e++a(o-Z) +B] on(s—2)’ -#}
{
af [e-o e+n+a(o-2)" +B) or}
eS
>
Case 2: a > 3,
12)-
afr [(a — 6)?(a +b) + 2a(a ~ b) + a(a — 3) + 36(b— 1)]
20)? 224
+u(»—F) -#} J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-192 war
2 {ir — b)?(a +b) + 2a(a — 6) + a(a — 3) + 30(6 - 1)]
2
20)? 224
* u(s-2) -=t
Combining the above arguments, we have 0 < 15 <2 <1 <1 <2 <2 9, a > b> 0, ry(U2(a.b))
By Lemma 3.4 and U2(a,b) is a bipartite graph, we have
Lemma 4.10. For n > 9,a > b > 0, q3(U2(a,b)) = 2
Lemma 4.11. Let G be a graph with n vertices. If G— {e1,e2} = HU Ka, where e1,¢2 €
E(G), and H = Ui(a,b) for some integers a,b with a+b=n—2—i,a>b>0, and
i=1,2, then S}(G) < e(G) +6.
Proof. By Lemma 4.8 and Lemma 4.10, the first three largest signless Laplacian eigen-
values of H UK» are 41(H),q2(H),2, which implies Sf(H U K2) = S}(H) + 2. Hence,
by Lemma 2.2 and Theorem 4.1,
S$(G) < S$(HU Ke) +28$(Ko) = SF(H) +6
241 5
< eH) + ( > )so= acre. a
For n > 11, define sixteen classes of bicyelic graphs with n vertices, denoted by
UL,U2,...,U!8, for which the structures of graphs in them are given in Pig. 3. For
n> 11, we also define three bicyclic graphs with n vertices, denoted by Ui7, U8, U}S,
see also Fig. 3.
Clearly, the graphs U}7, U8, U}9 are the special graphs in the UZ, U8, U8, respec-
tively.
For # = 0,1,2,3, let Ti, be the tree obtained by attaching i paths with two vertices to
the central vertex of Ky,,—9:-1, where n > 2i+1, see Pig. 4. In particular, T? = Ky
From Lemma 3.4 and Lemma 4.6 in [6], we have
nie
Lemma 4.12. (i) For n > 6, we have 1 < qo(T?) < 2.7,
n> 7, we have 1 < q(T) < 2.7, ST (Ti) < e(TB) + 2.
S} (12) < e(T2) +2. (ii) For 128. J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-132
‘
& @
fi
us ur us
Fig. 2. The structures of graphs in U!,U2,
‘Theorem 4.13. Let n, k be positive integers with n > 1 and 1 3 and 2 3, then let v7 and vg be the neighbor of » and
vz in A, respectively (v7 = vs if |A] = 3) (see Fig. 5). Let G3 be the component of
G — {vy07, v2vs} containing v7. Furthermore, if |A| = 2, we define e(Gs) = 0.
Case 1: e(G1) > 2 or e(G2) > 2 or e(Gs) > 2.
Without loss of generality, we suppose e(G1) > 2. Then G can be considered as
G1 ~ (G—G4) in which the inserted edges are v1 and vavs. Thus by the Inequality (1),
Theorem 4.1 and Lemma 4.5, Sf (G) < e(G) + 6.
Case 2: ¢(G1) < 1 and e(G2) < 1 and e(Gs) <1.
We show the structure of @ in each case in Table 1.
Now we show for any G € UlS, Ui, Sf(G) < e(G) + 6. Without loss of generali
suppose dy, 2 dy,. Note that n > 11, it implies dy, > 5.
Let G’ = G — {v103, e105, 0107} if |Al > 3 and G! = G — {vig 0105, 0109} if [A] = 2.
Let G4 be the component of G’ containing v;, G; be the component of G’ containing
vp. Then G’ = GyUGs and both G4, Gs are trees. Let nj = |V(Gi)|, where i = 4,5
and ny +7s =n. Obviously, Gy = 12, with ny > 3, which implies that q(Gs) = ma,
(Ga) = 1 130 J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-132
Table 1
‘The structure of G in each case.
0 ° 0 UL, UE
0 1 o vi, ut, US, Ue
1 o 0 U8, US, US, Us
1 1 0 ut, ut, ule, ult, Ul?
0 o 1 vue
0 1 1 ut, ue
1 o 1 ut, ue
1 1 1
Uy Unt UNS, Ue
Subease 2.1: G € UL UU2.
Then Gs ~ 72,, then the first three largest signless Laplacian eigenvalues of G’ are
ng,ns,1, that is, Sf(G’) = ng +ns5 +1=n+1. By Lemma 2.2,
S$(G) < S$ (C’) +38} (Ky) =n +146 =e(G) +6
Subcase 2.2: G € Uf_, Ul.
For each graph G, let e1, ¢2 be the edges as labeled in Fig. 2. Then the result follows
from Lemma 4.11.
Subcase 2.3: G € U/2, Ul. Then Gs ~
Subcase 2.3.1: ns = 5.
Then G € {U}?,UJ8,U29}. For each graph G, let e1, eg be the edges as labeled in
Fig. 2. Then the result follows from Lemma 4.11.
Subcase 2.3.2: ns > 6. Then by Lemma 4.12, we have 1 < qa(@s) < 2.7 <3 7. By Lemma 2.2 and (ii) of Lerma 4.12, we can prove
the result similar to the proof of Subcase 2.3.2.
Combining the above arguments, the result holds. O
By Lemma 2.5, Theorem 4.2, Theorem 4.7 and Theorem 4.13, we have
‘Theorem 4.14. Conjecture 1.1 is true for bicyclic graphs. J. Yang, L. You / Linear Algebra and its Applications 446 (2014) 115-192 131
5. Some remarks and problems
Note that a tree is an acyclic graph, and the upper bound given in Conjecture 1.1 is
not tight for trees since Inequality (1) holds. It is nature to consider whether the upper
bound given in Conjecture 1.1 is tight or not for all graphs. In this section, we discuss
the question for cyclic graphs, and propose some problems for further research.
Lemma 5.1. (See [7].) Let @ be a connected graph with n vertices and e(@) edges. Then
n(@) < 22 4 n—2, with equality if and only if @ is Kin—1 or Ky.
Proposition 5.2. Let n > 4, G be ac
S#(@) 4, we have
An+e— 2
2+e“V Ly ging MmteVoREERT—BIIO, SHC) < o(G) + (M3).
From Propositions 5.2 and 5.3, we immediately have the following remark,
Remark 5.4. Let n, k be positive integers with n > d and 1
