# A new upper bound on the largest normalized Laplacian eigenvalue

Description
A new upper bound on the largest normalized Laplacian eigenvalue
Categories
Published

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
P   R   E   P   R   I   N   T OperatorsandMatrices Preprint A NEW UPPER BOUND ON THE LARGESTNORMALIZED LAPLACIAN EIGENVALUE O SCAR  R OJO AND  R ICARDO  L. S OTO( Communicated by Richard A. Brualdi )  Abstract.  Let  G   be a simple undirected connected graph on  n  vertices. Suppose that the verticesof   G   are labelled 1 , 2 ,... , n .  Let  d  i  be the degree of the vertex  i .  The Randi´c matrix of   G   ,denoted by  R ,  is the  n × n  matrix whose  ( i ,  j ) − entry is  1 √  d  i d   j if the vertices  i  and  j  areadjacent and 0 otherwise. The normalized Laplacian matrix of   G   is  L    =  I  −  R ,  where  I   is the n × n  identity matrix. In this paper, by using an upper bound on the maximum modulus of thesubdominant Randi´c eigenvalues of   G   , we obtain an upper bound on the largest eigenvalue of  L   .  We also obtain an upper bound on the largest modulus of the negative Randi´c eigenvaluesand, from this bound, we improve the previous upper bound on the largest eigenvalue of   L   . 1. Introduction Let  G   = ( V  ,  E  )  be a simple undirected graph on  n  vertices. Some matrices on  G  are the adjacency matrix  A ,  the Laplacian matrix  L  =  D −  A  and the signless Laplacianmatrix  Q  =  D +  L ,  where  D  is the diagonal matrix of vertex degrees. It is well knownthat  L  and  Q  are positive semideﬁnite matrices and that  ( 0 , 1 )  is an eigenpair of   L where  1  is the all ones vector. Fiedler  proved that  G   is a connected graph if andonly if the second smallest eigenvalue of   L  is positive. This eigenvalue is called thealgebraic connectivity of   G   . The signless Laplacian matrix has recently attracted theattention of several researchers. Recent papers on this matrix are [5, 6, 7, 8, 9] andsome of its basic properties  are:1. For a connected graph, the smallest eigenvalue of   Q  is equal to 0 if and onlyif the graph is bipartite. In this case, 0 is a simple eigenvalue. Then, for aconnected graph, the smallest eigenvalue of   Q  is positive if and only if the graphis not bipartite.2. If   G   is a bipartite graph then  Q  and  L  have the same characteristic polynomial.  Mathematics subject classiﬁcation  (2010): 05C50, 15A48. Keywords and phrases : normalized Laplacian matrix, Randi´c matrix, upper bound, largest eigenvalue,subdominant eigenvalue. This research was supported by Project Fondecyt 1100072, Chile. It was ﬁnished when the ﬁrst author was a visitorat the Department of Mathematics, Universidade de Aveiro, Aveiro, Portugal.c      , ZagrebPaper OaM-0594 1     P   R   E   P   R   I   N   T 2  O. R OJO AND  R. L. S OTO Othermatricesonthegraph  G   arethenormalizedLaplacianmatrixandtheRandi´cmatrixof   G  .  Supposethattheverticesof   G   arelabelled 1 , 2 ,..., n .  Let  d  i  bethedegreeof the vertex  i .  Let  D − 12  be the diagonal matrix whose diagonal entries are1 √  d  1 , 1 √  d  2 ,... , 1 √  d  n whenever  d  i  = 0 .  If   d  i  = 0 for some  i  then the correspondingdiagonalentry of   D − 1   2  isdeﬁnedtobe 0 .  ThenormalizedLaplacianmatrixof   G   , denotedby  L    , was introducedby F. Chung  as L    =  D − 12  LD − 12 =  I  −  D − 12  AD − 12 .  (1)The eigenvalues of   L    are called the normalized Laplacian eigenvalues of   G  . From  ( 1 ) , we have  D 12 L    D 12 =  D −  A =  L and thus  D 12 L    D 12 1  =  L 1  =  0 1 . Hence 0 is an eigenvalue of   L    with eigenvector  D 12 1 . We recall the following results on  L     :1. The eigenvalues of   L    lie in the interval  [ 0 , 2 ] . 2. 0 is a simple eigenvalue of   L    if and only if   G   is connected.3. 2 is an eigenvalue of   L    if and only if a connected component of   G   is bipartiteand nontrivial.Among papers on  L    , we mention [10, 11, 13, 14] and .From now on, we assume that  G   is connected graph. Then  d  i  >  0 for all  i . Thenotation  i ∼  j  means that the vertices  i  and  j  are adjacent. The matrix  R =  D − 12  AD − 12 in  ( 1 )  is the Randi´c matrix of   G   in which the  ( i ,  j ) -entry is  1 √  d  i d   j if   i  ∼  j  and 0otherwise. Moreover  I  − L    =  R .The eigenvalues of   R  are called the Randi´c eigenvalues of   G   . Clearly  L    and  R  arebothreal symmetricmatrices. The Randi´cmatrixwas earlierstudied in connectionwiththe Randi´c index [1, 2, 18] and . Two recent papers on the Randi´c matrix are and .Throughout this paper0  =  λ  n  λ  n − 1  ...  λ  1 and ρ n  ρ n − 1  ...  ρ 1     P   R   E   P   R   I   N   T B OUND ON THE LARGEST NORMALIZED  L APLACIAN EIGENVALUE  3are the normalized Laplacian eigenvalues and the Randi´c eigenvalues of   G   , respec-tively. It follows that λ  i  =  1 − ρ n − i + 1  ( 1  i  n ) . If   M   is a nonnegative matrix then, by the Perron-Frobenius Theorem,  M   hasan 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 corre-sponding positive eigenvector, called the Perron vector of   M  .  Since  G   is a connectedgraph, Randi´c matrix of   G   is a irreducible nonnegative matrix. Let  v  =  D 12 1 .  Then v  =  √  d  1 , √  d  2 ,... , √  d  n  T  .  An easy computation shows that  R v  = v . Hence, 1 and  v  are the Perron root and the Perron vector of   R ,  respectively.Let  ∆  and  δ   be the largest and smallest vertex degrees of   G  ,  respectively, and let q n  be the smallest eigenvalue of   Q .A recent result involving the largest eigenvalue of   L    and the smallest eigenvalueof   Q  isT HEOREM  1.   Let   G   be a connected graph. Then 2 − q n δ    λ  1  2 − q n ∆  .  (2)We may consider 2 − q n ∆  as an upper bound on  λ  1 .  Observe that 2 − q n ∆  = 2 if andonly if   G   is a bipartite graph.In this paper, we search for a new upper bound on  λ  1  not exceeding the trivialupper bound 2 . 2. Searching for an upper bound on  λ  1 Since  ∑ ni = 1 ρ i  = tr  (  R ) =  0 ,  it follows that  ρ n  <  0 .  We have λ  1  =  1 − ρ n  =  1 + | ρ n | . In order to ﬁnd an upper bound on  λ  1  not exceeding 2 ,  we look for an upper bound on | ρ n |  not exceeding 1 . An eigenvalue of a nonnegative matrix  M   which is different from the Perron rootis called a subdominant eigenvalue of   M  .  Let  ξ  (  M  )  be the maximum modulus of the subdominant eigenvalues of   M  .  Special attention has been devoted to ﬁnd upperbounds on  ξ  (  M  ) . In , we can ﬁnd a uniﬁed presentation of results concerningupper bounds on  ξ  (  M  ) .  These upper bounds are important because  ξ  (  M  )  plays amajor role in convergence properties of powers of   M  .  Since λ  1  1 + ξ (  R ) ,  (3)we focus our attention on upper bounds on  ξ  (  R ) .  We recall the result [12, p. 295] :     P   R   E   P   R   I   N   T 4  O. R OJO AND  R. L. S OTO T HEOREM  2.  If M   = ( m i ,  j )  0  of order n × n has a positive eigenvector  w  = [ w 1 , w 2 ,... , w n ] T  corresponding to the spectral radius  ρ (  M  )  of M then ξ  (  M  )   12 max i <  jn ∑ k  = 1 w k   m i , k  w i − m  j , k  w  j  . where the maximum is taken over all pairs  ( i ,  j )  ,  1  i  <  j  n . In order to apply Theorem 2, it is convenient to observe that the Randi´c matrix of  G   is diagonally similar to the row stochastic matrix S   =  D − 12  RD 12 .  (4)The following lemma gives some immediate properties of   S  . L EMMA  1.  1. The  ( i ,  j ) − entry of S is  1 d  i if j ∼ i and   0  otherwise.2. S  1 = 1  where  1  is the all ones vector.3.  u  is an eigenvector for R corresponding to the eigenvalue  α   if and only if  D − 12 u  is an eigenvector for S corresponding to the eigenvalue  α  . 4. If   G   is an r  − regular graph then S   =  R . Let  N  i  be the set of neighbours of the vertex  v i  and let  |  N  i |  be the cardinality of   N  i . T HEOREM  3.  Let   G   be a simple undirected connected graph. If   λ  1  is the largest eigenvalue of   L    then | λ  1 |  2 − min i <  j    N  i ∩  N   j  max  d  i , d   j   (5) where the minimum is taken over all pairs  ( i ,  j )  ,  1  i  <  j  n . Proof.  We knowthattheRandi´cmatrixof   G   is similartotherowstochasticmatrix S   deﬁned in  ( 4 ) .  Then  ξ  (  R ) =  ξ  ( S  ) .  The eigenvector corresponding to the spectralof   S   is  w = 1 . Applying Theorem 2 to  S   = ( s i ,  j ) ,  we have ξ  ( S  )   12 max i <  jn ∑ k  = 1  s i , k  − s  j , k   =  12 max i <  j   ∑ k  ∈  N  i −  N   j 1 d  i +  ∑ k  ∈  N   j −  N  i 1 d   j +  ∑ k  ∈  N  i ∩  N   j  1 d  i −  1 d   j  =  12 max i <  j   N  i −  N   j  d  i +   N   j −  N  i  d   j +  ∑ k  ∈  N  i ∩  N   j  1 d  i −  1 d   j  =  12 max i <  j  2 −   N  i ∩  N   j  d  i −   N   j ∩  N  i  d   j +  ∑ k  ∈  N  i ∩  N   j  1 d  i −  1 d   j  .     P   R   E   P   R   I   N   T B OUND ON THE LARGEST NORMALIZED  L APLACIAN EIGENVALUE  5Suppose  d  i  =  max  d  i , d   j  .  In this case2 −   N  i ∩  N   j  d  i −   N   j ∩  N  i  d   j +  ∑ k  ∈  N  i ∩  N   j  1 d  i −  1 d   j  =  2 −   N  i ∩  N   j  d  i −   N   j ∩  N  i  d   j +   1 d   j −  1 d  i   N  i ∩  N   j =  2 − 2   N  i ∩  N   j  d  i . Similarly, if   d   j  =  max  d  i , d   j   then2 −   N  i ∩  N   j  d  i −   N   j ∩  N  i  d   j +  ∑ k  ∈  N  i ∩  N   j  1 d  i −  1 d   j  =  2 − 2   N   j ∩  N  i  d   j . Hence ξ  ( S  )   12 max i <  jn ∑ k  = 1  s i , k  − s  j , k   =  12 max i <  j  2 −  2   N   j ∩  N  i  max  d  i , d   j  =  1 − min i <  j    N  i ∩  N   j  max  d  i , d   j  Since  λ  1  1 + ξ (  R ) =  1 + ξ ( S  ) , the upper bound in  ( 5 )  follows.   R EMARK  1. If   G   is a bipartite graph then   N  i ∩  N   j  =  0 ,  for some  i  <  j ,  andconsequently the upper bound in  ( 5 )  is equal to 2 .  This is sufﬁcient condition but it isnot a necessary condition. In fact, there are other instances in which  N  i ∩  N   j  =  0 forsome  i  <  j .  One of them is given by a nonbipartite graph having a bridge. However, if min i <  j   N  i ∩  N   j   1 and  q n  <  1 then2 − min i <  j    N  i ∩  N   j  max  d  i , d   j  <  2 − q n ∆  .  (6)In fact q n  <  1    N  i ∩  N   j   for  i  <  j and q n ∆    1max  d  i , d   j   for  i  <  j .

Jan 10, 2019

#### THE REAL PURPOSE OF THOSE WHO DENY THE SCHOOL OF THOUGHTS, INTERPRETERS AND THE COMPANIONS

Jan 10, 2019
Search
Similar documents

### A new lower bound below the information rate of Wiener phase noise channel based on Kalman carrier recovery

View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us. Thanks to everyone for your continued support.

No, Thanks