• 検索結果がありません。

BERMAN† AND K.-H

N/A
N/A
Protected

Academic year: 2022

シェア "BERMAN† AND K.-H"

Copied!
12
0
0

読み込み中.... (全文を見る)

全文

(1)

ALGEBRAICCONNECTIVITY OFTREES WITHA PENDANT

EDGE OFINFINITE WEIGHT

A. BERMAN

AND K.-H. FÖRSTER

Abstract. Let

G

beaweighted graph. Let

v

bea vertexof

G

and let

G v ω

denote the graph

obtainedbyaddingavertex

u

andanedge

{v, u}

withweight

ω

to

G

.Thenthealgebraicconnectivity

µ(G v ω )

of

G v ω

isanondecreasingfunctionof

ω

andisboundedbythealgebraicconnectivity

µ(G)

of

G

.Thequestionofwhen

ω→∞ lim µ(G v ω )

isequalto

µ(G)

isconsideredandansweredinthecasethat

G

isatree.

Keywords. Weightedgraphs,Trees,Laplacianmatrix,Algebraicconnectivity,Pendantedge.

AMSsubjectclassications. 5C50,5C40,15A18.

1. Introduction. Aweightedgraphon

n

vertices isanundirectedsimplegraph

G

on

n

verticessuchthatwitheachedge

e

of

G

,thereisanassociatedpositivenumber

ω(e)

whichiscalledtheweight of

e

.

TheLaplacian matrix of aweighted graph

G

on

n

vertices isthe

n × n

matrix

L(G) = L = (l ij )

,whereforeach

i, j = 1, . . . , n,

l ij =

 

 

 

 

−ω(e)

if

i = j

and

e = {i, j},

0

if

i = j

and

i

isnotadjacentto

j,

k=i l ik

if

i = j.

Clearly

L

is asingular

M

-matrixandpositivesemidenite, so

λ 1 (L) = 0

,where

forasymmetricmatrix

A

wearrangetheeigenvaluesinnondecreasingorder

λ 1 (A) λ 2 (A) . . .

Fiedler [3] showed that

λ 2 (L)

is positive i

G

is connected and called it the

algebraicconnectivity of

G

. Thealgebraicconnectivityof

G

willbedenotedby

µ(G).

Inthispaper

G

alwaysdenotesaconnectedweightedgraphwithoutloops.

Let

G

beagraphwith

n

vertices. Let

v

beavertexof

G

andlet

G v ω

bethegraph

with

n + 1

verticesobtainedbyadding to

G

avertex

u

andanedge

e = {v, u}

with

weight

ω

.

Receivedbytheeditors21July2003.Acceptedforpublication24June2005. HandlingEditor:

StephenJ.Kirkland.

Department of Mathematics, Technion, Haifa 32000, Israel,([email protected]).

Research supportedbythe New York MetropolitanFund for Research at the Technion. Mostof

theworkwasdoneduringavisittoTUBerlin;partoftheworkwasdoneinFUBerlin,Charite-

BenjaminFranklin.

DepartmentofMathematics, Technical UniversityBerlin,Sekr. MA6-4,Strassedes17. Juni 136,10623Berlin,Germany([email protected]).

(2)

Theorem 1.1. The algebraic connectivity

µ(G v ω )

isa nondecreasing function of

ω

andfor every

ω

and

n > 1

µ(G v ω ) µ(G).

Proof. Let

L ω

be the Laplacian matrix of

G v ω

and let

0 < ω 1 ω 2

. Then

B = L ω 2 L ω 1

isasingularrankonepositivesemidenitematrix. By[7,Th. 4.3.1]

λ k ( L ω 1 ) λ k ( L ω 2 )

for

k = 1 . . . n,

andfor

k = 2, µ(G v ω 1 ) µ(G v ω 2 ).

Toshowthat

µ(G v ω )

isbounded, write

L ω

asthesumoftwoblockmatrices

L ω =

 

L(G) 0

0 0

  +

 

0 0

0 ω ω

ω ω

  ,

where

L(G)

is

n×n

andtheleftupperzeroblockinthesecondmatrixis

(n−1)×(n−1)

.

By[7,Th. 4.3.4(a),thecase

k = 2

],

µ ( G v ω ) = λ 2 ( L ω ) λ 3 ( L ( G ) (0))

= λ 2 (L(G)) = µ(G).

Remark 1.2. Thetheorem is essentiallyaconsequenceof Cor. 4.2 of[6]. It is

provedfortreesin [8].

Example 1.3. Forthecompletegraphs

K n , n > 1,

withallweightsequalto 1

ω→∞ lim µ((K n ) v ω ) = n + 1

2 < n = µ(K n ).

Example 1.4. Forthecycles

C n , n > 2 ,

withweightsequalto1

ω→∞ lim µ((C n ) v ω ) = µ(C n+1 ) < µ(C n ).

Example 1.5. Let

G

be thegraphobtainedfrom

K 4

bydeleting an edge and

letalltheweightsof

G

beequalto1. Ifthedegreeof

v

is3,

ω→∞ lim µ(G v ω ) = 2 = µ(G).

Ifthedegreeof

v

is2

ω→∞ lim µ(G v ω ) < µ(G).

Since

µ(G v ω )

isboundedby

µ(G)

,itisnaturaltoaskwhendoes

ω→∞ lim µ ( G v ω ) = µ ( G ) .

We answer this question in Section 3, in the case that

G

is a tree. The needed

(3)

2. Results on trees. Our paper relies heavily on the work of [12] so in this

sectionwedescribetheirmainresultsandbasicbackgroundontreesneededforthese

resultsandforthenextsection. Insomecaseswechangethenotationof[12].

Theorem 2.1. [4,Th. 3.11]Let

T

be a weighted tree with Laplacian matrix

L

and algebraic connectivity

µ

. Let

y

be an eigenvector of

L

associated with

µ

. Then

exactly oneofthe followingtwocasesoccur:

(a) Someentryof

y

is

0

.

(b) Allentries of

y

arenonzero.

In the rst casethere exists aunique vertex

c

suchthat

y c = 0

and

c

isadjacent to

avertex

d

with

y d = 0

. In the secondcasethereis aunique pair of vertices

i

and

j

adjacent in

T

suchthat

y i y j < 0

.

Definition 2.2. Aweightedtree

T

is saidtobeoftypeI withacharacteristic vertex

c

ifcase

(a)

ofTheorem2.1holds,andoftypeII withcharacteristicvertices

i

and

j

incase

(b)

. Weusealsothenotation

I c

in therstcaseand

II i,j

inthesecond

case.

Thenamecharacteristicverticeswascoinedin[11]byR.Merriswhoshowedthat

if

µ

isnotasimpleeigenvalue,thenallthecorrespondingeigenvectorsyieldthesame typeoftreeandthesamecharacteristicvertices.

Definition 2.3. Let

v

beavertex ofatree

T

. Let

L v

bethematrixobtained

bydeleting therowandcolumn of theLaplacianmatrixof

T

that correspond to

v

.

Thematrix

M v,T := L −1 v

iscalled thebottleneckmatrixof

T

at

v

.

In[9]and[10],itisshownthattheentryof

M v,T

thatcorrespondstothevertices

k

and

l

is

m kl = 1 ω(g)

wherethesummationisonalledges

g

thatlieontheintersectionofthepathbetween

k

and

v

and thepathbetween

l

and

v

. Thematrix

M v,T

is permutationallysimilar to ablock diagonalmatrix, wherethe numberofblocksis thedegreeof

v

and each

blockisapositivematrixwhichcorrespondsto auniquebranchat

v

.

Forvertices

u, v

ofatree

T

let

v u

denotethebranchof

T

at

v

,thatcontains

u

.

Wedenoteby

M v→u,T

theblockof

M v,T

thatcorrespondsto

v u

,andby

M v→u,T

thematrixobtainedfrom

M v,T

bydeleting therowsandthecolumns corresponding to

M v→u,T

.

Definition 2.4. A diagonal block of

M v,T

whose spectral radius is equal to

ρ(M v,T )

, where

ρ(A)

denotes thespectralradiusof thematrix

A

, iscalled aPerron

blockandthecorrespondingbranchof

T

at

v

iscalled aPerronbranch.

Theorem2.5. [9,Cor. 2.1]Let

T

beaweightedtree. Then

T

isoftypeIwitha

characteristic vertex

c

,if andonlyif at

c

,

T

has morethanone Perron branch.

Inthis case,

µ(T )

,the algebraic connectivity of

T

isequal to

ρ(M 1

c,T )

.

Let

e

be an edge of a graph

G

. Replace the weightat

e

by

ω

and denote the

resultinggraphby

G e ω

. Observethat since

e = {v, u}

isapendantedgeof

G v ω

,then

(G v ω ) e ω = G v ω

. Let

G e

denotethefamilyofweightedgraphs

{G e ω , ω > 0}

,andlet

G v

denotethefamilyofweightedgraphs

{G v ω , ω > 0}

.

(4)

Theorem 2.6. [12,Corollary1.1] Let

T

bea weighted treeandlet

e

be anedge

of

T

. Then thereexistsapositive number

ω 0

suchthatallthe trees

T ω e , ω 0 < ω <

,

areof the sametypeandhave the samecharacteristicvertices.

Thefollowingdenitionsareusedin [12].

Definition 2.7. Thefamilyoftrees

T e

isatypeItreeatinnitywithcharac-

teristicvertex

c

if thereexistsan

ω 0 > 0

suchthat forall

ω 0 , ∞)

,

T ω e

isoftype

I c

. Similarly,

T e

is a typeII tree at innity with characteristic vertices

i

and

j

if

thereexists an

ω 0 > 0

suchthat forall

ω [ω 0 , ∞)

,

T ω e

is oftype

II i,j

.

Wenowcanstatethemainresultof[12].

Theorem 2.8. [12,Th.1.8]Let

e = {v, u}

be anedge thatis notapendantedge

of a tree

T

. Let

T 1

and

T 2

be the resulting components arising from the deletion of

e

. Suppose

v T 1

,

u T 2

and

µ ( T 1 ) µ ( T 2 )

. Then

lim

ω→∞ µ ( T ω e ) = µ ( T 1 )

i

T 1

isa

treeoftypeIwith acharacteristicvertex, say,

c

,and oneof thefollowing conditions

holds:

(a)

T e

isoftype Iwithacharacteristicvertex

c

.

(b)

c

isincident to

e

and

ρ(M u,T 2 ) µ(T 1 1 )

.

Weconcludethe backgroundresultswiththe analogueofTheorem 2.5 fortype

IItreesandtwopropositionsthatwillbeusedinprovingthemainresult.

Theorem 2.9. [9,Th.1]Aweightedtree

T

isof typeIIiateveryvertex

T

has

auniquePerronbranch. Ifthe characteristicvertices,

i

and

j

,of

T

arejoinedby an

edge ofweight

θ

,thenthereexistsanumber

0 < γ < 1

,suchthat

ρ(M i→j,T γ

θ J ) = ρ(M j→i,T 1 γ θ J ), and µ ( T ) = 1

ρ(M i→j,T γ θ J) = 1

ρ(M j→i,T 1−γ θ J) ,

whereJdenotes an allonesmatrix.

Proposition2.10. [8,Cor. 1.1]Thecharacteristicverticesof

T ω v

lieonthe path

between thecharacteristicverticesof Tand

u

.

Proposition 2.11. [12, Claim 3.2] Let Tbea tree. Let

{ i k , j k }

beedges in T

withweights

α k

,for

k = 1 , 2

,suchthatthepathfrom

i 1

to

j 2

contains

j 1

and

i 2

,and

let

0 < γ 1 , γ 2 < 1

. Then

ρ(M j 1 →i 1 ,T γ 1

α 1 J ) < ρ(M j 1 →i 1 ,T ) < ρ(M j 2 →i 2 ,T γ 2 α 2 J ).

3. Assigninganarbitrarilylargeweighttoapendantedgeofatree. In

thissectionweconsiderthecasewhere

T

isatreeand

u

isapendantvertexof

T e

where

e = { v, u }

and

v T

.

In some sense the question in this case may be considered asa special case of

thediscussion in [12]. Todo this,

{u}

isto beconsidered asa "treewith algebraic

connectivity

"and the spectralradius of an empty matrix hasto be dened (for

(5)

Ourdiscussionisbasedontheanalysisofthelimitsofthebottleneckmatricesof

T ω v

when

ω

increasesto

;namely

M v,T v

ω = M v,T 1

ω

,

M u,T v

ω = (M v,T (0)) + 1 ω J

andif

s = v

isavertexof

T

,

M s,T ω v =

M s,T M (v) M (v)t m vv + ω 1

,

where

M (v)

isthecolumnof

M s,T

correspondingto

v

and

m vv

isthediagonalentry

of

M s,T

, corresponding to

v

. In particular, for all the branches of

T

at

s

that do

not contain

v

, the diagonal blocks of

M s,T v

ω

and of

M s,T

are the same. Denoting

ω→∞ lim M s,T v

ω

by

M s,T v

weseethatfor

s / e M s,T v

=

M s,T M (v) M (v)t m vv

(3.1)

and

M v,T v

= M u,T v

= M v,T (0) .

(3.2)

Thereadershouldnotbeconfusedbythefactthat

T v

denotesafamilyoftrees

while

M s,T v

denotesasinglematrix(upto permutationsimilarity).

Example3.1.

1/2 a

@ @

@ @

@ @

@

c 1 v ω u

1/2

b

~ ~

~ ~

~ ~

~

M v,T v

= M u,T v

=

 

3 1 1 0 1 3 1 0 1 1 1 0 0 0 0 0

 

,

M c,T v =

 

2 0 0 0 0 2 0 0 0 0 1 1 0 0 1 1

 

,

(6)

M a,T v = M b,T v =

 

4 2 2 2 2 2 2 2 2 2 3 3 2 2 3 3

 

.

Remark 3.2. Thematrices

M s,T v

are ofcourse singular, but theydo contain

informationon

lim ω→∞ µ(T ω v )

.

As in thecaseofnonsingularbottleneck matriceswe callthe diagonalblocks of

M s,T v

whosespectralradiusismaximal,Perronblocks. Thecorrespondingbranches of

T ω v

donotdependon

ω

;theywillbecalledthePerronbranchesofthefamily

T v

.

Lemma 3.3. If

M s,T v

hasmorethanonePerronblock, then

ω→∞ lim µ(T ω v ) = 1 ρ(M s,T v ) .

Proof. Considertheprincipal submatrix

L s,T v

ω

obtainedfromtheLaplacianma-

trixof

T ω v

bydeleting therowandcolumn correspondingto

s

. Then

M s,T v

= lim

ω→∞ (L s,T ω v ) −1 .

By[7,Th. 4.3.15]for

r = n 1, k = 1

and

k = 2

,

λ 1 (L s,T ω v ) µ(T ω v ) λ 2 (L s,T ω v ),

so

ω→∞ lim λ 1 (L s,T ω v ) lim

ω→∞ µ(T ω v ) lim

ω→∞ λ 2 (L s,T ω v ),

Since

M s,T v

hasatleasttwoPerronblocksweobtain

ω→∞ lim λ 2 (L s,T ω v ) = lim

ω→∞ λ 1 (L s,T ω v ) = ρ(M s,T v ).

Remark 3.4. Ifthere existsan

ω 0

such thattwoofthePerronblocksof

M s,T v

arePerronblocksof

M s,T ω v

for

ω ω 0

then

λ 2 (L s,T ω v ) = λ 1 (L s,T ω v )

for

ω ω 0 .

Lemma 3.5. Let

s

be a vertex of

T

. Suppose

M s,T v

has at least two Perron

blocks andlet

t

beanother vertexof

T

. Then

ρ(M t,T v ) > ρ(M s,T v )

(7)

Proof. Byassumption, thefamily

T v

has atleast twoPerron branchesat

s

, so

oneofthem,say

s x

,doesnotcontain

t

. Let

t s

bethebranchat

t

thatcontains

s

. Thenitcontainsthebranch

s x

,andweobtain

ρ(M t,T v ) ρ(M t→s,T v ) > ρ(M s→x,T v ) = ρ(M s,T v ),

wherethestrictinequalityfollowsfrom [1, Cor. 2.1.5]andthefact that

M s→x,T v

is

asubmatrixof

M t→s,T v

,whichispositive.

Corollary3.6. Thereisatmost onevertex, say

c

,suchthat

M c,T v

has more

thanonePerronblock.

Definition 3.7. Inthecasethatthere isavertex

c

suchthat

M c,T v

hasmore

thanonePerronblock,wewillsaythatthefamilyoftrees

T v

isatrii(treeininnity)

oftypeIc. Ifnosuch

c

existswesaythat

T v

isatriioftypeII.

Remark 3.8.

(a)If thetrees

T ω v

are oftypeI

c

for allsucientlylarge

ω

, then thefamily

T v

isatriioftypeI

c

(andalsoatypeI treeat innitywithcharacteristicvertex

c

). In

otherwords,if

T v

isatriioftypeII,thenforall

ω

largeenough,

T ω v

aretreesoftype

II.

(b)Supposethetrees

T ω v

areoftypeII

p,q

forallsucientlylarge

ω

,thenbythe

representationof

L ω

intheproofofTheorem 1.1andbyTheorem2.1,

{ p, q }

cannot

bethependantedge

{v, u}

.

(c) It is possible that

T ω v

are of type II for all suciently large

ω

(so

T v

is a

typeII treeat innity) but

T ω v

is atriiof typeI;see Lemma3.10and Subcase 4of

Example3.13in thefollowingdiscussion.

Remark3.9. TheproofofLemma 3.5showsthatif

T

isatreeoftypeIwitha

characteristicvertex

c

, thenforanyothervertex

s

of

T ρ ( M s,T ) > ρ ( M c,T ) .

(ThishasalreadybeenestablishedinProposition2of[9].)

ConsiderTheorem2.9 where

T ω v

isoftypeII

i,j

andtheweightoftheedge

{i, j}

is

θ

. Thenforevery

ω

(sucientlylarge)there existanumber

γ ω

,between0and1,

suchthat

µ(T ω v ) = 1

ρ(M i→j,T ω v γ θ ω J ) = 1 ρ ( M j→i,T v

ω 1−γ θ ω J ) .

Whathappenstothethenumber

γ ω

when

ω

goesto

? Weclaimthat

lim

ω→∞ γ ω

exists. Indeed,oneofthebranchescorrespondingto

M i→j,T v

ω

and

M j→i,T v

ω

doesnot

contain

u

. Suppose it is the second, so

M j→i,T v

ω = M j→i,T

. The numbers

µ ( T ω v )

increasetoalimit,seeTheorem1.1,sothenumbers

ρ(M j→i,T ω v 1−γ θ ω J)

decreaseto

alimit,whichmeansthatthenumbers1-

γ ω

increasetoalimit. Thislimitisatmost

1since

0 < γ ω < 1

.

Lemma 3.10. If the trees

T ω v

are of typeII, with characteristic vertices

i, j

for

ω 0 < ω <

,andif

γ = lim

ω→∞ γ ω = 0

,where

ρ(M i→j,T ω v γ θ ω J ) = ρ(M j→i,T ω v 1−γ θ ω J )

(8)

and

θ

isthe weightof theedge

{i, j}

,then

T v

isatrii oftypeI

i

. Similarly,if

γ = 1

then

T v

isatrii oftype I

j

.

Proof.

M j→i,T v

= (M ij,T v (0)) + 1 θ J

so if

γ = 0

, then

ρ(M i→j,T v ) = ρ(M j→i,T v 1 θ J ) = ρ(M ij,T v )

so

T e

isatriioftypeI

i

.

Corollary 3.11. If the trees

T ω v

are of type II and if

T v

isa trii of type II,

then

0 < γ = lim

ω→∞ γ ω < 1

.

Remark 3.12. Thetree

T

can beatree oftypeI with acharacteristicvertex, say

c

,or atreeoftypeII.Intherstcasethereare3possibilities:

1 T v

is atriioftypeI

c

,

2 T v

is atriioftypeI

s

, where

s = c

,

3 T v

is atriioftypeII.

Inthesecond casethere aretwopossibilities:

4 T v

is atriioftypeI

s

forsome

s

,

5 T v

is atriioftypeII.

Thefollowingexampledemonstratesthatallvesubcasesarepossible.

Example 3.13.

Subcase1

x @ @

@ @

@ @

@

c 1 v ω u

x

~ ~

~ ~

~ ~

~ 0 < x 1 2 .

Here

M c,T ω v =

 

1/x 0 0 0

0 1/x 0 0

0 0 1 1

0 0 1 1 + ω 1

 

,

sofor

x < 1 2 , T v

isatreeoftypeI withacharacteristicvertex

c

andfor

x = 2, 1

itis

onlyatriioftype

I c

.

Anotherexampleiswhen

c = v

x

@ @

@ @

@ @

@

c ω u

x

~ ~

~ ~

~ ~

~

.

(9)

Subcase2

x

@ @

@ @

@ @

@

c 10 s 1 v ω u

x

~ ~

~ ~

~ ~

~

where

ρ

 1 /x + 0 . 1 0 . 1 0 . 1 0.1 1/x + 0.1 0.1

0.1 0.1 0.1

 = 2 .

Subcase3

x

@ @

@ @

@ @

@

c 1 v ω u

x

~ ~

~ ~

~ ~

~ 1

2 < x 1,

or

1 c 1 v ω ◦.

Subcase4

1 x s 1 v ω ◦, ρ

1 + 1/x 1/x 1/x 1/x

= 2.

Subcase5

Herewesuggest3examples:

x @ @

@ @

@ @

@

1 ω

x

~ ~

~ ~

~ ~

~

x > 1,

x 1 ω

x / ∈ {1/2, 1},

(10)

1

@ @

@ @

@ @

@

v ω u

x

~ ~

~ ~

~ ~

~

x = 1 .

Wearenowreadytostateandprovethemainresult.

Theorem 3.14. Let

T

beatree. Then

ω→∞ lim µ(T ω v ) = µ(T )

(3.3)

if andonlyif

(a)

T

isatreeof typeIwith characteristic vertexsay

c

,

and

(b)

ρ(M c→u,T v ) ρ(M c,T ) = µ(T 1 )

.

Proof. We provethetheorem byconsidering the vesubcases of Remark 3.12,

andshowingthat(3), (a)and(b)holdinSubcase1andonlyinthiscase,i.e. ifand

onlyif

T

and

T v

areoftypeI

c

forsomevertex

c

.

Subcase 1: Obviously (a) holds. From (1) and (2) follows that

ρ(M c,T ω v ) ρ(M c,T )

. Butif

T

and

T ω v

areoftype

I c

,thenequalityholds. Thus

ρ(M c→u,T v ) ρ(M c,T ),

proving(b),and

µ(T ) = 1

ρ ( M c,T ) = 1 ρ ( M c,T v

ω ) = lim

ω→∞ µ(T ω v ),

proving(3). ThiscompletestheproofinSubcase1.

If

T

is atreeof type

I c

and(b) holds, then itfollowseasily that

T ω v

is atrii of

typeI

c

. Therefore(b) doesnot hold in Subcases 2and 3,while (a) obviouslydoes

nothold in Subcases4 and5. Nowwewill provethat (3) doesnothold in thelast

foursubcases.

Subcase2: Wehaveto showthat(3)doesnothold. Indeed

µ(T ) = 1

ρ(M c,T ) > 1 ρ(M s,T )

= 1

ρ(M s,T ω v )

,byLemma3.5

= lim

ω→∞ µ(T ω v )

, byLemma 3.3.

Subcase3: ByRemark 3.8(a)thetrees

T ω v

areforsucientlylarge

ω

oftype

II

,

sayoftype

II p,q

,seeTheorem2.6,andtheedge

{p, q}

of

T

hasweight

θ

,byRemark

(11)

3.8(b)itdoesnotdependon

ω

. Withoutlossofgenerality,

p

liesonthepathbetween

q

and

c

.

ByProposition2.10thevertices

p

and

q

lie onthepathbetween

c

and

u

. Let

i

beaneighborof

c

suchthat

c, p

and

q

lieonthepathbetween

i

and

u

and

c i

isa

Perronbranchof

T

. Thenweobtain

ω→∞ lim µ(T ω v ) = lim

ω→∞

1

ρ(M q→p,T ω v γ θ ω J )

,byTheorem2.9,

= lim

ω→∞

1

ρ(M q→p,T γ θ ω J )

,since

q p

isin

T

,

= 1

ρ(M q→p,T γ θ J)

,where

0 < γ < 1

,byCorollary3.11,

< 1

ρ(M c→i,T )

,byProposition2.11,

= 1

ρ ( M c,T )

, since

c i

isaPerronbranchof

T,

= µ(T )

,since

T

isoftype

I c ,

so(3)doesnothold.

Subcase4: Here againwehaveto showthat (3)doesnothold. Suppose

T

is of

type

II ij

, where

j

lies on thepathfrom

i

to

u

. Let

θ

and

γ

beasin Theorem 2.9.

Since

T ω v

is atriioftype

I s

, thebyProposition 2.10,

s

lies onthepathfrom

i

to

u

.

Therefore

ω→∞ lim µ(T ω v ) = lim

ω→∞

1

ρ(M s→i,T ω v ) = lim

ω→∞

1 ρ(M s→i,T )

1

ρ(M j→i,T ) < 1

ρ(M j→i,T γ θ J) = µ(T ).

Subcase5: HereTisof,say,typeII

ij

andfor

ω

largeenough,

T ω v

areof,say,type

II

pq

, whereby Proposition 2.10,we maytake, withoutlossofgenerality,

p

and

q

to

liebetween

i

and

q

. Let

θ

and

γ

beas in Theorem2.9 forthe edge

{i, j}

in

T

and

let

θ ˆ

and

γ ω

bethecorrespondingpairfortheedge

{p, q}

in

T ω v

. Observethat

θ ˆ

does

notdependon

ω

byRemark3.8(b). Let

γ ˆ = lim ω→∞ γ ω

. ByCorollary3.13wehave

0 < γ < 1

. Now

ω→∞ lim µ(T ω v ) = lim

ω→∞

1 ρ(M q→p,T ω v γ ˆ ω

θ J ) = lim

ω→∞

1 ρ(M q→p,T γ ˆ ω

θ J )

= 1

ρ ( M q→p,T γ ˆ θ ˆ J ) < 1

ρ ( M j→i,T γ θ J ) = µ(T ),

wheretheinequalityfollowsfromProposition2.11.

Acknowledgments. We are indebted to Mr. Felix Goldberg [5] for suggesting

Example1.5 which showsthat

G

doesnothaveto beatree for

lim

ω→∞ µ(G v ω ) = µ(G)

(12)

tohold. Thequestionofwhen does

lim

ω→∞ µ(G v ω ) = µ(G)

, when

G

isageneralgraph,

seems to be much more dicult than the one in the case that

G

is a tree. We

are grateful to thereferee for his orherimportant remarks and forsuggestingthat

Propositions 1.3and1.4, as wellasLemma 2.2of[2], maybeusefulin dealingwith

thegeneralcase.

REFERENCES

[1] A.Bermanand R.J.Plemmons.NonnegativeMatricesinthe MathematicalSciences,SIAM,

Philadelphia,1994.

[2] S.FallatandS.Kirkland.Extremizing algebraicconnectivitysubjecttographtheoretic con-

straints. ElectronicJournalofLinearAlgebra,3:4874,1998.

[3] M.Fiedler.Algebraicconnectivityofgraphs.CzechoslovakMathematicalJournal,23: 298305,

1973.

[4] M.Fiedler.Apropertyofeigenvectorsofnonnegativesymmetricmatricesanditsapplications

tographtheory.CzechoslovakMathematicalJournal,25: 619633,1975.

[5] F.Goldberg.Privatecommunication.

[6] R.Grone,R.Merris,and V.Sunder.TheLaplacianspectrumofagraph. SIAMJournal on

MatrixAnalysisandApplications,11: 218-238,1990.

[7] R.J.HornandC.R.Johnson.MatrixAnalysis.CambridgeUniversityPress,Cambridge,1985.

[8] S.Kirklandand M. Neumann.Algebraic connectivity ofweighted trees underperturbation.

LinearandMultilinearAlgebra,42: 187203,1997.

[9] S.Kirkland,M.Neumann,andB.Shader.CharacteristicverticesofweightedtreesviaPerron

values.LinearandMultilinearAlgebra,40: 311325,1996.

[10] S.Kirkland, M.Neumann,and B.Shader. Distancesinweightedtrees andgroupinverse of

LaplacianMatrices. SIAM Journal on MatrixAnalysis and Applications, 18: 827841,

1997.

[11] R.Merris.Characteristicverticesoftrees.LinearandMultilinearAlgebra,22: 115131,1987.

[12] J.J.MolitiernoandM.Neumann.Thealgebraicconnectivityoftwotreesconnectedbyanedge

ofinniteweight. ElectronicJournalofLinearAlgebra,8: 113,2001.

参照

関連したドキュメント

In this paper, the exact formulae for the weighted Szeged indices of generalized hierarchical product and Cartesian product of two graphs are obtained.. Keywords:

Weighted Sums, Strong Law of Large Numbers, Almost Se Convergence, Generzed Gaussian Random Variables, Random Elements in Banach Space, Schauder Basis.. AMS (MOS) SUBJECT

Keywords: point of complete accumulation, linearly Lindel¨of space, local com- pactness, first countability, κ-accessible diagonal. AMS Subject Classification: 54F99,

Keywords: Aull-paracompactness of Y in X, strong star-normality of Y in X AMS Subject Classification: 54D20,

Keywords: σ-finite measure space, measure preserving transformation, conserva- tive, ergodic, supremum of ergodic ratios, maximal and reverse maximal inequalities AMS

2010 Mathematics Subject Classification: 05A10, 11B65, 28B99, 11B68, 11B73 Keywords and phrases: Bernoulli numbers and polynomials, Euler numbers and polyno- mials, Genocchi numbers

Canonical forms for ∗ congruence, ∗ Congruence, Codimension, Matrix equations, Orbits, ∗ Palindromic pencils.. AMS

Factor Analysis : Pearson product moment correlations of each variable with other variables were obtained, yielding 8 x 8 matrix in 4 groups admitted