• No results found

The lexicographic method for the threshold cover problem

N/A
N/A
Protected

Academic year: 2023

Share "The lexicographic method for the threshold cover problem"

Copied!
15
0
0

Loading.... (view fulltext now)

Full text

(1)

Contents lists available atScienceDirect

Discrete Mathematics

journalhomepage:www.elsevier.com/locate/disc

The lexicographic method for the threshold cover problem

Mathew C. Francis

a

, Dalu Jacob

b

,

aIndianStatisticalInstitute,ChennaiCentre,India bIndianInstituteofScience,Bangalore,India

a rt i c l e i nf o a b s t ra c t

Articlehistory:

Received14March2022

Receivedinrevisedform28January2023 Accepted31January2023

Availableonlinexxxx

Keywords:

Thresholdcover Chainsubgraphcover Lexicographicmethod

Threshold graphs are aclassofgraphs that have manyequivalentdefinitions and have applications ininteger programmingand set packingproblems. Agraphis saidto have athresholdcoverofsizekifitsedgescanbecoveredusingkthreshold graphs.Chvátal andHammer,in1977,definedthethresholddimensionth(G)ofagraphGtobetheleast integerksuchthat G hasathresholdcoverofsizekand observedthat th(G)

χ

(G), where Gisasuitablyconstructedauxiliarygraph.Raschleand Simon(1995)[9] proved thatth(G)=

χ

(G)wheneverGisbipartite.Weshowhowthelexicographicmethodof Hell andHuang canbeused toobtainacompletelynewand, webelieve,simplerproof for thisresult.Forthe case whenG is asplitgraph,our methodyieldsaproof that is much shorterthanthe onesknowninthe literature.Ourmethods giveriseto asimple and straightforward algorithmtogeneratea2-threshold coverofaninput graph,ifone exists.

©2023ElsevierB.V.Allrightsreserved.

1. Introduction

We consideronly simple,undirectedand finitegraphs.We denote an edgebetweentwo vertices u and v ofa graph bythetwo-element set

{

u

,

v

}

,whichisusuallyabbreviatedtojustuv.Twoedgesab

,

cdinagraphG aresaid toforman alternating4-cycle ifad

,

bc

E

(

G

)

. Agraph G that doesnot containanypair ofedges that forman alternating 4-cycleis calledathresholdgraph;orequivalently,G is

(

2K2

,

P4

,

C4

)

-free [1].AgraphG

= (

V

,

E

)

issaidtobe coveredbythegraphs H1

,

H2

, . . . ,

HkifE

(

G

) =

E

(

H1

)

E

(

H2

) ∪ · · · ∪

E

(

Hk

)

.

Definition1(Thresholdcoverandthresholddimension).AgraphGissaidtohaveathresholdcoverofsizekifitcanbecovered bykthresholdgraphs.ThethresholddimensionofagraphG,denotedasth

(

G

)

,isdefinedtobethesmallestintegerksuch that Ghasathresholdcoverofsizek.

MahadevandPeled [8] giveacomprehensivesurveyofthresholdgraphsandtheirapplications.

Chvátal andHammer [1] showed thatthefact thata graphG hasth

(

G

)

k isequivalent tothefollowing:there exist klinearinequalitieson

|

V

(

G

) |

variablessuchthatthecharacteristicvectorofaset S

V

(

G

)

satisfiesalltheinequalitiesif

Apreliminaryversionofthispaper,claimingthesameresultasprovedinthiswork,appearedintheproceedingsoftheconferenceCALDAM2020.But theproofinthatversioncontainsaseriouserror,andthealgorithmmentionedinthatpapermayfailtoproducea2-thresholdcoveriftheinputgraph containsaparagliderasaninducedsubgraph.

*

Correspondingauthor.

E-mailaddresses:mathew@isichennai.res.in(M.C. Francis),dalu1991@gmail.com(D. Jacob).

https://doi.org/10.1016/j.disc.2023.113364 0012-365X/©2023ElsevierB.V.Allrightsreserved.

(2)

a

f g

b c

d e

ab ac

f g bf

c f e f

cg

dg bc

bd

be

cd de

(a) (b)

Fig. 1.(a) A graphG, and (b) the auxiliary graphGofG.

andonlyifS isanindependentsetofG(see [9] fordetails).TheyfurtherdefinedtheauxiliarygraphGcorresponding to agraphGasfollows.

Definition 2 (Auxiliarygraph).Given a graph G, the graph G has vertex set V

(

G

) =

E

(

G

)

and edge set E

(

G

) = {{

ab

,

cd

}:

ab

,

cd

E

(

G

)

suchthatab

,

cdformanalternating4-cycleinG

}

.

AgraphG andtheauxiliary graphG correspondingtoitisshowninFig.1.ChvátalandHammerobservedthatsince foranysubgraph HofG thatisathresholdgraph, E

(

H

)

isanindependentsetinG,thefollowinglowerboundonth

(

G

)

holds.

Lemma1(Chvátal-Hammer).th

(

G

)χ (

G

)

.

This gave rise to the question of whetherthere is anygraph G such that th

(

G

) > χ (

G

)

.Cozzens and Leibowitz [4]

showed the existence of such graphs. In particular, they showedthat forevery k

4, there exists a graph G such that

χ (

G

) =

k butth

(

G

) >

k.Whilethequestion ofwhethersuchgraphsexistfork

=

3 doesnotseemtohavereceivedmuch attentionandstillremainsopen(tothebestofourknowledge),thequestionforthecasek

=

2 wasstudiedquiteintensively (see [7]).IbarakiandPeled [6] showed,by meansofsomeveryinvolvedproofs,thatifG isasplitgraphorifG contains atmosttwo non-trivialcomponents,then

χ (

G

) =

2 ifandonlyifth

(

G

) =

2.Theyfurtherconjecturedthat foranygraph G,

χ (

G

) =

2

th

(

G

) =

2.Iftheconjectureheld,itwouldshowimmediatelythat graphshavingathresholdcoverofsize 2can berecognized inpolynomialtime,since theauxiliarygraph G canbe constructedanditsbipartitenesscheckedin polynomial time.Incontrast,Yannakakis [12] showed thatitisNP-completetorecognize graphshavingathresholdcover ofsize3.CozzensandHalsey [3] studiedsome propertiesofgraphshavingathresholdcoverofsize 2andshowedthatit can bedecided inpolynomial time whetherthecomplementofabipartite graphhasa thresholdcoverofsize 2.Finally, morethanadecadeafterthequestionwasfirstposed,RaschleandSimon [9] provedtheconjectureofIbarakiandPeledby extendingthemethodsin [6].

Theorem1(Raschle-Simon).ForanygraphG,

χ (

G

) =

2ifandonlyifth

(

G

) =

2.

This proof of Raschle andSimon is very technical and involves the use of a numberof complicated reductions and previouslyknownresults.Inthispaper,weprovideanew,andwebelieve,simpler,proofforTheorem1.

Weconstructanalgorithmthatgeneratesa2-thresholdcoverofaninputgraphGifGisbipartite.InSections3and4, wedescribethealgorithmanditsproofofcorrectnessforthecasewhentheinputgraphbelongstotheclassof“paraglider- free” graphs,which isa superclassofthe classof chordalgraphs. Sinceall splitgraphs are alsochordal graphs,we have a proof ofTheorem 1for thecaseofsplit graphs inSection 4itself. We believe thatthis prooffor splitgraphs ismuch simplerandshorterthantheproofforsplitgraphsgivenbyIbarakiandPeled [6] (ortheproofofRaschleandSimon [9] for generalgraphsthatbuildsupontheworkofIbarakiandPeled).InSection 5,weshowhowouralgorithmcanbemodified toworkforgeneralgraphs.Notethatforthecaseofgeneralgraphs,eventhoughthealgorithmremains simple,theproof ofitscorrectnessbecomesmoreinvolved.

Outlineofthealgorithm.LetG beanygraphsuchthatGisbipartite.Wewouldliketoconstructathresholdcoverofsize 2forG.Anaturalwaytoapproachtheproblemistocomputea2-coloringofG,whichcorrespondstoapartitionofthe edgesetofG intotwosets,sayE1andE2,andtrytoshowthatG1

= (

V

(

G

),

E1

)

andG2

= (

V

(

G

),

E2

)

arethresholdgraphs (and so

{

G1

,

G2

}

is a 2-thresholdcover of G). But thisapproach doesnot work since if we take an arbitrary 2-coloring of G,the graphs G1 and G2 need not necessarilybe thresholdgraphs (thiscan be easily seenin thecasewhen G isa completegraph,asthenG containsonlyisolatedvertices).Instead,ouralgorithmgeneratesaspecialkindof2-coloringof

(3)

G,whichisthenusedtoconstructa2-thresholdcoverofG.LetX denotethesetofisolatedverticesinG.Ouralgorithm computesasetY

X anda2-coloringofG

Y (wecallthisa“partial2-coloring”ofG)whichpartitionsE

(

G

) \

Y into twosets E1 andE2 such thatthegraphsG1

= (

V

(

G

),

E1

Y

)

andG2

= (

V

(

G

),

E2

Y

)

areboththresholdgraphs,thereby yielding a 2-thresholdcover of G. Note that as G need not be connected, even if X

= ∅

,there can be an exponential numberof2-coloringsofG andasnotedabove,notevery2-coloringgivesrisetoa2-thresholdcoverofG.Thealgorithm runsintime O

(|

V

(

G

)| + |

E

(

G

)|) =

O

(|

E

(

G

)|

2

)

.Inthecaseofsplitgraphs,andmoregenerallyparaglider-freegraphs,our algorithmdoesnotneedtoprocesstheisolatedverticesinG atall;insteaditjusttakesY

=

X.Inotherwords,itcomputes a 2-coloringofG

X whichpartitions E

(

G

) \

X intotwo sets E1 andE2 such thatthegraphs G1

= (

V

(

G

),

E1

X

)

and G2

= (

V

(

G

),

E2

X

)

areboththresholdgraphs.

TheChainSubgraphCoverProblem. Abipartite graph G

= (

A

,

B

,

E

)

is calleda chaingraph if itdoes not contain a pair ofedges whose endpointsinduce a2K2 in G.LetG

ˆ

be thesplitgraph obtainedfromG byaddingedges betweenevery pair ofverticesin A (or B). Itcan be seenthat G isa chaingraph ifandonlyif G

ˆ

isa thresholdgraph.A collectionof chaingraphs

{

H1

,

H2

, . . . ,

Hk

}

issaidtobeak-chainsubgraphcoverofabipartitegraphG ifitiscoveredby H1

,

H2

, . . . ,

Hk. Yannakakis [12] creditsMartinGolumbicforobservingthat a bipartitegraph G hasa k-chainsubgraph coverifandonly ifG

ˆ

hasak-thresholdcover.The problemofdecidingwhetherabipartite graphG can becovered byk chaingraphs,i.e.

whetherG hasak-chainsubgraphcover,isknownasthek-chainsubgraphcover(k-CSC)problem. Yannakakis [12] showed that 3-CSC isNP-complete, whichimplies that theproblem ofdeciding whether th

(

G

)

3 for an input graph G is also NP-complete. He also pointedout that using Golumbic’s observationand theresults of Ibarakiand Peled [6], the 2-CSC problemcanbesolved inpolynomial time,asitcan bereducedto theproblemofdeterminingwhethera splitgraphcan be coveredbytwothresholdgraphs.Thusouralgorithmforsplitgraphscanalsobe usedtocomputea2-chainsubgraph cover,ifoneexists,foran inputbipartitegraph Gintime O

( |

E

(

G

) |

2

)

(notethateventhough

|

E

(

G

ˆ ) | > |

E

(

G

) |

,thevertices inG

ˆ

correspondingtotheedgesin E

(

G

ˆ ) \

E

(

G

)

areallisolatedverticesandhencedonot needtobeputinG sinceour algorithmforsplitgraphsdoesnottakethemintoaccountanyway).NotethatMaandSpinrad [7] proposeamoreinvolved O

( |

V

(

G

) |

2

)

algorithmfortheproblem. However,ouralgorithm forsplit graphs,andhencethealgorithm forcomputinga 2-chainsubgraphcoverthatityields,isconsiderablysimplertoimplementthanthealgorithmsof [6,7,9,11].

Lex-BFSorderings.ALex-BFSorderingofagraphisan orderingoftheverticesofthegraphhavingthepropertythat itis possibleforaLexicographicBreadthFirstSearch(Lex-BFS)algorithmtovisittheverticesofthegraphinthatorder.ALex-BFS orderingisalsoaBFSordering—i.e.,abreadth-firstsearchalgorithmcanalsovisittheverticesinthatorder—butithassome additionalproperties.Lex-BFScanbeimplementedtorunintimelinearinthesizeoftheinputgraphandwasintroduced by Rose, TarjanandLueker [10] toconstructa linear-time algorithmforrecognizing chordalgraphs. Later,Lex-BFSbased algorithmswerediscoveredfortherecognitionofmanydifferentgraphclasses(see [2] forasurvey).

TheLexicographicMethod. We use a technique calledthe lexicographicmethod introduced by Hell and Huang [5], who demonstrated how thismethod can lead to shorter proofsand simplerrecognition algorithms forcertain problems that involveconstructingaspecific2-coloringofanauxiliarybipartitegraphthatcapturescertainrelationshipsamongtheedges ofthegraph.Themethodinvolvesfixinganordering

<

oftheverticesofthegraph,andthenprocessingtheedges inthe

“lexicographicorder”impliedbytheordering

<

.Weadaptthistechniquetoconstructapartial2-coloringofGthatcanbe usedtogeneratea2-thresholdcoverofG.Hell andHuang [5] startwithanarbitraryorderingoftheverticesofthegraph intheirrecognitionalgorithms forcomparability graphsandpropercircular-arcgraphs,butforthecaseofproperinterval graphs,theystartwitha“perfecteliminationordering”ofthegivengraph,whichshouldnecessarilyexistwhenthegraphis chordal(notethatproperintervalgraphsformasubclassofchordalgraphs).FromtheworkofRose,TarjanandLueker [10], it isknown that forchordal graphs, theperfectelimination orderings are exactlythe reversalsof Lex-BFSorderings.Thus therecognitionalgorithmforproperintervalgraphsbasedonthelexicographicmethodthatisgivenin [5] startswiththe reversalofaLex-BFSorderingoftheinputgraph.Asweshallsee,ourrecognitionalgorithmforgraphshavinga2-threshold coverstarts witha Lex-BFSorderof theinput graph.Whenit isknownthat theinput graphisa “paraglider-freegraph”

(definedinSection4),wecanevenstartwithanarbitraryorderingoftheverticesoftheinputgraph.

2. Preliminaries

LetG

= (

V

,

E

)

beanygraph.Recallthatedgesab

,

cd

E

(

G

)

formanalternating4-cycleifbc

,

da

E

(

G

)

.Inthiscase,we alsosaythata

,

b

,

c

,

d

,

a isanalternating4-cycleinG (alternating4-cyclesarecalled AC4sin [9]).Theedgesabandcdare saidto betheoppositeedgesofthealternating 4-cyclea

,

b

,

c

,

d

,

a.Thusforagraph G,theauxiliary graphG isthegraph withV

(

G

) =

E

(

G

)

andE

(

G

) = {{

ab

,

cd

}:

ab

,

cd

E

(

G

)

aretheoppositeedgesofanalternating4-cycleinG

}

.Notethatit followsfromthedefinitionof analternating 4-cycle thatifa

,

b

,

c

,

d

,

a isan alternating 4-cycle,then theverticesa

,

b

,

c

,

d are pairwise distinct.We shallrefer tothevertexof G corresponding to anedge ab

E

(

G

)

alternativelyas

{

a

,

b

}

orab, dependinguponthecontext.

OurgoalistoprovideanewproofforTheorem1.

(4)

a

b e

c d

(a)

a

b c

d

(b)

a

b c

d

(c)

Fig. 2.(a)AnA1-pentagon,(b)anA1-switchingpath,and(c)anA1-switchingcycle.Ineachofthefigures:adashedlinebetweentwoverticesindicates thattheyarenon-adjacent,athinlinerepresentsanedgethatmaybeineitherA1orA0,athicklineindicatesanedgeinA1,andagraylinerepresents anedgeinA2.

It is easy to see that

χ (

G

) =

1 if and only if th

(

G

) =

1. Therefore, by Lemma 1, it is enough to prove that if G is bipartite, then G can be coveredby two threshold graphs.In order toprove this, we find a specific 2-coloringof the non-trivialcomponentsofG (componentsofsizeatleast2)usingthelexicographicmethodofHellandHuang [5].

We saythat

(

A0

,

A1

,

A2

)

is avalid3-partitionof E

(

G

)

if

{

A0

,

A1

,

A2

}

isa partition of E

(

G

)

with theproperty that in anyalternating 4-cycle inG, oneofthe oppositeedges belongsto A1 andtheother to A2.Inother words, foranyedge

{

ab

,

cd

} ∈

E

(

G

)

,oneofab

,

cdisin A1andtheotherin A2.Notethatthismeanswhilesomeedgesin A1andA2 mayhave thepropertythattheydonotformanalternating4-cyclewithanyotheredge,everyedgeinA0definitelyhasthisproperty.

Givena valid3-partition

(

A0

,

A1

,

A2

)

of E

(

G

)

and A

∈ {

A1

,

A2

}

,we saythat a

,

b

,

c

,

d isan alternatingA-path ifa

=

d, ab

,

cd

A

A0,andbc

E

(

G

)

.Further,we saythata

,

b

,

c

,

d

,

e

,

f

,

a isan alternatingA-circuit ifa

=

d,ab

,

cd

,

e f

A

A0, andbc

,

de

,

f a

E

(

G

)

.

Observation1.Let

(

A0

,

A1

,

A2

)

beavalid3-partitionofE

(

G

)

andlet

{

A

,

A

} = {

A1

,

A2

}

. (a) Ifa

,

b

,

c

,

d isanalternatingA-path,thenad

E

(

G

)

.

(b) Ifa

,

b

,

c

,

d

,

e

,

f

,

a isanalternatingA-circuit,thene f

A andad

A.

Proof. Toprove (a), itjust needs to beobserved that ifad

E

(

G

)

,then a

,

b

,

c

,

d

,

a would bean alternating 4-cycle in G whoseoppositeedgesbothbelongto A

A0,whichcontradictsthefactthat

(

A0

,

A1

,

A2

)

isavalid3-partitionof E

(

G

)

.To prove (b),suppose thata

,

b

,

c

,

d

,

e

,

f

,

a isanalternating A-circuit.Sincea

,

b

,

c

,

dis analternating A-path,we haveby (a) thatad

E

(

G

)

.Thensincea

,

d

,

e

,

f

,

aisanalternating4-cycleinGande f

A

A0,itfollowsthate f

A andad

A.

Weshallusetheaboveobservationthroughoutthispaperwithoutreferringtoitexplicitly.

Let

(

A0

,

A1

,

A2

)

beavalid3-partitionofE

(

G

)

andlet

{

A

,

A

} = {

A1

,

A2

}

.Wesaythat

(

a

,

b

,

c

,

d

,

e

)

isan A-pentagoninG withrespectto

(

A0

,

A1

,

A2

)

ifa

,

b

,

c

,

d

,

e

V

(

G

)

,ac

,

ad

,

be

E

(

G

)

,ab

,

ae

A,bc

,

bd

,

ec

,

ed

Aandcd

A

A0.Weabbreviate thistojust“A-pentagon”whenthegraphG andthe3-partition

(

A0

,

A1

,

A2

)

ofGare clearfromthecontext.Wesaythat an A-pentagon

(

a

,

b

,

c

,

d

,

e

)

isastrictA-pentagonifcd

A.Wesaythat

(

a

,

b

,

c

,

d

,

e

)

isapentagon(resp.strictpentagon)ifit isan A-pentagon(resp.strict A-pentagon)forsome A

∈ {

A1

,

A2

}

.(Pentagonsaresimilartothe“A P5-s”in [9].)

Wesaythat

(

a

,

b

,

c

,

d

)

isan A-switchingpathinG withrespectto

(

A0

,

A1

,

A2

)

ifa

,

b

,

c

,

d

V

(

G

)

,ad

E

(

G

)

,ab

,

cd

A

A0, and bc

A. When the graph G andthe 3-partition

(

A0

,

A1

,

A2

)

of G are clear from the context, we abbreviate this to just “A-switching path”. We say that

(

a

,

b

,

c

,

d

)

is a strict A-switchingpath ifit is an A-switching path and in addition, ab

,

cd

A. We say that

(

a

,

b

,

c

,

d

)

is a switchingpath (resp. strictswitchingpath) ifit is an A-switching path (resp.strict A-switchingpath)forsome A

∈ {

A1

,

A2

}

.Wesaythat

(

a

,

b

,

c

,

d

)

isan A-switchingcycleinG withrespectto

(

A0

,

A1

,

A2

)

if ab

,

cd

A

A0andbc

,

ad

A.Asbefore,wesaythat

(

a

,

b

,

c

,

d

)

isaswitchingcycleinGwithrespectto

(

A0

,

A1

,

A2

)

ifthere exists A

∈ {

A1

,

A2

}

suchthat

(

a

,

b

,

c

,

d

)

isan A-switchingcycle.

Note that from thedefinitions of pentagons,switching paths and switchingcycles, it followsthat if

(

a

,

b

,

c

,

d

,

e

)

isa pentagon,thentheverticesa

,

b

,

c

,

d

,

earepairwisedistinct,andif

(

a

,

b

,

c

,

d

)

isaswitchingpathoraswitchingcycle,then the vertices a

,

b

,

c

,

d are pairwise distinct. Fig. 2 illustrates an A1-pentagon, an A1-switchingpath, andan A1-switching cycle.

Lemma2.Let

(

A0

,

A1

,

A2

)

beavalid3-partitionofE

(

G

)

.IftherearenoswitchingpathsandnoswitchingcyclesinG withrespectto

(

A0

,

A1

,

A2

)

thenth

(

G

) =

2.

Proof. Considerthegraphs H1

,

H2,having V

(

H1

) =

V

(

H2

) =

V

(

G

)

,E

(

H1

) =

A1

A0 andE

(

H2

) =

A2

A0.Weclaimthat H1 and H2 are both threshold graphs. Suppose forthe sake of contradictionthat Hi isnot a threshold graph forsome

(5)

i

∈ {

1

,

2

}

.Thenthereexistedgesab

,

cd

E

(

Hi

)

suchthatbc

,

ad

E

(

Hi

)

.Ifbc

,

ad

E

(

G

)

,thena

,

b

,

c

,

d

,

aisanalternating4- cycleinGwhoseoppositeedgesbothbelongtoAi

A0,whichcontradictsthefactthat

(

A0

,

A1

,

A2

)

isavalid3-partition.So wecanassumebysymmetrythatbc

E

(

G

)

.Sincebc

E

(

Hi

)

,bc

/

Ai

A0,whichimpliesthatbc

A3i.Nowifad

E

(

G

)

, then

(

a

,

b

,

c

,

d

)

isan Ai-switchingpath inG withrespectto

(

A0

,

A1

,

A2

)

,which isacontradiction. Ontheother hand,if ad

E

(

G

)

, thenad

A3i (sincead

E

(

Hi

)

), whichimpliesthat

(

a

,

b

,

c

,

d

)

is an Ai-switchingcyclein G withrespect to

(

A0

,

A1

,

A2

)

,whichisagainacontradiction.Thuswecanconcludethatboth H1 andH2arethresholdgraphs.

Lemma3.Let

(

A0

,

A1

,

A2

)

beavalid3-partitionofE

(

G

)

.Let

{

A

,

A

} = {

A1

,

A2

}

.Let

(

x

,

y

,

z

,

w

)

beanA-switchingpathinG andlet yz

E

(

G

)

besuchthatyz

,

zy

E

(

G

)

.Then,

(a) ifx

=

y,then

(

x

=

y

,

y

,

z

,

w

,

z

)

isanA-pentagonand (b) ifw

=

z,then

(

w

=

z

,

z

,

y

,

x

,

y

)

isanA-pentagon.

Proof. Since yz

Aand

{

yz

,

yz

} ∈

E

(

G

)

wehavethat yz

A.Supposethat x

=

y.Then y

, (

x

=

y

),

z

,

w

, (

x

=

y

),

z

,

y isan alternating A-circuit (notethat y

=

w asx

N

(

y

) \

N

(

w

)

),implying that y w

A.Thisfurtherimpliesthat z

=

w.

Then we also havealternating A-circuits z

,

y

,

z

,

w

,

x

,

y

,

z and z

, (

y

=

x

),

w

,

z

, (

y

=

x

),

y

,

z,implying that xy

A and zw

,

zz

A. Consequently,

(

x

=

y

,

y

,

z

,

w

,

z

)

is an A-pentagon. Since

(

w

,

z

,

y

,

x

)

is also an A-switching path, we can similarlyconcludethatifw

=

z,then

(

w

=

z

,

z

,

y

,

x

,

y

)

isan A-pentagon.

Wethenhavethefollowingcorollary.

Corollary1.Let

(

A0

,

A1

,

A2

)

beavalid3-partitionofE

(

G

)

.Supposethattherearenopentagons(resp.strictpentagons)inG with respectto

(

A0

,

A1

,

A2

)

.Let

(

x

,

y

,

z

,

w

)

beaswitchingpath(resp.strictswitchingpath)withrespectto

(

A0

,

A1

,

A2

)

.Letyz

E

(

G

)

besuchthatyz

,

zy

E

(

G

)

.Then,y

=

x andz

=

w.

Proof. Let

{

A

,

A

} = {

A1

,

A2

}

.Supposethattherearenopentagons(resp.strictpentagons)inGwithrespectto

(

A0

,

A1

,

A2

)

. Let

(

x

,

y

,

z

,

w

)

bean A-switchingpath (resp.strict A-switchingpath, andtherefore xy

,

zw

A). ByLemma 3, we know that if y

=

x then

(

x

=

y

,

y

,

z

,

w

,

z

)

isan A-pentagon (resp.a strict A-pentagon, as zw

A), andifz

=

w then

(

w

=

z

,

z

,

y

,

x

,

y

)

isan A-pentagon (resp.astrict A-pentagon,asxy

A).Sincetherearenopentagons(resp.strictpentagons), wecanconcludethat y

=

xandz

=

w.

Let

<

be an ordering of the verticesof G. Giventwo k-element subsets S

= {

s1

,

s2

, . . . ,

sk

}

and T

= {

t1

,

t2

, . . . ,

tk

}

of V

(

G

)

,wheres1

<

s2

< · · · <

sk andt1

<

t2

< · · · <

tk, S issaidtobe lexicographicallysmallerthan T,denotedby S

<

T,if sj

<

tjforsome j

∈ {

1

,

2

, . . . ,

k

}

,andsi

=

tiforall1

i

<

j.Intheusualway,weletS

T denotethefactthateitherS

<

T or S

=

T.Foraset S

V

(

G

)

,we abbreviatemin<S tojust minS.Notethat therelation

<

(“is lexicographically smaller than”)thatwehavedefinedonk-elementsubsetsofV

(

G

)

isatotalorder.Therefore,givenacollectionofk-elementsubsets ofV

(

G

)

,thelexicographicallysmallestoneamongthemiswell-defined.

Thefollowingobservationstatesawell-knownpropertyofLex-BFSorderings [2].

Observation2.Let

<

denoteaLex-BFSorderingofagraphG.Fora

,

b

,

c

V

(

G

)

,ifa

<

b

<

c,ab

/

E

(

G

)

andac

E

(

G

)

,thenthere existsx

V

(

G

)

suchthatx

<

a

<

b

<

c,xb

E

(

G

)

andxc

/

E

(

G

)

.

3. Thealgorithm

Let G be a graph such that G is bipartite. For two vertices ab

,

ab

V

(

G

)

(i.e. ab

,

ab

E

(

G

)

), we say that ab is lexicographicallysmallerthanab withrespecttoanordering

<

ofV

(

G

)

,if

{

a

,

b

} < {

a

,

b

}

.

Weshallnowconstructa partial2-coloringoftheverticesofG usingthecolors

{

1

,

2

}

bymeansofanalgorithm,and thenconstructavalid3-partitionofE

(

G

)

usingthispartial2-coloring.

PhaseI.ConstructaLex-BFSordering

<

ofG.

RecallthateveryvertexofG isatwo-elementsubsetofV

(

G

)

.

Phase II.For every non-trivial componentC ofG, perform the following operation:

Choose the lexicographically smallest vertex inC (with respect to the ordering

<

) and assign the color 1 to it. Extend this to a proper coloring ofCusing the colors

{

1

,

2

}

.

Note that after Phase II, every vertex of G that is in a non-trivial component has been colored either 1 or 2. For i

∈ {

1

,

2

}

,let Fi

= {

e

V

(

G

) :

eis coloredi

}

.Further,let F0 denotethesetofall isolated vertices(trivialcomponents) in

(6)

G.Clearly,F0isexactlythesetofuncoloredverticesofGandwehaveV

(

G

) =

F0

F1

F2.Notethatsincetheopposite edgesofanyalternating4-cycleinGcorrespondtoadjacentverticesinG,oneofthemreceivescolor1andtheothercolor 2inthepartial2-coloringofG constructedinPhaseII.Itfollowsthat

(

F0

,

F1

,

F2

)

isavalid3-partitionofE

(

G

)

.

Firstwenotethefollowinglemma.

Lemma4.IftherearenostrictpentagonsinG withrespectto

(

F0

,

F1

,

F2

)

,thentherearenostrictswitchingpathsinG withrespect to

(

F0

,

F1

,

F2

)

.

Proof. Supposethat G contains a strictswitchingpath.We saythata strictswitchingpath

(

a

,

b

,

c

,

d

)

islexicographically smallerthanastrictswitchingpath

(

a

,

b

,

c

,

d

)

if

{

a

,

b

,

c

,

d

} < {

a

,

b

,

c

,

d

}

.Let

(

a

,

b

,

c

,

d

)

bethelexicographicallysmallest strictswitchingpathinG.

Claim.

(

a

,

b

,

c

,

d

)

isnotastrictF1-switchingpath.

Supposeforthesake ofcontradictionthat

(

a

,

b

,

c

,

d

)

isastrict F1-switchingpath.LetC be thecomponentof G con- tainingbc.Letb0c0

,

b1c1

, . . . ,

bkck,whereb0

=

bandc0

=

c,beapathinC betweenbcandthelexicographicallysmallest vertexbkckin C.Weassume thatforeachi

∈ {

0

,

1

, . . . ,

k

1

}

,bici+1

,

cibi+1

E

(

G

)

.Asb0c0

F2,itfollowsthatbici

F2 foreacheveni andbici

F1 foreach oddi.SincebkckisthelexicographicallysmallestvertexinitscomponentinG,we knowthatbkck

F1,whichimpliesthatkisodd.

Weclaimthatbia

,

cid

F1 foreacheveni andbia

,

cid

F2 foreachoddi,where0

i

k.Weprovethisbyinduction on i. The case where i

=

0 is trivial as b0

=

b and c0

=

c. So let us assume that i

>

0. Consider the case where i is odd. As i

1 iseven, by the inductionhypothesis we have bi1a

,

ci1d

F1. Since bi1ci1

F2, we can observethat,

(

a

,

bi1

,

ci1

,

d

)

isastrict F1-switchingpath.ThenbyCorollary1,we havethata

=

bi andd

=

ci.Now thealternating F1- circuits bi

,

ci

,

bi1

,

a

,

d

,

ci1

,

bi andci

,

bi

,

ci1

,

d

,

a

,

bi1

,

ci implythat bia

,

cid

F2.Thecasewherei iseven issymmetric andhencetheclaim.

By the above claim,bka

,

ckd

F2. Since bkck

F1, we now havethat

(

a

,

bk

,

ck

,

d

)

is a strict F2-switching path.Since bkck

<

bc,wehavethat

{

a

,

bk

,

ck

,

d

} < {

a

,

b

,

c

,

d

}

,which isacontradictiontoour assumptionthat

(

a

,

b

,

c

,

d

)

isthelexico- graphicallysmalleststrictswitchingpathinG.Thisprovestheclaim.

Bythe aboveclaim, we havethat

(

a

,

b

,

c

,

d

)

is astrict F2-switchingpath.By thesymmetry betweena andd, wecan assumewithoutlossofgeneralitythata

<

d.

As bc

F1,thevertexbc belongstoanon-trivialcomponentofG.Thenthereexists aneighbor uv ofbc inG such thatbv

,

uc

E

(

G

)

.Asbc

F1,wehaveuv

F2.ByCorollary1,wehavethat u

=

a.Then a

,

b

,

v

,

u

,

c

,

d

,

aisanalternating F2-circuit,implyingthatau

F1.Asab

F2,weknowthatabisnotthelexicographicallysmallestvertexinitscomponent.

Leta0b0

,

a1b1

, . . . ,

akbkbeapathinGbetweenabandthelexicographicallysmallestvertexakbk initscomponent,where a0

=

a,b0

=

b,andfor0

i

<

k,aibi+1

,

ai+1bi

E

(

G

)

.Notethatfor0

i

k,aibi

F2 ifiisevenandaibi

F1 ifi isodd.

Sinceakbk

F1 (asitisthelexicographicallysmallestvertexinitscomponentinG),thisimpliesthatkisodd.

Weclaimthatfor0

i

k,aiu

,

bic

F1ifiisevenandaiu

,

bic

F2ifiisodd.Weprovethisbyinductiononi.Thebase casewheni

=

0 istrivial, sinceau

,

bc

F1.Leti

>

0 beodd.Bytheinductionhypothesis wehavethat ai1u

,

bi1c

F1. Sinceai1bi1

F2wecanobservethat

(

u

,

ai1

,

bi1

,

c

)

isastrict F1-switchingpath.ThereforebyCorollary1,wehavethat ai

=

u andbi

=

c.Then we have alternating F1-circuitsai

,

bi

,

ai1

,

u

,

c

,

bi1

,

ai andbi

,

ai

,

bi1

,

c

,

u

,

ai1

,

bi, implyingthat aiu

,

bic

F2.Thecasewheni isevenissymmetric.Thisprovesourclaim.Sincekisodd,we nowhavethataku

,

bkc

F2. Notethatnow

(

c

,

bk

,

ak

,

u

)

isastrict F2-switchingpath.

Suppose that d

<

b.Then we havethat a

<

d

<

b, wheread

E

(

G

)

andab

E

(

G

)

. Thereforeby Observation 2, there exists x

<

a suchthat xd

E

(

G

)

andxb

E

(

G

)

.Then x

,

d

,

a

,

b

,

xisan alternating 4-cycle inwhichab

F2,implyingthat xd

F1.Thenwehaveastrict F1-switchingpath

(

x

,

d

,

c

,

b

)

suchthat

{

x

,

d

,

c

,

b

} < {

a

,

b

,

c

,

d

}

,whichisacontradictiontothe choiceof

(

a

,

b

,

c

,

d

)

.Thereforewecanassumethatb

<

d.Sinceakbk

<

abanda

,

b

<

d,wehavethat

{

c

,

bk

,

ak

,

u

} < {

a

,

b

,

c

,

d

}

. As

(

c

,

bk

,

ak

,

u

)

isastrictswitchingpath,thiscontradictsthechoiceof

(

a

,

b

,

c

,

d

)

.

4. ProofofTheorem1forsplitgraphs

Let G be anygraphsuch that G is bipartite. Let

(

F0

,

F1

,

F2

)

be a valid3-partition of E

(

G

)

obtainedby running the algorithmofSection3onthegraphG.

Lemma5.IftherearenopentagonsinG withrespectto

(

F0

,

F1

,

F2

)

,thentherearenoswitchingpathsorswitchingcyclesinG with respectto

(

F0

,

F1

,

F2

)

.

Proof. Supposenot.Let

(

a

,

b

,

c

,

d

)

bea switchingpathin G withrespectto

(

F0

,

F1

,

F2

)

.Leti

∈ {

1

,

2

}

suchthat

(

a

,

b

,

c

,

d

)

is an Fi-switchingpath.Then we havead

E

(

G

)

,ab

,

cd

Fi

F0, andbc

F3i.Since bc

F3i, thereexists uv

E

(

G

)

References

Related documents

Sealed Tenders are invited from reputed manufactures/authorized dealers/suppliers for Supply, installation and commissioning of Supply, installation and commissioning of

In a slightly advanced 2.04 mm stage although the gut remains tubular,.the yent has shifted anteriorly and opens below the 11th myomere (Kuthalingam, 1959). In leptocephali of

INDEPENDENT MONITORING BOARD | RECOMMENDED ACTION.. Rationale: Repeatedly, in field surveys, from front-line polio workers, and in meeting after meeting, it has become clear that

Based on the assumption that revenue from additional carbon pricing would be transferred back to households as lump-sum payments, we estimate that the level of real GDP in 2030

Harmonization of requirements of national legislation on international road transport, including requirements for vehicles and road infrastructure ..... Promoting the implementation

Angola Benin Burkina Faso Burundi Central African Republic Chad Comoros Democratic Republic of the Congo Djibouti Eritrea Ethiopia Gambia Guinea Guinea-Bissau Haiti Lesotho

Daystar Downloaded from www.worldscientific.com by INDIAN INSTITUTE OF ASTROPHYSICS BANGALORE on 02/02/21.. Re-use and distribution is strictly not permitted, except for Open

The petitioner also seeks for a direction to the opposite parties to provide for the complete workable portal free from errors and glitches so as to enable