/
Use of MAXCUT for Ramsey Arrowing of Triangles Alexand Use of MAXCUT for Ramsey Arrowing of Triangles Alexand

Use of MAXCUT for Ramsey Arrowing of Triangles Alexand - PDF document

karlyn-bohler
karlyn-bohler . @karlyn-bohler
Follow
454 views
Uploaded On 2015-04-30

Use of MAXCUT for Ramsey Arrowing of Triangles Alexand - PPT Presentation

Lange Stanis law P Radziszowski Department of Computer Science Rochester Institute of Technology Rochester NY 14623 arl9577spr csritedu and Xiaodong Xu Guangxi Academy of Sciences Nanning Guangxi 530007 China xxdmathssinacom Abstract In 1967 Erd733o ID: 57886

Lange Stanis law

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "Use of MAXCUT for Ramsey Arrowing of Tri..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

1IntroductionGivenasimplegraphG,wewriteG!(a1;:::;ak)eandsaythatGarrows(a1;:::;ak)eifforeveryedgek-coloringofG,amonochro-maticKaiisforcedforsomecolori2f1;:::;kg.Likewise,forgraphsFandH,G!(F;H)eifforeveryedge2-coloringofG,amonochro-maticFisforcedinthe rstcolororamonochromaticHisforcedinthesecond.De neFe(a1;:::;ak;p)tobethesetofallgraphsthatarrow(a1;:::;ak)anddonotcontainKp;theyareoftencalledFolkmangraphs.TheedgeFolkmannumberFe(a1;:::;ak;p)isthesmallestorderofagraphthatisamemberofFe(a1;:::;ak;p).In1970,Folkman[6]showedthatfork�maxfs;tg,Fe(s;t;k)exists.TherelatedproblemofvertexFolkmannumbers,whereverticesarecoloredinsteadofedges,ismorestudied[15,17]thanedgeFolkmannumbers,butwewillnotbediscussingthem.In1967,Erd}osandHajnal[5]askedthequestion:Doesthereex-istaK4-freegraphthatisnottheunionoftwotriangle-freegraphs?ThisquestionisequivalenttoaskingfortheexistenceofaK4-freegraphsuchthatinanyedge2-coloring,amonochromatictriangleisforced.AfterFolkmanprovedtheexistenceofsuchagraph,thequestionthenbecameto ndhowsmallthisgraphcouldbe,orusingtheabovenotation,whatthevalueofFe(3;3;4)is.Priortothispa-per,thebestknownboundsforthiscasewere19Fe(3;3;4)941[20,4].FolkmannumbersarerelatedtoRamseynumbersR(s;t),de nedastheleastpositivensuchthatany2-coloringoftheedgesofKnyieldsamonochromaticKsinthe rstcolororamonochromaticKtinthesecond.Usingthearrowingoperator,itisclearthatR(s;t)isthesmallestnsuchthatKn!(s;t)e.TheknownvaluesandboundsforvarioustypesofRamseynumbersarecollectedandregularlyupdatedbythesecondauthor[19].Wewillbeusingstandardgraphtheorynotation:V(G)andE(G)forthevertexandedgesetsofgraphG,respectively.Acutisapartitionoftheverticesofagraphintotwosets,SV(G)and S=V(G)nS.Thesizeofacutisthenumberofedgesthatjointhetwosets,thatis,jffu;vg2E(G)ju2Sandv2 Sgj.MAX-CUTisawell-knownNP-hardcombinatorialoptimizationproblemwhichasksforthemaximumsizeofacutofagraph.2 upperbound.Lin[12]obtainedalowerboundon10in1972with-outthehelpofacomputer.All659graphson15verticeswit-nessingFe(3;3;5)=15[18]containK4,thusgivingthebound16Fe(3;3;4).In2007,twooftheauthorsofthispapergaveacomputer-freeproofof18Fe(3;3;4)andimprovedthelowerboundfurtherto19withthehelpofcomputations[20].ThelonghistoryofFe(3;3;4)isnotonlyinterestinginitselfbutalsogivesinsightintohowdiculttheproblemis.FindinggoodboundsonthesmallestorderofanyFolkmangraph(with xedparameters)seemstobedicult,andsomerelatedRamseygraphcoloringproblemsareNP-hardorlieevenhigherinthepolynomialhierarchy.Forexample,Burr[2]showedthatarrowing(3;3)eiscoNP-complete,andSchaefer[21]showedthatforgeneralgraphsF,G,andH,F!(G;H)isP2-complete.3ArrowingviaMAX-CUTBuildingo Spencer'sandothermethods,DudekandRodl[4]in2008showedhowtoconstructagraphHGfromagraphG,suchthatthemaximumsizeofacutofHGdetermineswhetherornotG!(3;3)e.TheyconstructthegraphHGasfollows.TheverticesofHGaretheedgesofG,sojV(HG)j=jE(G)j.Fore1;e22V(HG),ifedgesfe1;e2;e3gformatriangleinG,thenfe1;e2gisanedgeinHG.Lett4(G)denotethenumberoftrianglesingraphG.Clearly,jE(HG)j=3t4(G).LetMC(H)denotetheMAX-CUTvalueofgraphH.Theorem1(DudekandRodl[4]).G!(3;3)eifandonlyifMC(HG)2t4(G).ThereisaclearintuitionbehindTheorem1thatwewillnowdescribe.Anyedge2-coloringofGcorrespondstoabipartitionoftheverticesinHG.IfatrianglecoloredinGisnotmonochromatic,thenitsthreeedgeswhichareverticesinHGwillbeseparatedinthebipartition.Ifwetreatthisbipartitionasacut,thenthesizeofthecutwillcounteachtriangletwiceforthetwoedgesthatcrossit.4 stillshowarrowing.Weappliedmultiplestrategiesforremovingsetsofverticesandmostweresuccessful.Thisledtothefollowingtheorem:Theorem2.Fe(3;3;4)860.Proof.ForagraphGwithverticesZn,de neC=C(d;k)=fv2V(G)jv=idmodn;for0ikg.LetG=G941,d=2,k=81,andGCbethegraphinducedonV(G)nC(d;k).ThenGChas860vertices,73981edgesand542514triangles.Usingtheupperbound(1)andtheMATLABeigsfunction,weobtainMC(HGC)10849671085028=2t4(Gc):(2)Therefore,GC!(3;3)e.2Noneofthemethodsusedallowedfor82ormoreverticestoberemovedwithouttheupperboundonMCbecominglargerthan2t4.3.2Goemans-WilliamsonMethodTheGoemans-WilliamsonMAX-CUTapproximationalgorithm[8]isawell-known,polynomial-timealgorithmthatrelaxestheprob-lemtoasemide niteprogram(SDP).Itinvolvesthe rstuseofSDPincombinatorialapproximationandhassinceinspiredavarietyofothersuccessfulalgorithms(seeforexample[13]).Thisrandomizedalgorithmreturnsacutwithexpectedsizeatleast0.87856oftheoptimalvalue.However,inourcase,allthatisneededisthesolu-tiontotheSDP,asitgivesanupperboundonMC(H).AbriefdescriptionoftheGoemans-Williamsonrelaxationfollows.The rststepinrelaxingMAX-CUTistorepresenttheproblemasaquadraticintegerprogram.GivenagraphHwithV(H)=f1;:::;ngandnonnegativeweightswi;jforeachpairofverticesfi;jg,wecanwriteMC(H)asthefollowingobjectivefunction:Maximize1 2Xiwi;j(1�yiyj)(3)subjectto:yi2f�1;1gforalli2V(H):6 G 2t4(G) min SDP L(127;5) 19558 20181 20181 L(457;6) 347320 358204 358204 L(761;3) 694032 731858 731858 L(785;53) 857220 857220 857220 G786 857762 857843 857753 Table2:PotentialFe(3;3;4)graphsGandupperboundsonMC(HG),where\min"isthebound(1)and\SDP"istheso-lutionof(4)fromSDPLR-MC,SDPLR,andSBmethod.G786isthegraphofTheorem3.andSDPupperbounds.MultipleSDPsolversthatweredesignedtohandlelarge-scaleSDPandMAX-CUTproblemswereused.Specif-ically,wemadeuseoftwoversionsofSDPLRbySamuelBurer[1],bothusinglow-rankfactorization.SDPLR-MCisaversionofthesoft-warespeci callyfortheMAX-CUTrelaxation.TheregularsoftwareSDPLRismeantforanySDP.SBmethodbyChristophHelmberg[10]implementsaspectralbundlemethodandwasalsoappliedsuccess-fullyinourexperiments.Table2liststheresults.Inallcases,allthreesolversgavethesameresult.NotethatalthoughnoneofthecomputedupperboundsoftheL(n;s)graphsimplyarrowing(3;3)e,allSDPboundsmatchthoseoftheminimumeigenvaluebound.Thisisdistinctfromotherfamiliesofgraphs,includingthosein[4],astheSDPboundisusuallytighter.Thus,thesegraphsweregivenfurtherconsideration.L(127;5)wasgivenparticularattention,asitisthesamegraphasG127,whereV(G127)=Z127andE(G127)=ffx;ygjx�y 3mod127g(thatis,thegraphG(127;3)asde nedintheprevioussection).IthasbeenconjecturedbyExoothatG127!(3;3)e.Healsosuggestedthatsubgraphsinducedonlessthan100verticesofG127mayaswell.FormoreinformationonG127see[20].NumerousattemptsweremadeatmodifyingthesegraphsinhopesthatoneoftheMAX-CUTmethodswouldbeabletoprovearrowing.Indeed,wewereabletodosowithL(785;53).NoticethatalloftheupperboundsforMC(HL(785;53))are857220,thesameas2t4(L(785;53)).OurgoalwasthentoslightlymodifyL(785;53)8 5AcknowledgmentsThethirdauthorissupportedbytheGuangxiNaturalScienceFoun-dation(2011GXNSFA018142).WewouldliketothankG.RinaldiandL.GrippofortheirenthusiasticaidinthecomputationofMAX-CUTboundswiththeirSpeeDPalgorithm[9].References[1]SamuelBurerandRenatoD.C.Monteiro.Anonlinearpro-grammingalgorithmforsolvingsemide niteprogramsvialow-rankfactorization.MathematicalProgramming(SeriesB),95(2):329{357,February2003.Softwareavailableathttp://dollar.biz.uiowa.edu/~sburer.[2]StefanA.Burr.1976.ResultmentionedinbookbyM.GareyandD.Johnson.ComputersandIntractability:AGuidetotheTheoryofNP-Completeness,1979.W.H.FreemanandCompany.[3]ClaytonW.Commander.MaximumCutProblem,MAX-CUT.InChristodoulosFloudasandPanosPardalos,editors,Encyclo-pediaofOptimization,pages1991{1999.Springer,secondedi-tion,2009.[4]AndrzejDudekandVojtechRodl.OntheFolkmanNumberf(2;3;4).ExperimentalMathematics,17(1):63{67,2008.[5]PaulErd}osandAndrasHajnal.Researchproblem2{5.JournalofCombinatorialTheory,2:104,1967.[6]JonFolkman.Graphswithmonochromaticcompletesubgraphsineveryedgecoloring.SIAMJournalonAppliedMathematics,18(1):19{24,January1970.[7]PeterFranklandVojtechRodl.Largetriangle-freesubgraphsingraphswithoutK4.GraphsandCombinatorics,2:135{144,1986.10 [18]KonradPiwakowski,Stanis lawP.Radziszowski,andSebastianUrbanski.ComputationoftheFolkmanNumberFe(3;3;5).JournalofGraphTheory,32:41{49,1999.[19]Stanis lawP.Radziszowski.SmallRamseyNumbers.ElectronicJournalofCombinatorics,August2011.DynamicSurvey1,Revision#13.http://www.combinatorics.org.[20]Stanis lawP.RadziszowskiandXiaodongXu.OntheMostWantedFolkmanGraph.Geombinatorics,16(4):367{381,2007.[21]MarcusSchaefer.GraphRamseyTheoryandthePolynomialHierarchy.JournalofComputerandSystemSciences,62:290{322,2001.[22]JoelSpencer.Threehundredmillionpointssuce.JournalofCombinatorialTheory,SeriesA,49(2):210{217,1988.AlsoseeerratumbyM.HoveyinVol.50,p.323.12