Download pdf Fundamentals of computation theory 23rd international symposium fct 2021 athens greece

Page 1


Fundamentals of Computation Theory

23rd International Symposium FCT

2021 Athens Greece September 12 15

2021 Proceedings Lecture Notes in Computer Science 12867 Evripidis Bampis

(Editor)

Visit to download the full and correct content document: https://ebookmeta.com/product/fundamentals-of-computation-theory-23rd-internationa l-symposium-fct-2021-athens-greece-september-12-15-2021-proceedings-lecture-not es-in-computer-science-12867-evripidis-bampis-editor/

More products digital (pdf, epub, mobi) instant download maybe you interests ...

Computer Algebra in Scientific Computing 23rd

International Workshop CASC 2021 Sochi Russia September 13 17 2021 Proceedings Lecture Notes in Computer Science 12865 François Boulier (Editor)

https://ebookmeta.com/product/computer-algebra-in-scientificcomputing-23rd-international-workshop-casc-2021-sochi-russiaseptember-13-17-2021-proceedings-lecture-notes-in-computerscience-12865-francois-boulier-editor/

Learning and Intelligent Optimization 15th

International Conference LION 15 Athens Greece June 20 25 2021 Revised Selected Papers Lecture Notes in Computer Science Dimitris E. Simos (Editor)

https://ebookmeta.com/product/learning-and-intelligentoptimization-15th-international-conference-lion-15-athens-greecejune-20-25-2021-revised-selected-papers-lecture-notes-incomputer-science-dimitris-e-simos-editor/

Reversible Computation 13th International Conference RC 2021 Virtual Event July 7 8 2021 Proceedings Lecture Notes in Computer Science 12805 Shigeru Yamashita (Editor)

https://ebookmeta.com/product/reversible-computation-13thinternational-conference-rc-2021-virtual-eventjuly-7-8-2021-proceedings-lecture-notes-in-computerscience-12805-shigeru-yamashita-editor/

Static Analysis 28th International Symposium SAS 2021 Chicago IL USA October 17 19 2021 Proceedings Lecture Notes in Computer Science Cezara Dr■goi (Editor)

https://ebookmeta.com/product/static-analysis-28th-internationalsymposium-sas-2021-chicago-il-usa-october-17-19-2021-proceedingslecture-notes-in-computer-science-cezara-dragoi-editor/

Intelligent Computing Theories and Application 17th

International Conference ICIC 2021 Shenzhen China

August 12 15 2021 Proceedings Part III Lecture Notes in Computer Science De-Shuang Huang (Editor)

https://ebookmeta.com/product/intelligent-computing-theories-andapplication-17th-international-conference-icic-2021-shenzhenchina-august-12-15-2021-proceedings-part-iii-lecture-notes-incomputer-science-de-shuang-huang-editor/

Advances in Visual Computing 16th International

Symposium ISVC 2021 Virtual Event October 4 6 2021

Proceedings Part I Lecture Notes in Computer Science

George Bebis

https://ebookmeta.com/product/advances-in-visual-computing-16thinternational-symposium-isvc-2021-virtual-eventoctober-4-6-2021-proceedings-part-i-lecture-notes-in-computerscience-george-bebis/

Intelligent Computing Theories and Application 17th

International Conference ICIC 2021 Shenzhen China

August 12 15 2021 Proceedings Part I Lecture Notes in Computer Science Book 12836 2nd Edition Katharina Rein

https://ebookmeta.com/product/intelligent-computing-theories-andapplication-17th-international-conference-icic-2021-shenzhenchina-august-12-15-2021-proceedings-part-i-lecture-notes-incomputer-science-book-12836-2nd-edition-kat/

Computational Science and Its Applications ICCSA 2021

21st International Conference Cagliari Italy September 13 16 2021 Proceedings Part IV Lecture Notes in Computer Science Osvaldo Gervasi (Editor)

https://ebookmeta.com/product/computational-science-and-itsapplications-iccsa-2021-21st-international-conference-cagliariitaly-september-13-16-2021-proceedings-part-iv-lecture-notes-incomputer-science-osvaldo-gervasi-editor/

Computational Science and Its Applications ICCSA 2021

21st International Conference Cagliari Italy September 13 16 2021 Proceedings Part V Lecture Notes in Computer Science 12953 Osvaldo Gervasi (Editor)

https://ebookmeta.com/product/computational-science-and-itsapplications-iccsa-2021-21st-international-conference-cagliariitaly-september-13-16-2021-proceedings-part-v-lecture-notes-incomputer-science-12953-osvaldo-gervasi-edi/

ARCoSS

Fundamentals of Computation Theory

23rd International Symposium, FCT 2021 Athens, Greece, September 12–15, 2021 Proceedings

LectureNotesinComputerScience12867

FoundingEditors

GerhardGoos,Germany

JurisHartmanis,USA

EditorialBoardMembers

ElisaBertino,USA

WenGao,China

BernhardSteffen ,Germany

GerhardWoeginger ,Germany

MotiYung,USA

AdvancedResearchinComputingandSoftwareScience

SublineofLectureNotesinComputerScience

SublineSeriesEditors

GiorgioAusiello, UniversityofRome ‘LaSapienza’,Italy

VladimiroSassone, UniversityofSouthampton,UK

SublineAdvisoryBoard

SusanneAlbers, TUMunich,Germany

BenjaminC.Pierce, UniversityofPennsylvania,USA

BernhardSteffen , UniversityofDortmund,Germany

DengXiaotie, PekingUniversity,Beijing,China

JeannetteM.Wing, MicrosoftResearch,Redmond,WA,USA

Moreinformationaboutthissubseriesat http://www.springer.com/series/7407

EvripidisBampis • ArisPagourtzis(Eds.)

Fundamentals ofComputationTheory

23rdInternationalSymposium,FCT2021

Athens,Greece,September12–15,2021

Proceedings

Editors

EvripidisBampis

SorbonneUniversity

Paris,France

ISSN0302-9743ISSN1611-3349(electronic) LectureNotesinComputerScience

ISBN978-3-030-86592-4ISBN978-3-030-86593-1(eBook) https://doi.org/10.1007/978-3-030-86593-1

LNCSSublibrary:SL1 – TheoreticalComputerScienceandGeneralIssues

© SpringerNatureSwitzerlandAG2021

Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartofthe materialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation, broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodologynow knownorhereafterdeveloped.

Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant protectivelawsandregulationsandthereforefreeforgeneraluse.

Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbookare believedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsortheeditors giveawarranty,expressedorimplied,withrespecttothematerialcontainedhereinorforanyerrorsor omissionsthatmayhavebeenmade.Thepublisherremainsneutralwithregardtojurisdictionalclaimsin publishedmapsandinstitutionalaffiliations.

ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland

Preface

The23rdInternationalSymposiumonFundamentalsofComputationTheory(FCT 2021)washostedvirtuallybytheNationalTechnicalUniversityofAthensduetothe COVID-19pandemicduringSeptember12–15,2021.TheSymposiumonFundamentalsofComputationTheory(FCT)wasestablishedin1977forresearchersinterestedinallaspectsoftheoreticalcomputerscienceandinparticularalgorithms, complexity,andformalandlogicalmethods.FCTisabiennialconference.Previous symposiahavebeenheldinPoznan(Poland,1977),Wendisch-Rietz(Germany,1979), Szeged(Hungary,1981),Borgholm(Sweden,1983),Cottbus(Germany,1985),Kazan (Russia,1987),Szeged(Hungary,1989),Gosen-Berlin(Germany,1991),Szeged (Hungary,1993),Dresden(Germany,1995),Krakow(Poland,1997),Iasi(Romania, 1999),Riga(Latvia,2001),Malmö (Sweden,2003),Lübeck(Germany,2005), Budapest(Hungary,2007),Wroclaw(Poland,2009),Oslo(Norway,2011),Liverpool (UK,2013),Gdansk(Poland,2015),Bordeaux(France,2017),andCopenhagen (Denmark,2019).

TheProgramCommittee(PC)ofFCT2021received94submissions.EachsubmissionwasreviewedbyatleastthreePCmembersandsometrustedexternal reviewers,andevaluatedonitsquality,originality,andrelevancetothesymposium. ThePCselected30papersforpresentationattheconferenceandinclusioninthese proceedings.

FourinvitedtalksweregivenatFCT2021byConstantinosDaskalakis(MassachusettsInstituteofTechnology,USA),DanielMarx(MaxPlanckInstitutefor Informatics,Germany),ClaireMathieu(CNRSandUniversityofParis,France),and NobukoYoshida(ImperialCollege,UK).DavidRicherby(UniversityofEssex,UK) offeredaninvitedtutorial.

Thisvolumecontains,inadditiontothe30acceptedregularpapers,thepapers oftheinvitedtalksofClaireMathieuandNobukoYoshida,theabstractsoftheinvited talksofConstantinosDaskalakisandDanielMarx,andtheabstractoftheinvited tutorialofDavidRicherby.

TheProgramCommitteeselectedonecontributionforthebestpaperawardandtwo contributionsforthebeststudentpaperawards,allsponsoredbySpringer:

ThebestpaperawardwenttoMarcNeveling,JörgRothe,andRobinWeishauptfor theirpaper “ThePossibleWinnerProblemwithUncertainWeightsRevisited.”

Twopaperssharedthebeststudentpaperaward:(a) “FasterFPTAlgorithmsfor DeletiontoPairsofGraphClasses” byAshwinJacob,DiptapriyoMajumdar,and VenkateshRaman,and(b) “OnFindingSeparatorsinTemporalSplitandPermutationGraphs” byNicolasMaack,HendrikMolter,RolfNiedermeier,andMalte Renken.

WethanktheSteeringCommitteeanditschair,MarekKarpinski,forgivingusthe opportunitytoserveastheprogramchairsofFCT2021,andfortrustinguswiththe

responsibilitiesofselectingtheProgramCommittee,theconferenceprogram,and publications.

Wewouldliketothankalltheauthorswhorespondedtothecallforpapers,the invitedspeakers,themembersoftheProgramCommittee,andtheexternalreviewers fortheirdiligentworkinevaluatingthesubmissionsandfortheircontributionstothe electronicdiscussions.WewouldalsoliketothankthemembersoftheOrganizing CommitteeandthemembersoftheLocalArrangementsteamforthegreatjobthey havedone;specialthanksgotoDimitrisFotakis,IoannaProtekdikou,andAntonis Antonopoulos.

WewouldliketothankSpringerforpublishingtheproceedingsofFCT2021in theirARCoSS/LNCSseriesandfortheirsponsoringofthebestpaperawards.Weare thankfultothemembersoftheEditorialBoardof LectureNotesinComputerScience andtheeditorsatSpringerfortheirhelpthroughoutthepublicationprocess.Wealso acknowledgesupportfromtheInstituteofCommunicationandComputerSystems oftheSchoolofElectricalandComputerEngineeringoftheNationalTechnical UniversityofAthens,towardscoveringteleconferenceexpensesandregistrationcosts foranumberofstudents.Sponsorsthatprovidedsupportafterthepreparationofthese proceedingsappearonthewebpageoftheconference: https://www.corelab.ntua.gr/ fct2021/

TheEasyChairconferencesystemwasusedtomanagetheelectronicsubmissions, thereviewprocess,andtheelectronicProgramCommitteediscussions.Itmadeour taskmucheasier.

ThisvolumeisdedicatedtothefondmemoryofourfriendandcolleagueYannis Manoussakis,ProfessoratUniversityofParis-Saclay,France.Yannis,aspecialistin graphtheory,unexpectedlypassedawayearlierthisyearinhisbelovedhometownon Crete.Wewillalwaysrememberhimforhisopenheartandhisgreatpassionfor theoreticalcomputerscience.

July2021EvripidisBampis ArisPagourtzis

Organization

SteeringCommittee

BogdanChlebusAugustaUniversity,USA

MarekKarpinski(Chair)UniversityofBonn,Germany

AndrzejLingasLundUniversity,Sweden

MiklosSanthaCNRSandUniversityParisDiderot,France

EliUpfalBrownUniversity,USA

ProgramCommittee

EvripidisBampis(Co-chair)SorbonneUniversity,France

PetraBerenbrinkUniversityofHamburg,Germany

ArnaudCasteigtsUniversityofBordeaux,France

MarekChrobakUniversityofCalifornia,Riverside,USA

HansvanDitmarschCNRSandUniversityofLorraine,France

ThomasErlebachUniversityofLeicester,UK

BrunoEscoffierSorbonneUniversity,France

HenningFernauUniversityofTrier,Germany

DimitrisFotakisNationalTechnicalUniversityofAthens,Greece PierreFraigniaudCNRSandUniversityofParis,France

LeszekGasieniecUniversityofLiverpool,UK,andAugustaUniversity, USA

LaurentGourvesCNRSandParisDauphineUniversity,France GiuseppeF.ItalianoLUISSGuidoCarliUniversity,Italy

RalfKlasingCNRSandUniversityofBordeaux,France

AlexanderKononovSobolevInstituteofMathematicsandNovosibirskState University,Russia

AntoninKuceraMasarykUniversity,CzechRepublic

DietrichKuskeTechnischeUniversitätIlmenau,Germany

NikosLeonardosNationalandKapodistrianUniversityofAthens, Greece

MinmingLiCityUniversityofHongKong,HongKong ZsuzsannaLiptakUniversityofVerona,Italy

GiorgioLucarelliUniversityofLorraine,France

VangelisMarkakisAthensUniversityofEconomicsandBusiness,Greece NicoleMegowUniversityofBremen,Germany

AndrzejMurawskiUniversityofOxford,UK

ArisPagourtzis(Co-chair)NationalTechnicalUniversityofAthens,Greece CharisPapadopoulosUniversityofIoannina,Greece

IgorPotapovUniversityofLiverpool,UK

TomaszRadzikKing’sCollegeLondon,UK

MariaSernaUniversitatPolitecnicadeCatalunya,Spain

HadasShachnaiTechnion,Israel

VorapongSuppakitpaisarnUniversityofTokyo,Japan

NikosTzevelekosQueenMaryUniversityofLondon,UK

GuochuanZhangZhejiangUniversity,China

OrganizingCommittee

DimitrisFotakis(Co-chair)NationalTechnicalUniversityofAthens,Greece NikosLeonardosNationalandKapodistrianUniversityofAthens, Greece

ThanasisLianeasNationalTechnicalUniversityofAthens,Greece ArisPagourtzis(Co-chair)NationalTechnicalUniversityofAthens,Greece

AdditionalReviewers

FaisalAbu-Khzam

IoannisAnagnostides

AntonisAntonopoulos

AndreiAsinowski

MaxBannach

RémyBelmonte

NathalieBertrand

René VanBevern

ThereseBiedl

FelixBiermeier

DavideBilò

AhmadBiniaz

JohannaBjörklund

BenediktBollig

MichaëlCadilhac

OlivierCarton

ArmandoCastaneda

PyrrosChaidos

SankardeepChakraborty

PierreCharbit

HunterChase

VincentChau

LeroyChew

DmitryChistikov

DimitriosChristou

FerdinandoCicalese

FlorenceClerc

BrunoCourcelle

GeoffreyCruttwell

DominikD.Freydenberger

ClémentDallard

MinatiDe

YichaoDuan

SwanDubois

PavlosEfraimidis

MatthiasEnglert

LeahEpstein

VincentFagnon

QilongFeng

IreneFinocchi

FlorentFoucaud

ShayanGarani

RanGelles

MariosGeorgiou

ArchontiaGiannopoulou

AndreasGöbel

StefanGöller

RaduGrigore

NathanGrosshans

HermannGruber

LongkunGuo

SiddharthGupta

AnthonyGuttmann

ChristophHaase

ChristopherHahn

AraratHarutyunyan

StefanHoffmann

MarkusHolzer

HamedHosseinpour

LarsJaffke

ŁukaszJeż

SeungbumJo

JeromeJochems

DominikKaaser

AlkisKalavasis

PhillipKeldenich

EvgenyKiktenko

EunJungKim

JurijsKirillovs

BjørnKjos-Hanssen

OndřejKlíma

Sang-KiKo

TomaszKociumaka

AthanasiosKonstantinidis

RichardKorf

EvangelosKosinas

ArielKulik

ArnaudLabourel

DanielLeivant

OndrejLengal

JérômeLeroux

Yu-YangLin

ZhenweiLiu

ChristofLöding

DiptapriyoMajumdar

PasqualeMalacaria

AndreasMaletti

FlorinManea

NikolaosMelissinos

WolfgangMerkle

OthonMichail

SarahMinich

HendrikMolter

TobiasMömke

ClementMommessin

NelmaMoreira

AchourMostéfaoui

LarsNagel

VasileiosNakos

ReinoNiskanen

JanaNovotna

ThomasNowak

JanObdrzalek

PauloOliva

YotaOtachi

DominikPajak

PanagiotisPatsilinakos

SimonPerdrix

GiovanniPighizzini

GiuseppePrencipe

GabrielePuppis

SrinivasaRao-Satti

MikhailRaskin

DivyaRavi

RostislavRazumchik

AndrewRechnitzer

VojtechRehak

RebeccaReiffenhäuser

MarcRenault

ViniciusF.dosSantos

RamprasadSaptharishi

JayalalSarma

ChristianScheideler

DanielSchmand

SylvainSchmitz

VladimirShenmaier

SebastianSiebertz

BlerinaSinaimeri

MalteSkambath

DmitrySokolov

SrikanthSrinivasan

KonstantinosStavropoulos

FrankStephan

GerthStølting-Brodal

YuichiSudo

TillTantau

KonstantinosTsakalidis

ArtemTsikiridis

KeiUchizawa

WalterUnger

YuanhaoWei

AndreasWiese

xOrganization

NicolasWieseke

KyrillWinkler

PetraWolf

LirongXia

ChenyangXu

KuanYang

WeiYu TomvanderZanden

JingruZhang

PengZhang

RuilongZhang

XuZijian

PlenaryTalks

Min-MaxOptimization:FromvonNeumann toDeepLearningPlenaryTalks

MassachusettsInstituteofTechnology,Cambridge,MA,USA

Abstract. DeepLearningapplications,suchasGenerativeAdversarialNetworksandotheradversarialtrainingframeworks,motivatemin-maximizationof nonconvex-nonconcaveobjectives.Unliketheirconvex-concavecounterparts, however,forwhichamultitudeofequilibriumcomputationmethodsare available,nonconvex-nonconcaveobjectivesposesignificantoptimization challenges.Gradient-descentbasedmethodscommonlyfailtoidentifyequilibria,andevencomputinglocalapproximateequilibriahasremaineddaunting. Weshedlightonthischallengethroughacombinationofcomplexity-theoretic, game-theoreticandtopologicaltechniques,presentingobstaclesandopportunitiesforDeepLearningandGameTheorygoingforward.

(ThistalkisbasedonjointworkswithNoahGolowich,StratisSkoulakisand ManolisZampetakis)

TightComplexityResultsforAlgorithmsUsing TreeDecompositions

CISPAHelmholtzCenterforInformationSecurity,Saarbrücken,Germany

Abstract. Itiswellknownthathardalgorithmicproblemsongraphsareeasier tosolveifwearegivenalow-widthtreecompositionoftheinputgraph.For manyproblems,ifatreedecompositionofwidthkisavailable,algorithmswith runningtimeoftheformf(k)*poly(n)areknown;thatis,theproblemis fixed-parametertractable(FPT)parameterizedbythewidthofthegiven decomposition.Butwhatisthebestpossiblefunctionf(k)insuchanalgorithm? Inthepastdecade,aseriesofnewupperandlowerboundsgaveusatight understandingofthisquestionforparticularproblems.Thetalkwillgivea surveyoftheseresultsandsomenewdevelopments.

TheComplexityofCountingProblems (Tutorial)

UniversityofEssex,Colchester,UK

Abstract. Everycomputationaldecisionproblem(“IsthereanX?”)hasanaturalcountingvariant(“HowmanyX’sarethere?”).Moregenerally,computing weightedsumssuchasintegrals,expectationsandpartitionfunctionsinstatisticalphysicscanalsobeseenascountingproblems.

Thistutorialwillgiveanintroductiontothecomplexityofsolvingcounting problems,bothexactlyandapproximately.Iwillfocusonvariantsofconstraint satisfactionproblems.Thesearepowerfulenoughtonaturallyexpressmany importantproblems,butalsobeingrestrictedenoughtoallowtheircomputationalcomplexitytobeclassifiedcompletelyandelegantly.Nopriorknowledge ofcountingproblemswillbeassumed.

Contents

InvitedPapers

Two-SidedMatchingMarketswithStronglyCorrelatedPreferences........3 HugoGimbert,ClaireMathieu,andSimonMauras

CommunicatingFiniteStateMachinesandanExtensibleToolchain forMultipartySessionTypes..................................18

NobukoYoshida,FangyiZhou,andFranciscoFerreira

ContributedPapers

First-OrderLogicandItsInfinitaryQuantifierExtensionsover CountableWords..........................................39

BharatAdsul,SaptarshiSarkar,andA.V.Sreejith

FromSymmetrytoAsymmetry:GeneralizingTSPApproximations byParametrization..........................................53

LukasBehrendt,KatrinCasel,TobiasFriedrich, J.A.GregorLagodzinski,AlexanderLöser,andMarcusWilhelm

APoly-logCompetitivePosted-PriceAlgorithmforOnlineMetrical MatchingonaSpider.......................................67 MaxBender,JacobGilbert,andKirkPruhs

ComputationalComplexityofCoveringDisconnectedMultigraphs........85 JanBok,Jiří Fiala,NikolaJedličková,JanKratochvíl, andMichaelaSeifrtová

TheComplexityofBicriteriaTree-Depth..........................100 PiotrBorowiecki,DariuszDereniowski,andDorotaOsula

TS-ReconfigurationofDominatingSetsinCircleandCircular-ArcGraphs...114 NicolasBousquetandAliceJoffard

Bipartite3-RegularCountingProblemswithMixedSigns..............135 Jin-YiCai,AustenZ.Fan,andYinLiu

TheSatisfiabilityProblemforaQuantitativeFragmentofPCTL..........149 MiroslavChodilandAntonínKučera

BeyondtheBESTTheorem:FastAssessmentofEulerianTrails..........162 AlessioConte,RobertoGrossi,GrigoriosLoukides,NadiaPisanti, SolonP.Pissis,andGiuliaPunzi

xviiiContents

Linear-TimeMinimalCographEditing............................176 ChristopheCrespelle

RegularModelCheckingwithRegularRelations....................190 VrundaDave,TaylorDohmen,ShankaraNarayananKrishna, andAshutoshTrivedi

MinimumConsistentSubsetProblemforTrees......................204 SanjanaDey,AnilMaheshwari,andSubhasC.Nandy

ParameterizedComplexityofFindingSubgraphswithHereditaryProperties onHereditaryGraphClasses..................................217 DavidEppstein,SiddharthGupta,andElhamHavvaei

TheSpaceComplexityofSumLabelling..........................230 HenningFernauandKshitijGajjar

OnMinimizingRegularExpressionsWithoutKleeneStar..............245 HermannGruber,MarkusHolzer,andSimonWolfsteiner

ComputationalComplexityofComputingaQuasi-ProperEquilibrium......259 KristofferArnsfeltHansenandTroelsBjerreLund

ComputationalComplexityofSynchronizationUnderSparse RegularConstraints.........................................272 StefanHoffmann

OnDasgupta’sHierarchicalClusteringObjectiveandItsRelationtoOther GraphParameters..........................................287

SveinHøgemo,BenjaminBergougnoux,UlrikBrandes, ChristophePaul,andJanArneTelle

MengerianTemporalGraphsRevisited...........................301 AllenIbiapinaandAnaSilva

FasterFPTAlgorithmsforDeletiontoPairsofGraphClasses...........314 AshwinJacob,DiptapriyoMajumdar,andVenkateshRaman

FastAlgorithmsfortheRootedTripletDistanceBetweenCaterpillars......327 JesperJanssonandWingLikLee

DecidingTop-DownDeterminismofRegularTreeLanguages...........341 PeterLeupoldandSebastianManeth

PropositionalGossipProtocols.................................354 JosephLiveseyandDominikWojtczak

ComplexityofWordProblemsforHNN-Extensions..................371 MarkusLohrey

OnFindingSeparatorsinTemporalSplitandPermutationGraphs.........385 NicolasMaack,HendrikMolter,RolfNiedermeier,andMalteRenken

ThePossibleWinnerProblemwithUncertainWeightsRevisited..........399 MarcNeveling,JörgRothe,andRobinWeishaupt

StreamingDeletionProblemsParameterizedbyVertexCover............413 JelleJ.OostveenandErikJanvanLeeuwen

OntheHardnessoftheDeterminant:SumofRegular Set-MultilinearCircuits......................................427 S.RajaandG.V.SumukhaBharadwaj

ConcentrationoftheCollisionEstimator..........................440 MaciejSkorski

Valency-BasedConsensusUnderMessageAdversariesWithout Limit-Closure.............................................457 KyrillWinkler,UlrichSchmid,andThomasNowak

AuthorIndex ............................................475

InvitedPapers

Two-SidedMatchingMarkets withStronglyCorrelatedPreferences

1 CNRS,LaBRI,Bordeaux,France

hugo.gimbert@cnrs.fr

2 CNRS,IRIF,Paris,France

Claire.Mathieu@irif.fr

3 Universit´edeParis,IRIF,Paris,France

simon.mauras@irif.fr

Abstract. Stablematchinginacommunityconsistingofmenand womenisaclassicalcombinatorialproblemthathasbeenthesubject ofintensetheoreticalandempiricalstudysinceitsintroductionin1962 inaseminalpaperbyGaleandShapley,whodesignedthecelebrated “deferredacceptance”algorithmfortheproblem.

Intheinput,eachparticipantranksparticipantsoftheoppositetype, sotheinputconsistsofacollectionofpermutations,representingthe preferencelists.Abipartitematchingisunstableifsomeman-womanpair isblocking:bothstrictlyprefereachothertotheirpartnerinthematching.Stabilityisanimportanteconomicsconceptinmatchingmarkets fromtheviewpointofmanipulability.Theunicityofastablematching impliesnon-manipulability,andnear-unicityimplieslimitedmanipulability,thusthesearemathematicalpropertiesrelatedtothequalityof stablematchingalgorithms.

Thispaperisatheoreticalstudyoftheeffectofcorrelationsonapproximatemanipulabilityofstablematchingalgorithms.Ourapproachisto gobeyondworstcase,assumingthatsomeoftheinputpreferencelistsare drawnfromadistribution.Approximatemanipulabilityisapproached fromseveralangles:whenallstablepartnersofapersonhaveapproximatelythesamerank;orwhenmostpersonshaveauniquestable partner.

1Introduction

Intheclassicalstablematchingproblem,acertaincommunityconsistsofmen andwomen(allheterosexualandmonogamous)whereeachpersonranksthoseof theoppositesexinaccordancewithhisorherpreferencesforamarriagepartner (possiblydeclaringsomematchesasunacceptable).Ourobjectiveistomarry offthemembersofthecommunityinsuchawaythattheestablishedmatching is stable, i.e. suchthatthereisno blockingpair.Amanandawomanwhoare notmarriedtoeachotherformablockingpairiftheyprefereachothertotheir mates.

Intheirseminalpaper,GaleandShapley[11]designedthe men-proposing deferredacceptance procedure,wheremenproposewhilewomendisposes.This c SpringerNatureSwitzerlandAG2021

E.BampisandA.Pagourtzis(Eds.):FCT2021,LNCS12867,pp.3–17,2021. https://doi.org/10.1007/978-3-030-86593-1 1

4H.Gimbertetal.

algorithmalwaysoutputsamatchingwhichisstable,optimalformenandpessimalforwomen(intermsofrankofeachperson’spartner).Bysymmetry,there alsoexistsawomen-optimal/men-pessimalstablematching.GaleandShapley’s originalmotivationwastheassignmentofstudentstocolleges,asettingtowhich thealgorithmandresultsextend,andtheirapproachwassuccessfullyimplementedinmanymatchingmarkets;seeforexample[1, 2, 8, 29].

However,thereexistsinstanceswherethemen-optimalandwomen-optimal stablematchingsaredifferent,andevenextremecasesofinstancesinwhichevery man/womanpairbelongstosomestablematching.Thisraisesthequestionof whichmatchingtochoose[14, 15]andofpossiblestrategicbehavior[9, 10, 28].

Moreprecisely,ifawomanliesaboutherpreferencelist,thisgivesrisetonew stablematchings,whereshewillbenobetteroffthanshewouldbeinthe truewomen-optimalmatching.Thus,awomancanonlygainfromstrategic manipulationuptothemaximumdifferencebetweenherbestandworstpartners instablematchings.Bysymmetry,thisalsoimpliesthatthemenproposing deferredacceptanceprocedureisstrategy-proofformen(astheywillgettheir bestpossiblepartnerbytellingthetruth).

Fortunately,thereisempiricalevidencethatinmanyinstances,inpractice thestablematchingisessentiallyunique(aphenomenonoftenreferredtoas “core-convergence”);seeforexample[6, 16, 23, 29].Oneoftheempiricalexplanationsforcore-convergencegivenbyRothandPeransonin[29]isthatthe preferencelistsarecorrelated: “Onefactorthatstronglyinfluencesthesizeof thesetofstablematchingsisthecorrelationofpreferencesamongprogramsand amongapplicants.Whenpreferencesarehighlycorrelated(i.e.,whensimilarprogramstendtoagreewhicharethemostdesirableapplicants,andapplicantstend toagreewhicharethemostdesirableprograms),thesetofstablematchingsis small.”

Followingthatdirectionofenquiry,westudythecore-convergencephenomenon,inamodelwherepreferencesarestochastic.Whenpreferencesof womenarestronglycorrelated,Theorem 1 showsthattheexpecteddifferenceof rankbetweeneachwoman’sworstandbeststablepartnerisaconstant,hence theincentivestomanipulatearelimited.Ifadditionallythepreferencesofmen areuncorrelated,Theorem 2 showsthatmostwomenhaveauniquestablepartner,andthereforehavenoincentivestomanipulate.

1.1DefinitionsandMainTheorems

Matchings. Let M = {m1 ,...,mM } beasetof M men, W = {w1 ,...,wW } be asetof W women,and N =min(M,W ).Inamatching,eachpersoniseither single,ormatchedwithsomeoneoftheoppositesex.Formally,weseeamatching asafunction μ : M∪W→M∪W ,whichisself-inverse(μ2 =Id),where eachman m ispairedeitherwithawomanorhimself(μ(m) ∈W∪{m}),and symmetrically,eachwoman w ispairedwithamanorherself(μ(w ) ∈M∪{w }). PreferenceLists. Eachpersondeclareswhichmembersoftheoppositesexthey findacceptable,thengivesastrictlyorderedpreferencelistofthosemembers.

Two-SidedMatchingMarketswithStronglyCorrelatedPreferences5

Preferencelistsare complete whennooneisdeclaredunacceptable.Formally,we representthepreferencelistofaman m asatotalorder m over W∪{m},where w m m meansthatman m findswoman w acceptable,and w m w means thatman m preferswoman w towoman w .Similarlywedefinethepreference list w ofwoman w .

Stability. Aman-womanpair(m,w )isblockingamatching μ when m w μ(w ) and w m μ(m).Abusingnotations,observethat μ matchesaperson p with anunacceptablepartnerwhen p wouldprefertoremainsingle,thatiswhenthe pair(p,p)isblocking.Amatchingwithnoblockingpairisstable.Astablepair isapairwhichbelongstoatleastonestablematching.

RandomPreferences. Weconsideramodelwhereeachperson’ssetofacceptable partnersisdeterministic,andorderingsofacceptablepartnersaredrawnindependentlyfrom regular distributions.Whenunspecified,someone’sacceptable partnersand/ortheirorderingis adversarial,thatischosenbyanadversary whoknowstheinputmodelbutdoesnotknowtheoutcomeoftherandomcoin flips.

Definition1(Regulardistribution). Adistributionofpreferenceslistsis regular whenforeverysequenceofacceptablepartners a1 ,...,ak wehave P[a1

Intuitively,knowingthat a2 isrankedwellonlydecreasestheprobability that a1 beats a2 .Mostprobabilitydistributionsthathavebeenstudiedareregular.Inparticular,sortingacceptablepartnersbyscores(drawnindependently fromdistributionson R),yieldsaregulardistribution.Asanexampleofregulardistribution,westudypopularitypreferences,introducedbyImmorlicaand Madhian[17].

Definition2(Popularitypreferences). Whenawoman w has popularity preferences,shegivesapositivepopularity Dw (m) toeachacceptablepartner m. Wesee Dw asadistributionoverheracceptablepartners,scaledsothatitsums to1.Sheusesthisdistributiontodrawherfavouritepartner,thenhersecond favourite,andsoonuntilherleastfavouritepartner.

ThefollowingTheoremshowsthatundersomeassumptionseverywoman givesapproximatelythesameranktoallofherstablepartners.

Theorem1. Assumethateachwomanindependentlydrawsherpreferencelist fromaregulardistribution.Themen’spreferencelistsarearbitrary.Let uk be anupperboundontheoddsthatman mi+k isrankedbeforeman mi : ∀k ≥ 1,uk =max w,i P[mi+k w mi ] P[mi w mi+k ] w findsboth mi and mi+k acceptable

Thenforeachwomanwithatleastonestablepartner,inexpectationallof herstablepartnersarerankedwithin (1+2exp( k ≥1 kuk )) k ≥1 k 2 uk ofone anotherinherpreferencelist.

6H.Gimbertetal.

Theorem 1 ismostrelevantwhenthewomen’spreferencelistsarestrongly correlated,thatis,wheneverywoman’spreferencelistis“close”toasingle ranking m1 m2 ... mM .Thisclosenessismeasuredbytheoddsthatin someranking,somemanisrankedaheadofamanwho,intheranking m1 m2 ... mM ,wouldbe k slotsaheadofhim.

Wedetailbelowthreeexamplesofapplications,wheretheexpecteddifference ofranksbetweeneachwoman’sbestandworstpartnersis O (1),andthusher incentivestomisreportherpreferencesarelimited.

– Identicalpreferences. Ifallwomenranktheiracceptablepartnersusingamasterlist m1 m2 ··· mM ,thenall uk ’sareequalto0.ThenTheorem 1 statesthateachwomanhasauniquestablehusband,awell-knownresultfor thistypeofinstances.

– Preferencesfromidenticalpopularities. Assumethatwomenhavepopularity preferences(Definition 2)andthateachwomangivesman mi popularity2 i . Then uk =2 k andtheexpectedrankdifferenceisatmost O (1).

– Preferencesfromcorrelatedutilities. Assumethatwomenhavesimilarpreferences:eachwoman w givesman mi ascorethatisthesumofacommonvalue i andanidiosyncraticvalue η w i whichisnormallydistributed withmean0andvariance σ 2 ;shethensortsmenbyincreasingscores.Then uk ≤ maxw,i {2 P[η w i η w i+k >k ]}≤ 2e (k/2σ )2 andtheexpectedrank difference,byashortcalculation,isatmost4√πσ 3 (1+2e4σ 2 )= O (1).

Astrongernotionofapproximateincentivecompatibilityisnear-unicityofa stablematching,meaningthatmostpersonshaveeithernooroneuniquestable partner,andthushavenoincentivetomisreporttheirpreferences.Whendoes thathold?OneanswerisgivenbyTheorem 2.

Theorem2. Assumethateachwomanindependentlydrawsherpreferencelist fromaregulardistribution.Let uk beanupperboundontheoddsthatman mi+k isrankedbeforeman mi :

∀k ≥ 1,uk =max w,i

P[mi+k w mi ]

P

Furtherassumethatallpreferencesarecomplete,that uk =exp( Ω (k )),and thatmenhaveuniformlyrandompreferences.Then,inexpectationthefraction ofpersonswhohavemultiplestablepartnersconvergesto0.

NoticethatinthethreeexamplesofTheorem 1,thesequence(uk )k ≥1 is exponentiallydecreasing.TheassumptionsofTheorem 2 areminimalinthe sensethatremovingonewouldbringusbacktoacasewhereaconstantfraction ofwomanhavemultiplestablepartners.

– Preferencelistsofwomen. Ifweremovetheassumptionthat uk isexponentiallydecreasing,theconclusionnolongerholds:considerabalancedmarket balanced(M = W )inwhichbothmenandwomenhavecompleteuniformly randompreferences;thenmostwomenhave ∼ ln N stablehusbands[19, 25].

Two-SidedMatchingMarketswithStronglyCorrelatedPreferences7

Preferencelistsofmen. Assumethatmenhaverandompreferencebuiltas follows:startingfromtheordering w1 ,w2 ,...,wM ,eachpair(w2i 1 ,w2i )is swappedwithprobability1/2,forall i.Asymmetricdefinitionforwomen’s preferencessatisfythehypothesisofTheorem 2,with u1 =1and uk =0for all k ≥ 2.Thenthereisa1/8probabilitythatmen m2i 1 and m2i areboth stablepartnersofwomen w2i 1 and w2i ,forall i,henceaconstantexpected fractionofpersonswithmultiplestablepartners. – Incompletepreferences. Consideramarketdividedintogroupsofsize4of theform {m2i 1 ,m2i ,w2i 1 ,w2i },whereamanandawomanaremutually acceptableiftheybelongtothesamegroup.Onceagain,withconstantprobability, m2i 1 and m2i arebothstablepartnersofwomen w2i 1 and w2i

1.2RelatedWork

Analyzinginstancesthatarelessfar-fetchedthanintheworstcaseisthemotivationunderlyingthemodelofstochasticallygeneratedpreferencelists.Aseries ofpapers[19, 22, 24–26]studythemodelwhere N menand N womenhavecompleteuniformlyrandompreferences.Asymptotically,andinexpectation,the totalnumberofstablematchingsis ∼ e 1 N ln N ,inwhichafixedwomanhas ∼ ln N stablehusbands,whereherbeststablehusbandhasrank ∼ ln N andher worststablehusbandhasrank ∼ N/ ln N .

Thefirsttheoreticalexplanationsofthe“core-convergence”phenomenon wheregivenin[17]and[4],invariationsofthestandarduniformmodel.ImmorlicaandMahdian[17]considerthecasewheremenhaveconstantsizerandom preferences(truncatedpopularitypreferences).Ashlagi,KanoriaandLeshno[4], considerslightlyunbalancedmatchingmarkets(M<W ).Botharticlesprove thatthefractionofpersonswithseveralstablepartnerstendsto0asthemarket growslarge.Theorem 2 anditsproofincorporateideasfromthosetwopapers. Beyondstrong“core-convergence”,wheremostagentshaveauniquestable partner,onecanboundtheutilitygainbymanipulatingastablemechanism. Lee[21]considersamodelwithrandomcardinalutilities,andshowsthatagents receivealmostthesameutilityinallstablematchings.Kanoria,MinandQian [18],andAshlagi,Braverman,ThomasandZhao[3]studytherankofeach person’spartner,underthemenandwomenoptimalstablematchings,asa functionofthemarketimbalanceandthesizeofpreferencelists[18],orasa functionofeachperson’s(bounded)popularity[3].Theorem 1 canbecompared withsuchresults.

Beyondone-to-onematchings,schoolchoiceisanexampleofmany-to-one markets.KojimaandPathak[20]generalizeresultsfrom[17]andprovethatmost schoolshavenoincentivestomanipulate.AzevedoandLeshno[5]showthatlarge marketsconvergetoauniquestablematchinginamodelwithacontinuumof students.Tocounterbalancethosefindings,Bir´o,Hassidim,RommandShorer [7],andRheingans-Yoo[27]arguethatsocioeconomicstatusandgeographic preferencesmightunderminecore-convergence,thussomeincentivesremainin suchmarkets.

2StronglyCorrelatedPreferences:ProofofTheorem 1

Theorem1. Assumethateachwomanindependentlydrawsherpreferencelist fromaregulardistribution.Themen’spreferencelistsarearbitrary.Let uk be anupperboundontheoddsthatman mi+k isrankedbeforeman mi :

Thenforeachwomanwithatleastonestablepartner,inexpectationallof herstablepartnersarerankedwithin (1+2exp(

ofone anotherinherpreferencelist.

InSubsect. 2.1,wedefineapartitionofstablematchinginstancesinto blocks. Forstronglycorrelatedinstances,blocksprovidethestructuralinsighttostart theanalysis:inLemma 3,weusethemtoupper-boundthedifferenceofranks betweenawoman’sworstandbeststablepartnersbythesumof(1)thenumber x ofmencomingfromotherblocksandwhoareplacedbetweenstablehusbands inthewoman’spreferencelist,and(2)theblocksize.

Theanalysisrequiresadelicatehandlingofconditionalprobabilities.InSubsect. 2.2,weexplainhowtoconditiononthemen-optimalstablematching,when preferencesarerandom.

Subsection 2.3 analyzes(1).Themeninvolvedareoutofplacecomparedto theirpositionintheranking m1 ... mM ,andtheoddsofsucheventscanbe bounded,thankstotheassumptionthatdistributionsofpreferencesareregular. OurmaintechnicallemmathereisLemma 4.

Subsection 2.4 analyzes(2),theblocksizebyfirstgivingasimplegreedy algorithm(Algorithm 2)tocomputeablock.Eachofthetwolimitsofablock iscomputedbyasequenceof“jumps”,sothetotaldistancetraveledisasum ofjumpswhich,thankstoLemma 4 again,canbestochasticallydominatedby asum X ofindependentrandomvariables(seeLemma 7);thusitallreducesto analyzing X ,asimplemathematicalexercise(Lemma 8).

Finally,Subsect. 2.5 combinestheLemmaspreviouslyestablishedtoprove Theorem 1.

OuranalysisbuildsonTheorems 1 and 2,twofundamentalandwell-known results.

Theorem1(Adaptedfrom [11]). Algorithm 1 outputsastablematching μM inwhicheveryman(resp.woman)hashisbest(resp.herworst)stablepartner. Symmetrically,thereexistsastablematching μW inwhicheverywoman(resp. man)hasherbest(resp.hisworst)stablepartner.

Theorem2(Adaptedfrom [12]). Eachpersoniseithermatchedinallstable matchings,orsingleinallstablematchings.Inparticular,awomanismatched inallstablematchingsifandonlyifshereceivedatleastoneacceptableproposal duringAlgorithm 1.

Two-SidedMatchingMarketswithStronglyCorrelatedPreferences9

Algorithm1. MenProposingDeferredAcceptance.

Input: Preferencesofmen( m )m∈M andwomen( w )w ∈W

Initialization: Startwithanemptymatching µ

While aman m issingleandhasnotproposedtoeverywomanhefindsacceptable, do m proposestohisfavoritewoman w hehasnotproposedtoyet. If m is w ’sfavoriteacceptablemanamongallproposalsshereceived, w accepts m’sproposal,andrejectsherprevioushusbandifshewasmarried.

Output: Resultingmatching.

2.1SeparatorsandBlocks

Inthissubsection,wedefinetheblockstructureunderlyingouranalysis.

Definition3(separator). A separator isaset S ⊆M ofmensuchthatinthe men-optimalstablematching μM ,eachwomanmarriedtoamanin S prefers himtoallmenoutside S :

Lemma1. Givenaseparator S ⊆M,everystablematchingmatches S tothe samesetofwomen.

Proof. Let w ∈ μM (S )andlet m bethepartnerof w insomestablematching. Since μM isthewoman-pessimalstablematchingbyTheorem 1, w prefers m to μM (w ).Bydefinitionofseparators,thatimpliesthat m ∈ S .Hence,inevery stablematching μ,womenof μM (S )arematchedtomenin S .Byacardinality argument,menof S arematchedby μ to μM (S ).

Definition4(prefixseparator,block). A prefixseparator isaseparator S suchthat S = {m1 ,m2 ,...,mt } forsome 0 ≤ t ≤ N .Givenacollectionof b +1 prefixseparators Si = {m1 ,...,mti } with 0= t0 <t1 < <tb = N ,the i-th block istheset Bi = Sti \ Sti 1 with 1 ≤ i ≤ b

Abusingnotations,wewilldenote S astheprefixseparator t and B asthe block (ti 1 ,ti ]

Lemma2. Givenablock B ⊆M,everystablematchingmatches B tothesame setofwomen.

Proof. B equals Sti \ Sti 1 forsome i.ApplyingLemma 1 to Sti andto Sti 1 provestheLemma.

Lemma3. Considerawoman wn whoismatchedby μM andlet B =(l,r ] denoteherblock.Let x denotethenumberofmenfromabetterblockthatare rankedby wn betweenamanof B and mn :

x = |{i ≤ l |∃j>l,mj wn mi wn mn }|

10H.Gimbertetal.

Thenin wn ’spreferencelist,thedifferenceofranksbetween wn ’sworstandbest stablepartnersisatmost x + r l 1.

Proof. Since μM iswoman-pessimalbyTheorem 1, mn isthelaststablehusband in wn ’spreferencelist.Let mj denoteherbeststablehusband. In wn ’spreferencelist,theintervalfrom mj to mn containsmenfromherown block,pluspossiblysomeadditionalmen.Suchaman mi comesfromoutside herblock(l,r ]andsheprefershimto mn :since r isaprefixseparator,wemust have i ≤ l .Thus x countsthenumberofmenwhodonotbelongtoherblock butwhoinherpreferencelistarerankedbetween mj and mn

Ontheotherhand,thenumberofmenwhobelongtoherblockandwhoin herpreferencelistarerankedbetween mj and mn (inclusive)isatmost r l

Together,thedifferenceofranksbetween wn ’sworstandbeststablepartners isatmost x +(r l ) 1.SeeFig. 1 foranillustration.

l =2, r =8and x =1

mi with i ≤ l mi with l<i ≤ r

mi with r<i

x = |{|∃ ,

Fig.1. Preferencelistof wn ,with n =6.Theblockof wn isdefinedbyaleftseparator at l =2andarightseparatorat r =8.Colorswhite,grayandblackcorrespondsto blocks,andaredefinedinthelegend.Allstablepartnersof wn mustbegray.Menin blackareallrankedafter mn = µM (wn ).Thedifferenceinrankbetween wn ’sworst andbestpartnerisatmostthenumberofgraymen(here r l =6),minus1,plusthe numberofwhitemenrankedafteragraymanandbefore mn (here x =1).

2.2ConditioningontheManOptimalStableMatchingWhen PreferencesAreRandom

Westudythecasewhereeachpersondrawsherpreferencelistfromanarbitrary distribution.Thepreferencelistsarerandomvariables,thatareindependentbut notnecessarilyidenticallydistributed.

Intuitively,weusethe principleofdeferreddecision andconstructpreferencelistsinanonlinemanner.ByTheorem 1 theman-optimalstablematching μM iscomputedbyAlgorithm 1,andtheremainingrandomnesscanbeused forastochasticanalysisofeachperson’sstablepartners.Tobemoreformal, wedefinearandomvariable H,andinspectionofAlgorithm 1 showsthat H containsenoughinformationoneachperson’spreferencestorunAlgorithm 1 deterministically.

Definition5. Let H =(μM , (σm )m∈M , (πw )w ∈W ) denotetherandomvariable consistingof(1)theman-optimalstablematching μM ,(2)eachman’sranking ofthewomenhepreferstohispartnerin μM ,and(3)eachwoman’srankingof themenwhopreferhertotheirpartnerin μM

2.3AnalyzingtheNumber x ofMenfromOtherBlocks

Lemma4. Recallthesequence (uk )k ≥1 definedinthestatementofTheorem 1:

∀k ≥ 1,uk =max w,i P[mi+k w mi ] P[mi w mi+k ] w findsboth mi and mi+k acceptable

Let w beawoman.Givenasubsetofheracceptablemenandarankingofthat subset a1 w ··· w ap ,weconditionontheeventthatin w ’spreferencelist, a1 w ··· w ap holds.Let mi = a1 be w ’sfavoritemaninthatsubset.Let Ji bearandomvariable,equaltothehighest j ≥ i suchthatwoman w prefers mj to mi .Formally, Ji =max{j ≥ i | mj w mi }.Then,forall k ≥ 1,wehave P[Ji <i + k |

Proof. Ji isdeterminedby w ’spreferencelist.Weconstruct w ’spreferencelist usingthefollowingalgorithm:initiallyweknowherranking σA ofthesubset A = {a1 ,a2 ,...,ap } ofacceptablemen,and mi = a1 isherfavoriteamong those.Foreach j from N to i indecreasingorder,weinsert mj intotheranking accordingtothedistributionof w ’spreferencelist,stoppingassoonassome mj isrankedbefore mi (orwhen j = i isthatdoesnothappen).Thenthestep j ≥ i atwhichthisalgorithmstopsequals Ji .

Toanalyzethealgorithm,observethatateachstep j = N,N 1,...,we alreadyknow w ’srankingofthesubset S = {mj +1 ,...,mN }∪{a1 ,...,ap }∪ {menwhoarenotacceptableto w }.If mj isalreadyin S , w prefers mi to mj , thusthealgorithmcontinuesand Ji <j .Otherwisethealgorithminserts mj intotheexistingranking:bydefinitionofregulardistributions(Definition 1), theprobabilitythat mj beats mi giventherankingconstructedsofarisatmost theunconditionalprobability P[mj w mi ].

Bydefinitionof uj i ,wehave1

uj i ) 1 ≥ exp( uj i )

Summingoverallrankings σS of S thatarecompatiblewith σA andwith Ji ≤ j , P[Ji <j | Ji ≤ j ]= σS compatiblewith Ji ≤j andwith σA

Finally, P[Ji <j ]= N

).

RecallfromLemma 3 that r l 1+ x isanupperboundonthedifferenceof rankofwoman wn ’sworstandbeststablehusbands.Wefirstboundtheexpected valueoftherandomvariable x definedinLemma 3

12H.Gimbertetal.

Lemma5. Givenawoman wn ,definetherandomvariable x asinLemma 3: conditioningon H, x = |{i ≤ l |∃j>l,mj wn mi wn mn }| isthenumber ofmeninabetterblock,whocanberankedbetween wn ’sworstandbeststable husbands.Then E[x] ≤ k ≥1 kuk .

Proof. Startbyconditioningon H,andlet mn = a1 w a2 w ··· w ap be wn ’s rankingofmenwhopreferhertotheirpartnerin μM .Wedrawthepreference listsofeachwoman wi with i<n,anduseAlgorithm 2 tocomputethevalue of l .

Foreach i ≤ l ,weproceedasfollows.If mn wn mi ,then mi cannotbe rankedbetween wn ’sworstandbeststablepartners.Otherwise,weareina situationwhere mi wn a1 wn ··· wn ap .UsingnotationsfromLemma 4, w prefers mi toall mj with j>l ifandonlyif Ji <l +1.ByLemma 4 thisoccurs withprobabilityatleastexp( k ≥l+1 i uk ).Thus

Summingthisprobabilityforall

2.4AnalyzingtheBlockSize

Lemma6. Consider wn whoismatchedby μM .ThenAlgorithm 2 outputsthe blockcontaining wn .

Algorithm2. Computingablock

Initialization:

Computethemanoptimalstablematching µM

Relabelwomensothat wi denotesthewifeof mi in µM

Pickawoman wn whoismarriedin µM

Leftprefixseparator: initialize l ← n 1 while thereexists i ≤ l and j>l suchthat mj wi mi : l ← min{i ≤ l |∃j>l,mj wi mi }− 1.

Rightprefixseparator: initialize r ← n. while thereexists j>r and i ≤ r suchthat mj wi mi : r ← max{j>r |∃i ≤ r,mj wi mi }.

Output: (l,r ].

Proof. Algorithm 2 isunderstoodmosteasilybyfollowingitsexecutiononFig. 2 Algorithm 2 appliesaright-to-leftgreedymethodtofindthelargestprefixseparator l whichis ≤ n 1.Bydefinitionofprefixseparators,awitnessthatsome t isnotaprefixseparatorisapair(mj ,wi )where j>t ≥ i andwoman wi prefersman mj toherpartner: mj >wi mi .Thenthesamepairalsocertifies thatno t = t,t 1,t 2,...,i canbeaprefixseparatoreither,sothealgorithm

Two-SidedMatchingMarketswithStronglyCorrelatedPreferences13

jumpsto i 1andlooksforawitnessagain.Whenthereisnowitness,aprefix separatorhasbeenfound,thus l isthelargestprefixseparator ≤ n 1.Similarly, Algorithm 2 computesthesmallestprefixseparator r whichis ≥ n.Thus,by definitionofblocks,(l,r ]istheblockcontaining wn .

Women

Fig.2. Computingtheblockcontainingwoman wn .Theverticalblackedgescorrespondtothemen-optimalstablematching µM .Thereisalightgrayarc(mj ,wi )if j>i andwoman wi prefersman mj toherpartner: mj wi mi .Theprefixseparatorscorrespondtothesolidredverticallineswhichdonotintersectanygrayarc. Algorithm 2 appliesaright-to-leftgreedymethodtofindthelargestprefixseparator l whichis ≤ n 1,jumpingfromdashedredlinetodashedredline,andasimilar left-to-rightgreedymethodagaintofindthesmallestprefixseparator r whichis ≥ n. Thisdeterminestheblock(l,r ]containing n.(Colorfigureonline)

Definition6. Let X betherandomvariabledefinedasfollows.Let (Δt )t≥0 denoteasequenceofi.i.d.r.v.’stakingnon-negativeintegervalueswiththefollowingdistribution:

Then X = Δ0 + Δ1 + + ΔT 1 ,where T isthefirst t ≥ 0 suchthat Δt =0

TheproofsofthefollowingLemmascanbefoundin[13].

Lemma7. Givenawoman wn ,let (l,r ] denotetheblockcontaining n.Conditioningon H, l and r areintegerrandomvariable,suchthat r n and n 1 l arestochasticallydominatedby X .

Lemma8. Wehave E[X ] ≤ exp( k ≥1 kuk ) k ≥1 k 2 u

2.5PuttingEverythingTogether

Proof(ProofofTheorem 1). Withoutlossofgenerality,wemayassumethat N = M ≤ W andthateachmanismatchedintheman-optimalstablematching μM :toseethat,foreachman m weadda“virtual”woman w ashisleastfavorite acceptablepartner,suchthat m istheonlyacceptablepartnerof w .Amanis

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.