




版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請(qǐng)進(jìn)行舉報(bào)或認(rèn)領(lǐng)
文檔簡介
Contentslistsavailableat
ScienceDirect
ExpertSystemsWithApplications
journalhomepage:
/locate/eswa
ExpertSystemsWithApplications225(2023)120151
Maximumflowaccelerationbytraversingtreebasedtwo-boundarygraph contraction
WeiWei
a
,
b
,PengpengWang
a
,
b
,YaboDong
c
,
?
aKeyLaboratoryofGrainInformationProcessingandControl(HenanUniversityofTechnology),MinistryofEducation,Zhengzhou450001,China
bHenanProvincialKeyLaboratoryofGrainPhotoelectricDetectionandControl,HenanUniversityofTechnology,Zhengzhou450001,China
cCollegeofcomputerscienceandtechnology,ZhejiangUniversity,Hangzhou310027,China
ARTICLEINFO ABSTRACT
Keywords:
Tree-cutmappingBoundarynodeMaximumflowGraphcontractionOverlay
Themaximumflow(max-flow)problemissofundamentalthatthecorrespondingalgorithmshavemanyapplicationsinmanyscenarios.Acommonaccelerationstrategyistoreducegraphsizebycontractingsubgraphs,buttherearefewcandidatesubgraphswithnosizelimitthatcanguaranteeexactmax-flowsolutionsaftercontraction.Weaddanewsubgraphcontractionconditiontotheexistingconditionswithcorrespondingefficientlocatingalgorithm,andweproposeamax-flowaccelerationalgorithmusingtraversingtree-basedcontraction(TTCMax).TTCMaxcanworkefficientlyonlybyusingconnectivityinformation.Throughdepth-firsttraversing,itcancontracttwo-boundarygraphsofanysizeintoedges,andthenaccelerateexistingmax-flowalgorithmsusingcontractedgraphs.Large-scalerandomandreal-worldgraphexperimentstestitseffectandtheaccelerationratiocanbemorethan50timeswithlowgraphreductionoverhead,indicatingthattheproposedalgorithmcanbeusedasaneffectivecomplementtoexistingalgorithms.
Introduction
Asafundamentalandwidelyinvestigatedproblem,themaximumflow(max-flow)problemanditsdualproblem,theminimumcut(min-cut)problemhavemanyapplicationsandareoftenusedassubroutinesinotheralgorithms(
Sherman
,
2009
).Manyadvanceshavebeenmade
in
thedevelopmentofefficientalgorithmsforthisproblem(
Goldberg
&Rao
,
1998
).Fastsolutionscanbefoundthroughalgorithm-orientedacceleration(
Boykov&Kolmogorov
,
2004
;
Goldberg&Tarjan
,
2014
;
Verma&Dhruv
,
2012
),orgraphdata-orientedacceleration(
Gusfield
,
1990
;
Wei,Liu,&Zhang
,
2018
;
Zhang,Hua,Jiang,Zhang,&Chen
,
2011
;
Zhang,Xu,Hua,&Zhao
,
2012
;
Zhao,Su,Liu,&Zhang
,
2014
;
Zhao,Xu,Hua,&Zhang
,
2012
).Asfordata-orientedalgorithms,thecoreideaistoreducethesizeoftheinput,suchasreducingthesizeofthegraphbycontractingsubgraphstonodes.Therearemanycontraction-basedalgorithmsthatexploitvarioussubgraphcontractionconditions.Unfortunately,mostofthesealgorithmscanonlyobtainapproximatedvaluewhenusingsubgraphcontractionconditionwithnosizelimit.
Inthepaper,weproposealosslessalgorithmthatemploysasub-graphcontractionconditionwithnosizelimit,andatoyexampleisgivenin
Fig.
1
toclarifythedifferencebetweentheproposedmethodandexistingcontraction-basedmethods.Unlikeexistingnode-orientedcontractions,theproposedalgorithmcontractssubgraphsintoedges.
Thecorepartisaroutinethatcanefficientlysearchforthetwo-boundarygraph(TBG),thatis,thesubgraphcontainingtwoboundarynodesthatseparateitfromotherpartsofthewholegraph.Experimentsvalidateitseffectivenessbyusinglarge-scalegraphscontainingupto107nodes.
Theworkadvancesthestateoftheartbyintroducinganewlosslessandgeneralsize-freecontractionconditionwithahighlyefficientwaytolocatethecorrespondingsubgraphs.Specifically,thetheoreticalcon-tributionsinclude:(a)Weprovethatthecontractionoftwo-boundarygraphscanmaintaintheexactmax-flowresult,whichleadstothecontractionalgorithmthathasfewconstraints(e.g.,unlimitedsubgraphsize)andcanensureaccuratesolution.Theseconditionshavehardlybeenmetsimultaneouslybefore.(b)Tofindthetwo-boundarygraph,weproposetousethemappingbetweenthedepth-firsttraversaltreeandthenodecutset(NCS).ThemethodcanefficientlysearchfortheNCScontainingonlytwonodes(i.e.,twonodesoftheboundedgraph).
(c)WeproposethestrategyofselectingallreusableTBGsforagivennodepair.
Inaddition,thepracticalcontributionsare:(a)Toensurealgorithmgenerality,weproposetoidentifythebi-connectedcomponent(BCC)firstandthenTBGineachBCC,toensurealgorithmgenerality.SinceTBGpartiallyoverlapswithsomeexistingandinefficientsubgraph
?Correspondingauthor.
E-mailaddresses:
weiwei_ise@
(W.Wei),
wangpengpeng_do@
(P.Wang),
dongyb@
(Y.Dong).
/10.1016/j.eswa.2023.120151
Received10June2022;Receivedinrevisedform11March2023;Accepted12April2023
Availableonline20April2023
0957-4174/?2023ElsevierLtd.Allrightsreserved.
W.Weietal.
ExpertSystemsWithApplications225(2023)120151
2
Fig.1.ExamplesshowingthedifferencebetweenTTCMaxandexistingcontractionalgorithms(usedinSection
1
).
contractionconditions,itcannotbeapplieddirectly.(b)Toensurealgo-rithmefficiencyandavoidthehighcomplexityofenumeratingTBGs,weintroducealowcomplexityrandomsearchalgorithm,whichcanhelpfindenoughTBGsforaccelerationatasmallcost.Experimentalresultsshowthatthealgorithmcanaccelerateexistingalgorithmsingeneratedgraphfamiliesandreal-worldgraphs,whiletheaccelerationratiocanbeuptotwoordersofmagnitude,andthuscanbeusedasaneffectivecomplementtoexistingalgorithms.
Literaturereview
Classicaccelerationattemptsmainlyfocusedonalgorithmtuning.Existingalgorithmsevolveintwodirections:theaugmentingpath-basedalgorithms(
Jack&Richard
,
1972
)andpreflow-basedalgo-rithms(
Goldberg&Tarjan
,
1988
).Theaugmentingpath-basedalgo-rithmsfocusonhowtofindasuitablepathbetweensourceandsinknodesineachiterationofmax-flowcomputation,sothatthewholecomputationcanbeefficientlyaccelerated.
JackandRichard
(
1972
)
W.Weietal.
ExpertSystemsWithApplications225(2023)120151
providedselectionrulesthatcanhelpavoidimproperselectionofaugmentingpathsleadingtoseverecomputationaldifficulties.Anewalgorithmusingtheshortestaugmentingpathwasproposed.
Karzanov
(
1974
)explicitlyintroducedtheblockingflowmethod,whichaug-mentedalongamaximumsetofshortestpathsbyfindingablockingflow.Suchaugmentationincreasedtheshortestaugmentingpathlengthandthusacceleratedtheshortestaugmentingpathmethod.Unlikeaug-mentingpath-basedmethods,push-relabel-basedmethodsmaintainedapreflowanditerativelyappliedpushoperationstoupdatethepreflow,
and
usedrelabeloperationstoupdatenodedistances(
Goldberg&
Tarjan
,
1988
).Here,apreflowisdifferentfromaflowinthatthetotalamountflowingintoanodecantemporarilyexceedthetotalamountflowingout,andtheproposedalgorithmpushesthelocalflowexcesstowardsthesinkalongtheestimatedshortestpath.
Inadditiontotuningalgorithmsthatkeepexactmax-flowvalues,
there
isalsoresearchintotradingaccuracyforspeed.
Christiano,Kel-
ner,Madry,Spielman,andTeng
(
2011
)proposedanewapproximationalgorithmforthemax-flowproblem.Bysolvingalistofelectricalflowproblemswiththesolutionofalinearequationsystem,thealgorithmcanbefasterthantheprevious??(??3∕2)algorithms.
Sherman
(
2013
)proposedanewalgorithmusing??-congestion-approximators.Theal-gorithmmaintainedanarbitraryflowthatmayhavesomeresidualexcessanddeficits,andittookstepstominimizeapotentialfunctionmeasuringthecongestionofthecurrentflowplusanover-estimate
of
thecongestionrequiredtoroutetheresidualdemand.
Jonathan,
YinTat,Lorenzo,andAaron
(
2014
)introducedanewframeworkforapproximatelysolvingflowproblemsandappliedittoprovideasymptoticallyfasteralgorithmsformax-flowandmaximumconcur-rentmulti-commodityflowproblems.Thetechniquestheyusedareanon-Euclideangeneralizationofgradientdescentwithboundsonitsperformance,thedefinitionandefficientconstructionofanewtypeofflowsparsifier,andafastobliviousroutingschemeforflowproblems.
Anothermajorcategoryofmax-flowaccelerationalgorithmisbasedonpreprocessing,i.e.,reducingthegraphsizeforacceleration.Specialsubgraphsarecontractedtohelpreducethesizeofthegraph,butcanalsocauseachangeinthemax-flowvaluebetweennodepairs.Therearemanydifferentsubgraphcontractionconditionsthatcanbeusedtocontractthegraph.Thefirsttypeofcontractionconditionistomaintainthemax-flowvaluesintheoriginalgraph.
ScheuermannandRosenhahn
(
2011
)proposedanalgorithmforimagesegmentation,andthebasicideaistosimplifytheoriginalgraphwhilemaintainingthemax-flowproperties.ThealgorithmconstructsaSlimGraphbymergingnodesthatareconnectedbysimpleedges,andtheresultingslimGraphcanbeappliedwithstandardmax-flowalgorithms.
LiersandPardella
(
2011
)presentedflow-conservingconditionsunderwhichsubgraphscanbecontractedtoasinglenode.Aftercapacitynormalization,thealgorithmattemptedtoshrinkthemaxedgeamongalledgesbetweentwonodes,orbetweenthreenodes,orbetweenaspecialclusterofnodes.Basedontheconditions,theresearchersalsoproposedatwo-stepmax-flowalgorithmtosolvetheprobleminstancesfromphysicsandcomputervision.
Therearealsographcontractionconditionswithnosizelimitthatcansignificantlyreducethegraphsizebutattheexpenseofchangedmax-flowvalues.
Zhangetal.
(
2011
)examinedthegraphcontractionconditionofmaximalclique.Bycontractingtheappropriatemaximumcliqueinthenetwork,thealgorithmcanapproximatethemax-flowresultefficiently.
Zhangetal.
(
2012
)leveragedthegraphcommunityforacceleration.Thecommunitydetectionalgorithmisappliedto
detectallcommunitiesinanetwork,wherethesourceandsinknodes
wereestablishedtofindneighbor-node-setastocontracttheoriginalgraphformax-flowacceleration.
Existingresearchhasalsoacceleratedmax-flowcalculationbycon-structingahigh-leveloverlay.Theoverlaypathcanhelpaccelerateordirectlyobtainmax-flowresults.Awell-knownoverlayistheGomory–Hutree(
Gusfield
,
1990
).Itisbuiltbyorganizingallthe??nodesoftheoriginalgraphintoatree,andtheminimumedgecapacityinthepathbetweenanytwonodesisthemax-flowvaluebetweenthem.TheGomory–Hutreeisnotwidelyusedduetoitshighcomplexity,e.g.,it
needs
???1max-flowcalculationtobuildaGomory–Hutree.
Zhao
etal.
(
2014
)viewedtheoriginalgraphasalayergraphbetweenthesourceandsinknodes,e.g.,nodesatthesamedistancefromthesourcenodeformalayer.Themax-flowvaluecanbecalculatedasthesmallestofallthemax-flowvaluesbetweenadjacentlayers.
Weietal.
(
2018
)attemptedtodividetheoriginalgraphintosubgraphs(i.e.,thebi-connectedcomponent)withcutnodesasboundarynodes.Max-flowcalculationbetweenanygivennodepairinvolvesonlyaportionofcutnodesandsubgraphs,anditcanbedividedintoseveralsub-calculationsbetweenadjacentsubgraphs.
Alltheabovemax-flowcalculationscanbefurtheracceleratedby
algorithm
parallelization(
Baumstark,Blelloch,&Shun
,
2015
;
Jiadong,
Zhengyu,
&Bo
,
2012
;
Khatri,Samar,Behera,etal.
,
2022
).
Khatri
etal.
(
2022
)proposedseveraltechniquestoimprovetheperformanceofthepush-relabelmax-flowalgorithmonGPUs.Theythencombinethepush-relabelalgorithmwiththeirnewlyproposedpull-relabelalgo-rithmtoconstructapull-push-relabelmaxflowalgorithm.Experimentsusingreal-worldandsyntheticgraphsshowitssignificantacceleration
ratio
inthreerepresentativemaximumflowapplications.
Baumstark
etal.
(
2015
)foundthatthewell-knownhighestlabel-basedpush-relabelimplementationisnotoptimalfortheircollectedbenchmarkinstances.Instead,thefirst-in-first-outpush-relabelalgorithmappearstobemoresuitableandisimplementedasasharedmemory-basedparallelalgorithm.Aparallelversionofglobalrelabelingisusedtoimprovespeed.Experimentalresultsshowthesuperiorityofthisalgo-rithmoverthefastestexistingimplementation.
Jiadongetal.
(
2012
)proposedtwoGPU-basedmaximumflowalgorithms,wherethefirstisasynchronousandlockfreeandthesecondissynchronizedviatheprecoloringtechnique.ExperimentalresultsshowthattheGPU-basedalgorithmoutperformstheCPU-basedalgorithmbyaboutthreetimes,andinGPU-basedalgorithms,thesynchronousalgorithmoutperformstheasynchronousalgorithminsomesparsegraphs.
Model
Themax-flowproblem
Themax-flowproblembetweennodes??and??inadirectedgraphisformulatedinEq.(
1
).??isthecollectionofallnodesinthegraph,and
??isthecollectionofalldirectededgesinthegraph.Foradirectededge
??fromnode??to??,denote????≥0asthecapacityof??,and??(??,??)≤????or
| |
??(??)≤????astheamountofflowfrom??to??,themax-flowproblemistofindthemaximumsumofflowoutof??andinto??,asshowninEq.(
1
).Here???(??)={??∈??(??,??)∈??}and??+(??)={??∈??(??,??)∈??}
aretwosetsofadjacentnodesof??.Theconstraintsoftheproblemare:
(∑
(a)theflowvalueoneachedgeshallnotexceedthecapacityoftheedge,and(b)thevalueofallflowsintoanodemustbethesameasthatoutsidethenode.Thesolutionisthemax-flowvalue???(??,??)andtheflowvaluesatalledges.Wemainlyfocusontheprobleminconnectedundirectedgraphs,whichcanberepresentedasdirectedgraphsbytransformingeachundirectededgeintotwodirectededgesbetweenadjacentnodes.
arenotdividedintoanycommunities.Foreachcommunityfoundinthenetwork,thealgorithmwillvalidatethecontractconditionthatthetotalamountflowingintoacommunitycanflowthroughthecommunity,whichwillbecontractedifconditionshold.
Zhaoetal.
???(??,??)=??????
??∈??+(??)
?
??0≤????≤???????∈??(??)
??(??,??))
(1)
(
2012
)examinedthegraphcontractionconditioncalledneighbor-node-set,whichisonenodeplusitsneighboringnodes.Specialconditions
??.??.
3 ??
??∈∑???(??)
??(??,??)=
∑
??∈??+(??)
??(??,??)???∈???{??,??},(??)
W.Weietal.
ExpertSystemsWithApplications225(2023)120151
PAGE
4
Definition3.5.Intheoriginalgraph,TBGcontractionistoreplaceallnodesandedgesintheTBGasanedgebetweenitstwoBNs,andtheedgecapacityisthemax-flowvaluebetweenBNsthroughnodesinsidetheTBG.
TBGcanhelpspeedupmax-flowcalculationbecauseTBGcontrac-tiondoesnotchangethemaximumflowvaluebetweenanynodesoutsidetheTBG.
Lemma3.6.ForaTBG,themax-flowvaluebetweenthenodepairsoutsidetheTBGintheoriginalgraphisthesameasinthegraphwiththecontractedTBG.
Fig.2.AnexampleofTBGdefinedin
Definition
3.4
(Allsolid-linenodesandtheedgesconnectingthemconstituteTBG).
Thebi-connectedcomponenttree
Givenanundirectedgraphof??withthesetofallnodes??andthesetofalledges??,wefirstintroducethecoredefinitionsthatarerelatedtoouralgorithm.Ifnotmentioned,allthegraphsmentionedareundirectedgraphs.
Definition3.1.Acutnodeinagraphisanodewhoseremovalwilldividethegraphintoatleasttwodisconnectedsubgraphs.Bi-connectedcomponent(BCC)isthemaximumconnectedsubgraphinwhichanytwonodeshaveatleasttwodisjointpathsbetweenthem.
NotethatthecutnodeswillsplittheoriginalgraphintoBCCs.Anexampleisshownin
Fig.
4
(a),whereanygraphcanberepresentedasthegraphontheleft,withtheBCCtreeontheright.Itisobviousthatremovinganycutnodewilldisconnectthegraph.IthasbeenfoundthatanygivengraphcanbetransformedintoaBCCtreewithcutnodeconnectingtheseBCCs(
Hopcroft&Tarjan
,
1973
)(asshownontherightof
Fig.
4
(a)):
Theorem3.2.AnyconnectedgraphcanbetransformedintoatreeofBCCs.
Proof.Pleaserefertopaper(
Weietal.
,
2018
).
Thetwo-boundarygraph
Thedefinitionofboundarynodeisintroducedfirst.
Definition3.3. Foraconnectedsubgraph??′in??withnodeset
??′???andedgeset??′???,aboundarynode(BN)of??′isanode
??∈??′thathasatleastoneadjacentnode??′∈?????′.
Inthepaper,wefocusonthekindofsubgraphthatcontainsonlytwoBNs.
Definition3.4.Thetwo-boundarygraph(TBG)isaconnectedsub-graphin??withonlytwoBNs.
AnexampleofTBGisgivenin
Fig.
2
,whereallsolid-linenodes(andtheedgesbetweenthem)constituteTBG.AnyothernodeexceptBNs(nodes1and2)intheTBGiscalledtheinnernode(IN),i.e.,allINsandBNsin
Fig.
2
constituteTBG.AnynodeoutsidetheTBGiscalledtheouternode(ON)oftheTBG.
TBGisusefulbecausebycalculatingthemax-flowvaluebetweenitsBNsinsidetheTBG,wecancontracttheTBGasanedgebetweenitsBNs,andtheedgecapacityisjustthemax-flowvaluebetweenitsBNsinsidetheTBG:
Proof.
Forthemax-flowprobleminanundirectedgraph,itneedstobeconvertedtotheproblemintheequivalentdirectedgraph,thatis,theundirectededgebetweentwonodesisconvertedintotwodirectededgesofthetwonodeswiththesamecapacityastheundirectededgecapacity.Anyin-edgeflowisadirectedflowwiththesamedirectionasthedirectededge.
Denotethemax-flowfrom??to??as???(??,??),giventwonodes??and??
intheequivalentdirectedgraphcorrespondingtotheundirectedgraph,itisshownbelowthatforeachpairofadjacentnodes??and??,onlyoneedgeof(??,??)and(??,??)hasapositiveflowvalueinthefinalmax-flowresult.
Becausethemaximumvaluesofvariousaccuratemaximumflowalgorithmsareconsistent,weuseoneofthecommonlyusedmethods,theaugmentingpath-basedmethod,toillustratetheaboveconclusion.Formax-flowfrom??to??,thecoreideaistofindandfillasimplepathwithpositivevalueflowbetween??and??ineachiterationuntilnofurtherpathcanbefoundaftermanyiterations.Thepathfindingisappliedtotheresidualgraphupdatedineachiteration.Forexample,intheoriginalgraphforthedirectededge(??,??)withitscapacity????(??,??)andflowvalue
????(??,??)afteraniteration,intheresultingresidualgraphtherewillbeanedge(??,??)with??(??,??)=????(??,??)?????(??,??),andanedge(??,??)with??(??,??)=????(??,??)+????(??,??).
Notethatintheaugmentingpath-basedmethod,theresidualcapacityof(??,??)introducedbytheflowonthereverseedgeisusedatfirst,andusingthecapacityofoneedgeintheresidualgraphdecreasestheflowvalueonitsreverseedgeintheoriginalgraph.Thus,inoneiteration,sincethesimplepathallowsnoloopingintheresidualgraph,onlyoneofthetwoedges(??,??)and(??,??)willbeincludedinthesimplepathbetween??and
??.Soafterthefirstiteration,onlyoneedgehasflowintheoriginalgraphandassumethatitisthedirectededge(??,??)withitscapacity????(??,??)andflowvalue????(??,??).Assumingthatinthenextiterationtheedge(??,??)isincludedinthepathwithflow
??(??,??),if????(??,??)≥??(??,??),intheoriginalgraphitwillcausetheflowvalueof(??,??)todecreasei.e.,(??,??)withupdatedflowvalue
????(??,??)???(??,??)and(??,??)withflowvalue0.
Orif????(??,??)<??(??,??),thenintheoriginalgraphtheflowvalueof(??,??)willdecreaseto0andtheflowvalueof(??,??)willbecomepositive??(??,??)?????(??,??).Theresultissimilarifedge(??,??)isselected,i.e.,onlyoneedgeofthetwoedgesbetweentwonodeshasapositiveflowvalueintheupdatedoriginalgraphofeachiteration.
Thus,itisobviousthatintheresultingoriginalgraphofeachiteration(includingthefinalresultgivenbythefinaliteration),onlyoneedgeof(??,??)and(??,??)isusedtoholdtheflow.
Giventwonodes??and??intheequivalentdirectedgraphcorre-spondingtotheundirectedgraph,when???(??,??)isdeployedinthegraph,???(??,??)withthesamemax-flowvaluecanbedeployedinthegraphatthesametime.
Accordingtoitem
(2)
above,thereverseedgeofeachpositive-flowedgeisnotusedinthemax-flowresult.Itisobviousthatfor
eachedge(??,??)withpositiveflowvalue??(??,??)inthemax-flow
Itisclearthat????≤???(??,??)and????≤???(??,??)inanymax-flow
1 12 2 21
resultbetween??and??,bymovingtheflowinedge(??,??)toedge
(??,??)withsameflowvalue(whichisequivalenttoreversingtheinandoutflowsofeachnode),theper-nodeflowconstraints
resultbetween??and??outsidetheTBG.Becauseanydeployed
flowsmustbefeasibletoreachthetargetnode??inthemax-flowresult.Orif????>???(??,??)anditmustflowoutofTBGthrough??2,
? 1 12
inEq.(
1
)alsoholdanditwillgeneratethemax-flowresult??(??,??)
withsamemax-flowvaluefromnodes??to??.Andsinceeachin-edgeflowof???(??,??)sitsatadifferentedgefromitscounterpartin
???(??,??),???(??,??)and???(??,??)onlysharenodesandthustheirflowswillnotaffecteachother,itisclearthat???(??,??)and???(??,??)canbedeployedinthegraphatthesametime.
DenotetheboundarynodeofthegivenTBGas??1and??2.Denotethemax-flowvaluebetween??1and??2as?????????,the
whichwillgenerateanin-edgeflowdistributionwiththetotal
flowvaluebetween??1and??2insideTBGlargerthan???(??1,??2)
and???(??2,??1),whichcontradictsthemax-flowdefinition.
Asaresult,iftheTBGiscontractedastheundirectededge(or
directededgesinbothdirections)between??1and??2withthemax-flowvalue???(??1,??2)==???(??2,??1),theflowsinto??1or??2inthemax-flowresultcanbekeptunchangedandthecapacitylimitis
met,sotheotherpartsofthemax-flowresultarekeptthesame.
???????
(??1,??2)
Inaddition,themax-flowvaluebetween??and??willnotincrease
??(??2,??1)withthesamemax-flowvaluecanbedeployedinthe
TBGatthesametime.
Thisisobviousfromtheitem
(3)
above.
Denotethemax-flowfrom??to??as???(??,??),giventwonodes??and??intheequivalentdirectedgraphcorrespondingtotheundirectedgraph,itisshownbelowthatthereisnocycleflowpathinthe
asthecontractionwillnotgenerateanynewaugmentingpathbetween??and??.Ifthebottleneckedgetofindamoreaug-mentingpathisnotinsidetheTBG,TBGcontractionwillkeeptheotherpartofthemax-flowresultandthebottleneckedgeunchanged,andthusnonewaugmentingpathwillbegenerated.IfthebottleneckedgeisinsidetheTBG,theremustbe????==
max-flowresults.Thatis,intheoriginalgraphthedirectededge
???
1
or??2==???
(??1,??2) 1
(??2,??1)(orTBGcanindependentlyadjustthe
withpositiveflowwillnotformacycle.
Supposethatwhenasimplepathbetween??and??(i.e.,??)isfoundintheresidualgraphinoneiteration,therewillbeacycleflowpathintheupdatedoriginalgraph.Supposeintheoriginalgraphthepathsegmentcontainingtheoverlappednodesinthelooppathandthesimplepathisboundedbynodes??and??.Itisclearthatthereareatleasttwodirectedpathsinreversedirectionswithpositiveflowbetween??and??intheoriginalgraph,anddenotethemas????(??,??)and????(??,??)andsuppose????(??,??)asthepathsegmentinthefoundsimplepathintheresidualgraph,whichisalsothesimplepathintheoriginalgraph.So,beforethesimplepath??isfound,????(??,??)alreadyexistsintheoriginalgraphandthereiscertainlyasimplepathwithresidualcapacitiesbetween??and??intheresidualgraph(i.e.,??(??,??)).Andnotalledgeshaveresidualcapacityinthe????(??,??)appliedintheresidualgraphortherewillbeacycleflowpathbeforethesimplepath
??.Thatis,givenapath??(??,??)withresidualcapacitiesinallitsedges,thealgorithmchoosesthepath????(??,??)withsomeedgesthatdonothaveresidualcapacitiesintheresidualgraph,whichcontradictsthefactthatintheaugmentingpath-basedmethod,residualcapacitywillbeusedfirst.Thus,inthemax-flowresultoftheoriginalgraph,thedirectededgewithpositiveflowwillnotformacycle.
Inthemax-flowresultbetweenanynodepair,??and??outsidea
TBGwithBN??1and??2,representtherespectiveflowvalueintotheTBGthrough??1and??2as????and????,onlyoneof????and????
internalflowtoallowmoreflowthroughtheTBG),sobykeepingtheedgecapacityaftercontractionas???(??1,??2),thecontractionwillnotgenerateanewaugmentingpath.
Asaresult,TBGcontractionwillmaintainthemax-flowresultbeforecontractionandwillnotgenerateanewaugmentingpathtochangethemax-flowresult.
Thelemmaisproved.
Asaresult,formax-flowcalculationbetweenanynodepairoutsidetheTBG,theTBGcanbetreatedasanedgebetweenitsBNs.AndforTBGsthatdonotcontainthegivennodepair,theycanbecontractedonebyonewhilethemax-flowvalueofthenodepairremainsun-changedaftereachcontraction.Thatis,allTBGsthatdonotcontainthegivennodepaircanbecontractedatthesametime.
ItisalsofoundthatfortheTBGinsideaBCC,theremainingportionoftheBCCafterremovingtheTBGisstillaTBG,whichisdefinedastheinverseTBGoftheoriginalTBG:
Lemma3.7.InaBCCwithnodeset???andedgeset???,givenanyTBGinsideitwithnodeset??′????andedgeset??′????withthesetoftwoboundarynodes????.TheBCCportionoutsidetheTBGwithnodeset
??′′=??????′+????andedgeset??′′=??????′isstillaTBGwiththesameBNs.
Proof.BecausethereareatleasttwopathswithoutdisjointingnodesbetweenanytwonodesinaBCC,neighborsaandbofboundarynodes
isgreaterthan0.
1 2 1
2 ??1and??2outsidetheTBGhavepathsthatdonotpassthrough??1and
Thus,ifboth????and????arenotzerointhemax-flowresult,since
??2.Byadding??1and??2andtheiredgeswiththerespectiveneighbors
1 2 ??and??,itwillgenerateapathbetween??1and??2intheBCCportion
theflowintotheTBGfrom??2canonlyexitTBGthrough??1
溫馨提示
- 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請(qǐng)下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請(qǐng)聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會(huì)有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
- 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
- 5. 人人文庫網(wǎng)僅提供信息存儲(chǔ)空間,僅對(duì)用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對(duì)用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對(duì)任何下載內(nèi)容負(fù)責(zé)。
- 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請(qǐng)與我們聯(lián)系,我們立即糾正。
- 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時(shí)也不承擔(dān)用戶因使用這些下載資源對(duì)自己和他人造成任何形式的傷害或損失。
最新文檔
- 學(xué)習(xí)古代文學(xué)史的試題及答案
- 湖北省武漢市東西湖區(qū)2023-2024學(xué)年八年級(jí)下學(xué)期期中考試英語試題(含答案)
- 藥理學(xué)考試常見復(fù)習(xí)誤區(qū)試題及答案
- 統(tǒng)計(jì)學(xué)考試過程與結(jié)果評(píng)估試題及答案
- 2024年汽車維修工行業(yè)交流技巧試題及答案
- 食品檢測技術(shù)的發(fā)展趨勢(shì)試題及答案
- 2024年汽車維修工考試技巧提升
- 古代文學(xué)史作品分析試題及答案
- 庫房管理工作
- 時(shí)間管理大師65
- 消費(fèi)行為影響機(jī)制-深度研究
- 健康咨詢與服務(wù)推廣協(xié)議
- 護(hù)士N1晉級(jí)N2述職報(bào)告
- 中國糖尿病防治指南(2024版)解讀
- 食堂食材配送采購?fù)稑?biāo)方案(技術(shù)標(biāo))
- 山東省汶上縣市級(jí)名校2025屆中考生物全真模擬試卷含解析
- 2025年度智能硬件產(chǎn)品全國區(qū)域獨(dú)家代理合同3篇
- 辦公室安全知識(shí)培訓(xùn)課件
- 2025年四川省成都市青白江區(qū)招聘50人歷年高頻重點(diǎn)提升(共500題)附帶答案詳解
- 2025年浙江嘉興市眾業(yè)供電服務(wù)限公司招聘38人高頻重點(diǎn)提升(共500題)附帶答案詳解
- 中國技能大賽-第45屆世界技能大賽全國選拔賽“水處理技術(shù)”項(xiàng)目技術(shù)工作文件
評(píng)論
0/150
提交評(píng)論