ebook img

Further applications of the Container Method PDF

0.16 MB·
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Further applications of the Container Method

Further applications of the Container Method Jo´zsefBaloghandAdamZs.Wagner 6 1 0 2 n a J 8 2 AbstractRecently,Balogh–Morris–SamotijandSaxton–Thomasonprovedthathy- pergraphssatisfyingsomenaturalconditionshaveonlyfewindependentsets.Their ] O main results already have several applications. However, the methods of proving C these theorems are even more far reaching. The general idea is to describe some . familyofevents,whosecardinalityaprioricouldbelarge,onlywithafewcertifi- h cates.Here,weshowsomeapplicationsofthemethods,includingcountingC -free t 4 a graphs, considering the size of a maximumC -free subgraph of a random graph 4 m andcountingmetricspaceswithagivennumberofpoints.Additionally,wediscuss [ someconnectionswiththeSzemere´diRegularityLemma. 1 v 9 1 Introduction 0 8 7 Recently, an importanttrend in probabilistic combinatoricsis to transfer extremal 0 results in a dense environmentinto a sparse, random environment.The first main . 1 breakthrough is due to Conlon–Gowers [8] and Schacht [23]. Not much later, 0 Balogh–Morris–Samotij[2] and Saxton–Thomason[22] had a differentapproach, 6 1 whichnotonlyprovedmostoftheresultsofConlon–Gowers[8]andSchacht[23], : but also provided counting versions of these results. Additionally, there are some v furtherapplicationsofthemaintheoremsof[2]and[22].However,theproofmeth- i X r a Jo´zsefBalogh DepartmentofMathematicalSciences,UniversityofIllinoisatUrbana-Champaign,Urbana,Illi- nois61801,USAe-mail:[email protected]. ResearchispartiallysupportedbySimons Fellowship, NSA Grant H98230-15-1-0002, NSF CAREER Grant DMS-0745185, NSF Grant DMS-1500121,ArnoldO.BeckmanResearchAward(UIUCCampusResearchBoard15006)and MarieCurieFP7-PEOPLE-2012-IIF327763. AdamZs.Wagner DepartmentofMathematicalSciences,UniversityofIllinoisatUrbana-Champaign,Urbana,Illi- nois61801,USAe-mail:[email protected]. 1 2 Jo´zsefBaloghandAdamZs.Wagner odshavethepotentialtobemoreinfluentialthanthetheoremsthemselves.Thegen- eralidea,whichcanbetracedbacktotheclassicalpaperofKleitman–Winston[13] (andmoreexplicitlyinseveralpapersofSapozhenko)istodescribesomefamilyof events,whosecardinalityaprioricouldbelarge,onlywithafewcertificates. Here,wetrytooutlinewhatliesbehindthismethod.Becauseeachoftheabove mentionedfourpapers[8, 23, 2, 22] already gavea nice surveyof the field, addi- tionallyConlon[7],Ro¨dl–Schacht[20]andSamotij[21]havesurveypapersofthe topics, we choose a different route. Each of the following four sections discusses themethodwithsomeapplications.Threeofthesectionscontainnewresults. InSection2 we state an importantcorollaryof themainresultsof [2]and [22] andwilldiscussitsconnectionwiththeSzemere´diRegularityLemma[24]. (n) In Section 3 we estimate the volumeof the convexsubsetof [0,1] 2 represen- tingmetricspaceswithnpoints,andconsideradiscretevariantofthisproblemas well. Kozma–Meyerovitch–Peled–Samotij[15] consideredthe followingquestion: chooserandomlyandindependently n numbersfromtheinterval[0,1],labelling 2 the edges of a complete graph K , what is the probability that any three of them n (cid:0) (cid:1) formingatrianglewillsatisfythetriangleinequality?Theyusedentropyestimates toderiveanupperboundonthisprobabilityandnoticedthatSzemere´diRegularity Lemmacanbeusedtocountmetricspaceswhosealldistancesbelongtoadiscrete setofafixedsize. Mubayi and Terry [19] more recently pushed the discrete case into 0,1 -law { } typeresults.Usingthegeneralresultsof[2]and[22],weimprovetheresultsimplied bytheregularitylemma,andobtaingoodresultsforthecontinuouscase.Parallelto ourwork,Kozma,Meyerovitch,Morris,Peled andSamotij[16] practicallysolved thecontinuousversionoftheproblemusingmoreadvancedversionsofthemethods of[2]and[22].Inordertoavoidduplicatework,ourgoalinthissectionisclarity overpushingthemethodtoitslimit. It was probably Babai–Simonovits–Spencer [1] who first considered extremal problemsinrandomgraphs.Mostoftheirproposedproblemsareresolved,butthere wasnoprogressonthefollowing:Whatisthemaximumnumberoftheedgesofa C -freesubgraphoftherandomgraphG(n,p)when p=1/2?Inthecase p=o(1), 4 strongboundsweregivenbyKohayakawa–Kreuter–Steger[14].Here,inSection4, we improve on the trivial upper bound, noting that an n-vertexC -free graph can 4 haveatmost(1/2+o(1))n3/2edges. Theorem1.1 For every p (0,1), there is a c>0 that the largestC -free 4 ∈ subgraph of G(n,p) has at most (1/2 c)n3/2 edges w.h.p. In particular, if − p=0.5wecantakec=0.028. Kleitman–Winston[13]authoredoneofthefirstpapersinthefieldwhosemain ideawastofindsmallcertificatesoffamiliesofsetsinordertoprovethatthereare notmanyofthem.TheyprovedthatthenumberofC -freegraphsonnverticesisat 4 most2cn3/2 forc 1.081919.InSection5,weimprovetheconstantintheexponent ≈ FurtherapplicationsoftheContainerMethod 3 by using ideas found in Section 3. The improvement is tiny; the main aim is to demonstratethatthemethodhaspotentialapplicationsforotherproblemsaswell. 2 Connections to Szemere´di’s RegularityLemma In this section we describe some connections between the Szemere´di Regularity Lemma and the new counting method. Originally, without using the regularity lemma,Erdo˝s–Kleitman–Rothschild[10]estimatedthenumberofK -freegraphs. k Theorem2.1 Thereare2(1+o(1))·ex(n,Kk) Kk-freegraphsonnvertices,where ex(n,K )isthemaximumnumberoftheedgesofaK -freegraphonnvertices. k k Here, we sketch a well-knownproofof Erdo˝s–Frankl–Ro¨dl[9], using the Sze- mere´diRegularityLemma.WeapplytheRegularityLemmaforaK -freegraphG , k n which outputs a cluster graph R, wheret =O(1). Then we clean G , i.e., we re- t n move edges inside the clusters, between the sparse and the irregularpairs. Define C to be the blow-up of R to n vertices(getting back the same preimage vertices n t ofG ).ObservethatC containsallbuto(n2)edgesofG andC isK -free,hence n n n n k e(C ) ex(n,K ).NowwecancountthenumberofchoicesforG :Thenumberof n k n ≤ choices forC is O(1) nn, the numberof choices for E(G ) E(C ) givenC , is n n n n · ∩ atmost2ex(n,Kk),andthenumberofchoicesforE(Gn) E(Cn)givenCn,is2o(n2), − completingtheproofoftheresult. An important corollary of the main results of Balogh–Morris–Samotij [2] and Saxton–Thomason[22]isthefollowingcharacterizatonofK -freegraphs. k Proposition2.2 There is a t 2O(logn·n2−1/(k−1)) and a set G1,...,Gt of ≤ { } graphs,eachcontainingo(nk)copiesofK ,suchthatforeveryK -freegraph k k H thereisani [t]suchthatH G. i ∈ ⊆ Notethatitfollowsfromstandardresultsinextremalgraphtheorythateachsuch graphG necessarilycontainsatmost(1+o(1)) ex(n,K )edges. i k · Thisprovidesan evenshorterproofofTheorem2.1:Foreach K -freegraphH k thereisani [t]suchthatH Gi.Thenumberofchoicesforiis2O(logn·n2−1/(k−1)), andthenum∈berofsubgraphs⊆ofGiisatmost2(1+o(1))ex(n,Kk). TheproofusingtheRegularityLemmayieldstwovariantsofProposition2.2,one whichisweaker,asitgivest=2o(n2),thoughitisstrongenoughforthisparticular application,andonethatwecalltheSzemere´diApproximateContainerLemma. 4 Jo´zsefBaloghandAdamZs.Wagner Proposition2.3 Thereisat =2o(n2) andaset G ,...,G ofgraphs,each 1 t containing at most (1+o(1))ex(n,K ) edges, a{nd o(nk) co}pies of K , such k k thatforeveryK -freegraphH thereisani [t]suchthatH G. k i ∈ ⊆ Proposition2.4 There isat =O(1)anda set G ,...,G ofgraphs,each 1 t { } K -free,suchthatforeveryK -free graphH there isani [t]andapermu- k k ∈ tationofV(H),givinganisomorphicgraphH ,suchthat E(H ) E(G) = ′ ′ i | − | o(n2). Most of the tools fromthis section are likely to be usefulin countingmaximal K -freegraphs,wheretherearestillnosatisfactoryboundswhenr 4. r ≥ Problem2.5 WhatisthenumberofmaximalK -freegraphswithvertexset r [n]?Forr=3thiswasaquestionofErdo˝s,whichwassettledin[4]and[3]. 3 The number ofmetricspaces Ourgoalistoestimatethenumberofmetricspacesonnpoints,wherethedistance betweenanytwopointsliesin 1,...,r forsomer=r(n).Thisproblemwascon- { } sideredfirstbyKozma–Meyerovitch–Peled–Samotij[15],who,usingtheregularity lemmagaveanasymptoticboundonthenumberofsuchmetricspacesfora fixed constantr. Recently,Mubayi–Terry[19] provideda characterisationofthe typical structureofsuchmetricspacesforafixedconstantr,whilen ¥ .Wewillbemore → interestedinwhathappensifrisallowedtogrowasafunctionofn.Ourmainresult isthefollowing: Theorem3.1 Fixanarbitrarysmallconstante >0.If n1/3 r=O , log43+e n! thenthenumberofsuchmetricspacesis FurtherapplicationsoftheContainerMethod 5 r+1 (n2)+o(n2) . 2 (cid:24) (cid:25) Kozma–Meyerovitch–Peled–Samotij[15] pointed out that the discrete and the continuousproblemsarerelated.Theyconsideredthesamequestioninthecontinu- ouscasewithdistancesin[0,1].TheirentropybasedapproachyieldsTheorem3.1 forr<n1/8. At the end of this section we show how our results translate to the continuous setting. For a positive integer r define m(r)= r+1 . We will use an easy corollary of ⌈ 2 ⌉ Mubayi–Terry([19]Lemma4.9): Lemma3.2 LetA,B,C [r], allnon-empty.Supposethetriple A,B,C doesnot ⊂ { } contain a non-metric triangle – that is, every triple a,b,c:a A,b B,c C { ∈ ∈ ∈ } satisfies the triangle-inequality.Then if r is evenwe have A + B + C 3m(r), | | | | | |≤ andifrisoddwehave A + B + C 3m(r)+1. | | | | | |≤ LetH bethe3-uniformhypergraphwithvertexsetr rows,oneforeachcolor, and n columns, one for each edge of K . A vertex (i,f) of H corresponds to 2 n theeventthatthegraphedge f hascolori.ThreeverticesofH formahyperedge (cid:0) (cid:1) when the graph edge coordinates of the vertices form a triangle in K while the n ‘colors’ do not satisfy the triangle inequality. With other words, the hyperedges correspondtonon-metrictriangles,andindependentsetshavingexactlyonevertex fromeachcolumncorrespondtopointsofthemetricpolytope.Ourplanistoprove asupersaturationstatement,butfirstweneedtwolemmas. The first lemma we use is due to Fu¨redi [11]. For a graph G, write G2 for the “propersquare”ofG,i.e.,wherexyisanedgeifandonlyifthereisazsuchthatxz andzyareedgesinG.Writee(G)forthenumberofedgesinG. Lemma3.3 ForanygraphGwithnvertices,wehave e(G2) e(G) n/2 . ≥ −⌊ ⌋ ThesecondlemmaboundsthesizeofthelargestindependentsetinH. Lemma3.4 Let S V(H) have no empty columnsand containno edges in H. Thenifrisevenwe⊂have S m(r) n ,andifrisoddwehave S m(r) n +rn. | |≤ 2 | |≤ 2 Proof. Theevencasefollowsdirectl(cid:0)y(cid:1)fromLemma3.2,andwenotethatth(cid:0)is(cid:1)bound istight–letScontaintheinterval[r/2,r]fromeachcolumn.Theboundintheodd case is slightly more difficult, and we make no effort to establish a tight bound, whichshouldprobablybe S m(r) n +n/2. | |≤ 2 Letrbeodd,andletA,B,CbethreecolumnsthatformatriangleofS.Notethatif (cid:0) (cid:1) forsomek 1wehave A m(r)+kand B m(r)+kthen C m(r) 2k+1 ≥ | |≥ | |≥ | |≤ − ≤ m(r) kbyLemma3.2.WriteB forthesetofcolumnsinSoforderatleastm(r)+k k − 6 Jo´zsefBaloghandAdamZs.Wagner andwriteS forthesetofcolumnsinSoforderatmostm(r) k.LetG bethegraph k k − on[n]withedgesB .ThenbyLemma3.3weget k S e(G2) e(G ) n/2 B n. | k|≥ k ≥ k −⌊ ⌋≥| k|− Hence n (cid:229) r (cid:229) r (cid:229) r S m(r) = k(B B ) k(S S )= (B S ) nr, | |− 2 | k|−| k+1| − | k|−| k+1| | k|−| k| ≤ (cid:18) (cid:19) k=1 k=1 k=1 andtheresultfollows. ⊓⊔ Nowwearereadytoproveasupersaturation-likeresult. Lemma3.5 Lete >0andletS V(H)withnoemptycolumns. 1. Ifrisevenand S (1+e )⊂n m(r),thenScontainsatleast e n hyperedges. | |≥ 2 10 3 2. Ifr is odd,n>ne sufficientlylargeand S (1+e ) n m(r), thenScontains (cid:0) (cid:1) | |≥ 2 (cid:0) (cid:1) atleast e4 n hyperedges. 40000 3 (cid:0) (cid:1) Proof. Supposefi(cid:0)rs(cid:1)tthatriseven.Thenthereareatleast e n trianglesinGsuch 10 3 that the correspondingcolumns contain at least (1+e /10)3m(r) vertices from S. (cid:0) (cid:1) Indeed,ifthiswasnotthecase,then e n 3r+ n 3m(r)(1+e /10) n e (r+m(r)) 10 3 3 = m(r)+ > S, n 2 2 10 | | (cid:0) (cid:1) (cid:0) (cid:1)− (cid:18) (cid:19)(cid:18) (cid:19) whichisacontradiction.Hencepart1ofthelemmafollowsfromLemma3.2. Now suppose r is odd. Given T [n], write f (T) for the set of vertices of S S ⊂ containedinthe T columnsofH correspondingtotheedgesspannedbyT.Set |2| n =20/e , so thatby Lemma3.4, wheneverT [n] with T =n and f (T) 0 0 S m(r) n0 (1+e )(cid:0)then(cid:1)H[f (T)]containsahyper⊂edge. | | | |≥ 2 3 S First,weclaimthatthereareatleast e n choicesofT [n]with T =n and f (T(cid:0)) (cid:1) m(r) n0 (1+e ). 4 n0 ⊂ | | 0 | S |≥ 2 3 (cid:0) (cid:1) Indeed,ifthiswasnotthecase,thenwewouldhave (cid:0) (cid:1) e n n0 r+ n m(r) n0 (1+e ) n e r e m(r) S 4 n0 2 n0 2 3 = m(r)+ + | |≤ (cid:0) (cid:1)(cid:0) (cid:1) (cid:0)nn0−(cid:1)−22 (cid:0) (cid:1) (cid:18)2(cid:19)(cid:18) 4 3 (cid:19) (1) < n m(r)(1+(cid:0)e ), (cid:1) 2 (cid:18) (cid:19) whichisnotpossible.SothenumberofhyperedgescontainedinSisatleast e n n 3 e e 3 n e(H[S]) / − , ≥ 4 n n 3 ≥ 4203 3 (cid:18) 0(cid:19) (cid:18) 0− (cid:19) (cid:18) (cid:19) andtheresultfollows. ⊓⊔ FurtherapplicationsoftheContainerMethod 7 Write d¯fortheaveragedegreeof H, andfor j [3] definethe j-th maximum ∈ co-degree D =max e E(H):s e :s V(H)and s = j . j {|{ ∈ ⊂ }| ⊂ | | } Below,wewillmakeuseofaversionofthecontainertheoremof[2,22],thewayit wasformulatedbyMousset–Nenadov–Steger[18]. Theorem3.6 There exists a positive integerc such thatthe following holds for every positive integer N. Let H be a 3-uniform hypergraphof order N. Let0 p 1/(36c)and0<a <1besuchthatD (H,p) a /(27c),where ≤ ≤ ≤ 4D 2D D (H ,p)= 2 + 3. d¯p d¯p2 ThenthereexistsacollectionofcontainersC P(V(H))suchthat (i)everyindependentsetinH iscontained⊂insomeC C, (ii)forallC C wehavee(H[C]) a e(H ),and ∈ ∈ ≤ (iii)thenumberofcontainerssatisfies log C 39c(1+log(1/a ))Nplog(1/p). | |≤ Proof of Theorem 3.1. Let H be the hypergraph defined earlier, i.e., the 3- uniform hypergraph with vertex set formed by pairs of the r colors and the n 2 edgesofK ,with3-edgescorrespondingtonon-metrictriangles.Lete ,d >0bear- n bitrarilysmallconstantsandset p= 1 anda = 1010clog4+2d n.InH weha(cid:0)v(cid:1)e rlog2+d n n D nr2, D r, D =1, d¯ r2n/64and 1 2 3 ≤ ≤ ≥ 64r2log2+d n 64r2log4+2d n a D (H,p) 4 + . ≤ r2n 2r2n !≤ 27c ThenTheorem3.6providescontainerswith e(H[C]) a e(H) 104cr3n2log4+2d n, ≤ ≤ andthenumberofcontainersis c310rn2 logn logr loglogn log C · · · =o(n2). | |≤ rlog2+d n Nowassume n1/3 r=o . log(4+2d )/3n! 8 Jo´zsefBaloghandAdamZs.Wagner Thenthemaximumnumberofedgesinacontaineriso(n3),hencebyLemma3.5, andthefactthatausefulcontainerdoesnothaveanemptycolumn,wehaveforn largeenough, n V(C) <(1+e )m(r) . | | 2 (cid:18) (cid:19) Hence, the number of colourings in a container is at most (1+e )(n2)m(r)(n2) = m(r)(n2)+o(n2).Thelogarithmofthenumberofcontainersiso(n2).Thetotalnumber ofgoodcolouringsisatmostthenumberofcontainerstimesthemaximumnumber ofcolouringsinacontainer.Hencethetotalnumberofgoodcolouringsis m(r)(n2)+o(n2), asrequired. ⊓⊔ Now we turn our attention to the continuous setting. The set-up in [15] is as follows. Given a metric space with n points and all distances being in [0,1], we (n) regardthesetofdistancesasavectorin[0,1] 2 .Wewillcalltheunionofallsuchn (n) pointsin[0,1] 2 forallfinitemetricspacesthemetricpolytopeMn.Moreprecisely, the metric polytopeMn is the convexpolytopein R(n2) defined by the inequalities 0<d 1andd d +d . ij ij ik jk ≤ ≤ Notethatifa+b cthen ≥ a + b c . (2) ⌈ ⌉ ⌈ ⌉≥⌈ ⌉ Theorem3.7 Fixd >0constant.Thenforn>nd sufficientlylarge,wehave (vol(Mn))1/(n2)≤ 21+n611−d . Proof. First consider the discrete setting, colouring with r colours, where r is the evenintegerclosestton16−d2.W.l.o.g.d <1/4andset 1/p=n31−d4, a =300cnd −32, wherecistheconstantfromTheorem3.6.Then 1 1 n1/3 d /4 n2/3 d /2 D (H,p)<300 + 300 − + − a , (cid:18)rnp p2r2n(cid:19)≤ n7/6−d /2 n4/3−d !≤ andwegetcontainerswith e(H[C]) a n3r3 n3 1/6 d /4, − − ≤ ≤ FurtherapplicationsoftheContainerMethod 9 wherethenumberofcontainerssatisfies log C n2rplog3n n2 1/6 d /5. − − | |≤ ≤ Hence,byLemma3.5,thenumberofverticesinacontainerisatmost n V(C) (1+n 1/6 d /6) m(r). − − | |≤ 2 (cid:18) (cid:19) Thisimpliesthatthenumberofcolouringscontainedinacontainerisatmost (n) col(C)≤ V(n2C)! 2 ≤ (1+n−1/6−d /6)m(r) (n2)≤m(r)(n2)en2−1/6−d/7. (cid:16) (cid:17) (cid:0) (cid:1) Thatis,thetotalnumberX ofcolouringsisatmost X m(r)(n2)en2−1/6−d/7en2−1/6−d/5 m(r)(n2)en2−1/6−d/8. ≤ ≤ Now consider colouringsin the continuoussetting. Cut up each edge of the cube intorpieces.Wegetby(2)that (vol(Mn))1/(n2)≤ (m(r)+1)r((n2n2))en2−1/6−d/8!1/(n2) (n) 1/(n2) (n) 4 2 1 1 1 ≤ 2− 2 (cid:18)1+r(cid:19) ! (cid:18)1+n1/6+d /9(cid:19)≤ 2+n61−d , asrequired. ⊓⊔ Remark.Kozma,Meyerovitch,Morris,PeledandSamotij[16]usingastonger supersaturation result and a somewhat different container type of theorem, in- dependently, parallel to our work, improved the error term in Theorem 3.7 to (log2n)n 1/2,wherethe 1/2isbestpossibleasitwaspointedoutin[15]. − − It would be interesting to extend the above ideas to generalised metric spaces (note that there are several different definitions of these), hence we propose the followingpurposelyvaguequestion: Problem3.8 Itisanaturalquestiontoask:isthereaninterestingextension of the problem discussed in this section, when instead of requiring metric triangles,onewantsmetricd-dimensionalsimplicesforafixedd? 10 Jo´zsefBaloghandAdamZs.Wagner 4 The largestC -free subgraph ofarandom graph 4 NowweturnourattentiontoprovingTheorem1.1.Asbefore,foragraphGwrite G2 forthe“propersquare”ofG,i.e.,wherexyisanedgeifandonlyifthereisaz suchthatxzandzyareedgesinG. To simplify the technicality of the proof, we present here only the case when p=1/2;forlarger p we can use a properlychosensmaller c and the argumentis thesame. AssumethatH isthelargeC -freegraphthatweaimtofindinG(n,1/2).Anat- 4 uralapproachwouldbetodowhatFu¨redi[12]did,workingonaRamseytypeprob- lem.UsingtheKleitman–Winstonmethod,hegaveanupperboundonthenumber ofC -freegraphswithmedges.If pissufficientlysmall,thentheexpectednumber 4 of copies of such graphs in G(n,p) is o(1), proving the desired result. His upper boundwas, assumingthat m>2n4/3(log2n), that the numberof such graphsis at most(4n3/m2)m,hencetheexpectednumberofthemis(4pn3/m2)m=o(1),aslong asm>(1/2 a)n3/2and p<1/16 a,wherea>0issomesmallconstant. − − Theideaofourproofisthattheunionboundintheaboveargumentistoowaste- ful;hence,insteadofconsideringeachH separately,wewanttoshowthatforevery C -freesubgraphHwithmanyedges,thereexistssomecertificateinthegraph.Each 4 of these certificates will say that in a certain part of the graphone needs to select (1 o(1))d n3/2 edges of G(n,p) out of possible (1+o(1))d n3/2 pairs. Since the − numberof certificatesis muchsmaller than 1/(probabilitythatthe aboveunlikely eventholds),whichisoforder2cn3/2,wecannowshowusingtheunionboundthat w.h.p.thereisnosuchsubgraph. Soourfirsttaskistobuildsuchacertificate.FixaC -freegraphH withatleast 4 m>(1/2 c)n3/2 edges, where for the p=1/2 case we can set c=10 5. First − fixalinear−orderp ofthevertexsetofH,whichwewilluseasatiebreakeramong vertices.Wewillalsoneedasecondlinearordering,asfollows: Definition4.1 GivenagraphG,amin-degreeorderingisanordering v ,...,v 1 n { } ofV(G)suchthatforeachi [n],thevertexv isofminimumdegreeinG[v,...,v ]. i i n ∈ Whentherearemultiplesuchvertices,thenweletv bethefirstamongtheminthe i orderingp . Fix a min-degree ordering of H and let Y := v ,...,v and X =V(H) Y, 1 s { } − where we will choose s such that the minimum degree of H[X] is large. Now we fix F H[X] with the following properties. The first is that F is sparse, i.e., e(F) n3⊂/2/log2n.ThesecondisthattheindependentsetsinF2 approximatethe ≤ independentsets in H2[X] rather well. In particular, we choose F such that every largeindependentsetofH2[X]isina‘container’determinedbyF,wherethenum- ber of containers is at most r:=2n2/5log20n. The key observation is that for every v Y itsneighborhoodinX spansanindependentsetinH2[X],asH isC -free. j 4 ∈The certificate for H will be the vector [Y,F, d s , r s ], where d := { j}j=1 { j}j=1 j N(v ) X , and r r is the index of the container containing N(v ) X, and j j j | ∩ | ≤ ∩ s= Y .Thenumberofcertificatesisatmost2n n2/2 nn rn 22n3/2/logn. | | · n3/2/log2n · · ≤ (cid:0) (cid:1)

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.