ebook img

On equivalence between maximal and maximum antichains over range-restricted vectors with integral coordinates and the number of such maximal antichains PDF

0.07 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 On equivalence between maximal and maximum antichains over range-restricted vectors with integral coordinates and the number of such maximal antichains

NonamemanuscriptNo. (willbeinsertedbytheeditor) On equivalence between maximal and maximum antichains over range-restricted vectors with integral coordinates and the number of such maximal antichains Shen-FuTsai 7 1 Received:2017/01/24 0 2 n Abstract Consider a strict partially ordered set P consisting of all d-dimensional a vectors with integral coordinates restricted in a certain range. We found that any J maximalantichainisalsomaximum,andthemaximumsizehasasimpleexpression 4 intermsoftherange.Propertiesofthenumberofmaximalantichainsgiventherange 2 are explored. We present our proof, the application on other areas, and some open ] questions. O Keywords OrderedSet·(0−1)Matrix·ExtremalProblems C . h t 1 Introduction a m ConsiderastrictpartiallyorderedsetP consistingofalld-dimensionalvectorswith [ integral coordinates restricted in a certain range. The binary relation < is defined 1 as y=(y ,...,y )<x=(x ,...,x ) if y <x for every i∈[1,d]. Without loss of 1 d 1 d i i v generality,supposetherangeisgivenbyanpositiveintegervectorw=(w ,...,w ), 0 1 d meaningthatforeachdimesioni,thecoordinateofeveryvectorinP isrestrictedto 5 7 [1,wi].ThenumberofvectorsinP is(cid:213) di=1wi.AnantichainisasubsetofP such 6 thatforanytwoelementsitcontains,neither<oritsinverse>holds.Amaximalan- 0 tichainisanantichainthatisnotapropersubsetofanyotherantichain.Amaximum . 1 antichainisanantichainthathascardinalityatleastaslargeaseveryotherantichain. 0 7 IfinPwereplacethebinaryrelation<by≤,i.e.y=(y ,...,y )≤x=(x ,...,x ) 1 d 1 d 1 if y ≤x for every i∈[1,d], then the size of maximal antichain varies. For exam- : i i v ple for w=(2,2), {(1,1)} is a maximal antichain, and so is {(1,2),(2,1)}. Their i X sizes differ by 1. On the other hand, under < the only two maximal antichains are {(1,1),(2,1),(1,2)}and{(2,2),(2,1),(1,2)},bothwithcardinality3.Themainre- r a sultofthisworkisthatunder<,everymaximalantichainisalsomaximum,andthe Shen-FuTsai 7319143rdAve.NE,Redmond,WA,USA E-mail:[email protected] 2 Shen-FuTsai maximalsizeis(cid:213) d w −(cid:213) d (w −1). i=1 i i=1 i Becausewediscoveredtheresultwhileworkingonthefieldofmatrixexclusion1, we will present and proveour result using its terminology.The rest of the paper is organized as follows: in Section 2 we introduce matrix exclusion terminology and reviewsomeliteratures.InSection3weprovethatmaximalandmaximumantichains areequivalentinoursetting.WeexplorethenumberofmaximalantichainsinSection 4.InSection5weconcludewithsomeopenquestions. 2 Matrixexclusion A matrixis binaryif eachofits entriesiseither 0or 1.Allthe matricesmentioned inthispaperarebinary.Anentrywithvalue1isanoneentry;otherwiseitisazero entry.AbinarymatrixAcontainsanotherbinarymatrixPifwecanobtainabinary matrix A′ with the same dimensionality as P by deleting 0 or more columnsand 0 or morerowsfromA, and A′ ≥P forall i and j. Thedefinitionextendsnaturally ij ij tod-dimensionalmatrix– ineachdimensionwe delete0ormoreindicesandform amatrixwiththesamedimensionalityasP.AavoidsPifAdoesnotcontainP.We callPtheforbiddenmatrix.InthisworkweonlycareaboutP=I ,the2×...×2 2,d d-dimensionalidentitymatrix. Definition1 Let P be a d-dimensional matrix. ex(w,P) is the maximum number of one entries a w ×w ×...×w matrix can have while avoiding P. ex(n,P)≡ 1 2 d ex((n,n),P),ex(n,P,d)≡ex((n,...,n),P). In contrast to our result, the field of matrix exclusion has mostly focused on asymtotic behavior of ex(w,P), especially in two-dimensional matrices. Bienstock and Gyo¨ni[1] showed ex(n,P)=Q (na (n)) for some trapezoidalpatternswith four oneentries,wherea (n)istheextremelyslowgrowinginverseAckermannfunction. Fu¨redi and Hajnal[2] raised the question of what matrices have extremal functions linear in n? Subsequently an array of works emerged to respond to the quest, for example [3,4,5,6]. There are much fewer works on higher dimensional forbidden matrices.KlazarandMarcus[7]showedthatex(n,P,d)=O(nd−1)foranyk×...×k permutation matrix P. Geneson and Tian[8] extended the result to all d-dimension tuplepermutationmatrices. 3 Equivalencebetweenmaximalandmaximumantichains For notational simplicity we define 2-contain same as contain, except when d =1 amatrix2-containsI aslongasithasmultipleoneentries.Likewise2-avoidand 2,d ex2(w,P)followsnaturallyfrom2-containthusdefined.Theonlyreasontohave2- containdefinedistocovertheweirdbuttrivialedgecased=1inthemathematical 1 Afterconjecturingandprovingtheresult,theauthorpostthemontheforumofCrowdMathProject[9] underusernameparityhome,astheauthorwasoriginallyworkingonamatrixexclusiongameproposed ontheforumbymoderatorJesseGeneson. TitleSuppressedDuetoExcessiveLength 3 inductionusedbytheproof. Theorem1 ex2(w,I )=(cid:213) d w −(cid:213) d (w −1). Moreover, every maximal w × 2,d i=1 i i=1 i 1 ...×w matrix that 2-avoids I has exactly ex2(w,I ) one entries. A maximal d 2,d 2,d matrixisonesuchthatflippinganyzerotoonemakesit2-containI . 2,d Notethat(cid:213) d w −(cid:213) d (w −1)isthenumberofentrieswithatleastonecoordi- i=1 i i=1 i natebeing1,andalsothenumberofentrieswithatleastoneindexisuchthatx =w. i i TheresulthasimplicationonamatrixexclusiongameproposedbyGeneson[9]. Playedbym>1playersnumberedfrom0tom−1,thegamestartswithamatrixA filledwithzeroentries.TheplayerstaketurntoflipazeroinAtoone,andaplayer losesthegameifhisorhermovemakesAcontainapredefinedbinarymatrixP.Our resultimpliesthatifP=I ,player[((cid:213) d w −(cid:213) d (w −1)) modm]alwaysloses 2,d i=1 i i=1 i nomatterwhattheplayers’strategiesare. Proof Thetheoremistrivialwhend=1orw =1forsomeiin[1,d]. i Definition2 A d-row x=(x ,...,x ) of a d-dimensionalmatrix M is the set of 1 d−1 entrieswiththefirstd−1coordinatesbeingx ,...,x .Thereare(cid:213) d−1w d-rows, 1 d−1 i=1 i eachwithw entries. d Definition3 The x -th d-cross section is the set of entries with the last coordinate d beingx .Therearew d-crosssections,eachwith(cid:213) d−1w entries. d d i=1 i LetMbeamaximalw ×...×w matrixthat2-avoidsI .Wewillbeprimarily 1 d 2,d dealingwiththed-rowsofMthroughouttheproof. Definition4 Ad-rowx=(x ,...,x )isaancestorofd-rowy=(y ,...,y )if 1 d−1 1 d−1 x<y. We say y is a descendantof x. A(x) and D(x) are the sets of ancestors and descendantsofx,respectively. Ancestoranddescendantrelationsaretransitive,andthegraphsofthesetwore- lationsaredirectedacyclic.DefineM(x,x )=M . d x1,...,xd For a d-row x define l(x) and h(x) as the minimum and maximum y such that M(x,y)=1,respectively.Theyarewelldefinedifthereisanoneentryinx.Foraset of d-rowsZ, l(Z) and h(Z) are the smallest and largestd-coordinateof one entries inZ,respectively,withl(0/)=¥ andh(0/)=−¥ .ForZ6=0/,theyarewelldefinedif thereexistsz∈Zsuchthattheyarewelldefinedforz.Wefirstestablishthath(·)and l(·)arewelldefinedeverywhere,i.e.everyd-rowofMhassomeoneentry. Lemma1 Everyd-rowofMhassomeoneentry. 4 Shen-FuTsai Proof Givenad-rowx,ifA(x)isempty,thenM(x,w )=1.IfD(x)isempty,then d M(x,1)=1.Ifneithersetisempty,since(1,1,...,1)∈A(x),l(A(x))iswelldefined, and so is h(D(x)) as (w ,...,w )∈D(x). We then have l(A(x))≥h(D(x)), be- 1 d−1 causeotherwiseM2-containsI .Foranyy∈[h(D(x)),l(A(x))],havingM(x,y)=1 2,d doesnotmakeM2-containI . ⊓⊔ 2,d Lemma2 Givenanyd-rowx,theset{y: M(x,y)=1}isacontiguoussegment. Proof If M(x,y )=M(x,y )=1 for y ≤y , then setting M(x,y)=1 for any y∈ 1 2 1 2 [y ,y ]doesnotmakeM2-containI . ⊓⊔ 1 2 2,d w(M),thenumberofoneentriesinM,isthen(cid:229) h(x)−l(x)+1. x Lemma3 h(x)=min(w ,l(A(x))) (1) d Proof It is trivial when A(x)=0/, so we only discuss when A(x) is not empty. If h(x)>l(z)forsomez∈A(x),thenM 2-containsI .Ontheotherhand,ifh(x)< 2,d l(A(x)),settingM(x,l(A(x)))=1doesnotmakeM2-containI . ⊓⊔ 2,d Lemma4 l(x)=max(1,h(D(x))) (2) Proof It is trivial when D(x)=0/, so we only discuss when D(x) is not empty. If l(x)<h(z)forsomez∈D(x),thenM 2-containsI .Ontheotherhand,ifl(x)> 2,d h(D(x)),settingM(x,h(D(x))=1doesnotmakeM2-containI . ⊓⊔ 2,d Thetwolemmasaboveimplyl(x)≤l(z)ifD(x)⊂D(z)andsimilary,h(x)≤h(z) ifA(z)⊂A(x). We are now going to prove the theorem by mathematical induction on d and w throughthe followinglemmas. Assume the theoremholdsfor 1,2,...,d−1 di- d mensions,andforw=(w ,...,w ,1),(w ,...,w ,2),...,(w ,...,w ,w −1) 1 d−1 1 d−1 1 d−1 d withd dimensions. Lemma5 GivenamatrixM,ifh(·)andl(·)arewelldefinedandsatisfyEquation1 andEquation2foreveryd-rowx,thenM2-avoidsI andismaximal. 2,d Proof Clearly, if there is an one entry in some d-row x with d-coordinate y larger thanmin(w ,l(A(x))),thenyeithergoesbeyondw ,whichisimpossible,ory>l(z) d d forsomez∈A(x),andM 2-containsI .Similarlywe couldnothaveoneentryin 2,d anyd-rowxwithd-coordinatelessthanmax(1,h(D(x))). Ontheotherhand,ifallentriesinsomed-rowxhaved-coordinateissmallerthan min(w ,l(A(x))), setting M(x,min(w ,l(A(x)))) to 1 does not make M 2-contain d d I .Thesimilarargumentworksforl(x). ⊓⊔ 2,d TitleSuppressedDuetoExcessiveLength 5 We know that h(x)=w if A(x)=0/. In fact, if the converse is true things are d mucheasier.Let’sdenote{x: h(x)=w ,A(x)6=0/}byX(M),amoredifficultsetof d d-rowsthatwewilldealwithlater. Definition5 X(M)≡{x: h(x)=w ,A(x)6=0/} d Lemma6 IfX(M)=0/,thenw(M)=(cid:213) d w −(cid:213) d (w −1). i=1 i i=1 i Proof TransformMtoM′ byremovingthew -thd-crosssection.Thensinceh(x)= d w ifandonlyifA(x)=0/,wehave d w(M′)=w(M)−|{x: A(x)=0/}| WeclaimthatM′ isstillmaximal.First,weshowthattherearestilloneentryfor everyd-rowin{x: A(x)=0/}.IfD(x)=0/,thenM′(x,1)=1.IfD(x)6=0/,thensince h(z)<w foreachz∈D(x),l(x)<w inM,meaningM′stillhasoneentryind-row d d x.Sol(·)andh(·)arestillwelldefinedinM′.Second,sincetheonlychangeinh(·) and l(·) is the decrease of h(·) by 1 for {x: A(x)=0/}, Equation1 and Equation2 stillholdinM′,andsobyLemma5,M′ isstillmaximal.Byassumption w(M′)=w w ...w (w −1)−(w −1)(w −1)...(w −1)(w −2) 1 2 d−1 d 1 2 d−1 d A(x)=0/ ifandonlyifatleastoneofthecoordinatesx ,...,x is1.So 1 d−1 |{x: A(x)=0/}|=w w ...w −(w −1)(w −1)...(w −1) 1 2 d−1 1 2 d−1 Addingthemtogether,wehave d d (cid:213) (cid:213) w(M)= w − (w −1) i i i=1 i=1 ⊓⊔ Tocopewithd-rowsinX(M),wewillshowthatwecanalwaysconvertMtoM′ whichstill2-avoidsI maximally,with|X(M′)|<|X(M)|,andw(M)=w(M′).The 2,d firststep,inLemma7,istofindapairofd-rowsxandx′whichwecouldmanipulate justalittlebittoreduce|X(·)|whilekeepingmaximalityandw(·)intact. Wesaythata d-rowx=(x ,...,x )isasemi-ancestorofd-rowif y≤xand 1 d−1 x6=y. Lemma7 IfX(M)6=0/,thenthereared-rowsxandx′,suchthatx=(x′+1,...,x′ + 1 d−1 1),h(x)=w >l(x),andx′ hasnootherdescendantzwithh(z)=w . d d Proof Inthefirststagewefindad-rowxwith h(x)=w >l(x)andA(x)6=0/.We d startfromanarbitraryd-rowx∈X(M).Ifl(x)=w ,thenthereisdescendantzofx d withh(z)=w becausetheonlyotherpossibilityisD(x)=0/,whichleadstol(x)=1, d a contradiction.We set x to z and see if l(x)<w . We repeatthe process untilwe d findaxwithh(x)=w >l(x),whichisguaranteedtohavenon-emptyancestorset d 6 Shen-FuTsai A(x).Suchxexistsbecausethereareonlyfinitenumberofd-rows,andwheneverwe hitonewithoutdescendants,l(x)=1<w .Wethensetx′=(x −1,...,x −1). d 1 d−1 In the second stage, if x′ has no other descendant z with h(z)= w then we d are done. If it does, then since D(z)⊂D(x), l(z)≤l(x)<w , i.e. z qualifies the d requirement for x too. Moreover x is a semi-ancestor of z, so we set x to z and x′ =(x −1,...,x −1). We repeat the process until x′ has no other descendant 1 d−1 z with h(z)=w . This process terminates because in the sequence of x we found, d eachelementisasemi-ancestorofthesubsequentelement,andthissequencecould notbeinfinitelylong. ⊓⊔ Lemma8 If X(M)6=0/, we can convert M to another maximal matrix M′ that 2- avoidsI with|X(M′)|<|X(M)|andw(M′)=w(M). 2,d Proof By Lemma 7, we find d-rows x and x′ such that x=(x′ +1,...,x′ +1), 1 d−1 h(x)=w >l(x), and x′ has no other descendant z with h(z)=w . Construct M′ d d identicaltoMexcepth′(x)=w −1andl′(x′)=w −1.WehaveX(M′)=X(M)−x d d andw(M′)=w(M). WeclaimthatM′isstillamaximalmatrixthat2-avoidsI .Theonlychangewe 2,d make is to decrease h(x) and l(x′) by 1. Clearly Equation 1 is not violated by M′, becausex′isanancestorofx.TheonlypossibilitythatM′violatesEquation2isthat x′ has another descendant z with h(z)=w . But this is impossible because of the d choicewemakewithx′. ⊓⊔ Finally,givenanymatrixMthatmaximally2-avoidsI ,ifnecessarybyapplying 2,d Lemma8afinitenumberoftimeswecanconvertittoanothermaximalM′ withthe samenumberofoneentries,still2-avoidsI ,andwithX(M′)=0/.ThenbyLemma 2,d 6wehavew(M)=w(M′)=(cid:213) d w −(cid:213) d (w −1).Thisconcludesourproof. ⊓⊔ i=1 i i=1 i 4 Numberofmaximalantichains Afterprovingtheequivalenceofsuchmaximalandmaximumantichains,itisnatural toaskhowmanymaximalantichainstherearegivenrangevectorw.Inthissection weexplorethisquantity. Definition6 R(w)issetofmaximalantichainsoverrange-restrictedvectorswithin- tegralcoordinatesboundedbyw.Theorem1saysthateachelementofR(w)hassize (cid:213) d w −(cid:213) d (w −1). i=1 i i=1 i Forwwithlength2,|R(w)|canbeexpressedinbinomialcoefficient. Lemma9 |R([w1,w2])|=(cid:0)w1w+1w−21−2(cid:1) Wefoundapropertyof|R(w)|whenwcontainsa2. Lemma10 |R([w,2])|=R|(w)| TitleSuppressedDuetoExcessiveLength 7 Proof Wewillshowthatthereisbijectivemapping f :R(w)→R([w,2]).Denotethe lengthofvector[w,2]byd.ForamatrixMwithdimensionw ×...×w ×2,de- 1 d−1 fineS asthesetofitsd-rowsxwithM(x,y)=1ifandonlyifi≤y≤ j. i,j ForamatrixM∈R([w,2]),byLemma1andLemma2itsd-rowsarepartitioned intodisjointsetsS for1≤i≤ j≤2.WeclaimthatS ∈R(w).Thisisbecause i,j 1,2 d−1 (cid:213) w(M)=2|S |+|S |+|S |= w +|S | 1,2 1,1 2,2 i 1,2 i=1 Maximizationofw(M) is equivalentto maximizationof|S |, whichbydefini- 1,2 tionleadstoS ∈R(w). 1,2 Thefollowingdescribesaninjective f.ForamatrixN∈R(w),weformamatrix M by setting S =N and S =S =0/. Next, whenever there is a d-row x∈/ N 1,2 1,1 2,2 not in S nor S , we place it in either of them so that M still 2-avoidsI . Note 1,1 2,2 2,d thatx couldnotbeboththeancestoranddescendantofsomed-rowsin N,because otherwise N 2-containsI . Also, x must be either an ancestor or descendantof 2,d−1 some d-row in N, as otherwise N+x is a largerset that 2-avoidsI . Hence we 2,d−1 can always determine whether a d-row x∈/ N should be in S or S . Apparently 1,1 2,2 whenwearedoneMisamaximalmatrixthat2-avoidsI . 2,d f is surjective because we have established that given a matrix M ∈R([w,2]), S ∈R(w). ⊓⊔ 1,2 Acorollaryfollowsimmediately. Corollary1 Ifmax w ≤2,then|R(w)|=min w. i i i i 5 Openquestions Giventhesimpleexpression(cid:213) d w −(cid:213) d (w −1),itisnaturaltoaskifourresult i=1 i i=1 i canbeprovedwithsimplermethod,especiallywithoutthetechnicalandtrickyma- neuvers in Lemma 7 and 8. Also, is there a better and perhaps more intuitive way thanEquation1andEquation2tocharacterizeamaximalmatrixthat2-avoidsI ? 2,d One of the motivations of pursuing an alternative characterization is to derive the expressionmoregeneralthanSection4forthenumberofmaximalmatricesM that 2-avoidsI intermsofrangew=(w ,...,w ). 2,d 1 d References 1. D.Bienstock,E.Gyo¨ri,Anextremalproblemonsparse0-1matrices,SIAMJ.DiscreteMath.4(1991) 17-27. 2. Z.Fu¨redi,P.Hajnal,Davenport-Schinzeltheoryofmatrices,DiscreteMath.103(1992)233-251. 3. R.Fulek, Linearboundonextremal functions ofsomeforbidden patterns in0-1matrices, Discrete Math.309(6)(2009)1736-1739. 8 Shen-FuTsai 4. G.Tardos,On0-1matrices andsmallexcluded submatrices, J.Combin.TheorySer.A111(2005) 266-288. 5. B.Keszegh,Onlinearforbiddenmatrices,J.Combin.TheorySer.A116(1)(2009)232-241. 6. J.T.Geneson,Extremalfunctionsofforbiddendoublepermutationmatrices,J.Combin.TheorySer.A, 116(7)(2009),1235-1244. 7. M.KlazarandA.Marcus,ExtensionsofthelinearboundintheFuredi-Hajnalconjecture,Advancesin AppliedMathematics,38(2)(2007),258-266. 8. JesseT.Geneson and Peter M.Tian, Extremal Functions ofForbidden Multidimensional Matrices, CoRR,abs/1506.03874,(2015). 9. http://www.artofproblemsolving.com/polymath/mitprimes2016/f

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.