ebook img

A general renormalization procedure on the one-dimensional lattice and decay of correlations PDF

0.13 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 A general renormalization procedure on the one-dimensional lattice and decay of correlations

A GENERAL RENORMALIZATION PROCEDURE ON THE ONE-DIMENSIONAL LATTICE AND DECAY OF CORRELATIONS A.O.LOPES 7 1 0 Inst. deMatema´tica,UFRGS-PortoAlegre,Brasil 2 n a J 6 2 ABSTRACT. WepresentageneralformofRenormalizationoperatorRactingonpotentialsV: 0,1 N R. WeexhibittheanalyticalexpressionofthefixedpointpotentialV forsuchoperatorR. Thispo{tentia}lc→anbe ] expressedinanaturallywayintermsofacertainintegralovertheHausdorffprobabilityonaCantortypeset S ontheinterval[0,1]. ThisresultgeneralizesapreviousonebyA.Baraviera,R.LeplaideurandA.Lopes(see D [2])wherethefixedpointpotentialV wasofHofbauertype(see[16]). h. ForthepotentialsofHofbauertype(awellknowncaseofphasetransition)thedecayisliken−g,g >0(see [22][23],[14]or[10]) t a Amongotherthingswepresenttheestimationofthedecayofcorrelation oftheequilibrium probability m associatedtothefixedpotentialVofourgeneralrenormalizationprocedure.Insomecaseswegetpolynomial [ decaTyhleikpeotne−ngti,agls>g0w,eancdonisnidoethrehresreaadreecaeylefmasetnetrstohfanthnees−o√canl,lwedhfeanmnil→yo¥f.Walterspotentialson 0,1 N(see { } 1 [27])whichgeneralizes thepotentials consideredinitiallybyF.Hofbauerin[16]. Forthesepotentialssome v explicitexpressionsfortheeigenfunctionsareknown. 6 Inafinalsectionwealsoshowthatgivenanychoicedn 0ofrealnumbersvaryingwithn Nthere → ∈ 5 existapotentialgontheclassdefinedbyWalterswhichhasainvariantprobabilitywithsuchnumbersasthe 6 coefficientsofcorrelation(foracertainexplicitobservablefunction). 7 0 1. INTRODUCTION . 1 ItisaclassicalprobleminStatisticalMechanicsthestudyofphasetransitions(see[12],[13],[1],[15], 0 7 [28]and[29])forgeneralpotentials. Inseveralexamplesitisknownthatonthepointofphasetransition 1 one gets polynomialdecay of correlationfor the associated equilibriumprobability. As far as we know, : thereisnounifiedwayforunderstandingallkindsofphasetransitions. v i One important technique on the analysis of such class of problems is the use of the renormalization X operator (see for instance [5], [18], [4] and [8]). The potential which is fixed for the renormalization r operatorinmostofthecasesisquiterelevantfortheunderstandingoftheproblem. a PhasetransitionsinthesettingofThermodynamicFormalismhasbeenconsideredfromdifferentpoints ofview(seeforinstance[2],[14],[19],[20],[21],[7],[6],[22],[23],[9],[10],[11]and[17]). In[2]itwasintroducedarenormalizationoperator(actingonpotentials)ontheone-dimensionallattice N 0,1 andthefixedpointonegetsisthewellknownHofbauerpotential. Theequilibriumprobabilityfor { } suchpotentialexhibitpolynomialdecayofcorrelationforacertainclassofobservables. Here we consider a more general procedure of renormalization of potentials defined on the lattice N 0,1 . We want to exhibit a fixed point potential for such procedure. The fixed point potential will { } beonaclassoffunctionspreviouslyconsideredbyP.Waltersin[27](whichincludestheHofbauerpoten- tial). Weareparticularyinterestedinthedecayofcorrelation(foracertainnaturalfunction)associatedto theequilibriumprobabilityforsuchpotential. In some cases (first type) the fixed point will produce polynomialdecay of correlation and in others cases(secondtype)notofpolynomialtype(someofthemfasterthanne √n,whenn ¥ ). − → Inordertogiveanideatothereaderofthekindofresultswewillgetherewepresentanexample: in aparticularcasetherenormalizationoperatorR actsoncontinuouspotentialsU, suchthatU isconstant Date:January27,2017. 1 2 A.O.LOPES oncylindersetsoftheform0...01and1...10,forallQ N.ForeachfixedQthevaluesarethesameon ∈ Q Q bothcylindersets WedefineU =R(U )in|t{hze}followi|ng{zw}ay: supposexisoftheformx=(0...0,1,...),then 2 1 Q U (x) = R(U )(x)=U (0...0,1)+U (0...0,1).|{z} 2 1 1 1 3Q 3Q 2 − InthiscasethefixedpointpotentialV fortherenor|m{azli}zationoper|a{tozr}R isthefunction 1 V(x)= dn (t) − K (Q t)a Z − when x=(0...0,1,...), Q N, and where n is the Hausdorff probability on the classical Cantor set (a ∈ Q subsetoftheinterval[0,1])andwhichhasHausdorffdimensiona = log2. |{z} log3 Inseveraldistinctclassesofproblemsthefixedpointpotentialfortherenormalizationoperatorcanbe expressed on an integral form associated to a fractal probability (see for instance [5], [3], [18], [4] and [24]).Herewewillalsogetasimilarkindofexpression. We will present now the main definitions giving more details on the specific problems we want to analyze. We will consider a subclass of potentials g: 0,1 N R which were described in [27]. Using the notationofthatpaper,considerarealsequencea{, n } N→, andassumethata 0, whenn ¥ . Inour n n ∈ → → modela=0=c. Givena ,a ,...,a ,...<0defineh , j 1,insuchwaythat(a +a +...+a )=logh , j 1. 2 3 n j 2 3 j+1 j+1 Forthesequence1 h >0,n N,n≥ 1,h =1,weassumethat,h 0,and(cid:229) ¥ h g =W(≥b )>1 foranyb >1. ≥ n ∈ ≥ 1 n→ n=1 n log2 Particularcaseswhichweareinterestedarewhenh =e n1/2,h =e n1−log5 orh =n g,g >2. n − n − n − Considerthetwoinfinitecollectionsofcylindersetsgivenby L =000...01 and R =111...10, foralln 1. n n ≥ n n Usingtheseparametersweca|n{dzefi}neacontinuous|po{teznt}ialg=gb :W R, whichissimplydenote byg,inthefollowingway:foranyx W andb 1 → ∈ ≥ b a , ifx L , forsomen 2; n n ∈ ≥ b an, ifx Rn, forsomen 2; g(x)=−llooggWW((bb )),, iiffxx∈∈LR1;; ≥ 1 − ∈ ¥ ¥ Intheabovewearetakingcn=an0,,foralln. ifx∈{1 ,0 }. WearemainlyinterestedinpotentialsgwhicharenotofHo¨ldertype. Definition 1. Given A:W = 0,1 N R, the Ruelle operator L acts on functions y :W R in the A { } → → followingway d j (x)=L (y )(x)= (cid:229) eA(ax)y (ax). A a=1 BythiswemeanL (y )=j . A From the generalresult described in page 1341in [27] applied to our particular case, we get that the eigenfunctionj b oftheRuelleoperatorforA=b g,forb 1,issuchthat,foranyn 1 ≤ ≥ AGENERALRENORMALIZATIONPROCEDUREONTHEONE-DIMENSIONALLATTICEANDDECAYOFCORRELATIONS 3 ¥ h b j b (0n1...)=a 1+h1nb j(cid:229)=1l (nb+)jj!, (1) ¥ h b j b (1n0...)=b(b ) 1+h1nb j(cid:229)=1l (nb+)jj!, (2) j b (0¥ )=a and j b (1¥ )=b(b ). Theeigenfunctionisunique. Weassumethatforb =1wegetl (b )=1=a =b(b ) Whenb =1,itiseasytoseethatifitiswelldefinedtheprobabilityr ontheBoreliansofW ,suchthat, foranynaturalnumberq 1,wehavethat ≥ r 0q1 =h and r 1q0 =h . (3) q q then,r isaneigenprobabilityforthedualoftheRuelleoperatorforg. (cid:0) (cid:1) (cid:0) (cid:1) Therefore,fromageneralprocedure(see[25])onegetsthattheprobabilitym ,suchthat, ¥ ¥ m 0q1 =h j (0n1...)= (cid:229) h and m 1q0 =h j (1n0...)= (cid:229) h (4) q n+j q n+j j=0 j=0 (cid:0) (cid:1) (cid:0) (cid:1) istheequilibriumprobabilityforg. We will show in Theorem 6 that decay of correlation d , as a function of q N, of the equilibrium q ∈ probabilitym forthepotentialgandfortheindicatorofthecylinder0isoforder ¥ ¥ (cid:229) (cid:229) h . (k+q+s) s=1k=0 Given a certain sequence d 0, there exists (under mild conditions) a choice of h such that d = q n q (cid:229) ¥ (cid:229) ¥ h . Indeed, firs→t for the given sequence d , q N, take inductively c such that c = ds=1d k=0,n(k+Nq+.sI)nthisway(cid:229) ¥ c =d . q ∈ n n n−Nown+,1give∈nsuchsequencecj,=nq jN,taqkeanewsequenceh ,suchthat,h =c c ,r N.Inthis n r r r r+1 case(cid:229) ¥ h =c . ∈ − ∈ i=p i p Itfollowsthatforallqwegetd =(cid:229) ¥ (cid:229) ¥ h . q s=1 k=0 (k+q+s) Partoftheresultspresentedherewerea consequenceoffruitfuldiscussionswithA. BaravieraandR. Leplaideurwhichcontributedinanontrivialwaytoourreasoning.Ourthankstothem. Insections2and3wepresentresultsaboutthefixedpointpotentialforacertainrenormalizationopera- tor. Onsections4and5weestimatethedecayofcorrelationsfortheclassopotentialsofWalterstypeand weshowthatalargeclassoftypesofdecayofcorrelationsareattainedbysuchfamily. Ontheappendix weuseresultsoftheprevioussectionstoshowthatforsomeofthesefixedpointpotentialsthedecayisnot ofpurelypolynomialtype. 2. A GENERAL RENORMALIZATION- FIRST TYPE Thepotentials f whichweareinterestedareontheclass f(0n1z)=a , f(0¥ )=a, f(10n1z)=d , f(10¥ )=d, n n f(01n0z)=b , f(01¥ )=b, f(1n0z)=c , f(1¥ )=c, n n andisassumedthata a,b b,c candd d. n n n n → → → → Wewillassumethatb =d =b=d. n n Therenormalizationoperatorwillactonsuchsetoffunctions. Fixanaturalnumberk 2.Wewillgeneralizethekindofrenormalizationdescribedin[2]wherek=2. ≥ Nowtakeforn>1 H (0n1x)=0kn (k 2)1x k − − H (1n0y)=1kn (k 2)0y k − − and H (01mx)=01mx k 4 A.O.LOPES H (10my)=10my k ¥ ¥ ¥ ¥ andalsoH (0 )=0 ,H (1 )=1 . k k Nowdefinetherenormalizationoperatoras Rk(V)(x)=V(Hk(x))+V(s Hk(x))+V(s 2Hk(x))+...+V(s k−1Hk(x)). WewillshowthatitispossibletoexhibitafixedpointV fortherenormalizationoperatorR . k Theclassoffunctionsweconsiderhereinthesection1islargeenoughtobeabletogetfixedpointsV. Thismeanswewillbeabletofindasequencea ,n N,whichwilldefinesuchV n ∈ An interesting question is to estimate the decay of correlation of the equilibrium probability for the fixedpointpotentialV. Onecanshow(seesection5)thatforthecaseofrenormalizationoffirsttypethe potentialV issuchthatwegetpolynomialdecayofcorrelation. ¥ ¥ IfV isfixedfortherenormalizationoperatorR ,thenV(1 )=V(0 )=0=a=c. k Moreover, ¥ ¥ ¥ ¥ V(01 )=V(01 )+V(1 )+V(1 )=b and ¥ ¥ ¥ ¥ a =V(001 )=V(00...0001 )+V(00...0001 )+...+V(0001 )=a +a +...+a . 2 3 4 k+2 k+2 k+1 | {z } | {z } ¥ ¥ ¥ ¥ a =V(0001 )=V(00...0001 )+V(00...0001 )+...+V(00...0001 )=a +a +...+a . 3 k+3 k+4 2k+2 2k+2 2k+1 k+3 Aswewillseoncew|efi{zxth}evalueof|a2{azdb}thepotentialV|w{ilzlb}edetermined. TheWalterspotentialV whichisthefixedpointforR shouldsatisfyforn 2: k ≥ a =a +a +...+a n k(n 1)+2 k(n 1)+1 k(n 2)+3 − − − a =c =c +c +...+c n n k(n 1)+2 k(n 1)+1 k(n 2)+3 − − − and d =b =b=d n n areconstant. Nowwesolvea (2)asthesolutionof 2+a (2) a = log . 2 − 2+a (2) 1 − Now,a (n)isdefinedbyinductionforn 2, ≥ a =a =...=a , k(n 1)+2 k(n 1)+1 k(n 2)+3 − − − and a =ka (n)+(k 2). k(n 1)+2 − − Takinga =d ,n 2,oftheform n n ≥ n+a (n) a = log <0 n − n+a (n) 1 − onecanshowthatwegetthefixedpointsolutionoftheRenormalizationoperator. Wepointoutthata andbarefree. 2 OnecanshowthatinthiscaseforsuchV thereexistsg suchthatlogh g logn,n 1.Ifg >2,one n getspolynomialdecayofcorrelationoftheformn2 g similartotheHofba∼ue−rcase. This≥kindofquestion − waspreviouslyconsiderin[22][23][14][10]. Thiscanbealsoobtainedfromthegeneralproofofsection 5. AGENERALRENORMALIZATIONPROCEDUREONTHEONE-DIMENSIONALLATTICEANDDECAYOFCORRELATIONS 5 3. A GENERALRENORMALIZATION -SECOND TYPE We consider in this section a far more general generalization of the renormalization operator and we exhibitthefixedpointpotentialV. Wepointoutthatinmanyexamplesthefixedpointpotentialofarenormalizationoperatorisdescribed byan expressionobtainedbythe integralofa kernelwith respectto a certainfractalprobability(see for instance[5],[3],[18],[4]and[24]). Herewewillgetasimilarkindofresult. Let us denote L the cylinders[00...01] (with n zeros) and R the cylinders[11...10] (with n ones). n n Letsusdefineapotentialthatisa onL andb onR . n n n n Inordertodefinethesecondtyperenormalizationoperator,wefirstintroducethemapH. Letk bean integerlargerorequalthan2;takeh(0)=00...0andh(1)=11...1;hencewedefine k k H(x1,x2,x|3,{.z..)}=(h(x1)h(x2|)h{(zx3})...). Itiseasytoseethats kH=Hs . Hence H(00...01...)=00...01... say,ifx L thenH(x) L n kn ∈ ∈ n kn and | {z } | {z } H(11...10...)=11...10... say,ifx R thenH(x) R n kn ∈ ∈ n kn Wenowdefinetherenormalizationoperatorasthemaponfunctionsdefinedasfollows: fixl suchthat | {z } | {z } 2 l kandtheconstants0 c <c <...<c k. Then 1 2 l ≤ ≤ ≤ ≤ RV(x):=V(s c1H(x))+V(s c2H(x))+ +V(s clH(x)) ··· OnecanseethattheRenormalizationoperatoroflastsection(thecasel=k)isaparticularcaseofthis newone. TheequationaboveforthepotentialVcanberewritten,analogouslytotheexpressionsintheprevious section,as Ra =a +a + +a n kn−c1 kn−c2 ··· kn−cl (andasimilarexpressionholdstoc ). n WeareinterestedonfindingafixedpotentialV forsuchrenormalizationoperator.Forapotentialonthe classofWaltersthefixedpointequationis a =a +a + +a n kn−c1 kn−c2 ··· kn−cl Thenwegetthefollowingresult: Lemma2. GivenN 1then ≥ RNV(x)=(cid:229) V(s jHN(x)) j where jisoftheform j=b k0+b k1+b k2+ +b kN 1 0 1 2 N 1 − ··· − forb chosenontheset c ,c ,...,c . i 1 2 l { } Proof. Indeed,theexpressionitobviouslyvalidforN =1(bydefinitionoftheoperatorR). Now,admit thatitholdsforN;weneedtoshowthatthisimpliestheexpressionforRN+1V. But RN+1V =R(RNV)=RNs c1H+RNs c2H+ +RNs clH= ··· (cid:229) V(s jHNs c1H)+... j Nowitiseasytosee(alsobyinduction)that Hs PH =s PkH2 and HPs H=s kPH2(?)= s kPHP+1 6 A.O.LOPES Withthistwoexpressionsweget HNs cH=s ckNHN+1 ThenwecanshowthatRN+1V is (cid:229) V(s jHNs c1H)+...= j (cid:229) V(s js c1kNHN+1)+...=(cid:229) V(s j+c1kNHN+1)+... j j where j=b k0+b k1+b k2+ +b kN 1;hencetheexpressionaboveisindeed 0 1 2 N 1 − ··· − (cid:229) V(s JHN+1)+... J whereJ=d k0+d k1+d k2+ +d kN 1+d kN,showingthatRN+1 isasclaimed. (cid:3) 0 1 2 N 1 − N ··· − NowwewanttoshowthatwehaveafixedpointofR thatintheneighborhoodof0¥ behaveslike 1 V(00...01...) ∼− logl n nlogk Aswewillseeinamomentitwillbenaturaltoconsiderthe(Cantorlike)setK = K(l,k)whichisthe | {z } closureoftheset x= (cid:229) aik−i forai c1,...,cl ,n Z { ∈{ } ∈ } n 1 ≥ Notethatwhenk=lthesetKwillbetheinterval[0,1]. Example3. Considerthefollowingparticularcase. DefineH :W = 0,1 N W by: 3 { } → H ((0,...,0,1,...,10,...,0,1,...))=(0,...,0,1,...,10,...,0,1,...), 3 c1 c2 c3 3c1 c2 c3 and | {z } | {z }| {z } | {z } | {z }| {z } H ((1,...,1,0,...,01,...,1,1,...))=(1,...,1,0,...,01,...,1,1,...). 3 c1 c2 c3 3c1 c2 c3 Inthiscasetherenormalizationoperatoris | {z } | {z }| {z } | {z } | {z }| {z } R (V)(x)=V(H (x))+V(s H (x))+V(s 2H (x)) 3 3 3 3 Ifx=(0,...,0,1,..)orx=(1,...,1,0,..)denoteV(x)=log c1 .Notethatforeachc wegetthat c1−1 1 c1 c1 3c 3c 1 3c 2 c | {z } | {z1 } 1 1 1 log +log − +log − =log . 3c 1 3c 2 3c 3 c 1 1 1 1 1 − − − − InthiscaseV isafixedpointfortherenormalizationoperator. WecanchoosethesignalofV andthen,wetakeV intheform 1 1 c 1 V(1,...,1,0,..) =V(0,...,0,1,..)= dt= log c1 c1 −Z0 c1−t − c1−1 | {z } | {z } Inordertofindagoodguess(inthegeneralcase)forthefixedpointpotentialconsider 1 U(x)= , foranyQ Qa when,x=0,...,0,1,..),or,x=1,...,1,0,..). Q Q Then,byiterationofU weget: | {z } | {z } RNU(x)=(cid:229) 1 =(cid:229) 1 1 (QkN j)a kNa (Q j )a j − j −kN AGENERALRENORMALIZATIONPROCEDUREONTHEONE-DIMENSIONALLATTICEANDDECAYOFCORRELATIONS 7 for j=b k0+b k1+b k2+ +b kN 1 (andb c ,...,c ). 0 1 2 N 1 − i 1 l Fromtheaboveseemsnatu··r·altotr−ytofindafixed∈p{ointV oft}heform 1 V(x)= dn (t), − K (Q t)a Z − when(0...0,1,..),forsomeprobabilityn ontheinterval. Q Examp|le{z4.}Letusfixk=3,l=2andc =0,c =2. InthiscasetheCantorsetK wegenerateisindeed 1 2 theclassicalCantorset. WecanwriteK=K K whereK [0,1/3]andK [2/3,1]. 0 2 0 2 ∪ ⊂ ⊂ NotethatthetransformationT(x)=3x(mod1)istwotoonewhenrestrictedtoK. Moreover,T[0,1/3]=[0,1]=T[2/3,1]. Letn bethemeasuresupportedonK andconsidera =log2/log3,thatistheHausdorffdimensionof K. Note that the Radon-Nykodin derivative of n in each injective branch of T restricted to [0,1/3] and [2/3,1]isequalto2. Wewanttoshowthatthefunction 1 V(0...01)= dn (t) − K (Q t)a Q Z − isafixedpointoftherenormalizationo|p{ezra}tor,i.e., V(x)=V(H(x))+V(s 2H(x)) Ifx=0...01thiscorrespondstoshowthat Q |{z} 1 1 1 dn (t)= dn (t)+ dn (t) K (Q t)a K (3Q t)a K (3Q 2 t)a Z − Z − Z − − Notethatthesecondexpressionis 1 1 1 dn (t)+ dn (t) = 3a K (Q t/3)a K (Q 2/3 t/3)a (cid:18)Z − Z − − (cid:19) 1 1 1 dn (t)+ dn (t) 2 K (Q t/3)a K (Q 2/3 t/3)a (cid:18)Z − Z − − (cid:19) Weremindthat 32a =1. Inthefirstintegralwecanusethechangeofvariablesx=t/3,thenweget 1 1 dn (t)= 2dn (x) ZK (Q−t/3)a ZK0 (Q−x)a Inthesecondweusex=2/3+t/3andweget 1 1 dn (t)= 2dn (x). ZK (Q−2/3−t/3)a ZK2 (Q−x)a Therefore, 1 1 1 dn (t)+ dn (t) = 3a K (Q t/3)a K (Q 2/3 t/3)a (cid:18)Z − Z − − (cid:19) 1 1 1 2 dn (x)+ dn (x) = 3a (cid:18)ZK0 (Q−x)a ZK2 (Q−x)a (cid:19) 1 1 dn (x)= dn (x) ZK0∪K2 (Q−x)a ZK (Q−x)a asclaimed. 8 A.O.LOPES Theorem5. Letxoftheform: x=00...01...,orx=11...10...,thentheoperator Q Q RV(x):=V|(s{zc1H}(x))+V(s c2|H{(zx))}+ +V(s clH(x)) ··· hasafixedpointV oftheform 1 V(x)= dn (t) − K (Q t)a Z − where a = logl, and where n is the measure over K that maximizes the entropy for the transformation logk T(x)=kx(mod1)actingonK . Proof. Thereasoningissimilartothelastexample.Considera = logl. logk ThetransformationT(x)=kx(mod1)actingonK isl toone. Themaximalentropyprobabilityn has entropylogl. There exists l sets K1,K2,...,Kl [0,1] such that T(Kj K)=K, and by consideringT restricted to Kj=[c /k,(c +1)/k], j=1,2,...,l⊂,wegetaninjectivemap∩.Considertheinducedprobabilityn =T (n ) j j j ∗ onK . TheRadon-Nykodinderivativeofn withrespectton ineachsetKj isequaltol. j j Ifx=(0...0,1,...)(or,x=(1...1,0,...))wewanttoshowthat Q Q |{z}V(x)= |1{z}dn (t)= (cid:229) l 1 dn (t)=RV(x). −ZK (Q−t)a − j=1ZK ((kQ−cj)−t)a Foreach jwehavethat 1 1 1 dn (t)= dn (t)= dn (t). ZK ((kQ−cj)−t)a ZK ka (Q−cjk+t)a ZK l(Q−ckj −kt )a Foreach jweconsiderthechangeofcoordinatet t/konthesetKj. → Then,wegetthat 1 1 dn (t)= ldn (y). K (Q cj+t)a Kj (Q y)a Z − k Z − As 1 dn (t)= (cid:229) l 1 dn (t) ZK (Q−t)a j=1ZKj (Q−t)a wegettheclaim. (cid:3) Notethat 1 dn (t) Q a − K (Q t)a ∼− − Z − forlargeQ. Inthiscaseh n willbeofordere−n1−a .Whenk=5andl=2wewillgetthattheequilibrium stateforfixedpointV hasadecayfasterthanne √n (seeAppendix). − 4. THEASSOCIATED JACOBIAN Nowweareinterestedontheestimationofthedecayofcorrelationassociatedtoequilibriumprobabili- tiesassociatedtothefixedpointpotentialswegetviatherenormalizationoperator. Wewillshowthatitis notalwaysofpolynomialtype. Infact,wewillpresentthedecayofcorrelationforageneralpotentialof Walterstype. Supposeg is of Walter type. If we add a constantto a potentialits equilibriumstate will notchange. Then,wecanassumewithoutlostofgeneralitythatthepotentialghaspressurezero. Theassociated JacobianJ forthe potentialdefinedby g atthe inversetemperatureb =1 is suchthat logJ=g+logj log(j s ),wherej istheeigenfunctionoftheRuelleoperatorforg(see(2). − ◦ Wewilldefiner(q)forq 1. ≥ WesplittheexpressionoflogJ(x)insixcases: AGENERALRENORMALIZATIONPROCEDUREONTHEONE-DIMENSIONALLATTICEANDDECAYOFCORRELATIONS 9 a) forq 2andx L orx R wehave q q ≥ ∈ ∈ ¥ logJ(x)=b a +log 1+h (cid:229) h q q (n+q 1) n=2 − ! ¥ log 1+h (cid:229) h (q 1) (n+q 2) − − n=2 − ! :=b a +logr(q) logr(q 1). k − − q b) forx=0111...10... L wehave 1 ∈ ¥ z }| { logJ(x)= log 1+h (cid:229) h = log(r(q)). q (n+q 1) − n=2 − ! − q c) forx=1000...01... R wehave 1 ∈ ¥ z }| { logJ(x)= log 1+h (cid:229) h = log(r(q)). q (n+q 1) − n=2 − ! − ¥ ¥ d) forx=0 orx=1 wehaveJ(x)=1. ¥ ¥ e) forx=10 orx=01 wehaveJ(x)=0. Notethatr(1)h =W. 1 Theequilibriumprobabilitym forghasJacobianJand Ll∗ogJ(m )=m . 5. DECAY OF CORRELATION In this section we want to estimate the decay of correlation of the observable I for the equilibrium [0] probabilitym atthecriticalinversetemperatureT whereb =1= 1. T Weassumefromnowonthatb =1. Thatis,wewillestimatetheasymptoticbehaviorwithkofthedecayofcorrelationoftheobservableI 0 (I s q)[I m [0]]dm a (q), [0] [0] 1 W ◦ − ∼ Z wherem istheequilibriumprobabilityforg,whenb =1. Wewillshow,asaparticularcase,thatifh q g,withg >2,then,a (q)goestozerolikeq2 g. The q − − ∼ nextresultisquitegeneral. Theorem6. Thedecayofcorrelationasafunctionofqfortheequilibriumprobabilitym (forthepotential g)fortheindicatorofthecylinder0is ¥ ¥ (cid:229) (cid:229) h (k+q+s) s=1k=0 inthecaseiswelldefined. Proof:Thetechniqueissimilartotheoneemployedin[14]and[10]. Firstweneedtoestimatetheasymptoticlimit m [0] Lq (I )(01..). 1 − logJ [0] Weclaimthat q ¥ m [0] Lq (I )(01..) (cid:229) (cid:229) h . (5) 1 − logJ [0] ∼ j n=1 j=q+1 Wewillpresenttheproofofthisfactlater. 10 A.O.LOPES Now,forafixedqandanys 2weneedtoevaluatethedifference ≥ m [0] Lq (I )(00...01..). 1 − logJ [0] s s |{z} For fixed q and variable s, let Bs =Lq (I )(00...01..), A(t)=Lt (I )(01...) and finally denote q logJ [0] logJ [0] psn= r(sr+(sn))hhss+n,n=1,..,q−2,s∈N. z}|{ Wedenotea (q,s)= h (s+q)r(s+q). h sr(s) Onecanshow(seefigure3page1092andexpressioninthebottomofpage1090[14]whereasimilar caseisdescribed)thatforanys,q 2 ≥ r(s+1) r(s+2) r(s+q 2) Bs =A +eas+1 A +eas+1+as+2 A +...+eas+1+as+2+...+aq+s 2 − A + q q−1 r(s) q−2 r(s) q−3 − r(s) 1 r(s+q) eas+1+as+2+...+aq+s 1+aq+s = − r(s) Aq−1+r(s+r(1s))hh s(s+1)Aq−2+ r(sr+(s2))hhss+2Aq−3+...+r(s+qr−(s2))hhs(s+q−2)A1+r(s+r(qs))hh (ss+q) = Aq−1+r(s+r(1s))hh s(s+1)Aq−2+ r(sr+(s2))hhss+2Aq−3+...+r(s+qr−(s2))hhs(s+q−2)A1+a (n,s). (6) Thisisnotakindofrenewalequation. Letusnowconsiderforn 1fixed,ands 2thesequenceVs=m [0] Bs. Wealsointroduce,forn ≥1,thesequence≥sV =m [0] A ,annd 1 − n n 1 n ≥ − Uns=−m [0](r(s+r(1s))hh s(s+1) + r(sr+(s2))hhss+2+...+r(s+qr−(s2))hhs(s+q−2))−a (n,s). Fromtheequation(6)wededucethatforqfixed m [0] Lq I (00...01..) =m [0] Bs =Vs=V +V ps+V ps+...+V ps +Us. (7) 1 − logJ [0] − q q q−1 q−2 1 q−3 2 1 q−2 q s Rememberthat(4)|c{lazim}s m 0s1 =h j (0s1...)=h r(s). (8) s 1 s Foreachfixedqwewillhavetoest(cid:0)imat(cid:1)elaterforeachsthevalue m (0s1) Lq I (00...01..) m [0] = h r(s) Lq I (00...01..) m [0] = logJ [0] − s logJ [0] − s s (cid:2) (cid:3) (cid:2) (cid:3) |{z} |{z} V h r(s)+V r(1+s)h +V r(2+s)h +...+V r(s+q 2)h r(s+q)h . (9) q 1 s q 2 1+s q 3 2+s 1 s+q 2 s+q − − − − − − AsmentionedbeforetheRuelleoperatorL isthedual(intheL2(W ,B,m )sense)oftheKoopman logJ 1 operatorK (j )=j s ,usingthisduality,expressions(9)and(15)wegetthatforfixedqthat ◦

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.