A lattice approach to the conformal OSp(2S + 2|2S) supercoset sigma model Part I: Algebraic structures in the spin chain. The Brauer algebra. Constantin Candu(1) and Hubert Saleur(1,2) ServicedePhysiqueThe´orique,CEASaclay, GifSurYvette,91191,France(1) DepartmentofPhysicsandAstronomy,UniversityofSouthernCalifornia LosAngeles,CA90089,USA(2) 8 Abstract 0 Wedefine and study a latticemodel which weargue is inthe universality class of the OSp(2S +2|2S) 0 supercosetsigmamodelforalargerangeofvaluesofthecouplingconstantg2. Inthisfirstpaper,weanalyze 2 σ indetailsthesymmetriesofthislatticemodel,inparticularthedecompositionofthespaceofthequantumspin n chainV⊗LasabimoduleoverOSp(2S +2|2S)anditscommutant,theBraueralgebraB (2). Itturnsoutthat L a V⊗L isanonsemisimplemoduleforbothOSp(2S +2|2S)and B (2). Theresultsareusedinthecompanion J L papertoelucidatethestructureofthe(boundary)conformalfieldtheory. 2 ] 1 Introduction h t - p The solution of the AdS × S5 worldsheet string theory is one of the cornerstones of the AdS/CFT duality 5 e program.Despitecontinuouseffortandprogressonclassicalaspectsinparticular[1],andthegenerallyaccepted h presenceofbothintegrabilityandconformalinvariancesymmetries,mostaspectsofthequantumtheoryremain [ elusive. 1 Itisnaturaltotrytounderstandsomeaspectsofthisquantumtheorybyfirsttacklingsimplermodelswith v similarproperties. ThesocalledOSp(2S +2|2S)cosetmodel-specifically,asigmamodelonthesupersphere 0 OSp(2S + 2|2S)/OSp(2S + 1|2S) - is a very attractive candidate for such an exercise: like the AdS × S5 3 5 4 worldsheettheoryitisconformalinvariantanditstargetspaceisasupergroupcoset. Ofcourse,itlacksother 0 aspectssuchastheBRSTstructureofthestringtheory. . Apart from the string theory motivation, models such as the OSp(2S + 2|2S) coset model are extremely 1 0 interestingfromthepureconformalfieldtheorypointofview.Indeed,theyaresigmamodelswhicharemassless 8 withoutanykindoftopologicalterm,andforalargerangeofvaluesofthecouplingconstantg2. Tomakethings σ 0 morepreciseletusbrieflyremindthereaderofsomegeneralities. Superspheresigmamodelshavetargetsuper v: spacethesupersphereSR−1,2S :=OSp(R|2S)/OSp(R−1|2S)andcanbeviewedasa“supersymmetric”extension i ofthenonlinearO(N)sigmamodels(whichdiffersofcoursefromtheusualO(N)“supersymmetric”models). X Useascoordinatesarealscalarfield ar φ:=(φ1,...,φR+2S) wherethefirstRcomponentsarebosons,thelast2S onesfermions,andtheinvariantbilinearform φ.φ′ = J φiφ′j ij X whereJ istheorthosymplecticmetric I 0 0 R J = 0 0 −IS Idenotingtheidentity.Theunitsupersphereisdefin0edbIyS the0constraint φ.φ=1 Theactionofthesigmamodel(conventionsarethattheBoltzmannweightise−S)reads 1 S = d2x∂ φ.∂ φ 2g2 Z µ µ σ 1 TheperturbativeβfunctiondependsonlyonR−2S toallorders(see,e.g.,Ref[2]),andisthesameastheone oftheO(N)modelwithN :=R−2S.Physicscanbereliablyunderstoodfromthefirstorderbetafunction β(g2)=(R−2S −2)g4 +O(g6) σ σ σ The modelfor g2 positiveflows to strongcouplingfor R−2S > 2. Like in the ordinarysigma modelscase, σ thesymmetryisrestoredatlargelengthscales,andthefieldtheoryismassive. ForR−2S < 2meanwhile,the modelflowstoweakcoupling,andthesymmetryisspontaneouslybroken. Oneexpectsthisscenariotowork for g2 small enough, and the corresponding Goldstone phase to be separated from a non perturbative strong σ couplingphasebyacriticalpoint. The case we areinterested in hereis R−2S = 2, where the βfunctionvanishesto all ordersin perturba- tion theory, and the model is expectedto be conformalinvariant, at least for g2 small enough, the Goldstone σ phase beingreplacedby a phase with continuouslyvaryingexponentsnotunlikethe low temperatureKoster- litz Thouless phase. How the group symmetry combines with the (logarithmic)conformalsymmetry in such modelsislargelyunknown.Itisanessentialquestiontobesolvedbeforeanyseriousattemptstounderstanding universalityclassesinnoninteractingdisordered2Delectronicsystemscanbecontemplated[3]. TheOSp(2S +2|2S)cosetmodelwasconsideredinparticularintwopapersbyMannandPolchinskiusing the massless scattering and Bethe ansatz approaches. This is indeed a natural idea, since supersphere sigma modelsare in generalintegrable, and, when massive (ie R−2S > 2) can be describedby a scattering theory involvingparticlesinthefundamentalrepresentationofthegroup.TheS matrixiswellknown Sˇ(θ)=σ (θ)E+σ (θ)P+σ (θ)I 1 2 3 Here,Iistheidentity,Pisthegradedpermutationoperator,andEisproportionaltotheprojectorontheidentity representation.ForR,S arbitrary,factorizabilityrequiresthat 2iπ σ (θ) = − σ (θ) 1 (N−2)(iπ−θ) 2 2iπ σ (θ) = − σ (θ) 3 (N−2)θ 2 whereN = R−2S,whileσ itselfisdetermined,uptoCDDfactors,bycrossingsymmetryandunitarity. One 2 immediatelyobservesthatwhenN =2,theamplitudeσ cancelsout,leavingascatteringmatrixwithasimpler 2 tensorialstructure,sincethePoperatordisappears.Thiscorrespondstoaparticularpoint[4]onthesigmamodel criticalline(where,amongotherthings,thesymmetryisenhancedtoSU(2S +2|2S)),therestofwhichisnot directlyaccessiblebythisconstruction.1 Theideausedin[5] istoconsiderananalyticalcontinuationtoR,S real,andanapproachtoR−2S =2withproperscalingofthemass. Thoughinterestingresultswereobtained, theemphasisinthesepaperswasnotonconformalproperties. Anotherlineofattack,moresuitedtotheconformalaspects,waslaunchedbyReadandSaleurin2001[4], who proposedto use a lattice regularizationto controlthe integrablefeaturesof the model. They obtainedin thiswaythe spectrumofcriticalexponentsforseveralrelatedsigmamodelsonsupertargetspaces, including the OSp(2S +2|2S)cosetoneat a particular(critical)valueof the couplingg2. The resultsexhibitedseveral σ mysterious features, including a pattern of large degeneracies, and a set of values of the exponents covering (modulointegers)alltherationals. Intwo subsequentpapers[6, 7], itwasarguedfurtherthatmanyalgebraic propertiesoftheconformalfieldtheorycouldbeobtainedatthelatticelevelalready. Theseincludefusion,and thestructureofconformal“towers”(seebelowforfurtherdetails). The work we present in this paper and its companion is an attempt at understanding the conformal field theoreticdescriptionoftheOSp(2S+2|2S)modelforallvaluesofthecouplingbyusingalatticeregularization. Foremostinthelatticeapproachistheunderstandingofthealgebraicstructureofthelatticemodel-thealgebra defined by the local transfer matrices and its commutant. While in the cases discussed in [7] most necessary resultswerealreadyavailableinthemathematicalliterature,thesituationhereismuchmorecomplicated: ina fewwords,wehavetodeal,insteadoftheTemperleyLiebalgebra,withtheBraueralgebrawhoserepresentation theory,inthenonsemi-simplecase,isfarfromfullyunderstood. Animportantpartofourworkhasconsisted in filling up the necessary gaps of the literature, sometimes rigorously, but sometimes at the price of some 1ThecaseR=2,S =0isspecialandallowsforanextensionoftheSmatrixtothewholeO(2)criticalline. 2 conjectures.Thisalgebraicworkisthesubjectofthefirstpaper,whichwerealizemightbeabithardtoreadfor aphysicsreader. Wecapitalizeonthealgebraiceffortinthecompanionpaper,wheretheboundaryconformal fieldtheoryforthecosetsigmamodelisanalyzedthoroughly. InthesecondsectionofthispaperwediscussgeneralitiesaboutlatticeregularizationsofO(N)sigmamodels in 2 dimensions and define the model we shall be interested in. In section 3, the transfer matrix, the loop reformulationand the associated Brauer algebra are introducedand discussed. Section 4 is the main section, wherethefulldecompositionoftheHilbertspaceofthelatticemodelundertheactionofOSp(2S +2|2S)and B (2) is obtained. Our main resultcan be foundin eqs. (4.36) and (4.37). Section 5 discusses aspects of the L hamiltonianlimitandsection6containsconclusions. Technicalaspectsofrepresentationtheoryarediscussed furtherintheappendices. Forthereader’sconvenience,weprovideherealistofnotationsusedthroughoutthepaper: • osp(R|2S)istheLiesuperalgebraofthesupergroupOSp(R|2S) • B (N)istheBraueralgebraonLstringswithfugacityforloopsN =R−2S L • V = CR|2S isthemod2gradedvectorspaceCR⊕C2S withevenpartV = CR andoddpartV = C2S. R|2S 0 1 WeshalloftendroptheindicesR,2S inV . R|2S • V⊗L isconsideredasaleftosp(R|2S)andrightB (N)bimodule L • λ⊢ Lstandsfor“λisapartitionofL”andλ′isthepartitionλtransposed • Sym(L)andCSym(L)arethesymmetricgrouponLobjectsanditsgroupalgebra • T (q)istheTemperleyLiebalgebrawithfugacityforloopsq+q−1 L • dandDaregenericelementsofB (N)andOSp(R|2S) L • X = {µ ⊢ L−2k | k = 0,...,[L/2]}isthesetofpartitionslabelingtheweightsof B (N). X (S) ⊂ X L L L L selectsthoseofthemwhichdorealizeonV⊗L • Associateweightsλ,λ∗arelabelsofOSp(R|2S)irrepswhicharenonequivalent(identical)andbecome(split intotwo)isomorphic(nonequivalent)irrepsundertherestrictiontothepropersubgroupOSp+(R|2S)ofsu- permatriceswithsdetD=+1 • H (S) = {λ ∈ X (S) | λ ≤ S } is the set of hookshape partitionslabelingthe weightsof osp(R|2S) L L r+1 irrepsappearinginV⊗L andY (S)= H (S)∪H (S)∗ L L L • ∆ (µ)arestandardorgenericallyirreduciblerepresentationsofB (N) L L • S(λ),g(λ),G(µ),B (µ),D (j)areirreduciblerepresentationsofCSym(L),osp(R|2S),OSp(R|2S),B (N) L L L andT (q)respectively. L • Ig(λ),IG(µ),IB (µ)aredirectsummandsofV⊗L asaosp(R|2S),OSp(R|2S)andB (N)module L L • sc arethesupersymmetricgeneralizationofO(N)symmetricfunctions λ • χ′(d),χ (d),sch (D)arecharactersof∆ (µ),B (µ)andG(λ). µ µ λ L L 2 The OSp(R|2S) lattice models: generalities 2.1 The models andtheir loopreformulations LatticediscretizationsofO(N) sigmamodelshavehada longhistory. Thesimplestwayto gois obviouslyto introduce spins taking values in the target manifold - the sphere O(N)/O(N − 1) - on the sites of a discrete lattice, with an interaction energyof the Heisenbergtype E = −J S~ .S~ (where . stands for the bilinear <ij> i j O(N)-invariant quadratic form). This is however difficult to studyPtechnically, as the number of degrees of freedomoneachsiteisinfinite. Apossiblewaytogoistodiscretizethetargetspace,leadingtovarioustypesof 3 “cubicmodels”[8].Anotherwaywhichhasprovedespeciallyfruitfulintwodimensionshasbeentoreformulate theproblemofcalculatingthepartitionorcorrelationfunctionsgeometricallybyusingthetechniquesofhighor lowtemperatureexpansions,thusobtaininggraphswithcomplicatedinteractionrulesandweightsdeterminedby propertiesoftheunderlyinggroups.Thesimplestoftheseformulationsappearedin[9]wheretheauthorsstudied theO(N)modelonthehoneycomblatticeintwodimensions,andreplacedmoreovertheterm eβJS~i.S~jbyits <ij> considerablysimplerhightemperatureapproximation (1+KS~ .S~ ),K =βJ[10].ExpaQndingthebrackets, <ij> i j insaythecalculationofthepartitionfunction,onecanQdrawgraphsbyputtingabondbetweenneighboringsites iand jwheneverthetermS~ .S~ ispickedup.Theintegraloverspinvariablesleavesonlyloops,withafugacity i j equaltoNasthereareN colorsonecancontract.Notethatbecauseoftheverylowcoordinationnumberofthe honeycomblattice,onlyself-avoidingloopsareobtained. Thisleadstothewellknownself-avoidingloopgas partitionfunction: Z = KENL (2.1) SAL XG where the sum is taken over all configurationsG of self avoiding, mutually avoiding closed loops in number L, covering a total of E edges. Note that once an expression such as (2.1) is written down, it is possible to analyticallycontinuethedefinitionofthemodelforNanarbitraryrealnumber.Barringtheuseofsuperalgebras, onlyNintegergreaterorequaltoonehasawelldefinedmeaningasaspinmodel(thecaseN =1coincideswith theIsingmodel2). Intwodimensions,theMerminWagnertheorempreventsspontaneoussymmetrybreaking, so for N integer, critical behavior can only occur for N = 1,2. Analysis of the same beta function suggests howeverthatlatticemodelsdefinedbysuitableanalyticcontinuationshouldhaveaGoldstonelowtemperature phaseforallN < 2,thoughitsaysnothingaboutwhetherthisphasemightendbyasecondorfirstorderphase transition. Model (2.1) lacks interaction terms which would appear with less drastic choices of the lattice and the interactions: these are the terms where the loops intersect, either by going over the same edge, or over the same vertex, maybe many times. It has often been argued that such terms are irrelevant for the study of the criticalpointsoftheO(N)modelsintwodimensions.Mostoftheinteresthasfocusedonsuchcriticalpointsfor N ∈[−2,2],whichhavegeometricalapplications-inparticularthecaseN =0isrelatedwiththephysicsofself- avoidingwalks. Itturnsouthoweverthatintersectiontermsarecrucialfortheunderstandingoflowtemperature phases. Indeed,themodel(2.1)doeshaveasortofGoldstonephaseforN ∈[−2,2]calledthedensephase,but its propertiesare notgeneric, and destroyedbythe introductionof a smallamountof intersections. A simple waytoseethatthedensephaseisnotgenericisthattheexponentsatN = 2arealwaysthoseoftheKosterlitz Thoulesstransitionpoint: model(2.1)doesnotallowonetoenterthelowtemperaturephaseoftheXYmodel. Also, model(2.1) has a first order transition for N < −2, which is notthe behavior expectedfrom the sigma modelanalysis. Itwassuggestedin[11]thatmodel(2.1)canberepairedbyallowingforsomeintersections. Theminimal scenarioone can imagine is to define a similar modelon the square lattice, and allow for self intersectionsat verticesonly,soeithernone,oneortwoloopsgothroughthesamevertex.Theresultingobjectsareoftencalled trails. Thisgivesthenewpartitionfunction Z = KENLwI T XG′ wherethesumistakenoverallconfigurationsG′ ofclosedloops,whichvisitedgesofthelatticeatmostonce, innumberL,coveringatotalofE edges,withI intersections. Thephasediagramofthismodelhasnotbeenentirelyinvestigated. Itisexpectedthatatleastforwsmall enough, the critical behavior obtained with K = K , w = 0 is not changed (though K is), while the low c c temperaturebehaviorK >K willbe. c In [11] a yet slightly different version was considered corresponding, roughly, to the limit of very large K, whereallthe edgesof thelattice arecovered. Thepartitionfunctionofthisfullypackedtrailsmodelthen dependsononlytwoparameters: Z = NLwI (2.2) FPT XG′′ 2ThegroupO(2)isdifferentfromSO(2)≃U(1)becauseoftheadditionalZ2freedominchoosingthesignofthedeterminant. 4 Figure1: Dense intersectingloopcoveringofa lattice with annulusboundaryconditions. Illustrationof bulk (B),contractible(C),even(E)andodd(O)loops.Periodicimaginarytimerunsvertically a b c Figure2:VerticesoftheZ symmetricsixvertexmodel 2 and an example of allowed configuration is given in figure 2.1. Numerical and analytical argumentssuggest strongly that this model has all the generic properties of the O(N) sigma model in the spontaneously broken symmetry phase (such that one might derive from analytic continuation in N of the RG equations), for all N ≤2. Notethatexpression(2.2)canbeobtainedverynaturallyifinsteadofputtingthedegreesoffreedomonthe vertices, one putsthem on the edgesof the lattice. In thiscase, the minimalformof interactioninvolvestwo edgescrossingatonevertex.InvarianceundertheO(N)groupallowsforthreeinvarianttensorsasillustratedon thefigure8,whileisotropyandinvarianceunderanoverallscalechangeoftheBoltzmannweightsleavesone withasinglefreeparameter,thecrossingweightw. Graphicalrepresentationofthecontractionsontheinvariant tensorsreproduceseq.(2.2),aswillbediscussedbelow. ForN < 2,model(2.2)flowstoweakcouplingintheIR,andthereforeitisexpectedthatthecriticalprop- ertiesofthecorrespondinglowtemperature(Goldstone)phasedonotdependonw,afactcheckednumerically in[11]. Thecase N = 2isexpectedtobedifferent: asmentionedalreadyintheintroduction,thebetafunction of the correspondingsigma modelis exactly zero so the couplingconstantdoes notrenormalize. It is indeed easytoseethattheloopmodel(2.2)withN = 2isequivalenttothe6vertexmodelwitha= b= 1+w, c = 1. Considertheverticesofthe6vertexmodelasrepresentedonfigure2. Wechoseisotropicweightsa=b, c. We candecideto splittheverticesofaconfigurationintopiecesoforientedloopsasrepresentedonfigure3. For eachvertex,therearetwopossiblesplittings, andwe assumethattheyarechosenwith equalprobability. The loopsobtainedbyconnectingallthepiecestogetherprovideadensecoveringofthelattice,andcomewithtwo possibleorientations,henceafugacityoftwooncetheorientationsaresummedover. Theloopscanintersect, withaweightwgivenfromtheobviouscorrespondence: a b 1+w = = c c 2 andthus a2+b2−c2 2 ∆= =1− 2ab (1+w)2 WenotethatthereareindeedthreeinvarianttensorsforthecaseofO(2). Thecorrespondingprojectorsare E, 1(I −P), 1(I −E +P). Theyprojectrespectivelyontwoonedimensionalrepresentations,andonasingle 2 2 twodimensionalone. 5 = + = + = + Figure3: Themappingofthesixvertexmodelontotheorientedloops Theparameter∆coverstheinterval[−1,1]asw∈[0,∞].Changing∆iswellknowntochangetheexponents of6vertexmodel,andthereforeeq.(2.2)for N = 2exhibitsacriticalline,whichisinfactintheuniversality classofthecontinuousXY(O(2))modelinthelowtemperature(KosterlitzThouless)phase. All what was said so far can be easily generalized to the case of spins taking values on a supersphere OSp(R|2S)/OSp(R−1|2S). The fugacity of loops is now equal to R− 2S: this combination is the number ofbosonicminusthe numberof fermioniccoordinates,andfollowsfromthe usualfactthatwhen contracting fermionsalongaloop,aminussignisgenerated3,seesec.3.3. Theloopmodelformulationthereforeprovides a convenient graphical representation of the discrete supersphere sigma models for all R − 2S, in particular R−2S ≤2whereinterestingphysicsisexpectedtooccur. Thisphysicswasexploredin[11],andtheexpected resultswereobtainedforR−2S <2. ThepurposeofthispaperistoexplorethemorechallengingR−2S = 2 case. Ofcourse,atthenaivelevelofpartitionfunctionsandwithoutworryingaboutboundaryconditions,itlooks asifthereisnodifferencebetweentheO(N)spinmodelanditssuperspherecousinsprovidedR−2S = N. The pointisthattheobservablesofthemodelsaredifferentor,attheveryleast,comewithdifferentmultiplicities. Indeed, consider for instance correlation functionsof spin variables. In the O(2) case, the spin has only two componentsS1,S2, so one cannot build a totally antisymmetric tensor on three indices. This means that the correspondingoperator(whichhasa nicegeometricalinterpretationto be givenin the nextpaper)willnotbe presentinthiscase,thoughitwillbeintheOSp(2S +2|2S)modelwhenS >0. Notethatingeneral,correlators involvingspinswithin the first R bosonicandthe first 2S fermioniclabelswill be the same foranychoiceof groupOSp(R′|2S′)withR′−2S′ =R−2S andR′ ≥R.4 Thisisimmediatelyprovedbyperformingagraphical expansionofthecorrelator:variablesoutsideofthesetofthefirstRbosonicandthefirst2S fermioniclabelsare notgettingcontractedwiththespinsinthecorrelators,andcancelagainsteachotherintheloopcontractions. Astandardtricktoextractthefulloperatorcontentofamodelistostudythepartitionfunctionwithdifferent boundary conditions. Consider for instance the spin model on an annulus with some symmetry preserving boundaryconditionsinthespacedirection.Withwhatwewillcallperiodicboundaryconditions(corresponding totakingthesupertraceoftheevolutionoperator)inthetimedirection,representationsofOSp(2S +2|2S)will always be counted with their superdimension, and the partition function will be identical with the one of the O(2) case. But if we take antiperiodicboundaryconditions, we will get a modified partition function (in the sense of [4]) which is a trace over the Hilbert space instead of a supertrace, counts all observables with the multiplicities(notsupermultiplicities),andwillturnouttobeaverycomplexobject. A goodalgebraicunderstandingof the lattice modelwillbe essential to make furtherprogress, and, since theareaislargelyunexplored,thiswilloccupyusformostoftherestofthisfirstpaper. 3ThegeneralizationofresultsforO(N)modelstothecaseoforthosymplecticgroupsdatesbacktotheworkofParisiandSourlas[12]. 4Provided,ofcourse,thattheboundaryconditionsimposedontheR′bosonicand2S′fermionicdegreesoffreedomarethesamein bothcases. 6 3 Transfer matrices and algebra 3.1 Transfer matrices As discussed briefly in the introduction, the OSp(R|2S) spin model we consider is most easily defined on a squarelatticewithdegreesoffreedom(states)ontheedgesandinteractionstakingplaceatvertices. Thesetof statesoneveryedgeisacopyofthebasespaceVofthefundamentalOSp(R|2S)representation.Interactionsata vertexcanbeencodedinalocaltransfermatrixtactingonV⊗2andcommutingwiththeOSp(R|2S)supergroup action.Wecalltanintertwinerandwritet∈End V⊗2. OSp(R|2S) TheBoltzmannweightsofthemodelarecomponentsofthetransfermatrixalongabasisofintertwiners. A naturalchoiceof basis arethe projectorsontoOSp(R|2S)irreduciblerepresentationsappearingin the decom- positionofthetensorproductoftwofundamentalOSp(R|2S)representations. Tofindthemonecanapplythe same(anti)symmetrizationandtracesubstractiontechniquesusedforreducingO(N)tensorrepresentations. If e ,...,e isamod2gradedsetofbasisvectorsinV withgradingg,thedecompositionofV⊗2willread 1 R+2S 1 2J e ⊗e = e ⊗e +(−1)g(i)g(j)e ⊗e − ij Jkle ⊗e (3.1) i j 2 i j j i R−2S k l +1e ⊗e −(−1)g(i)g(j)e ⊗e + Jij Xk,l Jkle ⊗e. 2 i j j i R−2S k l (cid:16) (cid:17) Xk,l Here J is the OSp(R|2S)invarianttensor, Jij = (J−1) , and g(i) = 1 (resp. g(i) = 0) if i is fermionic(resp. ij ij bosonic). Each of the threetermson the l.h.s. of (3.1) transformsaccordingto an irrepsofOSp(R|2S),or, in otherwords,belongstoasimpleOSp(R|2S)module. IntroducetheidentityI,thegradedpermutationoperatorP(alsoknownasbraidoperator),andE theTem- perleyLieboperator(proportionaltotheprojectoronthetrivialrepresentation), I kl =δkδl, P kl =(−1)g(i)g(j)δkδl, E kl = JklJ . (3.2) ij i j ij i j ij ij IntermsofprojectorsontoirreducibleOSp(R|2S)modules,eq.(3.1)maybewritteninamoreelegantwayas 1 2 1 1 I = I+P− E + (I−P)+ E. 2 R−2S ! 2 R−2S LetPdenoteasusualtheinversionofspace,TtheinversionoftimeandCthechargeconjugationwiththe matrix J. One can check directly from definition (3.2) that P is C , P and T invariant, while E is P and 12 12 12 C T invariant. Moreover,E andI transformintoeachotherundertheπ/2rotationofthelatticeR,while E 12 12 andParerelatedbythecrossingsymmetryC T 1 1 R: Eijkl −→ Jjj′Ek′lij′Jk′k =Iijkl C1T1 : Eijkl −→ Jii′Ek′i′jlJk′k = Pijkl. TakeI,E andPasbasisofintertwinersinEnd V⊗2. Thelocaltransfermatrixgenerallydependson OSp(R|2S) threeindependentweightsw ,w andw . However,onahomogeneousandisotropiclatticeonecannormalize I E P w =w =1andleaveonlytheweightw=w . Finally,thelocaltransfermatrixtakestheform I E P t(w)=I+wP+E. (3.3) Ona diagonallattice withopenboundariesrepresentedinfig. 4choosethe timein verticaldirection. The notation of sites at a fixed time is such that the left edge i and right edge i+1 meet at vertex i. Let t(w) ∈ i End V⊗L denote a transfer matrix acting nontrivially only at vertex i according to eq. (3.3). From the OSp(R|2S) figureitisclearthatoddandeventimesareinequivalent. ThetransfermatrixT,propagatingonestepforward atequivalenttimes,maybewrittenasaproductT =YX ofonelayertransfermatrices [(L−1)/2] [L/2] X = t , Y = t , (3.4) 2i 2i−1 Yi=1 Yi=1 7 Y X Figure4:TheonelayertransfermatricesX andY representedonadiagonallatticeofwidth6. schematicallyshowninfig.4. The simplestway to define a partitionfunctionthatdependson the wholespectrumof the transfermatrix T is by taking the trace of T at a certain power β. Selecting other boundaryconditionswith some nontrivial symmetrygenerallyamountstorestrictingthewholespaceofstatesofthemodeltoasubspacecompatiblewith thesymmetryofchosenboundaryconditions. Whatexactlywe meanby “symmetryofboundaryconditions” willbeexplainedlaterinsec.3.3. Forthemomentletusjustsaythatitisconvenienttoconsideramoregeneral classofboundaryconditions,calledquasiperiodic,inwhichTβis“twisted”bytheactionofanelementDofthe supergroup.Definethequasiperiodicpartitionfunctiontobe Z =str D⊗LTβ. (3.5) D V⊗L Wemusttakethesupertraceineq.(3.5)ifwewantthequasiperiodicpartitionfunctiontobewelldefined. For instance,whenD= J2wegettheusualtracepartitionfunctionandwhenDequalstotheidentitymatrixweget thesupertracepartitionfunction. NotethatbecauseDisasupermatrix,thetensorproductinD⊗L hastobegraded,thatis D⊗2·η⊗ξ = D·η⊗D·ξ ⇒ D⊗2 ij =(−1)g(k)(g(j)+g(l))Di Dj kl k l (cid:16) (cid:17) Afterinsertingthelocaltransfermatrixfromeq.(3.3)ineq.(3.4)andexpandingthetransfermatrixT,the quasiperiodicpartitionfunctionreadsasasumofweightedproductsofE’sandP’s. Suchlinearlyindependent i i productsmustbeconsideredaswordsofatransfermatrixalgebra,whileintertwinersE andP aregenerators i i ofthisalgebra.InthenextsectionweidentifythisalgebraasarepresentationoftheBraueralgebra. 3.2 The Brauer algebra ForanabstractintroductiontotheBraueralgebraseeref.[13,14]whileinthecontextofosp(R|2S)centralizer algebraseeref.[15]. WecollectinthissectionsomewellknownfactsabouttheBraueralgebraweshallusein thenextsections. LetE andP,i=1,...,LactnontriviallyasE andPineq.(3.2)onlyatthesitesV ⊗V ofV⊗L. Onecan i i i i+1 checkthatforP andE sodefinedthefollowingrelationshold: i i P2 =1, E2 = NE, EP = PE = E, (3.6) i i i i i i i i PP = P P, EE = E E, EP = E P, i j j i i j j i i j j i PP P = P PP , EE E = E, i i±1 i i±1 i i±1 i i±1 i i PE E = P E, EE P = EP . i i±1 i i±1 i i i±1 i i i±1 Inthesecondlineoftheserelationsiand jaresupposedtobenonadjacentsites. Relations(3.6)(isoneofthemanywaysto)definetheB (N)Braueralgebra(alsodenotedsometimesbythe L namesof braid-monoidalgebraordegenerateBirman-Wenzel-Murakamialgebra[16, 17, 18]). Note thatthis algebradependsonasingle,generallycomplex,parameterN,andcontainsthemaybemorefamiliarTemperley Liebalgebra,generatedbyE’salone,andthesymmetricgroupalgebra,generatedbyP alone. i i For N fixedand Lbigenough,theOSp(R|2S)spinmodelsprovidehighlyunfaithfulrepresentationsofthe BraueralgebraB (N). Thisisbecause,inV⊗L,thegeneratorsP andE satisfyadditionalhigherorderrelations L i i Rontopof (3.6).5 Forasimpleexample,considertheO(2)spinmodelonalatticeofwidth3. Theprojector 5This situation is similar to what happens for models with SL(N) symmetry in the fundamental representation, and corresponding quotientsoftheHeckealgebra. 8 Figure5: ThegraphicalrepresentationofthewordP P E P inB (N). 5 3 1 2 6 1 L 1 i i+1 L 1 i i+1 L I E P i i Figure6:GraphicalrepresentationofgeneratorsI,E andP. i i onto the antisymmetric tensor of rank 3 is zero, thus, R contains the additional relation 1+ P P + P P = 1 2 2 1 P +P +P P P . OurspinmodelsingeneralproviderepresentationsofthequotientalgebrasB (N)/R. The 1 2 1 2 1 L setofrelationsRcanbeexplicitlydescribedforS =0,see[19,20]andreferencestherein,andwehavelittleto sayaboutthecaseS >0. The first step in understanding the spectrum of the transfer matrix T brings up the question of B (N) ir- L reduciblerepresentations, and of their multiplicities in T for a particular choice of R and S. This leads us to discussingsomeresultsabouttherepresentationtheoryoftheBraueralgebra. Themostnaturalrepresentationtobeginwithistheadjointrepresentation. Itadmitsadiagrammaticrepre- sentationintermsofgraphson2Lpointsinwhicheveryvertexhasdegree1. Usuallyoneordersthe2Lpoints ontwohorizontalparallellinesasshowninfig.5. LetB denotethevectorspacespannedonthe(2L−1)!!such L diagrams. Theproductd ∗d oftwodiagramsd andd isperformedbyputtingd ontopofd andreplacingeach 1 2 1 2 1 2 oftheloopsintheresultingdiagramwith N. Definediagrammaticallytheidentity I andthegeneratorsE, P i i asrepresentedinfig.6. Onecancheckthatthegraphicalrepresentationofgeneratorswiththemultiplication∗ ofdiagramssatisfyalloftheeqs.(3.6). TheleftactionofgeneratorsonB viathemultiplication∗ofdiagrams L providetheadjointrepresentationofB (N). L From the graphicalrepresentation we see that B has a series of invariantsubspaces B = BL ⊃ BL−2 ⊃ L L L L ··· ⊃ Bτ, where Bm is spanned on diagrams with fewer than m vertical lines and τ = L mod 2. The vector L L spacespannedondiagramswithexactlymverticallinesmaybedefinedasa B (N)modulebythecosetB′m = L L Bm/Bm−2. TheleftactionofB (N)onthismodulesmaybeseenasamodifiedmultiplication∗ ofdiagrams L L L m d ∗d , ifithasmverticallines d ∗ d = 1 2 1 m 2 0, otherwise Under the left action of the algebra theposition of horizontal lines in the bottom of a diagram does not change.Foragivenconfigurationofthehorizontallinesinthetopofadiagramandagivenpatternofintersec- tionsofverticallinesthereare(L−m−1)!!Cmpossibilitiesofchoosingtheconfigurationofhorizontallinesin L thebottomofthediagram.ThissimplymeansthatB′mdecomposesintoadirectsumof(L−m−1)!!Cmequiv- L L alentmodules. ThecosetrepresentativeB′′m oftheseequivalentleftmodulesisspannedonm!(L−m−1)!!Cm L L graphsonLpointswitheveryvertexhavingdegree0or1andalabelingwithnumbers1,...,moffreevertices. Anexampleofsuchalabeledgraphisshowninfig.7. Ifthelabellingsareomittedtheresultinggraphiscalled apartialdiagram. ThelabelingofthemfreepointsofalabeledgraphisapermutationπinthesymmetricgroupSym(m). The labeledgraphswillprovidearepresentationoftheBraueralgebra,whichisirreducibleforgenericvaluesofN, ifwetakethelabellingsπinanirreduciblerepresentationofSym(m). Wecallsuchrepresentationsgenerically irreducible. Letµbeapartitionofm,whichwewriteasµ ⊢ m. Inamorealgebraiclanguagethedefinitionof genericallyirreducibleleftmodulestranslatesto ∆ (µ):= B′′m⊗ S(µ), µ⊢m, (3.7) L L Sym(m) whereS(µ)isanirreducibleSym(m)module.Inviewoflaternumericalanalysiswegivebelowabasisin∆ (µ) L anddescribetheactionofB (N)onthisbasis. L 9 2 1 4 3 Figure7:Thetopofthediagramrepresentedinfig.5. Thelabelingpermutationis(2,1,4,3). Let p⊗ π denote the labeling of a partial diagram p with the permutation π, v ,...,v be a set of basis 1 fµ vectorsinS(µ)andρ (σ)bethematrixofthepermutationσintherepresentationρ . Anaturalbasisin∆ (µ) µ µ L isgivenbyallpairs p⊗v. Theactionofadiagramd ∈ B (N)onabasisvectoris i L d∗ p⊗ρ (σ−1)v, ifd∗phasmfreepoints d·p⊗v = m µ i (3.8) i 0, otherwise, where σ is the labeling of d ∗ g andg is the partial diagram p labeled with the identity permutation. The dimensionsd of∆ (µ)is f (L−m−1)!!Cm. µ L µ L Insimplewords,agenericallyirreduciblemoduleisaspanongraphsonLpoints,obtainedbychoosingm pointsamongL,pairingalltheothers(thisgivesthemultiplicity(L−m−1)!!sinceintersectionsareallowed), choosingforthemunpairedonesarepresentationofthepermutationgroupandsettingtozerotheactionofany Brauerdiagramthatreducesthenumbermofunpairedpoints. Thegenericallyirreduciblerepresentationslabeledbyµ ⊢ L−2k,k = 0,...,[L/2]appearinthedecompo- sitionoftheadjointrepresentationwithmultiplicitygivenbytheirdimensiond whenB (N)issemisimple. µ L Letusconcludewithafewwordsaboutthereducibilityofgenericallyirreduciblemodules∆ (µ).Forinteger L N and a numberof strings L > N the Brauer algebrais notsemisimple and, as a consequence,certain of the modules∆ (µ)becomereducible,thoughtheyremainindecomposable.6Theirreduciblecomponentsappearing L insuchreduciblemodules∆ (µ)arefarfrombeingunderstood(thesituationismuchworsethaninthecaseof L thenonsemisimpleTemperleyLiebalgebra[21,22]). Numericalcomputationsbasedonthediagonalizationof thetransfermatrixinthediagrammaticrepresentationofB (N)restrictedto∆ (µ)decreasesinefficiencyvery L L fastwithincreasingL, comparedtotheidealcasewherethetransfermatrixisrestrictedtoanirrepsof B (N). L ThisisbecauseforbigLandµfixedthenumberofirreduciblecomponentsin∆ (µ)“goeswild”andthereare L alotof“accidentaldegeneracies”inthespectrumofthetransfermatrixrestrictedto∆ (µ). L However, a significant progress in this direction has been recently made in [23, 24]. Let us note that, as described in [23], the content of (at least some) ∆ (µ)’s can be computed by repeated applications of Frobe- L nius reciprocity applied to the short exact sequence of [14] describing the structure of the induced modules B (N)⊗ ∆ (µ). L+1 BL(N) L In the end we recall the basic results for the Temperley Lieb algebra, to allow a quick comparison with Brauer. TemperleyLiebalgebradiagramsareasubsetofBraueralgebradiagramssubjecttotheconstraintthat no intersections between edges are allowed. The dimension of the algebra is given by the Catalan numbers (2L)!/L!(L+1)!. Themainlineofreasoningforfindinggenericallyirreduciblemodulesfollowsthesameway, except there is no available action of the symmetric group on vertical lines. Therefore, the analogue of the labeledgraphswillbethepartialdiagrams,inwhichnofreepointsmaybetrappedinsideanedge.Thenumber ofsuchgraphsisCn −Cn−2,wheren=(L−m)/2isthenumberofedges.Thegenericallyirreduciblemodules L−1 L−1 D (m)areparametrizedbythenumberm= L,L−2,... offreepointsinthegraphs. L ThepresentedfactsabouttheBraueralgebrashouldbeenoughtounderstandtheloopgasreformulationof OSp(R|2S)spinmodel,whichwegiveinthenextsection. 3.3 LoopreformulationofOSp(R|2S)spinmodels: the algebraicpointofview The emergence of dense intersecting loops becomes transparent if we take the local transfer matrices in the adjointrepresentationoftheBraueralgebra. Thissimplyamountstoreplacingineq.(3.3)thegeneratorsI, E i andP definedbyeq.(3.2)withthediagramsinfig.6. Theadjointlocaltransfermatrixisrepresentedinfig.8. i We now define a loop model on a diagonallattice represented in fig. 4, with reflecting boundarieson the left and right (ie, free boundary conditions in the space direction) and identified boundaries on the top and 6Weadoptherethephysicist’shabitofcallingindecomposableamodulewhichisreduciblethoughnotfullyreducible. Thereforethe setofindecomposablesdoesnotcontaintheirreducibles,unlikeinmostofthemathliterature. 10