Contents lists available atScienceDirect
Advances in Mathematics
www.elsevier.com/locate/aim
Composed inclusions of A
3and A
4subfactors
Zhengwei Liu
DepartmentofMathematics,VanderbiltUniversity,Nashville,TN37240, USA
a r t i c l e i n f o a b s t r a c t
Articlehistory:
Received18September2013 Accepted11March2015 Availableonline23April2015 CommunicatedbytheManaging EditorsofAIM
Keywords:
Subfactors Planaralgebras
Inthisarticle,weclassifyallstandardinvariantsthatcanarise froma composed inclusion of an A3 withan A4 subfactor.
Moreprecisely, ifN ⊂ P isanA3 subfactorandP ⊂ Mis anA4subfactor,thenonlyfourstandardinvariantscanarise fromthecomposedinclusionN ⊂ M.Weansweraquestion posedbyBischandHaagerupin1994.Thetechniquesofthis paperalsoshowthatthereareexactlyfourstandardinvariants forthecomposedinclusionoftwoA4subfactors.
© 2015ElsevierInc.All rights reserved.
1. Introduction
Jonesclassifiedtheindices ofsubfactorsoftypeII1 in[13].Itisgivenby {4 cos2(π
n), n= 3,4,· · ·} ∪[4,∞].
Fora subfactorN ⊂ M of typeII1 withfinite index,the Jonestower isa sequence of factors obtained by repeating the basic construction. The system of higher relative commutantsis called thestandardinvariantof thesubfactor[8,35].A subfactorissaid tobeoffinite depth,ifitsprincipalgraph isfinite.Thestandardinvariantisacomplete invariantofafinite depthsubfactor[35].
E-mailaddress:zhengwei.liu@vanderbilt.edu.
http://dx.doi.org/10.1016/j.aim.2015.03.017 0001-8708/© 2015ElsevierInc.All rights reserved.
SubfactorplanaralgebraswereintroducedbyJonesasadiagrammaticaxiomatization ofthestandardinvariant[17].OtheraxiomatizationsareknownasOcneanu’sparagroups [30]andPopa’sλ-lattices[37].EachsubfactorplanaralgebracontainsaTemperley–Lieb planar subalgebra which is generated by the sequence of Jones projections. When the index oftheTemperley–Liebsubfactorplanaralgebrais4cos2(n+1π ),itsprincipalgraph is theCoxeter–DynkindiagramAn.
Given two subfactors N ⊂ P and P ⊂ M, the composed inclusion N ⊂ P ⊂ M tells therelative position of these factors.The grouptype inclusion RH ⊂ R⊂ RK for outer actionsof finite groupsH and K on thehyperfinite factor Rof typeII1 was discussed byBischandHaagerup[5].
WeareinterestedinstudyingthecomposedinclusionoftwosubfactorsoftypeA,i.e., a subfactorN ⊂ M with anintermediate subfactor P, such thatthe principalgraphs of N ⊂ P andP ⊂ M aretypeA Coxeter–Dynkindiagrams. From theplanaralgebra point of view, the planaralgebra of N ⊂ M is a composition of two Temperley–Lieb subfactorplanaralgebras.Theirtensorproductiswellknown[17,23].Theirfreeproduct asaminimalcompositionwasdiscoveredbyBischandJones[6],calledtheFuss–Catalan subfactor planaralgebra.Ingeneral, thecompositionof two Temperley–Liebsubfactor planaralgebras isstillnotunderstood.
TheeasiestcaseisthecomposedinclusionoftwoA3subfactors.Inthiscase,theindex is 4,and such subfactorsare extended type D [8,36]. They also arise as agroup type inclusionRH⊂ R⊂ RK,whereH ∼=Z2 andK∼=Z2.
Thefirstnon-group-likecaseisthecomposedinclusionofanA3withanA4subfactor.
ItsprincipalgraphiscomputedbyBischandHaagerupintheirunpublishedmanuscript in1994. Eitheritisafreecomposedinclusion,thenitsplanaralgebraisFuss–Catalan;
or itsprincipalgraphisaBisch–Haagerupfishgraphas
.
Thentheyaskedwhetherthissequenceofgraphsaretheprincipalgraphsofsubfactors.
The first Bisch–Haagerup fish graph is the principal graph of the tensor product of an A3 and an A4 subfactor. By considering the flip on R⊗ R, Bisch and Haagerup constructedasubfactorwhoseprincipalgraphisthesecondBisch–Haagerupfishgraph.
Later Izumi generalised the Haagerup factor [1] while considering endomorphisms of Cuntzalgebras[10],andheconstructedaHaagerup–IzumisubfactorforthegroupZ4in his unpublishednotes,alsocalled the3Z4 subfactor[31].ThethirdBisch–Haagerupfish graphis theprincipalgraphof anintermediate subfactorofareducedsubfactorof the dual of3Z4 [12].It turnsouttheevenhalfisMoritaequivalentto theeven halfof 3Z4.
Inthispaper, weprovethefollowing classificationresult.
Theorem1.1.ThereareexactlyfoursubfactorplanaralgebrasasacompositionofanA3 with an A4 planaralgebra.
This answers the question posed by Bisch and Haagerup. When n ≥ 4, the nth Bisch–Haagerupfishgraphisnottheprincipalgraphofasubfactor.Inthemeanwhile, Izumi, Morrisonand Penneyshaveruled outthe4th–10th Bisch–Haagerup fishgraphs usingadifferentmethod,see[12].
Three of the four subfactor planar algebras have finite depth which are complete invariantsofsubfactorsofthehyperfinite factorof typeII1 [35].TheFuss–Catalanone hasinfinite depth.Itcanalsoberealised fromahyperfinitesubfactor[38].
Bysimilar techniques,wealsoprovethefollowingclassificationresult.
Theorem1.2. Thereare exactlyfoursubfactors planaralgebrasasacomposition oftwo A4 planaralgebras.
Ourclassificationresultis alsoimportant tothe classificationofsmall index subfac- tors. Theindex 3+√
5 isthenextfrontier after 5 where thesubfactorplanaralgebras werecompletedclassifiedrecently[18,28,26,11,32].Someinterestingexamplesand clas- sificationresultsareknownupto thisindex[27,25].
Now we sketchthe ideas of theproof. Following the spirit of [33,2], if theprincipal graphof asubfactorplanaralgebrais thenth Bisch–Haagerup fishgraph,then bythe embeddingtheorem[19],theplanaralgebraisembeddedinthegraph planaralgebra of itsprincipalgraph [14]. Bytheexistence ofa“normalizer” in theBisch–Haagerupfish graph, there will be a biprojection [3] inthe subfactor planar algebra, and the planar subalgebrageneratedbythebiprojectionisFuss–Catalan.Theimageofthebiprojection isdeterminedbytheuniquepossiblerefinedprincipalgraph,seeDefinition 3.14andThe- orem 3.26.Furthermore the planaralgebrais decomposed as anannular Fuss–Catalan module,similar to theTemperley–Lieb case, [15,20].Comparingthe principalgraphof this Fuss–Catalan subfactor planar algebra and the Bisch–Haagerup fishgraph, there isalowest weightvectorinthe orthogonalcomplementof Fuss–Catalan.It will satisfy somespecific relations, and there is a“unique” potential solution of these relations in thegraphplanaralgebra.
ThesimilarityofalltheBisch–Haagerupfishgraphsadmitsustocomputethecoeffi- cientsofloopsofthepotentialsolutionssimultaneously.Thecoefficientsoftwosequences ofloopshave periodicity5and20withrespectton.Comparingwiththecoefficientsof theothertwosequencesofloops,wewillruleoutthealltheBisch–Haagerupfishgraphs, exceptthefirstthree.
Theexistence ofthe firstthree follows from the constructionmentioned above.The uniquenessfollows fromthe“uniqueness”ofthepotentialsolution.
Furthermore we consider the composition of two A4 planar algebras in the same process.Inthislist,thereareexactlyfoursubfactorplanaralgebras.Theyallarisefrom reducedsubfactorsofthefourcompositionsofA3withA4.
Theskeintheoreticconstructionofthese subfactorplanaralgebras couldbe realised bytheFuss–CatalanJellyfishrelationsofageneratingvectorspace.
2. Background
Wereferthereaderto[16]forthedefinitionofplanaralgebras.
Notation 2.1. In aplanar tangle, we use a thickstring with a numberk to indicate k parallel strings.
A subfactor planar algebra S = {Sn,±}n∈N0 will be a spherical planar *-algebra over C, such that dim(Sn,±) < ∞, for all n, dim(S0,±) = 1, and the unnormal- ized Markovtrace induces apositivedefinite inner product of Sn,± [16,17]. Note that dim(S0,±)= 1,thenS0,± isisomorphicto Casafield.Itissphericalmeans
=
as anumberinC, foranyx∈S1,±. Theinner productof Sn,± definedas
< y, z >=tr(z∗y) = ,
the Markovtraceof z∗y,for any y,z∈ Sn,±,is positivedefinite.(The diagram of the multiplication z∗y is not the usual convention. The usual multiplication is from the bottomto thetop.)
A subfactor planaralgebrais alwaysunital, where unitalmeans any tangle without inner discscanbe identified as avector ofS. Note thatS0,± is isomorphic to C,the (shadedorunshaded)emptydiagramcanbeidentifiedasthenumber1 inC.Thevalue ofa(shadedorunshaded)closedstringisδ.Andδ−1 in Sn,+,denotedbyen−1, for n≥2,is thesequenceofJones projections.ThefilteredalgebrageneratedbyJones projections isthesmallestsubfactorplanaralgebra,wellknownas theTemperley–Lieb algebra,denotedbyTL(δ). Itsvectorscanbe writtenaslinearsumsoftangleswithout inner discs.
Notation 2.2. Wemayidentify Sm,− asasubspaceof Sm+1,+ by addingonestringto theleft.
Definition 2.3. Let us define the (1-string) coproduct of x ∈ Si,± and y ∈ Sj,±, for i,j ≥1,tobe
x∗y= ,
whenevertheshadingmatched.
Let us recall some facts about the embedding theorem. Then we generalize these resultstoprovetheembeddingtheoremforanintermediatesubfactorinthenextsection.
2.1. Principal graphs
SupposeN ⊂ MisanirreduciblesubfactoroftypeII1withfiniteindex.ThenL2(M) forms an irreducible (N,M) bimodule, denoted by X. Its conjugate X is an (M,N) bimodule.Thetensor productsX ⊗X⊗ · · · ⊗X, X⊗X⊗ · · · ⊗X, X⊗X⊗ · · · ⊗X andX⊗X⊗ · · · ⊗X aredecomposedintoirreduciblebimodulesover(N,N),(N,M), (M,N) and(M,M) respectively,where⊗isConnesfusionof bimodules.
Definition 2.4. The principal graph of the subfactor N ⊂ M is a bipartite graph. Its verticesareequivalenceclassesofirreduciblebimodulesover(N,N) and(N,M) inthe abovedecomposed inclusion.The numberof edges connecting two vertices, an (N,N) bimoduleY andan(N,M) bimoduleZ,isthemultiplicityoftheequivalenceclassofZ asasubbimoduleofY⊗X.Thevertexcorrespondingtothe(N,N) bimoduleL2(N) is markedbyastarsign.Thedimensionvectorofthebipartitegraphisafunctionλfrom thevertices of thegraphto R+. Its valueat avertex isdefinedto be thedimension of thecorrespondingbimodule.
Thedualprincipalgraphisdefinedsimilarlyfor(M,M) and(M,N) bimodules.
Remark 2.5. By Frobenius reciprocity, the multiplicity of Z in Y ⊗X equals to the multiplicityofY inZ⊗X.
2.2. Standardinvariants
ForanirreduciblesubfactorN ⊂ MoftypeII1 withfinite index,theJones toweris asequenceoffactors N ⊂ M⊂ M1 ⊂ M2⊂ ·· ·obtainedby repeatingthe basic con- struction.Thesystemofhigherrelativecommutants
C=N∩ N ⊂ N∩ M ⊂ N∩ M1 ⊂ N∩ M2 ⊂ · · ·
∪ ∪ ∪
C =M∩ M ⊂ M∩ M1 ⊂ M∩ M2 ⊂ · · · iscalledthestandardinvariantofthesubfactor[8,35].
ThereisanaturalisomorphismbetweenhomomorphismsofbimodulesX⊗X⊗· · ·⊗X, X⊗X⊗· · ·⊗X,X⊗X⊗· · ·⊗X andX⊗X⊗· · ·⊗X andthestandardinvariantofthe subfactor[4].Theequivalenceclassofaminimalprojectioncorrespondstoanirreducible bimodule.Sotheprincipalgraphtellshowminimalprojectionsaredecomposedafterthe inclusion.Thenwecandefinetheprincipalgraphforasubfactorplanaralgebrawithout thepresumedsubfactor.Thefollowingtwopropositionsarewell knowntoexperts.
Proposition 2.6.Suppose S isasubfactorplanaralgebra.If P1,P2 are minimalprojec- tionsof Sm,+,thenP1em+1,+,P2em+1,+ areminimalprojectionsof Sm+2,+.Moreover P1 andP2 are equivalent in Sm,+ if andonly if P1em+1 and P2em+1 are equivalent in Sm+2,+.
Proposition 2.7 (Frobenius reciprocity). Suppose S is a subfactor planar algebra. If P is a minimal projection of Sm,+ and Q is a minimal projection of Sm+1,+, then dim(PSm+1,+Q)= dim(P em+1Sm+2,+Q).
Bytheabovetwo propositions,theBrattelidiagram ofSm,+⊂Sm+1,+ isidentified as asubgraphof theBrattelidiagramof Sm+1,+ ⊂Sm+2,+. Soitmakessense to take thelimitoftheBrattelidiagramofSm,+ ⊂Sm+1,+ asmapproaches infinity.
Definition 2.8. The principalgraphof asubfactorplanar algebraS is thelimit of the Brattelidiagram of Sm,+ ⊂Sm+1,+.Thevertex correspondingto theidentityin S0,+
ismarkedbyastarsign.ThedimensionvectorλatavertexisdefinedtobetheMarkov traceoftheminimalprojectioncorresponding tothatvertex.
SimilarlythedualprincipalgraphofasubfactorplanaralgebraS isthelimitofthe Brattelidiagram of Sm,− ⊂Sm+1,−.Thevertex correspondingto theidentityin S0,−
ismarkedbyastarsign.ThedimensionvectorλatavertexisdefinedtobetheMarkov traceoftheminimalprojectioncorresponding tothatvertex.
The Brattelidiagram of Sm,+ ⊂Sm+1,+, as asubgraph ofthe Brattelidiagram of Sm+1,+⊂Sm+2,+,correspondstothetwo-sidedidealIm+1,+ ofSm+1,+ generatedby theJonesprojectionem.Sothetwographscoincideifandonlyif Sm+1,+=Im+1,+. Definition2.9.Forasubfactorplanaralgebra S,ifitsprincipalgraphisfinite,thenthe subfactorplanaralgebraissaidtobe finitedepth.Furthermoreitisofdepthm,ifmis thesmallestnumbersuchthat Sm+1,+=Sm+1,+emSm+1,+.
Definition 2.10.A vertex v intheprincipal graphhasdepthmifthe distance between v andthestar vertexis m.Thevertex hasmultiplicity nifthere arenlength-mpaths from thestarvertexto v.
A depth-m vertex in the principal graph corresponds to a central component in Sm,+/Im,+. The vertex has multiplicity n tells that the central component is an n bynmatrixalgebra.Thereforethedualofthevertex hasthesamemultiplicity.
2.3. Finite dimensionalinclusions
Wereferthereaderto Chapter3of[21]foradiscussionofinclusionsoffinitedimen- sionalvon Neumannalgebras.
Definition 2.11. Suppose A is a finite dimensional von Neumann algebra and τ is a trace on it. The dimension vector λτA is a function from the set of minimal central projections(orequivalence classesofminimal projectionsorirreducible representations upto unitaryequivalence) of A to C with following property, for any minimal central projectionz,λτA(z)=τ(x),wherex∈Aisaminimalprojectionwithcentralsupportz.
Thetraceofaminimalprojectiononlydependsonitsequivalenceclass,sothedimen- sionvectoriswelldefined.Ontheotherhand,givenafunctionfromtheset ofminimal centralprojectionsofAtoC,wecanconstructatraceofA,suchthatthecorresponding dimensionvectoristhegivenfunction.Sothemapλ→λτAisabijection.
LetusrecallsomefactsabouttheinclusionoffinitedimensionalvonNeumannalge- brasB0⊂ B1.
TheBratteli diagram Br for theinclusionB0 ⊂ B1 isa bipartitegraph. Itseven or oddverticesareindexedbytheequivalenceclassesofirreduciblerepresentationsofB0or B1 respectively.Thenumberofedgesconnectsavertex correspondingto anirreducible representation U ofB0 toa vertex corresponding toan irreducible representationV of B1isgiven bythemultiplicity ofU intherestrictionofV onB0.
Let Br± be the even/odd vertices of Br. The Bratteli diagram can be interpreted as the adjacency matrixΛ = ΛBB10 : L2(Br−)→ L2(Br+),where Λu,v is defined as the numberofedgesconnectsutov forany u∈Br+,v∈Br−.
Proposition 2.12. (See [13].) For the inclusion B0 ⊂ B1 and a trace τ on it, we have λτB0 = ΛλτB1.
Ifthetraceτ is afaithfulstate,thenby GNSconstructionwe willobtainarightB1
module L2(B1). And L2(B0) is identified as a subspaceof L2(B1). Let e be the Jones projectiononto thesubspaceL2(B0).LetB2 bethevonNeumannalgebra(B1∪ {e}). ThenweobtainatowerB0⊂ B1⊂ B2whichiscalledthebasicconstruction.Furthermore if thetracial state τ satisfies the condition Λ∗ΛλτB1 =μλτB1 for some scalarμ, then it issaidto be aMarkovtrace.Inthis casethescalarμis||Λ||2. Thenλτ =
λτB0 δλτB1
isa Perron–Frobeniuseigenvectorfor
0 Λ Λ∗ 0
.
Definition2.13.Wecallλτ thePerron–FrobeniuseigenvectorwithrespecttotheMarkov traceτ.
Remark 2.14. Theexistence of a Markovtrace for theinclusion B0 ⊂ B1 follows from the Perron–Frobenius theorem.The Markov traceis unique ifand onlyif the Bratteli diagram fortheinclusionB0⊂ B1isconnected.
Wewill seetheimportanceof theMarkovtracefromthefollowingproposition.
Proposition 2.15. If τ is a Markov trace for the inclusion B0 ⊂ B1, then τ extends uniquely to a trace on B2, still denoted by τ. Moreover τ is a Markov trace for the inclusionB1⊂ B2.
In this case, we may repeat the basic construction to obtain a sequence of finite dimensional vonNeumann algebras B0⊂ B1⊂ B2 ⊂ B3 ⊂ · · ·and asequence ofJones projectionse1,e2,e3· · ·.
2.4. Graphplanaralgebras
Given afiniteconnectedbipartitegraphΓ,itcanberealised astheBrattelidiagram of theinclusion of finite dimensionalvon Neumann algebras B0 ⊂ B1 with a (unique) Markov trace. Applying the basic construction, we will obtain the sequence of finite dimensionalvonNeumannalgebrasB0⊂ B1⊂ B2⊂ B3⊂ · · ·.Take Sm,+ tobeB0∩ Bm
and Sm,−tobeB1∩Bm+1.Then{Sm,±}formsaplanaralgebra,calledthegraphplanar algebra ofthebipartitegraphΓ.MoreoverSm,±hasanaturalbasisgivenbylength2m loopsofΓ.Wereferthereaderto[14,19]formoredetails.Wecite theconventionsused inSection3.4 of[19].
Definition 2.16. Let us define G = {Gm,±} to be the graph planar algebra of a finite connectedbipartitegraphΓ.LetλbethePerron–Frobeniuseigenvectorwithrespectto theMarkovtrace.
A vertexof theΓ correspondstoanequivalence classofminimal projections,soλis alsodefinedasafunctionfromV±toR+.IfΓ istheprincipalgraphofasubfactor,then itsdimensionvectorisamultipleofthePerron–Frobeniuseigenvector.Inthispaper,we onlyneedtheproportionofvaluesofλatvertices. Wedonothavetodistinguishthese two vectors.
Let V± be the sets of even/odd vertices of Γ, and let E be the sets of all edges of Γ directed from even to odd vertices. Then we have the source and target functions s : E → V+ and t : E → V−. For adirected edge ε ∈ E, we define ε∗ to be the same edge withanoppositedirection.Thesourcefunctions:E∗={ε∗|ε∈ E}→ V− and the target functiont:E∗→ V+ aredefinedass(ε∗)=t(ε) andt(ε∗)=s(ε).
Alength2mloopinGm,+ isdenotedby[ε1ε∗2· · ·ε2m−1ε∗2m] satisfying (i) t(εk)=s(ε∗k+1)=t(εk+1),foralloddk <2m;
(ii) t(ε∗k)=s(εk)=t(εk+1),forallevenk <2m;
(iii) t(ε∗2m)=s(ε2m)=t(ε1).
Thegraphplanaralgebraisalwaysunital.Theunshadedemptydiagram isgivenby
v∈V+v; and theshaded empty diagram is givenby
v∈V−v.It is worth mentioning thattheJonesprojectionisgivenby
e1=δ−1 =δ−1
s(ε1)=s(ε3)
λ(t(ε1))λ(t(ε3))
λ(s(ε1))λ(s(ε3))[ε1ε∗1ε3ε∗3].
NowletusdescribesomeelementaryactionsonG.
Theadjointoperation isdefinedas theanti-linearextensionof [ε1ε∗2· · ·ε2m−1ε∗2m]∗= [ε2mε∗2m−1· · ·ε2ε∗1].
ForGm,−,wehavesimilarconventions.
Forl1,l2∈Gm,+, l1= [ε1ε∗2· · ·ε2m−1ε∗2m],l2= [ξ1ξ2∗· · ·ξ2m−1ξ2m],wehave
= 1≤k≤mδεm+k,ξm+1−k[ε1ε∗2· · ·ε∗mξm+1· · ·ξ2m−1ξ2m∗ ] whenmis even;
1≤k≤mδεm+k,ξm+1−k[ε1ε∗2· · ·εmξ∗m+1· · ·ξ2m−1ξ2m∗ ] whenmis odd.
= s(ε)=s(εm)[ε1ε∗2· · ·ε∗mεε∗εm+1· · ·ε2m−1ε∗2m] whenmis even;
t(ε)=t(εm)[ε1ε∗2· · ·εmε∗εε∗m+1· · ·ε2m−1ε∗2m] whenmis odd.
=
δεm,εm+1
λ(s(εm))
λ(t(εm))[ε1ε∗2· · ·ε∗mεε∗εm+1· · ·ε2m−1ε∗2m] whenmis even;
δεm,εm+1
λ(t(εm))
λ(s(εm))[ε1ε∗2· · ·εmε∗εε∗m+1· · ·ε2m−1ε∗2m] whenmis odd.
Definition2.17.TheFouriertransformF:Gm,+→Gm,−,m>0 isdefinedasthelinear extensionof
F([ε1ε∗2· · ·ε2m−1ε∗2m]) =
⎧⎨
⎩
λ(s(ε2m)) λ(t(ε2m))
λ(s(εm))
λ(t(εm))[ε∗2mε1ε∗2· · ·ε2m−1] form even;
λ(s(ε2m)) λ(t(ε2m))
λ(t(εm))
λ(s(εm))[ε∗2mε1ε∗2· · ·ε2m−1] form odd.
Similarlyitisalsodefinedfrom Gm,− to Gm,+.
TheFouriertransformhasadiagrammatic interpretationas aone-clickrotation
.
Definition 2.18. Let us define ρ to be F2. Then ρ is defined from Gm,+ to Gm,+ as a two-click rotationform>0,
ρ([ε1ε∗2· · ·ε2m−1ε∗2m]) =
λ(s(ε2m)) λ(s(ε2m−1))
λ(s(εm))
λ(s(εm−1))[ε2m−1ε∗2mε1ε∗2· · ·ε2m−3ε∗2m−2].
It issimilarforGm,−.
Ingeneral,theactionofaplanartanglecouldbe realisedasacompositionofactions mentionedabove.Ithasaniceformula, seepage 11in[14].
2.5. Theembedding theorem
Foradepth2r(or2r+ 1) subfactorplanaralgebraS,wehave
Sm+1,+=Sm+1,+emSm+1,+=Sm,+em+1Sm,+, wheneverm≥2r+ 1.
So Sm−1,+ ⊂ Sm,+ ⊂ Sm+1,+ forms a basic construction. Note that the Bratteli diagram of S2r,+ ⊂ S2r+1,+ is the principal graph. So the graph planar algebra G of theprincipalgraphisgivenby
Gk,+=S2r,+ ∩S2r+k,+; Gk,−=S2r+1,+ ∩S2r+k+1,+.
Moreover the map Φ : S → G by adding 2r strings to the left preserves the planar algebrastructure.Itisnotobviousthattheleftconditionalexpectationispreserved.We havethefollowingembeddingtheorem,seeTheorem4.1 in[19].
Theorem 2.19. A finite depth subfactor planar algebra is naturally embedded into the graph planaralgebra of itsprincipalgraph.
Remark2.20.Theembeddingtheoremforgeneralcasesisprovedin[29].
2.6. Fuss–Catalan
TheFuss–CatalansubfactorplanaralgebraswerediscoveredbyBischandJonesasfree products of Temperley–Lieb subfactorplanar algebras while studying theintermediate
subfactorsof asubfactor[6].We referthe readerto [7,22] forthe definitionof thefree product of subfactor planar algebras. It has a nice diagrammatic interpretation. For two Temperley–Lieb subfactor planar algebras TL(δa) and TL(δb), their free product FC(δa,δb) is a subfactor planar algebra. A vector in FC(δa,δb)m,+ can be expressed as a linear sum of Fuss–Catalan diagrams, a diagram consisting of disjoint a,b-colour stringswhose boundarypointsareordered asabba abba · · ·abba
m
,mcopiesofabba,after the dollar sign. It is similar for a vector inFC(δa,δb)m,−, while the boundary points are ordered as baab baab · · ·baab
m
. For the action of a planar tangle ona simple tensor of Fuss–Catalan diagrams, first we replace each string of the planar tangle by a pair ofparallel a-colourandb-colour stringswhichmatchesthea,b-colourboundarypoints, then the output is gluing the new tangle with the input diagrams. If there is an a or b-colourclosed circle,thenitcontributesto ascalarδa orδb respectively.
The Fuss–Catalan subfactor planar algebraFC(δa,δb) is naturallyderived from an intermediatesubfactorofasubfactor.SupposeN ⊂ M isanirreduciblesubfactorwith finite index,and P is anintermediate subfactor. Then there aretwo Jones projections eN and eP acting on L2(M), and we have the basic construction N ⊂ P ⊂ M ⊂ P1 ⊂ M1.Repeating this process, wewill obtainasequenceof factorsN ⊂ P ⊂ M⊂ P1 ⊂ M1 ⊂ P2 ⊂ M2· · · and a sequence of Jones projections eN, eP, eM, eP1· · ·. The algebra generated by these Jones projections forms a planar algebra, denoted by FC(δa,δb), where δa =
[P :N] and δb =
[M:P]. Moreover eP ∈ FC(δa,δb)2,+
and eP1 ∈ FC(δa,δb)2,− could be expressed as and respectively.Inparticular, F(eP) isamultipleofeP1.
Definition2.21.ForanirreduciblesubfactorplanaralgebraS,aprojectionQ∈S2,+ is calledabiprojection,ifF(Q) isamultipleofaprojection.
If S istheplanaralgebrafor N ⊂ M, theneP ∈S2,+ isabiprojection.Conversely allthebiprojections inS2,+ arerealisedinthisway,see[6].
Proposition 2.22. If we identify S2,− as a subspace of S3,+ by adding a string to the left,thena biprojectionQ∈S2,+ willsatisfy QF(Q)=F(Q)Q,i.e.
= ,
called theexchangerelationof abiprojection.
Converselyifaself-adjointoperatorinS2,+ satisfiestheexchangerelation,thenitis amultipleofabiprojection.Thiscanbeprovedbyaddingcapsatproperpositions.We refer the reader to [23] for some other approaches to biprojections. The Fuss–Catalan subfactor planar algebra could also be viewed as the planar algebra generated by a biprojection withitsexchangerelation.
The planaralgebra of acomposed inclusion of an A3 with an A4 subfactoralways contains FC(δa,δb), where δa =√
2, δb = √5+12 , as aplanarsubalgebra. The principal graphanddual principalgraphofFC(δa,δb) aregivenas
.
3. Theembeddingtheoremforanintermediatesubfactor
A subfactor planaralgebrais embeddedin thegraphplanar algebraof itsprincipal graphbytheembeddingtheorem.Ifasubfactorplanaralgebracontainsabiprojection, then wehopetoknowtheimageofthebiprojection inthegraphplanaralgebra.Recall thattheimageoftheJonesprojectione1 isdeterminedbytheprincipalgraph,
δe1=
s(ε1)=s(ε3)
λ(t(ε1)) λ(s(ε1))
λ(t(ε3))
λ(s(ε3))[ε1ε∗1ε3ε∗3].
We willsee asimilarformula fortheimage ofthebiprojection.Itis determined bythe refined principalgraph. Therefined principalgraphis alreadyconsideredby Bisch and Haagerup for bimodules, by Bisch and Jones for planar algebras. For the embedding theorem,wewillusetheoneforplanaralgebras.
Thelopsided versionofembeddingtheorem foranintermediatesubfactorisinvolved in ageneralembedding theorem proved byMorrison in[29]. Toconsider thealgebraic structures,itisconvenienttoworkwiththesphericalversionoftheembeddingtheorem.
Their relationsaredescribed in[27].For convenience,we provethesphericalversion of embeddingtheorem,similarto theoneprovedbyJonesandPenneysin[19].
Inthissection,wealwaysassumeN ⊂ MisanirreduciblesubfactoroftypeII1with finite index, and P is an intermediate subfactor. If the subfactorhas an intermediate subfactor,thenitsplanaralgebrabecomesanN −P −Mplanaralgebra.ForN −P −M planaralgebras,wereferthereadertoChapter4in[9].Inthiscase,thesubfactorplanar algebracontainsabiprojectionP,andaplanartanglelabelled byP canbe replacedby
aFuss–Catalan planar tangle. In this paper, we will use planar tangles labelled by P, insteadofFuss–Catalanplanartangles.
3.1. Principal graphs
For the embedding theorem,we will consider the principalgraph of N ⊂ P ⊂ M. It refines the principal graph of N ⊂ M. Instead of a bipartite graph, it will be an (N,P,M)colouredgraph.Thefollowingdefinitionsandpropositionsarewellknownto experts[6,7,22,9].
Definition3.1.An(N,P,M)colouredgraphΓ isalocallyfinitegraph,suchthattheset V of itsverticesisdividedinto threedisjoint subsetsVN,VP andVM,and thesetE of itsedgesisdividedintotwodisjointsubsetsE+,E−.MoreovereveryedgeinE+connects avertexinVN tooneinVP andeveryedgeinE− connectsavertexinVP tooneinVM. Thenwedefinethesourcefunctions:E → VN∪ VMandthetargetfunctiont:E → VP
intheobviousway. Theoperation∗reversesthedirectionofanedge.
Definition3.2.Froman(N,P,M)colouredgraphΓ,wewillobtainan(N,M) coloured bipartitegraphΓasfollows,theN/McolouredverticesofΓareidenticaltotheN/M coloured vertices of Γ; for two vertices vn in VN and vm ∈ VM, the number of edges betweenvnandvminΓ isgivenbythenumberoflengthtwopaths fromvn tovmin Γ. ThegraphΓ issaid tobe thebipartitegraphinduced fromthe graphΓ.ThegraphΓ issaidtobe arefinementofthegraphΓ.
Remark3.3.HereweabusethenotationsofΓ andΓ whichareusuallyreservedforthe principalgraphandthedualprincipalgraphrespectively.
For afactor M of typeII1, ifN ⊂ P ⊂ M is a sequence of irreducible subfactors withfiniteindex,thenL2(P) formsanirreducible(N,P) bimodule,denotedbyX,and L2(M) formsanirreducible(P,M) bimodule,denotedbyY.TheirconjugatesX,Y are (P,N),(M,P) bimodulesrespectively.ThetensorproductsX⊗Y ⊗Y ⊗X⊗ · · · ⊗X, X⊗Y⊗Y⊗X⊗· · ·⊗X,X⊗Y⊗Y⊗X⊗· · ·⊗Y,X⊗Y⊗Y⊗X⊗· · ·⊗Y,aredecomposed intoirreducible bimodulesover(N,N),(N,P),(N,M) and(N,P) respectively.
Definition 3.4. The principal graph for the inclusion of factors N ⊂ P ⊂ M is an (N,P,M) coloured graph.Its vertices are equivalenceclasses of irreducible bimodules over (N,N), (N,P) and (N,M) in the above decomposed inclusion. The number of edgesconnectingtwovertices,an(N,N) (or(N,M))bimoduleU (orV)andan(N,P) bimoduleW,isthemultiplicityoftheequivalenceclassofU (orV)asasubbimoduleof W⊗X(orW⊗Y).Thevertexcorrespondingtotheirreducible(N,N) bimoduleL2(N) ismarked byastarsign∗.Thedimension vectoroftheprincipal graph isafunctionλ fromtheverticesofthegraphtoR+.Itsvalueatapointisdefinedtobethedimension ofthecorrespondingbimodule.
Similarlythedualprincipalgraphfortheinclusionoffactorsisdefinedbyconsidering thedecomposedinclusionof(M,M),(M,P),(M,N) bimodules.
There is another principal graphgiven by decomposing (P,N), (P,P) and(P,M) bimodules underinclusions,butitisnotneededinthispaper.
Proposition 3.5. The (dual) principal graph fortheinclusionof factors N ⊂ P ⊂ M is a refinementofthe(dual) principalgraph of thesubfactorN ⊂ M.
Proof. If followsfrom thedefinition andthe factthat X⊗Y is the(N,M) bimodule L2(M). 2
Let δa be
[P :N],the dimension ofX, and δb be
[M:P], the dimensionof Y. Then byFrobeniusreciprocitytheorem,wehavethefollowing proposition.
Proposition 3.6. For the principal graph of factors N ⊂ P ⊂ M and the dimension vector λ,wehave
δaλ(u) =
ε∈E+,s(ε)=u
λ(t(ε)), ∀u∈ VN; δbλ(w) =
ε∈E−,s(ε)=w
λ(t(ε)), ∀w∈ VM; δaλ(v) =
ε∈E+,t(ε)=v
λ(s(ε)), ∀v∈ VP; δbλ(v) =
ε∈E−,t(ε)=v
λ(s(ε)), ∀v∈ VP.
Definition3.7. Foran(N,P,M) colouredgraphΓ,ifthereexits afunctionλ:V →R+ satisfying the proposition mentioned above, then we call it a graph with parame- ter (δa,δb).
Proposition 3.8. The principal graph of factors N ⊂ P ⊂ M isagraph with parameter (
[P:N],
[M:P]).ConsequentlyifN ⊂ Mhasfinitedepth,thentheprincipalgraph of N ⊂ P ⊂ M isfinite.
Proof. The first statement follows from the definition. Note that the dimension of a bimodule isatleast1.Bythisrestriction,N ⊂ Mhasfinitedepthimpliestheprincipal graphofN ⊂ P ⊂ Misfinite. 2
3.2. Standardinvariants
Wewilldefinetherefined(dual)principalgraphforasubfactorplanaralgebrawitha biprojection.Thisdefinitioncoincideswiththedefinitiongivenbybimodules,butwedo notneedthisfactinthispaper.GivenN ⊂ P ⊂ M,therearetwoJonesprojectionseN andeP actingonL2(M).ThenwehavethebasicconstructionN ⊂ P ⊂ M⊂ P1⊂ M1. Repeating thisprocess,wewillobtainasequenceoffactorsN ⊂ P ⊂ M⊂ P1⊂ M1⊂
P2⊂ M2· · ·andasequenceofJonesprojectionseN,eP,eM,eP1· · ·.Thenthestandard invariantisrefinedas
C=N∩ N ⊂ N∩ P ⊂ N∩ M ⊂ N∩ P1 ⊂ N∩ M1 ⊂ · · ·
∪ ∪ ∪
C=P∩ P ⊂ P∩ M ⊂ P∩ P1 ⊂ P∩ M1 ⊂ · · ·
∪ ∪ ∪
C =M∩ M ⊂ M∩ P1 ⊂ M∩ M1 ⊂ · · · For Fuss–Catalan, the corresponding Bratteli diagram is described by the middle patterns,see pages114–115in[6].
We hope to define the refined principal graph as the limit of the Bratteli diagram Brk of N∩ Mk−2 ⊂ N∩ Pk−1 ⊂ N∩ Mk−1. To show thelimit is well defined, we needtoprovethatBrk isidentifiedasasubgraphofBrk+1.Todefineitforasubfactor planar algebrawith a biprojection without the presumed factors, we need to do some translationsmotivatedbythefact
N∩ Pk=N∩(Mk∩ {ePk}) = (N∩ Mk)∩ {ePk}.
Definition 3.9. Let S = Sm,± be a subfactor planar algebra. Let e1,e2,· · · be the sequenceofJonesprojections.
Suppose p1 is a biprojection in S2,+. Then we obtain another sequence of Jones projections p1,p2,p3,· · ·, corresponding to the intermediate subfactors, precisely p2 in S2,− ⊂S3,+isamultipleofF(p1),andpk isobtainedbyaddingtwostringsontheleft sideofpk−2.
Form≥1,letusdefine Sm,+ tobe Sm,+∩ {pm} and Sm,− tobe Sm,−∩ {pm+1}. Remark 3.10. If we interpret one string of a planar diagram as a pair of a/b-colour strings ,thensequenceofJonesprojectionsp1,p2,p3,· · ·,canbeinterpretedasfollow- inga/b-colour diagrams[6],
.
Thefollowingpropositionshowsthatthesubspace Sm,± of Sm,±consistsofdiagrams withana/b-colour throughstringontherightmost.
Proposition3.11.ForX ∈Sm,+,m≥1,wehave
Xpm=pmX ⇐⇒ F(X) =F(X)pm.
ThatmeansSm,+ istheinvariantsubspaceofSm,+ underthe“right action”of the biprojection.Diagrammaticallyitsconsistsofvectorswithonea/b-colourthroughstring ontherightmost.
Proof. IfpmX =Xpm, thentaketheaction givenbytheplanartangle ,we haveF(X)=F(X)pm.
Form odd,ifF(X)=F(X)pm,thenX =X∗ F(p1),i.e.
X = .
Bytheexchangerelationofthebiprojection,wehave
= .
SopmX =Xpm.
Form even,theproofissimilar. 2
Note that Sm−1,+ is in the commutant of pm. So we have the inclusion of finite dimensionalvonNeumannalgebras
S0,+⊂S1,+ ⊂S1,+ ⊂S2,+ ⊂S2,+ ⊂ · · ·.
ThenweobtaintheBrattelidiagramBrmfortheinclusionSm−1,+⊂Sm,+ ⊂Sm,+.To takethelimitofBrm,weneedtoprovethatBrmisidentified asasubgraphofBrm+1. Proposition 3.12. If P1, P2 are minimal projections of Sm,+ . Then P1pm, P2pm are minimalprojectionsof Sm+1,+ .MoreoverP1andP2 areequivalentin Sm,+ ifandonly if P1pm andP2pm areequivalent in Sm+1,+ .
This propositionisprovedsimilar toProposition 2.6.
Proposition 3.13(Frobenius reciprocity).
(1) For a minimal projection P ∈ Sm−1,+ and a minimal projection Q ∈ Sm,+ , we have that Qpm is a minimal projection of Sm+1,+ , P em is a minimal projection
of Sm+1,+,and
dim(P(Sm,+ )Q) = dim(P em(Sm+1,+)Qpm).
(2) Foraminimalprojection P ∈Sm,+ andaminimalprojectionQ ∈Sm,+,wehave Ppm isaminimalprojectionof Sm+1,+ ,and
dim(P(Sm,+)Q) = dim(Ppm(Sm+1,+ )Q).
Proof. (1)Considerthemaps
φ1= :Sm,+→Sm+1,+, φ2= :Sm+1,+→Sm,+.
Formodd,ifX∈P(Sm,+ )Q,thenbyProposition 3.11,wehaveX=P(X∗ F(p1))Q for some X ∈ Sm,+. So φ1(X) ∈ P em(Sm+1,+)Qpm. On the other hand, if Y ∈ P em(Sm+1,+)Qpm, then φ2(Y) ∈ P(Sm,+ )Q. While φ1 ◦φ2 is the identity map on P em(Sm+1,+)Qpmandφ2◦φ1istheidentitymaponP(Sm,+ )Q.Sodim(P(Sm,+)Q)= dim(Ppm(Sm+1,+ )Q).
Formeven,theproofissimilar.
(2)ThisisthesameasProposition 2.7. 2
ByProposition (2.6)(3.13), the Brattelidiagram Brm is identified as a subgraph of Brm+1.
Definition3.14.Letusdefinetherefinedprincipalgraphof S withrespecttothebipro- jectionp1 to be thelimit of theBratteli diagram of Sm,+ ⊂Sm+1,+ ⊂ Sm+1,+. The vertexcorresponding totheidentityinS0,+ ismarked byastarsign.
Similarly let us define the refined dual principal graph of S with respect to the biprojection p1 to be thelimitof the Brattelidiagram ofSm,− ⊂Sm+1, − ⊂Sm+1,−. Thevertexcorresponding totheidentityinS0,− ismarked byastarsign.
Therefinedprincipalgraphisan(N,P,M) colouredgraph.TheN,P, Mcoloured vertices are given by equivalence classes of minimal projections of S2m,−, S2m+1, −, S2m+1,−respectivelyasmapproachesinfinity.Similarlytherefineddualprincipalgraph isan(M,P,N) colouredgraph.
Definition3.15.Thedimensionvector λoftheprincipalgraphisdefinedasfollows, for an N or M coloured vertex, its value is the Markov trace of the minimal projection correspondingtothatvertex;foraP colouredvertexv,supposeQ∈Sm,+ isaminimal projection corresponding to v. Then λ(v) = δ−a1tr(Q), when m is even, where δa = tr(p1);λ(v)=δ−b1tr(Q),whenmisodd,whereδb=δδ−a1.
Remark 3.16.Anelement in Sm,+ has ana/b-colour throughstringon therightmost.
When wecompute thedimension vectorfor aminimal projectionin Sm,+ , thatstring shouldbe omitted.Sothere isafactorδa−1 orδ−1b .
NotethatthedimensionvectorsatisfiesProposition 3.6.Sotherefinedprincipalgraph is a graph with parameter (δa,δb). If the Bratteli diagram of Sm,+ ⊂ Sm+1,+ is the same as thatof Sm+1,+ ⊂Sm+2,+, i.e.S has finite depth,then Brm+1 =Brm+2 by therestrictionofthedimensionvector.Inparticular,theBrattelidiagramof Sm+1,+ ⊂ Sm+1,+isthesameasthatof Sm+1,+⊂Sm+2,+ .So Sm+1 ⊂Sm+1,+⊂Sm+2,+ forms a basicconstruction,and pm+1 isthe Jonesprojection. Whileapplying the embedding theorem, the image of theJones projection canbe expressed as alinear sum ofloops.
Wewill seetheformulalater.
ThesubfactorplanaralgebraFC(√
2,1+2√5) containsatrace-2 biprojection.Consid- eringthemiddlepatternofitsminimalprojections,seepages114–115in[6],wehaveits refined principalgraph,as
; and itsrefineddualprincipalgraphas
,
where the black, mixed, white points are N,P,M coloured vertices. We will discuss moreaboutthesegraphsinSection4.1.
3.3. Finitedimensional inclusions
Now given an inclusionof finite dimensional von Neumannalgebras B0 ⊂ B1 ⊂ B2, similarlywemayconsideritsBrattelidiagram,adjacencymatrixes,Markovtraceifthere exists one,andthebasicconstruction.
Definition3.17.TheBrattelidiagramBrfortheinclusionB0⊂ B1⊂ B2isa(B0,B1,B2) coloured graph.ItsBi colouredvertices areindexed bytheminimal centralprojections (or equivalently the irreducible representations) of Bi, for i = 0,1,2.The subgraph of Br consistingof B0, B1 colouredvertices andtheedges connectingthemisthesameas theBrattelidiagramfortheinclusionB0⊂ B1.ThesubgraphofBr consistingofB1,B2
colouredverticesandtheedgesconnectingthemisthesameas theBrattelidiagramfor theinclusionB1⊂ B2.