A universal Vassiliev invariant of links onΣis a mapV: LpΣqÑCpΣq±8m0CpmqpΣqof filtered vector spaces such that
VGrλIdCpΣq. (4.2)
We refer to (4.2) as the defining equation of a universal Vassiliev invariant; it implies the following corollary to Proposition 4.2.
Proposition 4.3. If V is a universal Vassiliev invariant for Σ, then λ:CpΣq Ñ LGrpΣq is an isomorphism of graded Poisson algebras with inverseλ1VGr.
Assume for the remainder of this chapter thatBΣ H. The construction in [AMR2] of a universal Vassiliev invariant of links on Σbuilds on the construction in [B] of a universal Vassiliev invariant of non-associative tangles in the standard square S II with top It1uand bottomIt0u; by this we mean a family of filtered, linear maps parametrized by all boundary markings onS:
V:LpS;Bpq,Bpq
qÑCpS;B ,Bq
and satisfying the obvious analogue of (4.2). It will be useful to have refined versions of the chord tangle spaces. To be specific, let T S be a tangle (chord tangle without chords), and defineCpS;TqCpS;B T,BTqto be the (homogeneous) subspace generated by chord tangles with skeletonT. Aperturbation of the skeletonis any elementpPmqPCpS;Tq such that P0 T. Also, we allow the second factor ofSto shrink and stretch so that the spaces of tangles in this square are closed under composition. In particular, this means thatCpS;ÒÒqis an algebra with composition as multiplication, the unit being the trivial tangle Ò Ò. To define V we fix, once and for all, two parameters, theassociator Φ P
CpS;ÒÒÒqand theR-matrix R PCpS;ÒÒqsubject to various conditions (cf. [B]); we mention a couple of them. BothΦandRare perturbations of their skeletons; this implies that they are invertible elements in the algebras they belong to. Also,Rsatisfies the identity
RR1 higher degree terms. (4.3)
One constructs an element C P CpS;Òqin terms ofΦ; it is a perturbation of its skeleton.
Bending Cappropriately, it may be regarded as a member of either of the spacesCpS; x
q
andCpS;xq; we put V
R, V
R1 , (4.4a)
V
Φ, V
Φ1, (4.4b)
V C, V C. (4.4c)
Any non-associative tangle inSmay be obtained from the above six elementary ones by ca-bling, extension and composition, so requiring thatVis compatible with these operations on non-associative tangles and chord tangles, of course, (over)determinesV. Only the care-ful choice ofpΦ,Rqensures that this procedure leads to well-defined mapsV:LpS;Bpq,Bpq
q
4.3 Universal Vassiliev Invariants 39
Figure 4.2: A partition ofΣ0,3.
· · · ·
Figure 4.3: The hexagon tangle.
Ñ CpS;B ,Bq. That V is filtered and satisfies the defining equation of a universal Vas-siliev invariant follows from (4.3) and the fact thatΦ,Rand Care perturbations of their skeletons. Note that ifTSis a non-associative tangle, then
VpTqPCpS;πpTqq. (4.5)
The above construction yields, of course, a universal Vassiliev invariant of non-associative tangles in any embedded squareSΣ; we use this below.
Now assume thatΣis not itself a square. The universal Vassiliev invariant of links on Σdepends not only onpΦ,Rqbut also on apartitionofΣ. A partitionPis determined by a finite collection of embedded intervalspIk,BIkqpΣ,BΣqchosen such that cuttingΣalong these intervals results in a decomposition
ΣpYiSiqYpYjHjq
consisting of squaresSi and hexagons Hj. The sides of these polygons are alternately an interval Ik and a piece ofBΣ. A possible partition of the three-holed sphereΣ0,3 is illus-trated in Figure 4.2. For technical reasons we assume that no two hexagons are adjacent;
in particular, the decomposition contains at least one square. (By the Euler characteristic the number of hexagons is constant, but we will not use that). Also part of the structure is a choice of top and bottom on all polygons; for a square this means that one of the two embedded intervals bordering it is the top of that square and the other one is the bottom, whereas for a hexagon either the top or the bottom consists of two of theIkbordering it and the opposite side is the remaining one of these intervals. The choice of tops and bottoms must be consistent, i.e., result in an unambiguous direction ‘up’ onΣ.
LetLbe a link onΣ. By isotopy we assume thatLis in general position with respect to the embedded intervals and that each intersectionLXHj looks like Figure 4.3. (possibly turned upside down). Now choose parenthesizations of all the setsLXIk subject to the condition that the parenthesization of the top (bottom) endpoints of a hexagon is the union of the parenthesizations of the bottom (top) endpoints. In this way all intersectionsLXSi are non-associative tangles, and we can define
VPpLq
¹
i
VpLXSiq
¹
j
πpLXHjqPCpΣq where the product is composition of chord tangles.
Theorem 4.4 (Andersen, Mattes & Reshetikhin). The map VP:LpΣqÑ CpΣqis a universal Vassiliev invariant.
Notice that for any linkLonΣ
VPpLqPCpΣ;πpLqq (4.6)
40 Chapter 4Quantization of the Algebra of Chord Diagrams
Sl S Sr
Figure 4.4: A squareS0containing another squareS.
because of (4.5). By definition of the universal Vassiliev invariant of non-associative tangles in a square it is also clear thatVPis compatible with cabling of links and chord diagrams.
It is natural to ask howVPdepends onP. On one hand, we can refinePby bisecting one of the squaresSi with an extra embedded interval running parallel to the top and bottom ofSi. This modification leavesVPunchanged since the universal Vassiliev invariant forSis compatible with composition. On the other hand, we can consider the action ofΓ pΣqon the setPpΣqof all partitions ofΣ; given a mapgPΓ pΣq, the imagegpPqof the embedded intervals constitutingPis another partition. Evidently
VgpPqpgpLqqgpVPpLqq, LPLpΣq. (4.7) We now introduce a useful computational tool, namely universal Vassiliev invariants of non-associative tangles inΣS. They also depend on a partitionPofΣ, now required to becompatiblewithSin the sense that there exists a squareS0inPcontainingSas depicted in Figure 4.4. For technical reasons we also assume thatS0is not adjacent to a hexagon; this is no restriction since we can refineP. We begin with the easy case whenS S0(so that Sl Sr Hq. For a tangleTPLpΣS0;BpqpΣS0q,Bpq
pΣS0qqwe proceed as for links and define
VPpTq
¹
i0
VpTXSiq
¹
j
πpTXHjqPCpΣS0;B pΣS0q,BpΣS0qq
remembering, of course, that the parentheses on TXIpΣS0q are already fixed to be
B pq
pΣS0q. Adhering to the rule for parentheses in top and bottom intervals of hexagons is no problem since these polygons do not neighbourS0.
The general case builds on the first one. LetT P LpΣS;BpqpΣSq,Bpq
pΣSqq. Put Tl TXSl, and choose parentheses onTlXI pSlqand TlXIpSlqto obtain a boundary markingpBpqSl,BpqSlqonSl; in this wayTlPLpSl;BpqSl,BpqSlq. Similarly for the right hand square. Define boundary markings onS0by
B
pqS0ppB
pqSlB
pqSqBpqSrq, BpqS0ppB
pq
SlB
pq
SqBpqSrq
so that TXpΣS0qcan be regarded as an element inLpΣS0;BpqpΣS0q,Bpq
pΣS0qq. Put
VPpTqVPpTXpΣS0qqVpTlqVpTrqPCpΣS;B pΣSq,BpΣSqq.
That these maps are well-defined universal Vassiliev invariants of tangles inΣSis proved much like Theorem 4.4 (cf. [AMR2]). Compatibility with cabling is immediate from the construction as in the case of links on Σ. For (compatible) non-associative tangles TS P
LpS;Bpq,Bpq
qandTΣS PLpΣS;Bpq,Bpqq, the compositionTΣSTSis a link onΣ, and the three kinds of universal Vassiliev invariants fit together in
VPpTΣSTSqVPpTΣSqVpTSq
4.4-Products and Standard Situations 41
as one readily deduces from the definitions. This formula is ubiquitous in calculations in the sequel.
4.4
-Products and Standard Situations
For partitionsPofΣwe consider the completed map VPVP:LpΣqÑCpΣq.
By Proposition 4.3 and Remark 3.3, Theorem 3.11 applies to S CpΣq,F LpΣq and VVP:
Theorem 4.5 (Andersen, Mattes & Reshetikhin). For any partition P ofΣthere is a-product
Pwith coefficients
crpD,EqpVPpVP1pDqVP1pEqqqpm1 m2 rq for chord diagrams D and E with m1and m2chords, respectively.
Different partitions may yield different-products as we shall see, but at least we have Theorem 4.6. If P1and P2are two partitions ofΣ, then the endomorphismτofCpΣqrrhss deter-mined by
τpDq
¸
r
pVP2VP11pDqqpm rqhr, DPCpmqpΣq is an equivalence fromP1toP2.
Proof. Theorem 3.12 applies sincepVP1qGrλpVP2qGr. l Remark 4.7. From formula (4.6) follows immediately that the AMR-products and the equivalences between them preserve the skeletons of the chord diagrams. Therefore the above two theorems also hold for CpΣ;Gqif we simply carry along the representations associated to core components in the definitions.
Remark 4.8. With a little more effort one can show thatPpreserves more than skeletons;
ifDandEare chord diagrams thencrpD,Eqis a linear combination of chord diagrams each of which is obtained fromDEby addingrchords appropriately. Formally, this is proved by generalizing the results about non-associative tangles in the complement of an embedded squareSΣto the case of two disjoint embedded squares assumed (by isotopy) to contain the ‘non-trivial’ parts ofD, respectivelyE.
Proposition 4.9. The mapPpΣqÑpCpΣqqgiven by PÞÑPisΓ pΣq-equivariant.
Proof. LetPbe a partition ofΣ, and letg PΓ pΣq. For chord diagramsDandEwe have by (4.7) and the analogous identity forVP1
gVPpVP1pDqVP1pEqqVgpPqpgpVP1pDqVP1pEqqq
VgpPqpgVP1pDqgVP1pEqq
VgpPqpVg1
pPqpgpDqqVg1
pPqpgpEqqq.
It follows from Theorem 4.5 thatgpDPEqgpDqgpPqgpEq; this is exactly the statement
gPgpPq, cf. Proposition 3.14. l
42 Chapter 4Quantization of the Algebra of Chord Diagrams
4.4.1 Standard Situations
We shall encounter a couple of standard situations in computations involvingP,PPPpΣq and the equivalences between these-products. The common set-up of the standard situ-ations is as follows: There is an embedded squarepS;I ,Iq Σwith boundary mark-ing pBpq,Bpq
q, and we have compatible elements L P CpmqpS;B ,Bqand T P Cpm1qpΣ S;B,B q. We writeDTLPCpm1 mqpΣq.
In the first standard situationEis a chord diagram onΣwithm2chords, and we want to calculateDPEwherePis some partition ofΣ. By a homotopy we may assume firstly thatSis contained in the interior of a squareS0fromPand secondly thatEis represented by an element E P CpΣSqso thatE EH P CpΣq. We refinePwith two embedded intervals as illustrated in Figure 4.5 in order to makePcompatible withS. Having settled these technical issues, we derive
VP VP1pDqVP1pEqVP VP1pTLqVP1pEHq
VPprVP1pTqV1pLqsrVP1pEqHsq
VPrpVP1pTqVP1pEqqV1pLqs
VP VP1pTqVP1pEqL so that
crpD,Eq VP VP1pDqVP1pEq pm1 m m2 rq
VP VP1pTqVP1pEqLpm1 m m2 rq
VP VP1pTqVP1pEq pm1 m2 rqL.
Of course, we can reverse the roles ofDandEand get a parallel result. A first application of this standard situation yields
Theorem 4.10 (Andersen, Mattes & Reshetikhin). A subspaceI CpΣq spanned by local relations is a-ideal with respect toPfor any partition P ofΣ.
Proof. Previously (cf. 2.4.1) we noted that I is a Poisson ideal so the statement of the theorem makes sense. To prove it we must verify condition (3.13). Consider a generator DPI. There exists an embedded squarepS;I ,IqΣwith boundary markingpB ,Bq such that D T°iλiLi; hereLi P CpS,B ,Bqand λi P C are the chord tangles and the scalars defining the relevant local relation, andT PCpm1qpΣS;B,B qis an arbitrary
S I+(S) I−(S)
I−(S0) I+(S0)
Figure 4.5: Two intervals inS0refiningP.
4.4-Products and Standard Situations 43
chord tangle. LettingE be a chord diagram with m2 chords and choosing an arbitrary parenthesization onpB ,Bq, the standard situation yields
crpD,Eq
¸
i
λicrpTLi,Eq
¸
i
λi VP VP1pTqVP1pEqpm1 m2 rqLi
VP VP1pTqVP1pEq pm1 m2 rq
¸
i
λiLi PI
as desired. In the same way,crpE,DqPI. l
In the second standard situation we aim to computeτpDqPCpΣqrrhsswhereτis the canon-ical equivalence fromP1toP2for partitionsP1andP2ofΣ, cf. Theorem 4.6. By homotopy and the refinement procedure illustrated in Figure 4.5, we may assume thatSis compatible with bothP1andP2. We get
VP2VP1
1
pDqVP2VP1
1
pTLqVP2pVP1
1
pTqV1pLqqpVP2VP1
1
pTqqL so that
τrpDqpVP2VP11pDqqpm1 m rqpVP2VP11pTqLqpm1 m rqpVP2VP11pTqqpm1 rqL.
Not surprisingly the first application of the second standard situation is the following re-sult.
Theorem 4.11. LetI CpΣqbe a subspace spanned by local relations, and let P1and P2be two partitions ofΣ. The canonical equivalenceτ:CpΣqrrhssÑCpΣqrrhssfromP1 toP2 descends to CpΣq{Ito yield an equivalence between the induced-products.
Proof. This is analogous to the proof of Theorem 4.10; in the notation of that proof we deduce
τrpDq
¸
i
λiτrpTLiq
¸
i
λipVP2VP11pTqqpm1 rqLipVP2VP11pTqqpm1 rq
¸
i
λiLi PI
as required, cf. (3.16). l
Applying the preceding two theorems to the loop relation (2.21), we obtain
Theorem 4.12. The AMR-products onCpΣqand the canonical equivalences between them de-scend to the Poisson loop algebraZs,fpΣqvia the resolving map Rs,f:CpΣqÑZs,fpΣq.
Chapter 5
Quantization of the Moduli Space
In this chapter we prove, under the assumptionBΣH, that the-productsP, PPPpΣq onCpΣqand the canonical equivalences between them descend toOpMpΣ;GqqifGis one of the groups GLnpCqand SLnpCq. These results are achieved by presentingOpMpΣ;Gqq as an explicit quotient of CpΣq; the description we give is also valid in the case whereΣ is closed. Obtaining it relies on Sikora’s work [S]; in the general linear case we adapt the methods of his paper to derive a parallel version of its main result, and in the special linear case we simply translate the main result into our context.
We round off the chapter with Andersen’s explicit formula forP in the abelian case GGL1pCq[A]; it is a corollary thatPis independent ofPandΓ pΣq-invariant. We also provide counterexamples illustrating that this corollary fails in general.
5.1 The General Linear Case
We consider the groupGGLnpCqequipped with the orthogonal structure BpX,YqTrpXYq, X,YPglnpCq.
That is, we fixtPglnpCqbglnpCqto be the Ad-invariant symmetric tensor corresponding to the pairingB. Colouring all core components of chord diagrams with the defining rep-resentationιIdGLnpCqof GLnpCqyields a Poisson homomorphismCpΣqÑCpΣ; GLnpCqq. We write
Ψ: CpΣqÑCpΣ; GLnpCqqÝΨÑt OpMpΣ; GLnpCqqq for the composite Poisson homomorphism, cf. Theorem 2.22.
Theorem 5.1. Assume thatBΣH. For any partition P ofΣthe-productPonCpΣqdescends viaΨto a-product onOpMpΣ; GLnpCqqq.
Theorem 5.2. Assume thatBΣ H, and let P1 and P2be two partitions ofΣ. The canonical equivalence fromP1toP2onCpΣqdescends viaΨtoOpMpΣ; GLnpCqqqto yield an equivalence between the induced-products.
Remark 5.3. Theorem 5.1 was also stated in [AMR2]. The proof appearing below roughly follows the outline of the justification supplied in that paper. The primary deviation is that we shall not claim that the kernel ofΨis generated by local relations.
44
5.1 The General Linear Case 45
5.1.1 The Relevant Loop Relation
LetEi,j P glnpCqbe the matrix whose sole non-zero entry is a 1 in thepi,jqth entry, and define
Bi,jEi,j Ej,i, Bi,jEi,jEj,i; 1¤i j¤n.
These matrices along withEi,i,i1, . . . ,nare readily seen to constitute an orthogonal basis forglnpCq. Recalling Remark 2.8, we normalize suitably and perform a simple calculation to obtain
pιbιqptqt
¸
i,j
Ei,jbEj,iPEndpCnqbEndpCnq
which under the isomorphism EndpCnqbEndpCnq EndpCnbCnqcorresponds to the transposition of the factors. HenceΨsatisfies the relation (cf. (2.24b))
(5.1)
that is, the loop relation (2.21) with parametersps,fqp1, 0q. Thus we derive a triangle of Poisson homomorphisms
CpΣq
R1,0 Ψ
Z1,0pΣq
Ψ OpMpΣ; GLnpCqqq
The Poisson structure on the loop algebra will not occupy us in the study of KerΨ, so we agree to writeZpΣqZ1,0pΣq.
5.1.2 The Universal GLn-Representation
Our strategy is to introduce a commutative complex algebraRnpΣq RpΣ; GLnpCqq en-dowed with a GLnpCq-action such thatΨfactors through the algebra of fixed points:
RnpΣq
GLnpCq
ZpΣq Ψ OpMpΣ; GLnpCqqq
A universal property definesRnpΣq; it admits a representationρΣ ρΣ,GLnpCq: π1pΣqÑ
GLnpRnpΣqq (the universal GLn-representation of π1pΣq) such that for any representation ρ: π1pΣq Ñ GLnpAq, Abeing a commutative complex algebra, there exists a unique ho-momorphismhρ:RnpΣqÑAfitting into the diagram
GLnpRnpΣqq
GLnphρq
π1pΣq
ρΣ
ρ GLnpAq
(5.2)
46 Chapter 5Quantization of the Moduli Space
Here is an explicit construction ofRnpΣq. Letxgλ,λP Λ| rµ,µ P Mybe a presentationP ofπ1pΣqsatisfying that all relationsrµare written as products of generatorsgλ. LetQnpΛq
be the polynomial algebraCrxλi,j,dλswhereλPΛandi,j1, . . . ,n. Define matricesAλ
pxi,jλq P MnpQnpΛqq,λ P Λ. HereMn denotes the functor assigning the complex algebra MnpRqofnnmatrices to a commutative complex algebraR; note that Ris included in MnpRqas the central subalgebra of scalar matrices. LetIpPqQnpΛqbe the ideal generated bydλdetAλ1 and all entries inAλ1Aλk1 for each relationrµgλ1gλk. Set
RnpΣ;PqRpΣ; GLnpCq,PqQnpΛq L
IpPq; q:QnpΛqÑRnpΣ;Pq.
We now prove thatRnpΣ;Pqsatisfies the universal property, implying in particular that dif-ferent presentations ofπ1pΣqyield canonically isomorphic algebras (all denoted byRnpΣq).
The formulas
detpMnpqqpAλqqqpdetAλq, qpdλqqpdetAλq1
prove thatMnpqqpAλqPGLnpRnpΣ;Pqq, and it is clear that we have a representation ρΣρΣ,GLnpCq,P:π1pΣqÑGLnpRnpΣqq; ρΣpgλqMnpqqpAλq. (5.3) IfAis a commutative complex algebra admitting a representationρ:π1pΣqÑGLnpAq, we definehρ:QnpΛqÑAby
hρpxλi,jqρpgλqi,j, hρpdλqdetρpgλq
1. (5.4)
SinceMnphρqpAλqρpgλq, it follows thatIpPqKerhρ; the induced maphρ:RnpΣqÑA is obviously the unique homomorphism making the triangle (5.2) commutative.
The action of GLnpCqonRnpΣqis the prime application of the universal property. Ele-mentsAPGLnpCqgive rise to representations
A1ρΣA:π1pΣqÑGLnpRnpΣqq
and the corresponding endomorphisms A:RnpΣqÑRnpΣqdefine a GLnpCq-action; this is a direct consequence of the uniqueness of (5.2). By (5.4) and (5.3) we have
Aqpxλi,jqpA1MnpqqpAλqAqi,j, AqpdλqdetpMnpqqpAλqq
1
qpdλq. (5.5) We extend the action of GLnpCqtoMnpRnpΣqqby
AMApAMi,jqA1, MPMnpRnpΣqq, APGLnpCq. (5.6) Consequently,
Lemma 5.4. The inclusionRnpΣqMnpRnpΣqqis equivariant.
We lift the GLnpCq-actions toQnpΛqand MnpQnpΛqq. By definition we can regardQnpΛq
as the algebra of polynomial functions frompMnpCqCqΛtoC, and henceMnpQnpΛqqas the algebra of polynomial functions frompMnpCqCqΛ toMnpCq. Since GLnpCqacts on MnpCqby conjugation and trivially onC, it acts on the productpMnpCqCqΛ. The induced actions on the function sets MapppMnpCqCqΛ,Cqand MapppMnpCqCqΛ,MnpCqq pre-serve the property of being polynomial, thereby defining the desired actions denoted also by.
5.1 The General Linear Case 47
Lemma 5.5. We have dλ, detAλPQnpΛq
GLnpCqand AλPMnpQnpΛqq
GLnpCq.
Proof. This is immediate when regarding dλ, detAλ and Aλ as functions on pMnpCq
CqΛ. l
The analogues of (5.6) and Lemma 5.4 hold:
Lemma 5.6. For any FPMnpQnpΛqqand APGLnpCqwe have AFApAFi,jqA1 Proof. Thinking ofFandFi,jas functions we derive
pAFqppMλ,sλqλqAFpA1pMλ,sλqλqA1
ApFi,jpA1pMλ,sλqλqqA1
AppAFi,jqppMλ,sλqλqqA1
wherepMλ,sλqλPpMnpCqCqΛ. l
Corollary 5.7. The inclusion QnpΛqMnpQnpΛqqis equivariant.
All four GLnpCq-actions are related by
Proposition 5.8. The maps in the commutative square MnpQnpΛqq
Mnpqq
Tr
MnpRnpΣqq
Tr
QnpΛq
q RnpΣq
are equivariant.
Proof. The traces are invariant under conjugation with complex matrices and hence equiv-ariant by (5.6) and Lemma 5.6, respectively. The equivariance ofqneed only be verified on the generatorsxλi,j,dλ. FixAPGLnpCq. For anypMλ,sλqλPpMnpCqCqΛwe have
pAxi,jλ0qppMλ,sλqλqxλi,j0pA1pMλ,sλqλq
xλi,j0ppA1MλA,sλqλq
pA1Mλ0Aqi,j
pA1Aλ0ppMλ,sλqλqAqi,j
ppA1Aλ0AqppMλ,sλqλqqi,j
so that
qpAxλi,j0qqppA1Aλ0Aqi,jqpA1MnpqqpAλ0qAqi,j Aqpxλi,j0q
48 Chapter 5Quantization of the Moduli Space
by (5.5). The elements dλ are invariant by Lemma 5.5, so (5.5) also takes care of those.
RegardingMnpqq, we derive forMPMnpQnpΛqq
MnpqqpAMqMnpqqpApAMi,jqA1q
ApqpAMi,jqqA1
ApApqpMi,jqqqA1
AMnpqqpMq
by Lemma 5.6, the equivariance ofqand formula (5.6). l
Proposition 5.9. The image of the universalGLn-representationρΣ: π1pΣq Ñ MnpRnpΣqqis invariant under the action ofGLnpCq.
Proof. It suffices to consider a generator gλ. The matrix Aλ P MnpQnpΛqq is invariant by Lemma 5.5; the result now follows from the previous proposition since MnpqqpAλq
ρΣpgλqis then invariant, too. l
Remark 5.10. Before continuing the investigation of RnpΣq, we explore its relationship withMpΣ; GLnpCqq. We think of GLnpCqas an affine subset ofMnpCqCCn2 1, namely
GLnpCqtpA,dqPMnpCqC|ddetA1u.
Recalling the construction of the combinatorial complex KP (cf. 2.1), we infer (at least in the case |Λ| 8) that the vanishing set of the idealIpPq QnpΛq OppMnpCqCqΛq is exactlyApKP; GLnpCqqGLnpCqEpKPqGLnpCqΛ pMnpCqCqΛ. By Hilbert’s Null-stellensatz, restriction of functions provides an isomorphism
QnpΛq La
IpPqÑOpApKP; GLnpCqqq.
The former space is, of course, nothing butRnpΣ;Pq{?0, so we have a commutative trian-gle
QnpΛq
q RnpΣ;Pq
p
OpApKP; GLnpCqqq
where Kerp?0. It also follows thatpis a GLnpCq-equivariant surjection since the other two maps in the diagram enjoy this property.
5.1.3 Diagrams and Relative Diagrams
Recall that a diagram onΣis simply the homotopy class of a map from a finite collection of oriented circles toΣ, or, in other words, a set of conjugacy classes in π1pΣq. We need a relative version of this concept; a relative diagram Dis the unit interval I union a finite collection of oriented circles, and arelative diagram onΣis a map f: DÑ ΣIsuch that fpiqpx0,iq,i0, 1, regarded up to homotopy relBI. Post-composing with the projection p: ΣI Ñ Σ(a homotopy equivalence), one sees that such an object is nothing but an
5.1 The General Linear Case 49
γ1 γ2
γ3
Figure 5.1: A decorated diagram.
γ1
γ2
Figure 5.2: A decorated relative diagram.
element ofπ1pΣqtogether with a finite set of conjugacy classes in this group. LetZpΣ,x0q
denote the complex vector space freely generated by relative diagrams onΣ; it is equipped with a natural algebra structure: For relative diagramsfi:DiÑΣI,i1, 2 we define
DD1YD2
L
tB I1BI2u (5.7)
and f f1f2: DÑΣIby fpdq
#
px, 1{2tq, dPD1^f1pdqpx,tq
px, 1{2t 1{2q, dPD2^f2pdqpx,tq The unit for the multiplication is
e:IÑΣI, eptqpx0,tq.
It is convenient to represent (relative) diagrams onΣ bydecorated (relative) diagrams. By this we mean (relative) diagrams along each component of which one or more elements ofπ1pΣqare written. We give a couple of examples of how decorated (relative) diagrams represent (relative) diagrams onΣ. The decorated diagram in Figure 5.1 determines the diagram onΣgiven by a mapS1ÑΣrepresenting the conjugacy class ofγ1γ2γ3Pπ1pΣq. Similarly, the relative diagram onΣrepresented by the decorated relative diagram in Figure 5.2 is defined by a map f: I Ñ ΣI such that the loop pf is in the homotopy class γ1γ2 P π1pΣq. How to interpret general decorated (relative) diagrams is obvious from these examples. Decoration of a component with 1 P π1pΣqis sometimes suppressed in the notation. If the component in question is the interval of a relative diagram, it may be omitted entirely; the potential confusion with a (relative) decorated diagram is non-serious as we shall later.
Remark 5.11. It is obvious that two decorated (relative) diagrams represent the same (rel-ative) diagram onΣif and only if there exists a bijection between the circles of the two dia-grams such that the products along corresponding circles are conjugate elements ofπ1pΣq, and, in the relative case, the products along the two intervals are equal.
Multiplication of (relative) diagrams onΣ lifts to the setting of decorated (relative) dia-grams in the obvious way; take the union of all components carrying along the decoration, and glue the intervals in the relative case (cf. (5.7)).
Parts (local and non-local on Σ) of decorated (relative) diagrams play an important role in the sequel. The ubiquitous example is the braid (over- and undercrossings being ignored)Bσcorresponding to a permutationσPSm; we depict it as (notice the notation for bundles of strands):
Bσ σ
· · ·
· · ·
σ
50 Chapter 5Quantization of the Moduli Space
For example, ifσp1, 2, 3qPS3we have Bσ
We define idealsInpΣqIpΣ; GLnpCqqZpΣqandInpΣ;x0qZpΣ,x0qto be generated by the following three kinds of expressions:
1 n (5.8a)
¸
σPSn 1
ǫpσq σ (5.8b)
¸
σPSn
ǫpσq γ σ
¸
τPSn
ǫpτq γ−1 τ pn!q2 (5.8c)
The first two relations are local onΣ, whereas the third one is not. We write DnpΣqZpΣqLInpΣq, DnpΣ,x0qZpΣ,x0q
L
InpΣ;x0q
for the quotient algebras. Certain elementary decorated (relative) diagrams deserve special attention:
Lγ γ , ELγ γ 1 , Eγ γ
For easy reference we record the following simple fact about them.
Proposition 5.12. DnpΣqis generated by Lγ,γ P π1pΣq, andDnpΣ,x0qis generated by Eg1 λ , λPΛand ELγ,γPπ1pΣq.
The algebrasDnpΣqandDnpΣ,x0qare related by a pair of maps. In one directionι: ZpΣqÑ ZpΣ,x0qis given on decorated diagrams by simply adding an interval decorated by 1 P π1pΣq. This is well-defined on the level of diagrams onΣand clearly induces an algebra homomorphismι: DnpΣq Ñ DnpΣ,x0q; its image is central, so we may viewDnpΣ,x0qas an algebra overDnpΣq. On the other hand, closing up the interval of a decorated relative diagram to a circle and thereby obtaining a decorated diagram results in a mapZpΣ,x0qÑ
ZpΣq; it descends to a linear map Tr : DnpΣ,x0qÑDnpΣq. By relation (5.8a) we have TrιnId : DnpΣqÑDnpΣq.
In particular,ιembedsDnpΣqas a subalgebra ofDnpΣ,x0q; this justifies the aforementioned convention of occasionally omitting a trivially decorated interval of a decorated relative diagram, and we often suppressιin the notation.