aa r X i v : . [ h e p - t h ] J un Quantization via Mirror Symmetry
Sergei Gukov
Abstract.
When combined with mirror symmetry, the A -model approach toquantization leads to a fairly simple and tractable problem. The most inter-esting part of the problem then becomes finding the mirror of the coisotropicbrane. We illustrate how it can be addressed in a number of interesting ex-amples related to representation theory and gauge theory, in which mirrorgeometry is naturally associated with the Langlands dual group. Hyperholo-morphic sheaves and ( B, B, B ) branes play an important role in the B -modelapproach to quantization. Contents
1. Introduction 12. Quantization is an art 33. B -model approach to quantization 134. Quantization of Chern-Simons theory 285. The Verlinde formula via mirror symmetry 34References 43
1. Introduction
Anyone who is not shocked by quantum theoryhas not understood a single word.
Niels BohrThe quantization problem of a symplectic manifold (
M, ω ) can be approachedvia the topological A -model of Y , a complexification of M [ GW ]. In this approach,the Hilbert space H obtained by quantization of ( M, ω ) is the space of open stringstates between two A -branes, B ′ and B cc ,(1.1) H = space of ( B cc , B ′ ) strings , where B ′ is an ordinary Lagrangian A -brane, and B cc is a space-filling coisotropic A -brane. More formally, we can write (1.1) as the space of morphisms(1.2) H = Hom( B cc , B ′ )between two objects, B cc and B ′ , in the Fukaya category of Y . Prepared for the Takagi Lectures 2010.
In general, in a Fukaya category the space of morphisms between two La-grangian objects, B and B , is given by the symplectic Floer homology, HF ∗ symp ( B , B ).Therefore, if both of our objects in (1.2) were familiar Lagrangian objects, the spaceof morphisms H would be obtained by counting their intersection points and ana-lyzing pseudo-holomorphic disks with boundary on B and B .However, our situation is more complicated and more interesting due to thefact that one of the objects, namely B cc , is coisotropic. As a result, the space ofmorphisms (1.2) is not “local” (in a sense that it does not localize to a set of pointsin Y ) and, according to [ GW ], is the Hilbert space obtained by quantizing ( M, ω ).Put differently, the results of [
AZ, GW ] can be interpreted as a statement thatthe space of morphisms between two objects, at least one of which is coisotropic, isclosely related to quantization. More generally, the study of coisotropic branes and their role in the constructionof the Fukaya category is an outstanding interesting problem. Although we will nottry to solve it in the present paper, we will be able to gain some insights by usingmirror symmetry.The computation of the space of morphisms (1.2) can be simplified if the space Y happens to admit additional structures. For example, if Y is hyper-K¨ahler then itis often instructive to look at B cc and B ′ from the vantage point of all three complexstructures, I , J , and K = IJ , as well as the corresponding symplectic structures ω I , ω J , and ω K . Even though originally we were interested in B cc and B ′ as objectsin the Fukaya category often they can be defined as half-BPS boundary conditionsin the N = (4 ,
4) sigma-model of Y , which means that they are also A -branes forsome other A -model of Y , and B -branes for a certain B -model of Y . In particular,the latter implies that (1.2) can be also computed in the B -model of Y :(1.3) H = Ext ∗ Y ( B cc , B ′ ) . Another example of a useful structure is a Calabi-Yau structure. In such case,if Y admits a Calabi-Yau metric, one can approach the computation of (1.2) in themirror B -model:(1.4) H = Ext ∗ e Y ( e B cc , e B ′ ) , where e B cc is the mirror of B cc , and e B ′ is the mirror of B ′ . As we explain below, thehyper-K¨ahler structure on Y and mirror symmetry can both be very useful tools inunderstanding quantization via categories of A - and B -branes. However, combiningthese tools together can double their power!We start our discussion in the next section with a friendly introduction to thequantization problem. Our goal is to explain why this problem is interesting andwhy it is hard. Along the way, we often illustrate the general ideas and key conceptswith concrete and (hopefully) simple examples, many of which have applications torepresentation theory and gauge theory. After recalling the A -model approach toquantization in section 2.3, we reformulate the problem in the mirror B -model andillustrate it in a number of examples in section 3. To be more precise, for this one needs a little bit more: the restriction of the curvature of theChan-Paton bundle of B cc to the subspace of Y where B cc and B ′ have common support shouldbe non-trivial. In the special case when the restriction is trivial the space of morphisms (1.2) isstill very interesting and leads to a theory of D -modules (as opposed to quantization), see [ KW ]. UANTIZATION VIA MIRROR SYMMETRY 3
One of our examples is so rich and important that it deserves a separate section.Thus, in section 4 we apply the mirror approach to quantization of Chern-Simonsgauge theory, where the classical phase space M is the moduli space, M flat ( G, C ),of flat connections on a Riemann surface C . One interesting feature of this exampleis that, for a compact Lie group G , the coisotropic brane B cc is defined only for adiscrete set of symplectic structures on Y , indexed by an integer number k calledthe “level.” Quantization of M leads to a finite-dimensional Hilbert space H , whosedimension is given by the celebrated Verlinde formula [ V ]. In general, the Verlindeformula has the following form:(1.5) dim H = a n k n + a n − k n − + . . . + a k + a , where a i are rational numbers. One novelty of our approach is that it offers aninterpretation of the coefficients a i in terms of branes on moduli spaces of Higgsbundles. The coefficients a n , a n − , . . . determine the asymptotic behavior of thispolynomial in the “classical” limit ~ = k →
0. Similarly, the coefficients a , a , . . . determine the behavior of the polynomial (1.5) in the opposite, “very quantum”regime ~ = k → ∞ which, as we explain in section 4, corresponds to the classicallimit L ~ = − ~ → L G .As a result, the coefficients a n , a n − , . . . have a simple interpretation (in termsof classical geometry of Y ) and are easier to determine in the A -model based on themoduli space of Higgs bundles with the structure group G . On the other hand, thecoefficients a , a , . . . have a simpler interpretation and are easier to determine inthe dual B -model, based on the moduli space of Higgs bundles for the Langlandsdual group L G . In section 5, we present a derivation of the Verlinde formula usingthis approach in a concrete example.
2. Quantization is an art
Very interesting theory — it makes no sense at all.
Groucho Marx (about Quantum Mechanics)The basic problem of quantization begins with a symplectic manifold M , calleda classical “phase space,” equipped with a symplectic form ω . By quantizing ( M, ω )one can mean a number of different things, but usually one is asking for a machinerythat allows to turn the following “classical” objects into their “quantum” analogs:(2.1) (
M, ω ) H (= Hilbert space)alg. of functions on M alg. A ~ of operators on H f
7→ O f : H → H
Lagrangian submanifolds vectors L ⊂ M ψ ∈ H symplectomorphisms automorphismsof M of A ~ There are various interrelations between the classical structures on the left-handside of this list, which should be reflected in their quantum counterparts (the right-hand side). Moreover, depending on specific applications, one can put more itemsto this “wish list”; here we listed only the standard ones.
SERGEI GUKOV
Since the only input data is (
M, ω ) it is not surprising that all of the itemson the left-hand side of (2.1) are the standard gadgets in symplectic geometry.Therefore, quantization can be regarded as a program of constructing a “quantumversion” of symplectic geometry.Another area where the input data is a symplectic manifold is mirror symmetry.Much like the problem of quantization, it starts with a symplectic manifold andconstructs the Gromov-Witten invariants, the Fuakaya category, and many otherinteresting invariants, some of which are even called “quantum” ( e.g. quantumcohomology). Is there any relation between these two problems?As we explain below, the answer is “yes” and the quantization problem canindeed be reformulated as a certain problem in mirror symmetry, however, notin the most naive and obvious way. In particular, the problem of quantizing asymplectic manifold (
M, ω ) can be directly related to a problem in the Fukayacategory of another symplectic manifold, namely a complexification of (
M, ω ).However, before we are ready to review the results of [ GW ] and formulatethem in terms of mirror symmetry, we need to explain some of the delicate featuresof quantization and to introduce important examples. Since the real dimension ofa symplectic manifold is always even, the simplest non-trivial example is either a2-sphere, M = S , or a 2-dimensional plane, M = R (depending on whether weprefer compact or non-compact manifolds). Example . Quantization of M = S One can represent (
M, ω ) as a unit sphere in a 3-dimensional space R ,(2.2) x + y + z = 1 , with a symplectic form(2.3) ω = 14 π ~ dx ∧ dyz . While it may not be immediately obvious, the 2-form ω is invariant underthe SO (3) symmetry of eq. (2.2). Indeed, using a relation between theCartesian coordinates ( x, y, z ) and the spherical coordinates ( r, θ, ϕ ), x = r sin θ cos ϕy = r sin θ sin ϕz = r cos θ one can write (2.3) as a multiple of the standard volume form on a 2-sphere, ω = π ~ sin θ dθ ∧ dϕ . According to textbooks, quantization of ( M, ω ) givesa finite-dimensional Hilbert space H , such that(2.4) dim H = Z M ω = 1 ~ . In particular, dim H must be an integer and this shows that quantizationof ( M, ω ) is possible only for discrete values of the parameter ~ :(2.5) ~ − ∈ Z . UANTIZATION VIA MIRROR SYMMETRY 5
In what follows, we consider a variety of interesting examples in which sym-plectic manifolds come from problems in representation theory on one hand, andfrom gauge theory and low-dimensional topology, on the other. These examplesprovide an excellent laboratory for the quantization problem.As a necessary preliminary to both groups of examples, we introduce the fol-lowing notations that will be used throughout the rest of this paper: G = (simple) compact Lie group, G C = complexification of G , G R = real form of G C .In particular, G R may be equal to G . For concreteness, one can keep in mind asimple example of G = SU (2), for which G C = SL (2 , C ) and G R can be either acompact real form SU (2) (equal to G ) or a split real form SL (2 , R ).Now, we can proceed to some interesting examples of symplectic manifolds.As we mentioned earlier, a large supply comes from representation theory. Let O R ( λ ) = G R · λ be a coadjoint orbit through an element λ ∈ g ∗ R , where g R = Lie( G R )and g ∗ R denotes its dual. To avoid cluttering, we often write O R ( λ ) simply as O R .Then, any such coadjoint orbit is an example of a symplectic manifold [ K1 ]. Indeed, M = O R comes equipped with the Kostant-Kirillov-Souriau Poisson structure /symplectic structure that can be written explicitly(2.6) π = ω − = f ijk X k ∂ i ∧ ∂ j in terms of the structure constants f ijk of g R . Example . G R = SU (2)In this case, the stabilizer of a generic element λ ∈ g ∗ R is a one-dimensional (abelian) subgroup of G R = SU (2), so that O R = SU (2) /U (1) ∼ = S is simply a 2-sphere, as in our previous example. Moreover, sincethe structure constants are given by the totally antisymmetric symbol, f ijk = ǫ ijk , the Kirillov-Kostant symplectic form (2.6) coincides withthe one written in (2.3) if we identify the three-dimensional space R parametrized by ( x, y, z ) with (the dual of) the Lie algebra g R = su (2).Therefore, these two examples are in fact identical.Since the classical phase space M = O R enjoys the action of the symmetrygroup G R , this property should be reflected in its quantum counterpart. Namely,the Hilbert space H obtained from quantization of ( M, ω ) should carry a unitaryrepresentation of the group G R . This is the basic idea of the orbit method.However, there have always been some puzzles with this approach to represen-tations of real groups, which can serve us as important lessons for understandingthe quantization problem: • there exist unitary representations that don’t appear to correspond toorbits; • conversely, there are real orbits that don’t seem to correspond to unitaryrepresentations. SERGEI GUKOV
An example of the first problem occurs even in the basic case of the real group G R = SL (2 , R ) and the complementary series representations. To illustrate thesecond phenomenon, one can take G R to be a real group of Cartan type B N , i.e. G R = SO ( p, q ) with p + q = 2 N + 1. The minimal orbit O min of B N is a nicesymplectic manifold of (real) dimension 4 N −
4, for any values of p and q . On theother hand, the corresponding representation of SO ( p, q ) exists only if p ≤ q ≤
3, and does not exist if p, q ≥
4. This curious observation [ Vo ] follows from arather lengthy algebra and cries out for a simple geometric interpretation!In other words, it would be desirable to have a set of simple topological and/orgeometric criteria that, starting with a symplectic manifold ( M, ω ), would tell usbeforehand whether it should be quantizable or not. Such criteria naturally emergein the brane quantization approach [ GW ] where both of the aforementioned is-sues can be resolved at the cost of of replacing classical geometric objects (namely,coadjoint orbits) with their “stringy” analogs (branes). In particular, in the caseof B N one finds that, while the minimal orbit exists for any values of p and q , thecorresponding brane exists only if p ≤ q ≤
3. (In general, the condition is thatthe second Stieffel-Whitney class w ( M ) ∈ H ( M ; Z ) must be a mod 2 reductionof a torsion class in the integral cohomology of M .)Our second class of examples (in fact, also related to representation theory)comes from gauge theory and low-dimensional topology. Namely, let us considerChern-Simons gauge theory with a real gauge group G R (that may be compact ornot). The key ingredients in any gauge theory include a gauge connection A andthe partial differential equations (PDEs) that it obeys. In the context of Chern-Simons theory, the relevant equations are the flatness equations and, according toAtiyah and Bott [ AB ], the moduli space of flat connections on a compact oriented2-manifold is a finite-dimensional symplectic manifold (possibly singular).Specifically, let A be a connection on a G R bundle E → C over a genus- g Riemann surface C . Then, the moduli space of flat connections on C , M = M flat ( G R , C ), is the space of homomorphisms π ( C ) → G R modulo gauge transfor-mations ( i.e. modulo conjugation). In order to get a better idea of what this spacelooks like, we can describe it more concretely by introducing G R -valued holonomies, A i , B j , i, j = 1 , . . . , g of the gauge connection over a complete basis of A -cycles and B -cycles. Then, the space M = M flat ( G R , C ) can be viewed as a space of solutionsto the equation(2.7) A B A − B − . . . A g B g A − g B − g = modulo conjugation by G R . In total, the group elements A i and B j contain2 g dim G R real parameters, so that generically, for g >
1, after imposing theequation (2.7) and dividing by conjugation we obtain a space of real dimensiondim M = 2( g −
1) dim G R .The space M = M flat ( G R , C ) comes equipped with a natural symplectic form(2.8) ω = 14 π ~ Z C Tr δA ∧ δA , where the parameter k := ~ is called the “level.” What does one find in quantizing( M, ω )? In particular, what is the Hilbert space H ? What is the dimension of H ?The answer to these questions turns out to be surprisingly rich, and dependsin a crucial way on the choice of G R . If G R = G is compact, the space ( M, ω ) is
UANTIZATION VIA MIRROR SYMMETRY 7 quantizable only for integer values of the level,(2.9) k = 1 ~ ∈ Z In this case, the corresponding Hilbert space H is finite-dimensional, and dim H isa polynomial in k , whose leading coefficient equals the volume of M with respectto the symplectic form ω , cf. (2.4),(2.10) dim H = Z M ω n n ! + . . . , where dim M = 2 n . Specifically, H is the space of conformal blocks in the WZWmodel at level k , and the dimension of H is given by the celebrated Verlinde formula[ V ] (see [ B ] for a nice review). This is only the beginning of a very beautiful storythat leads to the Witten-Reshetikhin-Turaev invariants of knots and 3-manifolds. Example . G = SU ( N )The dimension of H is given by the following explicit formula:(2.11) dim H = (cid:18) Nk + N (cid:19) g X S ∐ T =[1 ,k + N ] | S | = N Y s ∈ St ∈ T | π s − tk + N | g − . Notice, from this formula it is completely non-obvious that dim H is apolynomial in k , e.g. for G = SU (2) and g = 2 it gives:(2.12) dim H = 16 k + k + 116 k + 1 . Here, the leading term equals Vol( M ) = Vol( C P ) = k , in agreementwith (2.10) and the well-known fact M ∼ = C P for G = SU (2) and g = 2.The story is very different if G R is non-compact. In this case, the Hilbertspace H is infinite-dimensional, and much less is known about the correspondingquantum group invariants. In particular, the analogs of the Witten-Reshetikhin-Turaev invariants are still waiting to be discovered.The two general classes of examples considered here — based on coadjointorbits and moduli spaces of flat connections — are actually much closer related thanone might think. Indeed, a coadjoint orbit O R (more precisely, the correspondingconjugacy class C R ⊂ G R ) naturally appears as a “local model” for M flat ( G R , C )if we take C to be a punctured disc, see e.g. [ GW1 ]. In fact, these two classes ofexamples can be naturally combined in a larger family by picking a set of “markedpoints” p i , i = 1 , . . . , h on the Riemann surface C and requiring the gauge field A to have certain singularities at the points p i . Equivalently, one can remove thepoints p i and study Riemann surfaces with punctures (or boundary components).For ease of exposition, let us consider a Riemann surface with only one puncture p ∈ C , around which the gauge field has a holonomy(2.13) V = Hol p ( A ) ∈ C R , that takes values in a prescribed conjugacy class C R ⊂ G R . In this way, associatinga conjugacy class to a puncture, we obtain the moduli space M = M flat ( G R , C ; C R )of flat connections on C \ p . As in (2.7), this moduli space can be described ratherexplicitly as a space of solutions to the equation(2.14) A B A − B − . . . A g B g A − g B − g = V ,
SERGEI GUKOV modulo conjugation by G R .The moduli space M = M flat ( G R , C ; C R ) is a symplectic manifold, with thesymplectic form ω given by the general formula (2.8). At least when V is sufficientlyclose to , it has the structure of a symplectic fibration(2.15) C R → M flat ( G R , C ; C R ) ↓ ı M flat ( G R , C )Furthermore, the symplectic form on M flat ( G R , C ; C R ) is(2.16) ω = ı ∗ ω M + ω C , where ω M is the symplectic form on M flat ( G R , C ) and ω C restricts to the Kostant-Kirillov-Souriau symplectic form (2.6) on each fiber of the symplectic fibration (2.15). Example . G = SU (2)Unitary irreducible representations of G = SU (2) can be labeled eitherby the highest weight λ ∈ Z ≥ or, equivalently, by the dimension d = λ + 1.In the physics literature, a representation R λ = S λ C is often called thespin- j representation, where j = λ/
2. As in (2.13), we can associate arepresentation R λ to a marked point p ∈ C by making a puncture, suchthat on a small loop around p the gauge field has a holonomy conjugate to:(2.17) V = exp 2 πi (cid:18) α − α (cid:19) , with α = λ k . Then, for a Riemann surface of genus g with h punctures,the Verlinde formula is(2.18) dim H = (cid:18) k + 22 (cid:19) g − k +1 X j =1 (cid:18) sin πjk + 2 (cid:19) − g − h h Y i =1 sin πj ( λ i + 1) k + 2 . Believe it or not, this is an integer!Now, once we introduced a good supply of interesting symplectic manifolds, weshall return to our original problem of quantization of (
M, ω ). The quantizationproblem can be approached in many different ways. Each approach has its advan-tages and disadvantages, but in the end all methods are expected to yield the sameresult. Below we give a brief overview of various methods, quickly specializing tothe brane quantization approach that will be used in the rest of this paper.
In geometric quantization [
K2, S ], in orderto produce the desired items on the right-hand side of (2.1) one first needs tointroduce some extra data that is not supplied with the symplectic manifold (
M, ω ).Then, of course, one needs to show that the result is, in a suitable sense, independenton these auxiliary choices. (This last step turns out to be the most difficult onealmost in every approach to quantization.)The first piece of extra data — which one needs to introduce not only in geo-metric quantization, but more or less in any approach to quantization — is a choice The representations with even λ are also SO (3) representations. UANTIZATION VIA MIRROR SYMMETRY 9 of line bundle,
L → M , called the “prequantum line bundle” with a unitary con-nection of curvature ω . Note, a prequantum line bundle L → M only exists for(2.19) [ ω ] ∈ H ( M ; Z ) , which can lead to a quantization of ~ − ( i.e. a restriction of ~ to a discrete set ofvalues in C ∗ ). In fact, we already saw this phenomenon in our examples in (2.5)and (2.9).The second choice of extra data is more delicate: it is a choice of polarizationthat, on local charts, corresponds to representing M as a cotangent bundle T ∗ U .It is this second step where geometric quantization faces serious difficulties. Evenif one can locally represent M ≃ T ∗ U in every chart, such choices may not agreeglobally. Moreover, showing that the answer is independent of such choices becomesa rather difficult task. Deformation quantization involves no aux-iliary choices [
BFFLS ]. However, it is not a quantization in the sense of (2.1). In-deed, it does not construct the Hilbert space H and gives only a formal deformationof the ring of functions on M . In deformation quantization, there is no quantizationcondition on the parameter ~ . As in other quantization methods, the approachof [ GW ] starts with a number of auxiliary choices that we summarize below: • Y = complexification of M , i.e. a complex manifold equipped with acomplex structure that we shall call J and an antiholomorphic involution τ : Y → Y , such that τ ∗ J = − J and M is contained in the fixed pointset of τ , • Ω = (non-degenerate) holomorphic 2-form, such that τ ∗ Ω = Ω andΩ | M = ω , • L → Y unitary line bundle (extending the “prequantum line bundle” L → M ) with a connection of curvature Re Ω.Of course, these data need to be consistent. For example, we need to ask for τ tolift to an action on L → Y , such that τ | M = id, etc. To summarize, the basic idea is to pass from the original symplectic manifold(
M, ω ) and the prequantum line bundle L (that we often regard as a part of theinitial data) to the complexification ( Y, Ω) and L . Then, the problem of quantizing( M, ω ) can be formulated as a problem in the A -model / Fukaya category of Y withsymplectic structure(2.20) ω Y = − Im Ω . Note, the symplectic structure ω Y is not a part of the original data, and appearsonly after we complexify the original phase space M .Before we explain how all the desired items on the right-hand side of (2.1) canbe produced in the A -model of ( Y, ω Y ), it is important to emphasize that in thisapproach to quantization the focus shifts from M to Y , so that Y takes the center After embedding the quantization problem in the A -model of Y , it is natural to replacethe latter condition with a slightly more general one, τ : M → M . Among other things, thisgeneralization turns out to be important for finding “missing” coadjoint orbits corresponding tothe complementary series representations [ GW ]. of the stage. Then, from the vantage point of Y it may be natural to considerclose cousins of the original quantization problem suggested by the analysis of the A -model / Fukaya category of Y . For example, one can reduce the list of theauxiliary choices (see above) by omitting the involution τ , which is needed only forunitarity. If one does not require this extra structure (namely, the Hermitian innerproduct on H ), then it suffices to introduce the complex symplectic manifold ( Y, Ω)with a line bundle
L → Y , and no involution τ . From these data alone one canconstruct a space H and an algebra A ~ that acts on it. Later we consider examplesof such situations related to representations theory.We shall illustrate this approach to “quantization via complexification” in avariety of examples introduced earlier in this section; in particular, we apply it to M = O R and M = M flat ( G R , C ). Although these examples have a very differentflavor and come from completely different areas of physics and mathematics, theyare closely related to representation theory of real groups and, at the most basiclevel, the complexification of M can be understood as passing from a real group G R to its complexification G C . Example . Quantization of M = O R A coadjoint orbit O R of a real group G R admits an obvious complexi-fication, namely a complex coadjoint orbit of G C :(2.21) Y = O C . For example, the real coadjoint orbit (2.2) of G R = SU (2) has a complexi-fication Y = O C described by the same equation x + y + z = 1, where x , y , and z are now complex variables. Moreover, eq. (2.3) written in termsof ( x, y, z ) ∈ C defines a holomorphic symplectic form Ω on Y .Similarly, there is an obvious complexification of the moduli space of flat con-nections, M flat ( G R , C ). Example . Quantization of M = M flat ( G R , C )This symplectic manifold M admits an obvious complexification:(2.22) Y = M flat ( G C , C ) , the moduli space of flat G C connections on C . Much like M itself, the space Y can be explicitly described as a space of G C -valued holonomies A i and B j that satisfy (2.7), modulo conjugation by G C . It comes equipped witha holomorphic symplectic form Ω which, in terms of the g C -valued gaugeconnection, has the familiar form (2.8).In these examples, it is easy to verify that the holomorphic symplectic form Ω re-stricts to ω on M ⊂ Y .Now let us return to the quantization problem of ( M, ω ) and explain how thedesired items on the right-hand side of (2.1) can be produced in the approachof [ GW ]. The Hilbert space H is constructed as the space of morphisms (space ofopen strings),(2.23) H = Hom( B cc , B ′ ) , where B cc and B ′ are objects (branes) of the Fukaya category of ( Y, ω Y ). UANTIZATION VIA MIRROR SYMMETRY 11
In general, typical objects of the Fukaya category of (
Y, ω Y ) are Lagrangiansubmanifolds of Y equipped with flat unitary vector bundles and, in our setup, B ′ is exactly such an object. Specifically, we define B ′ to be an A -brane supported on M ⊂ Y . Indeed, according to our definitions,(2.24) ω Y | M = 0 , so that M is a Lagrangian submanifold of Y with respect to ω Y .Less familiar examples of A -branes (objects of the Fukaya category) are coisotropicsubmanifolds of ( Y, ω Y ) equipped with non-flat vector bundles (a.k.a. Chan-Patonbundles) with unitary connection that obeys certain conditions [ KO ]. In the sim-plest case of rank-1 coisotropic objects supported on all of Y , the condition on thecurvature 2-form F is(2.25) ( ω − Y F ) = − . In our approach to quantization, B cc is an example of such object, namely the so-called canonical coisotropic brane associated to a complexification of ( M, ω ) in acanonical way [ GW ].To summarize, after we choose a complexification of ( M, ω ) and the extensionof the prequantum line bundle L , we can define two canonical objects in the Fukayacategory of ( Y, ω Y ): B ′ = Lagrangian A -brane supported on M ⊂ Y B cc = coisotropic A -brane supported on Y and endowed with a unitary linebundle L with a connection of curvature F = Re ΩIn particular, it is easy to verify that, with our definition of ω Y , the curvature2-form F indeed obeys the required condition (2.25).Given two objects B ′ and B cc , it is natural to consider the spaces of morphisms(spaces of open strings) in the A -model of ( Y, ω Y ). As we already stated in (2.23),the space of ( B cc , B ′ ) strings gives the Hilbert space H associated with the quanti-zation of ( M, ω ). Just like in other quantization methods, one needs to show thatit is independent on the auxiliary choices (which, among other things, involve thechoice of complex structure J on Y ), i.e. to construct a flat connection on the H -bundle over the space of such choices. In a closely related context, this problemhas been studied in the mathematical physics literature [ CV, D ], and leads to abeautiful story that involves integrable systems and tt ∗ equations. Example . Quantization of M = T In this problem, M = T admits an obvious complexification,(2.26) Y ∼ = C ∗ × C ∗ , and the resulting Hilbert space H should not depend, among other things,on the choice of complex structure on M = T . If we denote by t ∈ T the corresponding complex structure parameter, then the states in H are simply theta functions of order k ,(2.27) ϑ r ( z ; t ) = ∞ X n = −∞ exp (cid:18) πitk ( kn + r ) + 2 πi ( kn + r ) z (cid:19) r ∈ Z /k Z , where k := dim H = ~ , cf. (2.9). It is easy to see from (2.27) that ϑ r ( z ; t )are quasi-periodic, ϑ r ( z + a + bt ; t ) = exp (cid:0) − πikb t − πikbz (cid:1) ϑ r ( z ; t ) a, b ∈ Z and obey the heat equation(2.28) (cid:18) ∂∂t − ~ πi ∂ ∂z (cid:19) ϑ r ( z ; t ) = 0 , which gives a connection on a bundle H → T . This example will beapproached from a different viewpoint in section 3.2.Furthermore, in brane quantization the involution τ leads to a Hermitian innerproduct on H . It is not necessarily positive definite; a necessary condition is that τ fixes M pointwise outside of a compact support. This slight generalization ofthe condition that M belongs to the fixed point set of τ is important e.g. forconstructing the complementary series representations, where τ acts non-triviallyon M .The space of ( B cc , B cc ) strings, on the other hand, gives an associative butnon-commutative algebra,(2.29) A ~ = Hom( B cc , B cc ) . Note, this algebra depends only on ( Y, Ω) and not on M . (In our examples, itmeans that the same algebra A ~ acts on Hilbert spaces obtained in quantization of M = M flat ( G R , C ) for different real forms G R of G C , and similarly for M = O R .)In fact, we can think of the algebra A ~ as arising from the deformation quantizationof Y .The path integral of the quantum mechanics on M also has an elegant realiza-tion in the A -model approach, see [ W3 ] for details. While highly desirable, a completeset of geometric criteria that determine which symplectic manifolds are quantizable(and which are not) is not known at present. Most likely, such criteria shouldinclude the condition (2.19) that controls the existence of the prequantum linebundle L (and sometimes leads to a quantization of ~ ). However, this conditionalone is clearly not enough, and — even as some of our examples suggest — thereshould be further criteria which determine whether ( M, ω ) is quantizable or not.From the viewpoint of brane quantization, (
M, ω ) is expected to be quantizablewhenever it admits a complexification, such that (
Y, ω Y ) has a “good” A -model /Fukaya category. A precise necessary condition for this is not known (in part,since the present understanding of the Fukaya category is incomplete). A sufficientcondition, though, is that ( Y, ω Y ) admits a complete Calabi-Yau metric g , for which Indeed, in the classical limit the norm of a state ψ ∈ H is roughly h ψ, ψ i = Z M ψ ( τx ) ψ ( x ) . It is positive definite only if τ fixes M pointwise. UANTIZATION VIA MIRROR SYMMETRY 13 ω Y is a K¨ahler form, i.e. K = g − ω Y is an integrable complex structure. Oneindication that such criteria are on a right track is that, in deformation quantization,one encounters similar conditions that tell us whether A ~ is an actual deformationof the algebra of holomorphic functions on ( Y, Ω), with a complex parameter ~ (notjust a formal variable). Example . Quantization of M = S In (2.2) we represented S as a unit sphere in R . Its complexification,(2.30) x + y + z = 1 , ( x, y, z ) ∈ C admits a complete Calabi-Yau metric (the Eguchi-Hanson metric) and adeformation of the ring of functions with a complex parameter ~ (not justa formal deformation). Both of these properties fail for a complex surface,(2.31) x + y + z = 1 , ( x, y, z ) ∈ C that can be viewed as an alternative complexification of M = S .Besides the requirement for ( Y, ω Y ) to have a good A -model / Fukaya category,one needs B ′ and B cc to exist in order to solve the original quantization problem, i.e. to compute the spaces of morphisms (2.23) and (2.29). Fortunately, the existenceof B ′ and B cc can be expressed in terms of concrete geometric criteria, which canbe useful even in other quantization methods. Specifically, the brane B ′ supportedon M exists whenever M admits a flat Spin c structure, and the brane B cc existswhenever [Re Ω] ∈ H ( Y ; Z ), cf. (2.19). B -model approach to quantization Gott w¨urfelt nicht!
Albert EinsteinIn section 2.3 we reviewed the A -model approach to quantization [ GW ], whereto a classical symplectic manifold ( M, ω ) one associates a Fukaya category of A -branes and the quantization (2.1) is achieved by studying the space of morphismsbetween two branes B cc and B ′ . It is important to emphasize, however, that theFukaya category in question is not that of the original symplectic manifold ( M, ω ).Rather, it is the Fukaya category of Y , a complexification of M , considered with anew symplectic form (2.20) that didn’t exist prior to complexification.Another area of physics & mathematics where Fukaya categories are of majorimportance is mirror symmetry. In general, mirror symmetry relates the A -modelof a symplectic manifold Y to the B -model of a complex manifold e Y , called themirror of Y . In mirror symmetry, however, the Fukaya category and A -model areusually considered to be the ‘difficult’ side of the correspondence, and it is oftenconvenient to use mirror symmetry to map the problem to the simpler B -modelside.In our present context, this map is described by the homological mirror symme-try conjecture [ K ] that relates the derived Fukaya category of Y and the (bounded)derived category of coherent sheaves on e Y . Specifically, the conjecture says thatthere exists a functor:(3.1) Φ mirror : Fuk ( Y ) ∼ −→ D b ( e Y ) such that it is the equivalence of triangulated categories. In the rest of this paperour goal will be to apply this map to the A -model of ( Y, ω Y ) described in section2.3 and thereby to reformulate the quantization problem entirely in terms of the B -model.In particular, mirror symmetry maps our A -branes B cc and B ′ to the dual B -branes: e B ′ = Φ mirror ( B ′ )(3.2) e B cc = Φ mirror ( B cc )whose geometry we wish to explore. Furthermore, mirror symmetry provides adual description of the Hilbert space H and the algebra of quantum operators A ~ in terms of Ext-groups of the dual objects e B ′ and e B cc . For example, it identifiesthe space of morphisms (2.23) with(3.3) H = Ext ∗ e Y ( e B cc , e B ′ ) , which can be analyzed using the standard tools of algebraic geometry. Thus, theEuler characteristic of (3.3) can be easily computed in the B -model with the helpof the Grothendieck-Riemann-Roch theorem:(3.4) X k ( − k dim Ext k e Y ( e B cc , e B ′ ) = Z e Y ch( e B cc ) ∗ ∧ ch( e B ′ ) ∧ Td( e Y ) , where ch( e B ′ ) (resp. ch( e B cc )) is the Chern character of e B ′ (resp. e B cc ), Td( e Y ) isthe Todd class of e Y , and ω ∗ denotes ( − p +1 ω for any 2 p -form ω . In applications,we will often use (3.4) to compute the dimension of the Hilbert space H (whendim H < ∞ ).What are the mirror objects e B ′ and e B cc ? Does the mirror of the canonicalcoisotropic A -brane B cc admit a ‘canonical’ definition in D b ( e Y ) ( i.e. in the B -model of e Y )? What is the role of ~ in the B -model of e Y ? In order to answer theseand other questions about the B -model approach to quantization of ( M, ω ), it isuseful to have a good geometric description of the mirror transform (3.1). One suchdescription was proposed in 1996 by Strominger, Yau, and Zaslow [
SYZ ] (see also[
BJSV ]), who argued that mirror Calabi-Yau manifolds Y and e Y should fiber overthe same base manifold B ,(3.5) Y e Y π ց ւ e π B with generic fibers F b = π − ( b ) and e F b = e π − ( b ), b ∈ B , being dual tori, in thesense that F b = H ( e F b , U (1)) and e F b = H ( F b , U (1)). Moreover, the fibers F b and e F b should be (special) Lagrangian submanifolds in Y and e Y , respectively.This way of looking at mirror symmetry can be very useful in understandinghow the functor (3.1) acts on the A -branes B ′ and B cc , which ultimately will lead usto a reformulation of the quantization problem in the mirror B -model. In particular,as we explain below, the fate of the coisotropic brane B cc depends in a crucial way We remind that, by definition, a middle-dimensional submanifold M ⊂ Y is called La-grangian if the symplectic form ω Y vanishes on M , and is special Lagrangian if, in addition, theimaginary part of the holomorphic volume form on Y vanishes when restricted to M . UANTIZATION VIA MIRROR SYMMETRY 15 on whether the restriction of the symplectic form F = Re Ω on Y to a generic fiber F of the SYZ fibration (3.5) is trivial or not. When it is trivial, the coisotropicbrane B cc transforms under mirror symmetry to a brane e B cc supported on a middle-dimensional submanifold of e Y , namely on a section of the dual SYZ fibration. (Anexample of such situation was considered e.g. in [ KW ].)In contrast, when F | F is non-trivial, the story becomes more interesting andmore complicated. In this case — which will be our subject here — mirror symmetrytransforms the coisotropic brane B cc into a B -brane e B cc ∈ D b ( e Y ) supported on allof e Y . Furthermore, in general e B cc is a brane of a fairly high rank. In fact, usingthe SYZ picture (3.5) we conclude that the rank of e B cc is given by(3.6) rank( e B cc ) = Vol( F )where Vol( F ) is the volume of the SYZ fiber F computed with respect to thesymplectic form F = Re Ω on Y ,(3.7) rank( e B cc ) = Z F F n n ! , and dim R F = dim C Y = 2 n . (Remember, that in our context Y is always acomplex symplectic manifold.) Notice, the formula (3.6) also applies to the simplercase where F | F is trivial.In what follows, we shall illustrate (3.6) in a variety of concrete examples.However, there is also a general argument based on (3.5) that we wish to sketchhere since it will be very useful in later applications. In the A -model approachto quantization, B ′ is a Lagrangian brane on ( Y, ω Y ). Since the fiber of the SYZfibration (3.5) is Lagrangian with respect to ω Y = − Im Ω, we can choose B ′ to be aLagrangian brane supported on a generic fiber F b ⊂ Y and equipped with a unitaryflat line bundle. In this simple warm-up example we know exactly what the dualobject e B ′ is. It is the skyscraper sheaf O p ∈ D b ( e Y ) of a point p ∈ e Y , such that e π ( p ) = b . For this reason, e B ′ = O p is often called a “zero-brane” or “D0-brane” on e Y . Summarizing,(3.8) Φ mirror : B F → B p , where we used slightly more intuitive notations B F and B p for this type of A -branesand B -branes, respectively.In fact, the mirror pair of branes in (3.8) was an important part of the originalmotivation in [ SYZ ] that led to the proposed picture (3.5). One way to see that B F and B p should be mirror to each other is to consider their self-Homs. For a B -brane B p = O p on e Y , we have(3.9) Ext ∗ e Y ( B p , B p ) ∼ = Λ ∗ T p e Y ∼ = H ∗ ( T n , C ) . As a gradede vector space, it is isomorphic to the Floer cohomology of F ∼ = T n ,which describes the self-Homs of the A -brane B F :(3.10) HF ∗ ( B F , B F ) ∼ = H ∗ ( F , C ) , hence, justifying (3.8).Now, once we understand the duality (3.8) between branes B F and B p , we canuse it to “probe” the geometry of e B cc . Namely, as suggested earlier, we can use B F for the A -brane B ′ (and, hence, B p for the mirror B -brane e B ′ ) to compute thespace of morphisms H (= space of open strings) between B F and B cc , just like wedid it a moment ago for the brane B F itself. According to (2.23), in the A -model of( Y, ω Y ) the space H = Hom( B cc , B F ) is obtained by quantizing the support of B F ,with the symplectic form ω = Re Ω | F . Since the support of B F is an abelian variety F ∼ = T n , its quantization is well understood and leads to a space of θ -functions, cf. (2.27), of dimension (3.11) dim H = Z F F n n ! , where we used F = Re Ω. This gives us the right-hand side of (3.7). On the otherhand, calculating the dimension of H in the B -model of e Y with the help of (3.4)we obtain dim H = dim Ext ∗ e Y ( e B cc , B p ) = rank( e B cc ), which is precisely the left-handside of (3.7). This calculation concludes a useful exercise that will also serve us asa practice example for studying H in the A -model of Y and in the B -model of e Y . ( B, B, B ) branes and hyperholomorphic bundles. In the A -modelapproach to quantization, the classical phase space ( M, ω ) is replaced by a complexsymplectic manifold ( Y, Ω) which, by definition, comes equipped with two symplec-tic forms that we call F = Re Ω and ω Y = − Im Ω, and a complex structure J thatrelates them.Now we wish to focus on a particularly nice situation where both symplecticforms F and ω Y are K¨ahler with respect to some complex structures I and K = IJ , so that Y is a hyper-K¨ahler manifold (this happens e.g. if M is a K¨ahlermanifold). Then, using the standard notations ω I , ω J , ω K for the three K¨ahlerforms corresponding to the complex structures I , J , and K , we can write theholomorphic symplectic form Ω as(3.12) i Ω = ω K + iω I , where, according to the conventions of section 2.3,(3.13) F = ω I , ω Y = ω K . What about the objects B cc and B ′ that play a central role in the A -modelapproach to quantization? As we already mentioned in the Introduction, they tendto be automatically compatible with the hyper-K¨ahler structure on Y , when itexists. Namely, defined as half-BPS boundary conditions in the N = (4 ,
4) sigma-model of Y , they often preserve supersymmetry in two different A -models of Y ,with respect to different symplectic forms, say ω J and ω K , and also in a B -modelof the third complex structure, I . Following the terminology introduced in [ KW ],we call such objects “branes of type ( B, A, A ).”A quick remark on the notation is on order. On a hyper-K¨ahler manifold Y thechoice of what we call the complex structures I , J , and K (and the correspondingK¨ahler forms ω I , ω J , and ω K ) is, of course, entirely up to us. In fact, I , J , and K are part of the entire sphere S = C P of complex structures on Y ,(3.14) I = aI + bJ + cK , parametrized by ( a, b, c ) ∈ R with a + b + c = 1. Therefore, when in a favorablesituation we say that B cc and B ′ are holomorphic in complex structure I — which Notice, eq. (2.10) is exact in this case.
UANTIZATION VIA MIRROR SYMMETRY 17 makes them branes of type (
B, A, A ) — this choice is quite random, except that itsorientation with respect to the fiber of the SYZ fibration (3.5) is very important.Throughout the paper, we adopt the convention that, when Y is hyper-K¨ahler, thefiber F is always holomorphic in complex structure I (and Lagrangian with respectto ω J and ω K ). In other words, the fiber itself is an object (brane) of type ( B, A, A ).This makes the complex structure I and, hence, the branes of type ( B, A, A ) a bitspecial among others.A typical example of a (
B, A, A ) brane is a middle-dimensional submanifoldof Y that is holomorphic in complex structure I and Lagrangian with respect toboth ω J and ω K . In fact, B ′ is a good example of such an object, when Y ishyper-K¨ahler. Example . (
B, A, A ) branes on Y = R Locally, the geometry of every hyper-K¨ahler manifold looks like aquaternionic n -pane, H n . In the simplest case n = 1, we may identifya point ( x , x , x , x ) ∈ R with a quaternion q ∈ H :(3.15) q = x + i x + j x + k x where i = j = k = ijk = −
1. The three complex structures I , J , K acton R ∼ = H by left multiplication by i , j , k , and the corresponding K¨ahlerforms are(3.16) i ω I + j ω J + k ω K = − dq ∧ dq where q = x − i x − j x − k x is the conjugate quaternion. Explicitly, ω I = dx ∧ dx + dx ∧ dx ω J = dx ∧ dx − dx ∧ dx (3.17) ω K = dx ∧ dx + dx ∧ dx Simple examples of (
B, A, A ) branes are branes supported on f ( z, w ) = 0,where f ( z, w ) is a holomorphic function of z = x + ix and w = x + ix .With the above definitions, it is easy to verify that these submanifolds arecomplex for complex structure I and Lagrangian for ω J and ω K .The second key ingredient, the coisotropic brane B cc , is an A -brane on Y withrespect to ω K , but at the same time it is a B -brane in complex structure I . In fact,in the B -model of ( Y, I ) the brane B cc corresponds to a holomorphic line bundlethat, abusing notations a little, we also denote L → Y , with the first Chern class(3.18) c ( L ) = ω I . Therefore, when Y admits a hyper-K¨ahler structure, the Hilbert space H is simplygiven by (1.3) and can be analyzed in complex structure I using the tools of alge-braic geometry.Our next goal is to see whether the extra structure of Y being a hyper-K¨ahlermanifold and B ′ , B cc being branes of type ( B, A, A ) can help us to identify themirror objects e B ′ , e B cc . As explained in [ KW ] and as we illustrate in many examples below, in general a brane of type ( B, A, A ) transforms under mirror symmetry intoa brane of type (
B, B, B ), i.e. a B -brane for all complex structures on e Y :Φ mirror : ( B, A, A ) branes −→ ( B, B, B ) branesThis statement depends, of course, in a crucial way on the fact that the fibers ofthe SYZ fibration (3.5) are also of type (
B, A, A ), i.e. the fibration is holomorphicin complex structure I and Lagrangian for ω J and ω K .In particular, since the fibration (3.5) is assumed to be holomorphic in complexstructure I , mirror symmetry transforms holomorphic objects into holomorphicobjects and, as a result, does not change the type of branes in complex structure I .(This, of course, is not the case in other complex structures.) Moreover, from thevantage point of the complex structure I , the SYZ duality along the fibers F canbe described as a Fourier-Mukai transform:(3.19) Φ FM : D b ( Y, I ) ∼ −→ D b ( e Y , e I ) , where, to avoid confusion, we made explicit the choice of complex structures. Thispoint of view can be very helpful in identifying the mirror objects (3.2) dual toour branes B ′ and B cc . As long as they are B -branes in complex structure I , theirmirrors can be obtained by the following general formula(3.20) e B = R e p ∗ ( p ∗ B ⊗ P ) , which describes explicitly the action of the functor (3.19) on a brane B ∈ D b ( Y, I ).Here, P is the relative Poincar´e line bundle on Z := Y × B e Y , and Z p ւ e p ց Y e Y In particular, the Chern character of the mirror (
B, B, B ) brane e B is given by(3.21) ch( e B ) = e p ∗ (ch( P ) ∧ p ∗ (ch( B ))) . Although the viewpoint of complex structure I is extremely useful (and we shallreturn to it later), now we wish to proceed with a more democratic approach where e B cc and e B ′ are considered as objects of type ( B, B, B ). In particular, our goal is tounderstand what this extra structure really means and what it can be good for.The simplest example of a (
B, B, B ) brane on e Y is a hyperholomorphic bundle E , i.e. a holomorphic bundle compatible with the hyper-K¨ahler structure on e Y ,in the sense that E admits a Hermitian connection ∇ with a curvature F ∇ ∈ Λ ( e Y ,
End( E )) which is of Hodge type (1 ,
1) with respect to all complex structures.A stable bundle E is hyperholomorphic if and only if its Chern classes c and c are SU (2)-invariant, with respect to the natural SU (2) action on the cohomology,see e.g. [ Ve ]:(3.22) E hyperholomorphic ⇔ c ( E ) , c ( E ) SU (2)-invariantThis simple criterion is our first indication that the study of ( B, B, B ) branes isclosely related to the study of SU (2) action on the cohomology of e Y . UANTIZATION VIA MIRROR SYMMETRY 19
For instance, if e Y = R as in our previous example, then the left multiplication q u · q by a unit quaternion u ( uu = 1) is an isometry of the flat hyper-K¨ahlermetric ds = dqdq on e Y = R and rotates the three K¨ahler forms (3.17). This givesa rather explicit local model for SU (2) action on the cohomology of e Y . In general,when we apply the criterion (3.22) to the Chern character ch( e B ) of the brane e B weshall often use the fact that a differential form ω on a hyper-K¨ahler manifold e Y is SU (2)-invariant if and only if it is of Hodge type ( p, p ) with respect to all complexstructures on e Y .A larger class of examples of ( B, B, B ) branes on e Y can be obtained by con-sidering hyperholomorphic sheaves, i.e. coherent sheaves compatible with a hyper-K¨ahler structure, in the same sense as hyperholomorphic bundles are holomorphicbundles compatible with a hyper-K¨ahler structure. Example . (
B, B, B ) branes on e Y = T ∗ S In quantization of M = S we encountered the Eguchi-Hanson met-ric on a complex surface (2.30) which, up to a hyper-K¨ahler rotation andirrelevant technicalities, is essentially self-mirror. Therefore, as a first ap-proximation to e Y we can take a locally asymptotically flat hyper-K¨ahlermetric on T ∗ S , for which the K¨ahler forms can be written explicitly ω I = e ∧ e + e ∧ e ω J = e ∧ e − e ∧ e (3.23) ω K = e ∧ e + e ∧ e in the orthonormal basis e = f − / dr , e = r f / ( dψ − cos θdϕ ) , e = r dθ , e = r θdϕ with f ( r ) = 1 − r r . This metric admits a normalisable anti-self-dual har-monic 2-form(3.24) ̟ = 1 r ( e ∧ e − e ∧ e ) , which, according to (3.22), can represent the first Chern class of a ( B, B, B )brane e B . Indeed, the 2-form (3.24) is of type (1 ,
1) with respect to allcomplex structures on e Y ∼ = T ∗ S and is orthogonal to all three K¨ahlerforms (3.23).This example is the simplest case of the following infinite family of hyper-K¨ahler metrics on T ∗ C P n discovered by E. Calabi [ C ] (who also introduced theterm “hyper-K¨ahler”). Example . (
B, B, B ) branes on e Y = T ∗ C P n In an orthonormal basis of 1-forms, the Calabi metric has the standardform ds = n X i =1 3 X a =0 e ( i ) a ⊗ e ( i ) a with the K¨ahler forms, cf. (3.23), ω I = n X i =1 (cid:16) e ( i )0 ∧ e ( i )1 + e ( i )2 ∧ e ( i )3 (cid:17) ω J = n X i =1 (cid:16) e ( i )0 ∧ e ( i )2 − e ( i )1 ∧ e ( i )3 (cid:17) (3.25) ω K = n X i =1 (cid:16) e ( i )0 ∧ e ( i )3 + e ( i )1 ∧ e ( i )2 (cid:17) These K¨ahler forms are rotated by the SU (2) symmetry, under which thebasis 1-forms transform as doublets: e ( i )0 + ie ( i )1 e ( i )2 − ie ( i )3 ! and e ( i )2 + ie ( i )3 − e ( i )0 + ie ( i )1 ! . From these one can construct singlets, i.e. SU (2)-invariant forms on e Y ,which include the following harmonic 2-form [ CGLP ]:(3.26) ̟ = 1 r (cid:16) e (1)0 ∧ e (1)1 − e (1)2 ∧ e (1)3 (cid:17) + 1 r n X i =2 (cid:16) e ( i )0 ∧ e ( i )1 − e ( i )2 ∧ e ( i )3 (cid:17) . This harmonic 2-form is not normalisable (except for n = 1, when it reducesto (3.24)), but it is regular and square-integrable at r = r .The special case ( n = 3) of this last example shows up in the B -model approachto quantization of M = M flat ( G, C ), with G = SO (3) and C of genus g = 2; seecomments below (2.12) and section 4. In particular, the SU (2)-invariant 2-form(3.26) turns out to be essentially the first Chern class of the ( B, B, B ) brane e B cc .Another way to construct a ( B, B, B ) brane is to take an ideal sheaf of a triana-lytic subvariety of e Y . Trianalytic subvarieties have an action of quaternion algebrain the tangent bundle. In particular, the real dimension of such subvarieties isdivisible by 4. By analogy with hyperholomorphic bundles (sheaves) they can becharacterized by the following criterion, similar to (3.22): if S ⊂ e Y is a closed ana-lytic subvariety of ( e Y , e I ) and [ S ] ∈ H i ( e Y ) is SU (2)-invariant, then S is trianalytic.Trianalytic subvarieties are quite rare; for example, a Hilbert scheme of a generic K e Y of (real) dimension 4.Clearly, in these examples ch( e B ) has components only in degree 0, 2, and 4, sothat (3.22) provides a non-trivial constraint only on a degree-2 component, i.e. onthe first Chern class of e B . On the other hand, if ω I , ω J , ω K , ̟ , . . . , ̟ k is anorthonormal basis in H ( e Y ), then the vectors ̟ , . . . , ̟ k are SU (2)-invariant, andin the natural SU (2)-invariant decomposition(3.27) H ( e Y ) = H ( e Y ) ⊕ H ( e Y ) UANTIZATION VIA MIRROR SYMMETRY 21 we have dim H ( e Y ) = 3 and H ( e Y ) ∼ = H − ( e Y ), where H ( e Y ) = H ( e Y ) ⊕ H − ( e Y )is the standard decomposition of H ( e Y ) according to the eigenvalues of the Hodge ∗ operator. In this section, we apply the general formalism describedabove to a simple model, where the SYZ fibration (3.5) is actually trivial. Specifi-cally, we take(3.28) Y = B × F where B = R and F = T . This model can be regarded as a quantization of M = T , cf. (2.26). Indeed, if we choose B ′ = B F to be a Lagrangian brane supportedon M = F and B cc to be a coisotropic brane with the appropriate Chan-Patonbundle L , we obtain precisely the setup of section 2.3. Note, this Lagrangian brane B ′ is the one we also used in (3.8) to prove the general formula (3.6). In particular,in (3.11) we already calculated the dimension of the corresponding Hilbert space H .As a warm-up to more interesting models, we wish to show explicitly in thisexample that the branes B ′ and B cc are compatible with the hyper-K¨ahler structureon Y and to use this information to find the mirror ( B, B, B ) branes e B ′ and e B cc .In order to do this, however, we first need to introduce the complex structures I , J , K , and the corresponding K¨ahler forms on Y . These are essentially written in(3.17). Let b , b be a basis of 1-forms on the base B , and f , f (resp. e f , e f ) bea basis of 1-forms on the fiber F (resp. the dual fiber e F ). Then, the K¨ahler formson Y are ω I = 1 ~ ( b ∧ b + f ∧ f ) ω J = 1 ~ ( b ∧ f − b ∧ f )(3.29) ω K = 1 ~ ( b ∧ f + b ∧ f )where, compared to (3.17), we introduced the parameter ~ relevant for quantization.Dualizing the fiber F , we obtain the mirror manifold(3.30) e Y = B × e F , which, of course, is also a trivial SYZ fibration, with the fiber e F = H ( F , U (1)) ∼ = T . Moreover, since the U (1) isometry of Y (that acts in a natural way by trans-lations along the SYZ fibers) is tri-holomorphic, the duality certainly does notspoil the hyper-K¨ahler structure. Hence, the resulting mirror manifold e Y is alsohyper-K¨ahler, and the corresponding K¨ahler forms e ω I = 1 ~ b ∧ b + ~ e f ∧ e f e ω J = b ∧ e f − b ∧ e f (3.31) e ω K = b ∧ e f + b ∧ e f can be obtained from (3.29) simply by replacing f i → ~ e f i . Note, in particu-lar, that in the K¨ahler metric corresponding to the complex structure I , we haveVol( F ) ∼ ~ − n , where n = dim C F , and Vol( e F ) ∼ ~ n . This property holds true for more general mirror pairs, Y and e Y .Now, we wish to identify the mirror ( B, B, B ) branes e B ′ and, most impor-tantly, e B cc . The brane e B ′ is easy to identify and, in fact, we already took careof it in our earlier discussion: as summarized in (3.8), mirror symmetry maps aLagrangian brane B F to a skyscraper sheaf B p = O p ∈ D b ( e Y ). Therefore, if wechoose B ′ = B F , as in our approach to quantization of M = T , then the mirror B -brane is e B ′ = B p . It has the Chern character(3.32) ch( e B ′ ) = − b ∧ b ∧ e f ∧ e f , which is consistent with (3.21) and is manifestly invariant under the SU (2) actionon the cohomology of e Y . (Clearly, the degree-0 form and the volume form are SU (2)-invariant on any hyper-K¨ahler manifold e Y .)Identifying the mirror of the coisotropic ( B, A, A ) brane B cc is more interesting.On general grounds, we know that it should be an object of type ( B, B, B ), i.e. holomorphic in all complex structures on e Y , and, according to (3.6), should haverank( e B cc ) = R F ω I = ~ . Therefore, we expect(3.33) ch( e B cc ) = 1 ~ + . . . , where the rest of the terms (denoted by ellipsis) should be invariant under the SU (2) action on the cohomology of e Y . Besides the 0-form and the volume form(which are always SU (2)-invariant), such terms may contain any linear combinationof the anti-self-dual 2-forms on e Y :(3.34) 1 ~ b ∧ b − ~ e f ∧ e f , b ∧ e f + b ∧ e f , b ∧ e f − b ∧ e f , which, according to (3.27), are precisely the generators of the SU (2)-invariant partof the cohomology H ( e Y ). (It is easy to verify that all of the forms in (3.34) areorthogonal to the K¨ahler forms (3.31).) Of course, this structure alone does notuniquely determine ch( e B cc ), but it is amusing to see how close we managed to getto the correct answer.In order to compute ch( e B cc ) more systematically, we can treat the coisotropicbrane B cc as a B -brane in complex structure I , where it corresponds to a holomor-phic line bundle L with the first Chern class (3.18). Then, substituting ch( B cc ) =exp( ω I ) into the general formula (3.21),(3.35) ch( e B cc ) = Z F ch( P ) ∧ p ∗ (ch( B cc )) , where P is a complex line bundle on Z = B × F × e F defined by its first Chernclass,(3.36) c ( P ) = X i =1 f i ∧ e f i , we obtain the Chern character of the mirror ( B, B, B ) brane e B cc :(3.37) ch( e B cc ) = 1 ~ + 1 ~ b ∧ b − e f ∧ e f − ~ b ∧ b ∧ e f ∧ e f . We leave this as an exercise.
UANTIZATION VIA MIRROR SYMMETRY 23
Note, the degree-0 term in this expression is precisely what we found in (3.33) andthe first Chern class is (a multiple of) one of the SU (2)-invariant 2-forms (3.34).Therefore, we conclude that, in the present example, e B cc is a hyperholomorphicbundle on e Y with the Chern character (3.37) which, in accordance with the criterion(3.22), is invariant under the SU (2) action on the cohomology of e Y .Now, if we wish to return to the original quantization problem, there is a simpleway to obtain the Hilbert space H associated with the quantization of M = T directly in the B -model of e Y . In the present case, only Ext e Y ( e B cc , e B ′ ) contributesto (3.3) and its dimension can be found with the help of the Grothendieck-Riemann-Roch theorem (3.4):(3.38) dim H = dim Ext e Y ( e B cc , e B ′ ) = Z e Y ch( e B cc ) ∗ ∧ ch( e B ′ ) = 1 ~ , where we used (3.32) and (3.37). In fact, as we already pointed out earlier, thedimension of H in this example was already computed in (3.11) when we discussedthe rank of e B cc .In our next example, we consider a mirror pair, Y and e Y , also of (real) dimen-sion 4, but with a non-trivial SYZ fibration (3.5). ( B, A, A ) and ( B, B, B ) branes on K . The first non-trivial exampleof a (compact) hyper-K¨ahler manifold is a K Y of dim C Y >
1, for which the homological mirrorsymmetry (3.1) is actually a theorem. In this case, the mirror manifold e Y is also a K Y and e Y .We remind that, topologically, a K b = 1, b = 22, and b = 1. Itscohomology group H ( Y, Z ) is an even unimodular lattice of signature (3 , , = ( − E ) ⊕ ( − E ) ⊕ U ⊕ U ⊕ U , where U denotes the two-dimensional even unimodular lattice U ∼ = II , with theintersection form(3.40) (cid:18) (cid:19) and E is the root lattice of the Lie algebra of the same name. To make a contactwith the SYZ approach to mirror symmetry (3.5), we choose Y (and e Y ) to be anelliptically fibered K i.e. there is a map(3.41) π : Y → B , whose general fibers are smooth elliptic curves F ∼ = T , and B ∼ = C P .In order to find the precise map between ( B, A, A ) branes on Y and ( B, B, B )branes on e Y , it is convenient to work in complex structure I (resp. e I ) where themirror map (3.1) is simply the Fourier-Mukai transform (3.19). Then, on bothsides of mirror symmetry we deal with B -branes, which can be described in termsof coherent sheaves on Y and e Y , respectively. Given a sheaf B on Y (similarly, on e Y ) we write its Chern class as a triple(3.42) (rank( B ) , c ( B ) , c ( B )) ∈ Z × NS( Y ) × Z where NS( Y ) is the N´eron-Severi lattice of Y , i.e. a sublattice of H ( Y, Z ) spannedby the cohomology classes dual to algebraic cycles of Y . As a group, NS( Y ) isisomorphic to the Picard group of Y , that is the group of algebraic equivalenceclasses of holomorphic line bundles over Y . The rank of the N´eron-Severi lattice,denoted by ρ Y varies between 0 and 20, and by the Hodge index theorem, thesignature of NS( Y ) is (1 , ρ Y − K ρ Y = 0, but forelliptic K the Picard number ρ Y is at least 2.Indeed, there are two special classes F , B ∈ NS( Y ) associated to the ellipticfiber and the section. These classes are independent and span a rank-2 sublatticein (3.39) with the intersection form (in the basis { F , B } ):(3.43) (cid:18) − (cid:19) It is easy to see that the null vectors e = F and e = F + B generate thetwo-dimensional hyperbolic lattice U = h e , e i with the intersection form (3.40).Mirror symmetry identifies this lattice with another copy of the two-dimensionalhyperbolic lattice, U ∼ = H ( e Y , Z ) ⊕ H ( e Y , Z ). Indeed, as described in (3.42) theChern classes of branes on Y and e Y take values in the lattice Z × NS × Z , which, forgeneric K U ⊕ U , and mirror symmetry acts by exchanging thetwo copies of U .Our next goal is to see more explicitly how mirror symmetry acts on particularbranes. Of course, we are especially interested in the coisotropic brane B cc , whichin the B -model of ( Y, I ) corresponds to a holomorphic line bundle
L → Y with thefirst Chern class (3.18). Clearly, this line bundle (and the brane B cc ) can only existif c ( L ) = ω I is an element in NS( Y ), in other words, only if (3.44) ω I = k B + k ′ F for a pair of integer numbers k > k ′ ≫ L , we obtain the dual bundle (sheaf) on e Y with the Cherncharacter, cf. (3.21), ch ( e B cc ) = k ch ( e B cc ) = ( k ′ − k ) k e F − e B (3.45) ch ( e B cc ) = − k ′ In general, this answer describes a higher rank object e B cc on e Y and has a structuresimilar to (3.37). Compared to (3.37), however, it has some extra “corrections” dueto non-trivial geometry of the fibration (3.41) in our present example.Now, let us take a closer look at the properties of the ( B, B, B ) brane e B cc .First, we recall that the moduli space of coherent semi-stable shaves on e Y = K M ]:(3.46) dim R M ( B ) = 2 v + 4 A generic elliptically fibered K ρ Y = 2. The Picard number canjump further to ρ Y > Y ,either if there are rational curves in the singular fibers of the fibration (3.41), or if the rank of theMordell-Weil group jumps. Here, we assume a generic situation with ρ Y = 2. UANTIZATION VIA MIRROR SYMMETRY 25 where v = v ( B ) is the charge vector of a brane B (= the Mukai vector of thecorresponding coherent sheaf):(3.47) v ( B ) := ch( B ) q Td( e Y ) = r + c + ℓ ∈ H ⊕ H ⊕ H D D D v = c − rℓ , where c is the inner product on H ( e Y , Z ) ∼ = Γ , . Applying this to the brane e B cc with the Chern character (3.45) and using (3.43), we find dim M ( e B cc ) = 0. Weexpect this to be a general feature of the mirror of the canonical coisotropic brane. Conjecture 3.1.
The brane e B cc is always rigid.Returning to the original quantization problem, now we are ready to quantizeany symplectic 2-manifold M ⊂ Y , on which ω = ω I | M is non-degenerate. Asusual, we take B ′ to be a Lagrangian brane supported on M and, to make use ofthe hyper-K¨ahler structure on Y , we choose M to be analytic in complex structure I and Lagrangian for ω J and ω K . Then, B ′ is a brane of type ( B, A, A ) supported on a holomorphic curve in the homology class M = n F F + n B B , whose genusfollows from the adjunction formula(3.49) 2 g ( M ) − M · M , and the intersection pairing (3.43). Applying the Fourier-Mukai transform (3.20), itis easy to see that the mirror (
B, B, B ) brane e B ′ is described by a hyperholomorphicsheaf on e Y with the Chern character(3.50) ch( e B ′ ) = ( n B , , − n F ) . This answer is manifestly invariant under the SU (2) action on the cohomology of e Y ,in perfect agreement with the general criterion (3.22). To make contact with thequantization of a 2-torus T considered in (2.26) and then in more detail in section3.2, we can take M = F to be a copy of the fiber. Then, the B -model approach(3.3) leads to a Hilbert space H of dimension dim H = k = ~ , in agreement withearlier results (2.10), (3.11), and (3.38) that we already rederived several times inthis paper from various angles. ~ . In the original quantization problem, ~ determinesthe norm of the symplectic form ω on the symplectic manifold M . After embeddingthe quantization problem in the A -model of Y , the closed 2-form ω and, therefore,the parameter ~ acquire a new interpretation. Namely, ω becomes (the restrictionto M ⊂ Y of) the curvature 2-form F of a unitary line bundle L → Y , the Chan-Paton bundle of B cc . Mirror symmetry maps the coisotropic brane B cc to a B -brane e B cc , and ~ determines the topology of its Chan-Paton bundle, cf. (3.37) and (3.45).Since the definition of the branes B cc and e B cc is intimately tied with the geom-etry of Y and e Y , the parameter ~ also admits a purely geometric interpretation, As in the earlier discussion, we assume that Y is a generic elliptically fibered K Y ) of rank ρ Y = 2 generated by the classes F and B . either as a symplectic structure of Y or, via mirror symmetry, as a complex struc-ture of e Y . In our toy model of section 3.2 this is easy to see from the explicitformulas (3.29) and (3.31).As illustrated in (2.1), after quantization the parameter ~ enters the definitionof various quantum objects, such as H and A ~ . In particular, when the phasespace M is compact, the Hilbert space H is finite-dimensional, and ~ determinesthe dimension of H , as in the volume integrals (2.4), (2.10) or (3.11).What happens if M is non-compact? For example, in our toy model of section3.2 we could just as well take M to be a copy of B = R (embedded in Y = B × F in an obvious way). Then, the Hilbert space H is infinite-dimensional, and theclosest to dim H is the trace,(3.51) Tr H e − βH , that one can define by introducing a Hamiltonian H and a parameter β . Classically, H is simply a function on M . According to the general principle (2.1), after quan-tization it becomes an operator on H , and the partition function (3.51) encodes thespectrum of H . Note, when M is compact, (3.51) gives the dimension of H if weset β or H to zero. Example . Quantization of M = R This problem can be realized as a special case of our toy model insection 3.2, if we take M = B . In the A -model approach, the symplecticform ω is the restriction to M ⊂ Y of the K¨ahler form F = ω I ,(3.52) ω = ω I | M = 1 ~ dx ∧ dx , where x and x are linear coordinates on B , cf. (3.17) and (3.29). Intro-ducing the Hamiltonian H = ( x + x ), we obtain a classical example ofa quantum system, namely the quantum harmonic oscillator. The eigen-values of this Hamiltonian are(3.53) H i = (cid:18) i + 12 (cid:19) ~ , i = 0 , , , . . . so that one can easily perform the sum in (3.51) to obtain the partitionfunction(3.54) Tr H e − βH = e − β ~ / − e − β ~ = 12 sinh( β ~ / . Note, that ~ appears only in a combination with β .Just as in the finite-dimensional case, the trace (3.51) is closely related to avolume integral of the form(3.55) Z M e F − βH = Z M F n n ! e − βH that, in favorable situations, one can also interpret as the “equivariant volume”of M . Indeed, if M (resp. Y ) admits a circle action, which is Hamiltonian withmoment map H : M → R , then the combination F − βH that appears in the The action of G on M is called Hamiltonian with moment map µ : M → g ∗ if d h β, µ i = − ι ( V β ) · Ω for every β ∈ g , where h , i denotes the pairing between g and g ∗ . This implies, amongother things, that the zeroes of the vector field V β are precisely the critical points of h β, µ i . UANTIZATION VIA MIRROR SYMMETRY 27 exponent of (3.55) can be interpreted as the equivariant symplectic form on M (resp. Y ), provided we identify β with the generator of the base ring(3.56) H ∗ S (pt) = H ∗ ( C P ∞ ) = C [ β ] . Indeed, since the S action is Hamiltonian and F is closed, we have ( d − ι ( V β ))( F − βH ) = 0, so that F − βH is a closed equivariant form. In fact, it is the equivari-ant first Chern class of a complex line bundle L compatible with the circle action.Therefore, the integrand of (3.55) is simply the equivariant Chern character of L .Other characteristic classes also can be extended to S -equivariant forms. For exam-ple, the Todd class — which often accompanies Chern characters in our integrationformulas — can be combined with ch( L , β ) to produce an equivariant version of theintegral in the Riemann-Roch formula (3.4),(3.57) Z M ch( L , β ) ∧ Td(
M, β ) , that, in the equivariant setting, computes the S -equivariant index of the Spin c Dirac operator /∂ L , twisted by L [ BGV ].The equivariant integrals (3.55) and (3.57) localize on the fixed points of thecircle group action ( i.e. zeroes of the corresponding vector field V ). Thus, theDuistermaat-Heckman formula asserts that the equivariant symplectic volume (3.55)can be written as a sum of local contributions of the fixed points (which, for sim-plicity, we assume to be isolated):(3.58) Z M F n n ! e − βH = X p ∈ zeroes of V e − βH ( p ) β n e ( p ) , where e ( p ) = w . . . w n is the product of the weights of the S action on T p M .Similarly, by the Atiyah-Segal-Singer equivariant index theorem [ ASS ], the S -equivariant index of the Spin c Dirac operator /∂ L can be expressed as a localizationof the integral (3.57),(3.59) index S ( /∂ L ) = X s Z F s ch( L , β ) Td( F s , β ) Q (1 − e − x i − βw i ) , where the sum runs over connected components of the fixed point set of S , and x i , i = 1 , . . . , codim F s , are the formal Chern roots of the normal bundle of F s . Example . Quantization of M = R Continuing with our previous example, we wish to study the space H of open strings between the coisotropic brane B cc on Y = B × F andthe Lagrangian A -brane B ′ supported on M = B × { pt } . Unlike thespace discussed in section 3.2, H is infinite-dimensional now, so we shallanalyze it using the equivariant technique and compare the result with(3.54). To do this, we consider the standard action of the circle group S on M = R , generated by the vector field V = x ∂ x − x ∂ x . Clearly,the origin ( x , x ) = (0 ,
0) is an isolated fixed point of this S action, and H = ( x + x ) is the Hamiltonian function for the vector field V and thesymplectic form (3.52). Now, the equivariant volume (3.55) can be easilyevaluated:(3.60) Z M e F − βH = 12 π ~ Z R e − β ( x + x ) dx dx = 1 β ~ , and the result agrees, of course, with what one could find by using thethe Duistermaat-Heckman formula (3.58). Notice, the expression (3.60)describes the β ~ → χ S ( B cc , B ′ ) = 11 − e − β ~ . Both (3.60) and (3.61) depend only on the combination β ~ .In general, the equivariant symplectic volume (3.55) describes the asymptoticbehavior (as ~ →
0) of the trace (3.51), which can be viewed as a regularized versionof dim H . This is similar to the role volume of M plays in (2.10). In order to geta better approximation to (3.51), one can consider the equivariant index (3.59) ofa Dirac operator which, roughly speaking, is a “square root” of the second orderoperator whose spectrum is described by (3.51), cf. (3.54) and (3.61) in our toymodel example. The upshot is that the equivariant cohomology of Y (resp. e Y ) fitsin well with the A -model (resp. B -model) approach to quantization and can be avery useful tool, especially when H is infinite-dimensional.
4. Quantization of Chern-Simons theory
I think I can safely say that nobody understandsquantum mechanics.
Richard FeynmanThe Hilbert space of Chern-Simons theory on a Riemann surface C is obtainedby quantizing the moduli space of flat connections [ W ]. Therefore, we take(4.1) M = M flat ( G, C ) . As explained in (2.22), this space has a natural complexification obtained by re-placing the compact Lie group G by its complexification G C .The resulting space, Y = M flat ( G C , C ) is, in fact, a hyper-K¨ahler manifold andcan be realized as the moduli space of Higgs bundles on C with structure group G (also known as the Hitchin moduli space) [ Hi ]:(4.2) Y ∼ = M H ( G, C ) . In order to approach the quantization problem of M via mirror symmetry, we firstneed to find a mirror e Y of Y . A nice fact that will be useful to us is that e Y is alsoa Hitchin moduli space, M H ( L G, C ), but for the Langlands dual group L G . In fact, Sometimes, in the literature M is “quantized” by attaching to it the virtual vector space Q ( M ) := ker /∂ L − coker /∂ L , whose equivariant character is (3.59). This space, however, should notbe confused with H . Throughout the paper we tacitly suppress one important detail: in general, the modulispace M (resp. its complexification Y ) has several connected components, which correspond togauge bundles E → C of different topology. Specifically, these connected components are labeledby an element of π ( G ) that was denoted by ξ in [ GW1 ]. Mirror symmetry maps ξ to an elementof Z ( L G ) ∼ = π ( G ) which, similarly, labels flat B -fields on e Y = M H ( L G, C ). For example, for G = SU (2) we have L G = SO (3) and there are two choices, classified by Z ( G ) ∼ = π ( L G ) ∼ = Z . If M is identified, by a theorem of Narasimhan and Seshadri, with the moduli space of (semi-)stablerank-2 bundles over C , then the two choices of ξ ∈ Z correspond to bundles of even (resp. odd)degree. In this paper we tacitly make the choice ξ = 0, which corresponds to stable G C bundlesof even degree. UANTIZATION VIA MIRROR SYMMETRY 29 the mirror manifolds M H ( G, C ) and M H ( L G, C ) fiber over the same vector space B (under the Hitchin maps), and the generic fibers are dual tori [ BJSV, HT ] (sothese two fibrations give us an example of SYZ T-duality (3.5)):(4.3) Y = M H ( G, C ) M H ( L G, C ) = e Y ց ւ B This fibration is holomorphic in complex structure I and Lagrangian with respectto both ω J and ω K , so that B and F are of type ( B, A, A ) in the terminology ofsection 3.1.There is also a version of this story for Riemann surfaces with punctures, whichhave M = M flat ( G, C ; C ) as the classical phase space. (To avoid cluttering, as insection 2 we write most of the formulas for a Riemann surface with a single punc-ture.) As described in (2.15), this moduli space has the structure of a symplectic fi-bration with the fiber C and symplectic form (2.16). Much like M = M flat ( G, C ; C ),its natural complexification Y = M flat ( G C , C ; C C ) combines (2.21) and (2.22) in asingle moduli space of flat G C connections on C , such that the holonomy of theconnection around the puncture takes values in a prescribed conjugacy class C C .Under mirror symmetry, this condition is replaced by a similar condition, but forthe Langlands dual group L G C .Namely, in this case, the mirror manifold e Y is the moduli space of semi-stableparabolic Higgs bundles on C with the structure group L G , cf. (4.3). In particular,it is a hyper-K¨ahler manifold and, in complex structure J , can be identified withthe moduli space M flat ( L G C , C ; L C C ), where L C C ⊂ L G C is a complex conjugacyclass dual to C C ,(4.4) Φ mirror : C C → L C C . This map is rather non-trivial. It preserves the dimension of conjugacy classes, aswell as some other invariants described in [
GW2 ]. Example . G = Sp (2 N )In section 2, we mentioned the minimal orbit of B N . For balance, nowlet us consider a group G C of Cartan type C N . The minimal conjugacyclass C min , i.e. the conjugacy class in G C of the smallest dimension, isthe class of a unipotent element U = exp( u ), where u ij = ν i ν j is a rank-1 symmetric matrix. It is parametrized by a vector ν , defined up to asymmetry ν → − ν , so that(4.5) C min ∼ = C N / Z . The dual conjugacy class L C min ⊂ L G C of B N is the 2 N -dimensional con-jugacy class of a semisimple element(4.6) L S = diag(+1 , − , − , . . . , − . Note, that our conventions for
I, J, K and ω I , ω J , ω K here agree with [ Hi ] and differ from[ GW ] by a cyclic rotation of three complex structures I → J → K → I . In general, the holonomy V ∈ G C can be written as V = SU , where S issemisimple, U is unipotent, and S commutes with U . The duality map (4.4) trans-forms the semisimple and unipotent data in a non-trivial way. Conjecture 4.1 ([ GW2 ]) . Let C C be a unipotent conjugacy class (or a semisimpleconjugacy class obtained by a deformation of C C ). Then, the parameter π log L S of the dual conjugacy class L C C is equal to the central character of (any) G R -representation obtained by quantizing C R ⊂ C C . B ′ . Now let us introduce branes.(For simplicity, we avoid punctures until section 5.) In the A -model approach,quantization of M is achieved by studying the space of open strings (= space ofmorphisms) between two A -branes, B ′ and B cc , defined in section 2.3. The La-grangian A -brane B ′ is supported on M ⊂ Y , while the coisotropic brane B cc issupported on all of Y and carries a non-trivial Chan-Paton line bundle L withcurvature F = Re Ω. Both of these branes turn out to be “automatically” compat-ible with the hyper-K¨ahler structure on Y , thus, providing another illustration of aphenomenon that we observed in some examples before. Namely, B ′ and B cc bothhappen to be branes of type ( B, A, A ).In the case of B ′ this follows from the fact that M = M flat ( G, C ) is a componentof the fixed point set of the involution τ : Y → Y that changes the sign of the Higgsfield [ Hi ]:(4.7) τ : ( A, φ ) ( A, − φ ) . This involution is holomorphic in complex structure I and antiholomorphic incomplex structures J and K , so that M is analytic in complex structure I andLagrangian with respect to ω J and ω K . In the conventions (3.12) - (3.13), it meansthat B ′ is not only a good A -brane in the A -model of Y with ω Y = ω K , but alsocan be viewed as a B -brane in the B -model of ( Y, I ), or else as an A -brane in the A -model of ( Y, ω J ). Using (2.25) and (3.18) one can easily verify that the canonicalcoisotropic brane B cc is also a brane of type ( B, A, A ).As pointed out in (1.3), we can take advantage of the fact that B ′ and B cc arecompatible with the hyper-K¨ahler structure on Y , and approach the problem fromthe vantage point of the complex structure I , in which B cc simply corresponds toa holomorphic line bundle L → Y with the first Chern class c ( L ) = ω I . Then, thedimension of the Hilbert space H can be computed with the help of the Hirzebruch-Riemann-Roch formula similar to (3.4),(4.8) X i ( − i dim H i ( M, L ) = Z M ch( L ) ∧ Td( M ) . The spaces H i ( M, L ) are trivial for i >
0, so that (4.8) gives(4.9) dim H = dim H ( M, L ) = Z M e ω I ∧ Td( M ) , which, for G = SU ( N ), indeed leads to the Verlinde formula (2.11). However, ouraim here is to approach the quantization of M = M flat ( G, C ) and, in particular,the calculation of dim H via mirror symmetry. Recall (from section 2.3) that the involution τ (antiholomorphic in complex structure J ,in which Ω is holomorphic) is needed for unitarity. UANTIZATION VIA MIRROR SYMMETRY 31
Mirror symmetry maps (
B, A, A ) branes B ′ and B cc on Y into ( B, B, B ) branes e B ′ and e B cc on e Y . In other words, e B ′ and e B cc are good B -branes with respectto all complex structures on the dual moduli space e Y = M H ( L G, C ). Followingour discussion in section 3.1, we expect that they correspond to hyperholomorphicbundles or sheaves on e Y , which (with a small abuse of notations) we also denoteby e B ′ and e B cc .To get an idea of what e B ′ looks like, it is instructive to start with a simpleexample of abelian gauge theory with gauge group G = U (1). In this case, theHitchin fibration (4.3) is trivial, and(4.10) Y = B × F , just like in our “toy model” considered in section 3.2. In fact, the toy model ofsection 3.2 arises as a special case of the present discussion when C is a Riemannsurface of genus g = 1. More generally, for G = U (1) we have(4.11) F = Jac( C ) , B = H ( C ; K C )where Jac( C ) is the Jacobian of C and K C is the canonical bundle. Furthermore,in this case B ′ is a Lagrangian brane supported on M = F . (As noted earlier, theHitchin fiber F is always of type ( B, A, A ), and so is the brane B ′ = B F .) Wealready discussed the mirror transform of such branes in (3.8). Indeed, dualizingthe fiber F we obtain a mirror brane(4.12) e B ′ = B p supported at a point p ∈ e Y = B × e F . Clearly, this is a brane of type ( B, B, B );in any B -model of e Y ( i.e. for any complex structure on e Y ) it corresponds to theskyscraper sheaf O p ∈ D b ( e Y ).When the gauge group G is non-abelian, the mirror brane e B ′ is also a 0-brane, ina sense that its support is a point on e Y = M H ( L G, C ), but now it has a non-trivial“inner structure.” Specifically, the (
B, B, B ) brane e B ′ is supported at the “mostsingular point” ( A, φ ) = (0 ,
0) on M H ( L G, C ), with a pole for the L g C -valued fields σ and σ that corresponds to the principal (a.k.a. regular) su (2) embedding [ W2,FG ]:(4.13) ρ princ : su (2) → L g In particular, the mirror of the Lagrangian brane B ′ supported on M = M flat ( G, C )does not admit a simple geometric description in the B -model of e Y = M H ( L G, C ),roughly speaking, because all the information about M is now clumped at the“most singular point” of e Y . In order to give a proper description of the ( B, B, B )brane e B ′ one needs either to introduce ramification (as in section 5 below) or toextend the B -model of e Y to account for the fields σ and σ . We will not attempt toformulate such a description here and, instead, focus on those ( B, B, B ) branes on e Y that can be described in the language of hyperholomorphic bundles or sheaves.In fact, we expect e B cc to be a nice example of a ( B, B, B ) brane that correspondsto a hyperholomorphic bundle on e Y = M H ( L G, C ). As for the Lagrangian brane B ′ , we can consider close cousins of the brane supported on the moduli space offlat connections, M = M flat ( G, C ). Namely, we can take B ′ to be a Lagrangianbrane supported on another component of the fixed point set of the involution (4.7).Clearly, all such branes are automatically of type ( B, A, A ), and some of them even admit a nice geometric interpretation as branes supported on M = M flat ( G R , C ),for other real forms G R of the complex group G C [ Hi ]. As reviewed in (2.7) - (2.8),these moduli spaces provide an excellent laboratory for the quantization problem,with many applications to gauge theory. ( B, A, A ) brane B cc . Now let us consider themirror transform of the canonical coisotropic (
B, A, A ) brane B cc with a Chan-Patonbundle of curvature F = ω I . Among all coisotropic branes on Y ∼ = M H ( G, C ), thisbrane has a number of special properties. In order to describe them in detail, letus introduce a complexified K¨ahler form(4.14) ω I + iB = 1 ~ ω ∗ where ω ∗ is the image in de Rham cohomology of a generator of H ( Y, Z ) ∼ = Z .One peculiar property of the ( B, A, A ) brane B cc is that it exists (with B = 0) onlyfor discrete values of ~ :(4.15) ~ = 1 k , where k ∈ Z is the “level.” This agrees well with the fact (2.9) that M = M flat ( G, C ) should be quantizable precisely for these values of ~ . Mirror symmetry(4.3) maps B cc to e B cc and acts on the parameter ~ as (4.16) ~ → L ~ = − ~ where, for simplicity, we assumed that g is simply laced. It follows that, just like B cc , its mirror e B cc exists only for a discrete set of values of ~ , namely(4.17) L ~ = − k . How and why this happens can be seen already in a basic example of abelian gaugetheory (which, in the special case of g = 1, was discussed in detail in section 3.2, andhas a straightforward generalization to arbitrary genus, based on (4.10) - (4.11)).To learn more about the brane e B cc and about the B -model approach to quan-tization of M = M flat ( G, C ) we can examine the problem from the viewpoint ofcomplex structure I . As explained in section 3.1, in the B -model of ( Y, I ) the orig-inal brane B cc corresponds to a complex line bundle L → Y with the first Chernclass (3.18), i.e. with the Chern character(4.18) ch( B cc ) = exp( ω I ) . Moreover, in complex structure I mirror symmetry (3.1) is simply the Fourier-Mukai transform (3.19), and (3.21) gives the Chern character of the dual brane e B cc .Although in general e B cc turns out to be a higher rank ( B, B, B ) brane, we expectthat it corresponds to an ordinary hyperholomorphic bundle on e Y (as opposed toa more complicated object in D b ( e Y )) with the Chern character ch( e B cc ).For concreteness, let us take G = SU (2). (We will try, however, to focus ongeneral properties of B cc that will have a clear analog for other groups.) Then, evenwithout getting too much into details of the geometry of M H ( L G, C ), we can say The setup of the present section arises in topological gauge theory on a 4-manifold R × C .In that context, the complex parameter ~ is identified with the coupling constant of the four-dimensional gauge theory [ KW ]. UANTIZATION VIA MIRROR SYMMETRY 33 that the answer for ch( e B cc ) must have the following structure, familiar from (3.37)and (3.45):(4.19) ch( e B cc ) = 2 g ~ g − + . . . − B . Here, the first term (of degree 0) is the well-known expression [
BNR ] for the volumeof the Hitchin fiber, Vol( F ) = R F e ω I , which according to (3.6) determines the rankof the mirror of the coisotropic brane B cc . The last term (to be discussed shortly)does not depend on ~ , while the remaining terms (denoted by ellipsis) all appearwith negative powers of ~ . In particular, all the terms in (4.19), except for the lastone, vanish in the “extreme quantum” limit:(4.20) ~ → ∞ . There is a simple explanation for this. Indeed, in the limit (4.20) the curvature F = ω I of the Chan-Paton bundle of B cc goes to zero, and B cc becomes a rank-1brane supported on all of Y with a trivial Chan-Paton bundle. As an object ofthe derived category D b ( Y, I ) this limiting brane is simply the structure sheaf O Y .Even though it preserves different supersymmetry, ( B, B, B ) instead of (
B, A, A ),the brane B = O Y can help us to understand what happens to the leading term inch( B cc ) = 1 + O ( ~ − ) under the mirror map. Indeed, the mirror of B = O Y is a( B, A, A ) brane e B on e Y supported on a section of the dual Hitchin fibration (4.3),whose homology class accounts for the last term in (4.19).If we naively try to compute the dimension of the Hilbert space H from thepartial answer (4.19), as we did e.g. in (3.38), we obtain an expression that lookslike(4.21) dim H = vol( F ) · k g − + . . . + 1 , where the first (resp. last) term comes from the corresponding term in (4.19). Tomake the meaning of these terms more transparent, we used (4.15) to replace ~ by k , and wrote vol( F ) for the volume of the fiber F with respect to the normalizedsymplectic form ω ∗ introduced in (4.14). Even though our derivation of (4.21) wassomewhat heuristic since we treated e B ′ as an ordinary zero-brane (4.12) ignoringthe pole (4.13), the answer (4.21) does capture correctly certain aspects of theVerlinde formula.For example, it is clear that (4.21) is a polynomial in k (because ch( e B cc ) isa polynomial in ~ − ). Furthermore, the last term in (4.21) is the constant termof this polynomial. This follows from the fact that the last term of (4.19) is theonly term in ch( e B cc ) independent of ~ . This agrees well with the behavior of theVerlinde formula in the extreme quantum limit (4.20):(4.22) dim H k → −→ , valid for any genus g , cf. (2.12) for g = 2. However, as that genus-2 example alsoillustrates, even though the polynomial (4.21) has the correct degree, the coefficientof the leading term (for large k ) is not what we expect it to be. Indeed, accordingto (2.10), the (coefficient of the) leading term in the Verlinde formula should bethe volume of M = M flat ( G, C ), not the volume of F . This is also clear from theviewpoint (4.9) of the B -model of ( Y, I ).The volumes of M and F are quite different. For example, for G = SU (2)the normalized volume of M , computed with respect to the symplectic form ω ∗ , is genus vol( M ) vol( F ) g = 2 g = 3 g = 4 g = 5 . . . . . . . . . Table 1.
Listed here are the volumes of the moduli space M and the Hitchin fiber F for G = SU (2) and small values of g , computed with respect to the symplectic form ω ∗ . given by the following formula [ W1 ]:(4.23) vol( M ) = 2 · ζ (2 g − π ) g − which is quite different from a (much simpler) expression for the volume of F . Thefirst few values of (4.23) are listed in Table 1.The fact that (4.21) correctly captures the constant term of the Verlinde for-mula, and not the top-degree term, should not be surprising. After all, (4.19)correctly captures the part of ch( e B cc ) that does not depend on ~ , whereas there areseveral terms in ch( e B cc ) — suppressed in (4.19) — that contribute to the leadingbehavior of dim H ∼ k g − . The number of such terms grows quickly with thegenus g of the Riemann surface C .
5. The Verlinde formula via mirror symmetry
The more success the quantum theory hasthe sillier it looks.
Albert EinsteinNow, let us analyze the space H of section 4 and its dimension more carefully.In particular, we wish to see how various terms in the Verlinde formula (1.5) arise inthe B -model approach, where the target space is the moduli space of Higgs bundleson C with the structure group L G .To keep our discussion concrete and simple at the same time, we take G = SU (2) and focus on a specific example, to which we secretely prepared ourselves insection 3. Namely, we take C to be a torus with a single puncture, around whichthe gauge field has a holonomy (2.17) labeled by a weight λ . Then, much like inexamples considered in section 3, we have dim R M = dim C Y = 2 and the Verlindeformula (2.18) gives:(5.1) dim H = k − λ + 1for even values of λ and sufficiently large k . In other words, in this case the Verlindeformula is a simple polynomial of degree n = 1 with only two non-trivial coefficients, a and a in the notations of (1.5). Nevertheless, understanding these coefficientsvia branes will be an illuminating and enjoyable exercise.First, let us summarize the relevant geometry of the space M = M flat ( G, C ),its complexification Y = M flat ( G C , C ), and the mirror e Y = M flat ( L G C , C ), cf. UANTIZATION VIA MIRROR SYMMETRY 35 (2.22) and (4.3). Since the fundamental group of a 2-torus with one puncture is afree group of rank two:(5.2) π ( C ) = h a, b, c | aba − b − = c i , its SU (2) and SL (2 , C ) character varieties M and Y admit a very simple and explicitdescription (2.14). For instance, the space Y of homomorphisms ρ : π ( C ) → SL (2 , C ) with a suitable boundary condition (2.17) at the puncture can be describedrather explicitly as an affine hypersurface in C with coordinates x = tr( ρ ( a )) y = tr( ρ ( b )) z = tr( ρ ( ab ))defined by(5.3) Y : x + y + z + xyz = tr V + 2 . For a surface Y defined by the zero locus of a polynomial f ( x, y, z ) the holomorphicform (3.12) can be written as(5.4) Ω = 14 π ~ dx ∧ dy∂f /∂z = 14 π ~ dx ∧ dyxy + 2 z . When tr V = 2 ( i.e. α = 0), we obtain a cubic surface with four simple singularitiesof type A (double points) at(5.5) ( − , − , − , ( − , , , (2 , , − , and (2 , − , . This singular surface, called the Cayley cubic, is simply a Z quotient of C ∗ × C ∗ :(5.6) Y = ( C ∗ × C ∗ ) / Z . A more direct way to see this is to note that, for the special value of the holonomyparameter α = 0, we have V = and the defining equation (2.14) reduces to thatof a 2-torus without punctures, cf. (2.7). On the other hand, since the fundamentalgroup of a torus is abelian, π ( T ) = Z × Z , the holonomies of the complexified gaugeconnection A = A + iφ around the A - and B -cycles of T can be simultaneouslyconjugated to a maximal torus T C ⊂ G C . Hence, M flat ( G C , T ) = ( T C × T C ) / W where W is the Weyl group. In the present case, this gives (5.6) because T C = C ∗ and W = Z .Now, it is clear that, for α = 0, the space Y in the present example is simplya Z quotient of that in section 3.2:(5.7) Y = ( T × R ) / Z . Moreover, the real slice M = T / Z , sometimes called the “pillow case,” is themoduli space of flat SU (2) connections on the (punctured) torus. Turning on theholonomy parameter α removes the four Z singularities and deforms (5.7) into asmooth complex surface (5.3). This is the viewpoint of complex structure J , inwhich Y is identified with the moduli space of flat SL (2 , C ) connections on C .Since we are interested in branes of type ( B, A, A ) let us consider what happensin complex structure I , in which M ⊂ Y is holomorphic and Y is naturally identifiedwith the moduli space of semi-stable parabolic Higgs bundles on C , cf. (4.2).In complex structure I , the parameter α is a K¨ahler structure parameter. For α = 0 the four C / Z orbifold singularities of Y are resolved, and we denote by D i , i = 1 , . . . ,
4, the corresponding exceptional divisors. As a result, Y has homology(for generic values of α ):(5.8) H ( Y ) ∼ = Z . One can also deform the C / Z singularities by turning on a complex structureparameter β + iγ , which in the present example corresponds to introducing a polefor the Higgs field φ at the puncture p ∈ C . This leads to a closely related model,recently considered in [ FW ]. Since such complex structure deformations createexceptional cycles which are not holomorphic in complex structure I , we shallmainly focus on the situation with α = 0 and β = γ = 0.In complex structure I , the surface Y has the structure of the elliptic fibra-tion (4.3). Indeed, if z and w are complex coordinates on T and R in (5.7), thenthere is a map π : Y → B , sending ( z, w ) b := w and exceptional divisorsto zero. The generic fibers of this map are F ∼ = T and the only singular fiber isthe “nilpotent cone” N := π − (0), which in the present case has five irreduciblecomponents (all rational):(5.9) N = M ∪ [ i =1 D i . The homology classes of M and D i are independent and generate H ( Y ). A quicklook at the intersection numbers shows that Y is indeed an elliptic fibration withone singular fiber over b = 0 of Kodaira type I ∗ , i.e. with the intersection form e D (in the basis { M, D , . . . , D } ):(5.10) − − − − − This intersection matrix has only one null vector, which therefore must be identifiedwith the class of the elliptic fiber,(5.11) [ F ] = 2[ M ] + X i =1 [ D i ] , and implies the following relation among the volumes:(5.12) Vol( F ) = 2Vol( M ) + X i =1 Vol( D i ) . Indeed, when α = 0 we have Vol( D i ) = 0 and this equation simply expresses thefact that F is a double cover of M , which is clear from the Z quotient (5.7). Apartfrom this multiplicity factor (due to a singularity), the relation (5.11) is the familiarstatement that different fibers of π : Y → B are homologous. In general, the hyper-K¨ahler metric on Y depends on a triple of “moment maps” ( α, β, γ ),such that for generic values of these parameters the exceptional cycles are holomorphic in complexstructure I = αI + βJ + γK p α + β + γ . UANTIZATION VIA MIRROR SYMMETRY 37
At this stage, we have everything we need to verify the Verlinde formula (5.1).In the A -model of Y , B ′ is a Lagrangian brane supported on M ⊂ Y and B cc is acoisotropic brane with a Chan-Paton bundle L of curvature F = ω I . Away formthe singular fiber over b = 0, the brane B cc (resp. its dual e B cc ) is essentially thesame as the one considered in section 3.2. (For simplicity, one can keep in mindthe special case α = 0, for which Y is given by the Z quotient (5.7).) The onlyimportant effect of the Z quotient is that F is a double cover of M and ~ = k ,so that(5.13) Vol( F ) = Z F ω I = 2 k , This relation actually holds for all values of α , as can be easily verified by a directevaluation of the period integral of the holomorphic 2-form (5.4):(5.14) Z F Ω = 1(2 πi ) ~ Z Z Z | x | = | y | = | z | =1 dx ∧ dy ∧ dzf ( x, y, z ) = 1 ~ . Similarly, we find (5.15) Z M Ω = 1 ~ (cid:18) − α (cid:19) = k − λ , where α = λ k was used in the last equality, cf. (2.17). Then, this gives us thecorrect answer for the dimension of H , consistent with the Verlinde formula (5.1),(5.16) dim H = Vol( M ) + 1 = k − λ + 1 , where we also used (4.9) and Td( M ) = e c ( M ) / b A ( M ).Note, according to the general formula (3.6), the volume of the elliptic fiber(5.13) determines the rank of the mirror of the coisotropic brane B cc ,(5.17) rank( e B cc ) = 2 k . This is our first hint that the mirror of the Lagrangian brane B ′ should be a“fractional brane.” Indeed, if e B ′ were a regular zero-brane on e Y represented bya skyscraper sheaf O p ∈ D b ( e Y ), as in (4.12), it would contribute 2 k (instead of k )to the dimension of H , cf. (3.38). Therefore, we expect that e B ′ should be roughlya “half” of the ordinary zero-brane B p supported at a generic point p ∈ e Y . As weshall see below, this is indeed the case.Before we proceed, let us remark that the relations (5.9) - (5.12) have analogsfor more general moduli spaces of (parabolic) Higgs bundles. For example, for SU (2) Higgs bundles on a Riemann surface of genus g the nilpotent cone has g irreducible components, each of complex dimension 3 g − Hi, T ]:(5.18) N = M ∪ g − [ i =1 D i . Here, M is the classical phase space of SU (2) Chern-Simons gauge theory on C ,and each D i is the locus of those stable Higgs bundles E = E φ → E ⊗ K C whichhave a unique subbundle L i of degree (1 − i ) killed by the non-zero Higgs field φ .Moreover, in this case, the middle dimensional homology H g − ( Y ) has dimension g and is freely generated by the homology classes of irreducible components of the Notice, when α = 0 we have Vol( F ) = 2Vol( M ), in agreement with (5.12). nilpotent cone (5.18). Similarly, the analog of (5.11) is the following relation (dueto T. Hausel),(5.19) [ F ] = X i dim( F i ) [ D i ] , where F i are the connected components of the fixed point set of the circle action( A, φ ) → ( A, e iξ φ ), and D i are the corresponding components of the nilpotent cone(with D ≡ M ), see [ Hi, T, Ha, HT ] for further details.Returning to our basic example of a genus 1 curve C with one puncture, let usconsider the B -model of the mirror variety e Y which, according to (4.3), we identifywith the moduli space of parabolic Higgs bundles for the Langlands dual group L G = SO (3). Much like Y itself, its mirror e Y is an elliptic fibration e π : e Y → B ,with generic fibers e F b = H ( F b , U (1)) ∼ = T and one singular fiber over b = 0.Since in general mirror symmetry for Calabi-Yau 2-folds is believed to preserve theKodaira type of singular fibers, we expect that e Y also has a singular fiber of type I ∗ over b = 0. (Of course, the singularities of e Y may be only partially resolvedsince mirror symmetry exchanges complex and symplectic structures.) In order toshow that this is indeed a correct guess, in the present example it is convenient toconstruct the moduli space e Y = M H ( L G, C ) as a quotient of Y = M H ( G, C ),(5.20) e Y ∼ = Y / Ξ , which follows from the well-known isomorphism SO (3) ∼ = SU (2) / Z . Here, Ξ = Z × Z is the “group of sign changes” generated by twists of the underlying gaugebundle E → C by line bundles of order 2. The elements of this group act on the SL (2 , C ) character variety (5.3) by reflections ( x, y, z ) ( ± x, ± y, ± z ) with an evennumber of minus signs, see e.g. [ Go ]. The resulting quotient e Y = Y /
Ξ is an ellipticsurface with three C / Z orbifold singularities located at ( x, y, z ) = ( √ V , , x , y , and z . All of these pointslie on the singular fiber of e π : e Y → B , namely on the zero fiber e N = e π − (0).In complex structure e J , the singular surface e Y can be represented as a zerolocus of a cubic in C , similar to (5.3),(5.21) e Y : x + y + z + xyz = 2 a ( x + y + z ) + (4 − a − a ) , where a = 2 − tr V . Branes on this particular family of singular cubic surfaceswere studied in a closely related context in [ Gu ]. In the new coordinates, theorbifold singularities of e Y are located at ( x, y, z ) = ( − tr V, ,
2) and two otherpoints obtained by permutations of x , y , and z . Notice, when tr V = − A collide and, in fact, e Y develops a worse singularityof type D at ( x, y, z ) = (2 , , α = 0 ( i.e. a = 0) the mirror geometry (5.21) developsthe fourth A singularity at ( x, y, z ) = ( − , − , − α both Y and e Y take the form of the Cayley cubic (5.6), with possible values of the B -field equal to 0 or in the direction of each exceptional divisor [ A ]. Which valuesare realized in our problem can be easily determined via the connection with [ Gu ],where the same sigma-model played an important role in the gauge theory approachto knot homologies. Thus, in order to understand the basic operations in knottheory (the skein relations) one needs to study the special case of a four-puncturedsphere C P \ { p , p , p , p } , and the family of cubic surfaces (5.21) is precisely UANTIZATION VIA MIRROR SYMMETRY 39 A A D * α ~ Y Figure 1.
Singularities of e Y : a ) four simple singularities of type A when α = 0, b ) one singularity of type D when α = , and c )three A singularities for all other values 0 < α < .the moduli space of flat SL (2 , C ) connections on a four-punctured sphere withholonomies V i , such that tr V i = a for all i = 1 , . . . , SL (2 , C ) connections on C P \ { p , p , p , p } with holonomy parameters(5.22) α = α −
12 and α = α = α = α , where we labeled each holonomy V i by a parameter α i as in (2.17). This fact canbe used to determine the value of the B -field in the mirror B -model of e Y . Indeed,the duality maps each holonomy parameter α i to the “quantum” parameter η i associated with the i -th puncture [ GW1 ],(5.23) Φ mirror : α i → η i . In the mirror B -model with the target space e Y , the quantum parameters η i describethe flux of the B -field through the corresponding 2-cycles e D i . This, together with(5.22), determines the value of the B -field: B =: X i =1 η i e D i (5.24) = (cid:18) α − (cid:19) e D + α e D + α e D + α e D . The non-trivial B -field has an important effect in the B -model of e Y . In particular,one should remember that, in the presence of a B -field, the Chern character ch( e B )always appears in a gauge invariant combination e − B ch( e B ) and the charge vectorof a brane e B is given by the modified Mukai vector,(5.25) v ( B ) = e − B ch( B ) q Td( e Y ) , instead of (3.47).As the parameter α gradually varies from 0 to , the geometry of e Y interpolatesbetween the two extreme cases depicted in Figure 1, so that three A singularities remain unresolved. With our choice of conventions, the corresponding exceptionaldivisors are e D , e D , and e D . In other words, the 2-cycles e D , e D , and e D havezero volume with respect to all K¨ahler forms on e Y , whereas Vol( f M ) and Vol( e D )vary with α in such a way that their linear combination (5.12) gives the volume ofthe elliptic fiber,(5.26) Vol( e F ) = 2Vol( f M ) + Vol( e D ) , and remains constant (independent of α ). Indeed, for the cubic surface (5.21) withreal values of a , only e ω I has non-zero periods over these 2-cycles, and an easycomputation analogous to (5.14) - (5.15) shows that, besides (5.26), the periodsobey the following relation,(5.27) Z e D e ω I = 2 α Z e F e ω I , which will be useful to us below. Note, in particular, that at α = 0 we haveVol( e D ) = 0, whereas at α = the volume of f M vanishes.Now, let us discuss ( B, B, B ) branes on e Y , in particular, the branes e B ′ and e B cc which are of major importance in the B -model approach to quantization of M = M flat ( G, C ). Starting with (3.8), by now we encountered several times oneparticular (
B, B, B ) brane, namely a regular zero-brane B p , which is dual to a( B, A, A ) brane B F supported on a generic fiber of π : Y → B .In addition, the category of B -branes on e Y contains “fractional” zero-branessupported at the orbifold singularities of e Y . Specifically, the spectrum of branes atthe Kleinian quotient singularity C / Γ by a finite group Γ ⊂ SL (2 , C ) is describedby D b Γ ( C ), and fractional branes correspond to the simple objects of this category:(5.28) B i = ̺ i ⊗ O p . Here, ̺ i are irreducible representations of Γ and O p is the skyscraper sheaf sup-ported at the origin of C . The category of fractional branes is equipped with anaction of the tensor category Rep(Γ). For example, if Γ = Z , as in our model with0 ≤ α < , then at every orbifold point there are two fractional branes B + and B − of charge v ( B ± ) = (0 , ± , ), permuted by the sign representation of Γ = Z andleft invariant by the action of the trivial representation of Γ = Z , cf. [ DM, FW ].Note, in this case, each fractional brane carries only a half of the zero-brane charge v ( B p ) = (0 , , v ( B + ) + v ( B − ) = v ( B p ) . More generally, in the equivariant category D b Γ ( C ), the zero-brane B p correspondsto ̺ reg ⊗ O p , where ̺ reg is the regular representation of Γ. The representation ̺ reg is reducible and, according to a fundamental theorem of finite group theory,decomposes as ̺ reg = ⊕ i d i ̺ i , where d i = dim( ̺ i ) and dim( ̺ reg ) = | Γ | . Therefore,in terms of the fractional branes (5.28), we have(5.30) B p = M ̺ i ∈ Irr(Γ) d i B i . This provides us with a good supply of (
B, B, B ) branes localized at the orbifoldsingularities of e Y . Our conventions are such that ̺ always denotes the trivial representation. UANTIZATION VIA MIRROR SYMMETRY 41
In order to describe (
B, B, B ) branes on e Y for generic values of α , one needs tounderstand what happens to the fractional branes B i under the minimal resolutionof the Kleinian quotient singularity C / Γ. The answer comes from the followingequivalence (the derived McKay correspondence)(5.31) D b ( X ) ∼ = D b Γ ( C ) , where X denotes the minimal resolution. According to [ KV ], in the derived cate-gory of X , the simple objects (5.28) are represented by B = O P d i D i , (5.32) B i = O D i ( − , i = 1 , where D i are the exceptional divisors. In particular, in the derived category of X it is easy to see that each fractional brane B i is a spherical object, i.e. (5.33) Ext ∗ X ( B i , B i ) ∼ = H ∗ ( C P , C ) . This gives yet another reason to identify the fractional branes on e Y with the dualsof Lagrangian A -branes supported on irreducible components of the singular fiber(5.9), since each component is a copy of C P , cf. [ FW ]. (Remember, the first hintcame from (5.17), which was then further supported by (5.29) and the fact that e Y has orbifold singularities.)In order to identify the mirror ( B, B, B ) branes more carefully, it is convenientto start at α = . As we explained earlier, at this special value of α the hypersurface(5.21) develops a singularity of type D which, luckily, is also a quotient singularity C / Γ by the binary dihedral group Γ = BD , whose action on C is generated bythe two elements,(5.34) γ = (cid:18) ξ ξ − (cid:19) and γ = (cid:18) − (cid:19) , with ξ = exp( πi/ BD has one 2-dimensional irreducible rep-resentation ̺ and four 1-dimensional irreducible representations ̺ i , i = 1 , . . . , D singularity on e Y , thezero-brane B p = Φ mirror ( B F ) is reducible and decomposes as(5.35) B p = 2 B ⊕ M i =1 B i . Comparing this to (5.11), one is led to identify B with e B ′ and B i , i = 1 , . . . , B, A, A ) branes supported on the exceptionaldivisors D i ⊂ Y .Indeed, since all of these ( B, A, A ) branes are supported on irreducible compo-nents of the singular fiber (5.9) over b = 0, we expect the mirror ( B, B, B ) branesto be also supported on various components of the singular fiber of e Y . In par-ticular, their Chern characters must be linear combinations of the Poincar´e dualsof the 2-cycles f M , e D , . . . , e D and, of course, the class of a point ch( B p ) = − p . The homology classes of f M and e D i generate H ( e Y ) ∼ = Z with the intersection form (5.10)and obey an analog of the relation (5.11). Specifically, from (5.32) we find(5.36) ch( e B ′ ) = 12 (cid:16) − e F + e D + e D + e D + e D (cid:17) According to (5.24) and (5.25), as a function of α the 0-brane charge of the brane e B ′ = B is equal to − ( η + η + η + η ) = − α . This fact plays an important role inthe application to the Verlinde formula. Indeed, together with (5.17), it determinesthe leading contribution to the dimension of H = Ext ∗ e Y ( e B cc , e B ′ ) which, accordingto (3.4), is given by(5.37) dim H = Z e Y ch( e B cc ) ∗ ∧ ch( e B ′ ) ∧ Td( e Y ) = k − λ + 1and matches exactly (5.1) if for e B cc we take the sheaf on e Y that descends from thehyperholomorphic sheaf on C ∗ × C ∗ described in section 3.2.Indeed, in section 3.2 we discussed a hyperholomorphic sheaf on C ∗ × C ∗ withthe Chern character (3.37), which is invariant under the Z action. Hence, it canbe thought of as a Z -equivariant coherent sheaf with the trivial Z -equivariantstructure that defines a coherent sheaf e B cc on e Y via the functor(5.38) Φ : D b Z ( C ∗ × C ∗ ) ∼ −→ D b ( e Y ) , which is an equivalence [ BKR ] between the bounded derived category of coherentsheaves on e Y and the bounded derived category of Z -equivariant coherent sheaveson C ∗ × C ∗ . To be more specific, the functor (5.38) is obtained by considering thefollowing commutative diagram(5.39) Z p −→ C ∗ × C ∗ κ y y κ ′ e Y p ′ −→ ( C ∗ × C ∗ ) / Z in which p and p ′ are birational, κ and κ ′ are finite of degree 2, and κ is flat (see[ BKR ] for further details and examples).Finally, we note that, with little extra work, one can extend the analysis in thepresent section to reproduce the Verlinde formula for the four-punctured sphere.In that case, the mirror manifolds Y and e Y are also cubic surfaces, similar to (5.3)and (5.21), which in general depend on four holonomy parameters α , α , α , and α (that we already found useful in our discussion here). They exhibit a moreelaborate structure of singularities ( cf. Figure 1) and a rich spectrum of branesthat account for the intricate structure of the Verlinde formula (2.18).
Acknowledgments
I would like to thank the organizing committee of the TakagiLectures for inviting me, and to acknowledge helpful discussions with E. Frenkel,T. Hausel, and E. Witten. I also would like to thank E. Witten for collaborationon the A -model approach to quantization reviewed in section 2. This work issupported in part by DOE Grant DE-FG03-92-ER40701 and in part by NSF GrantPHY-0757647. Opinions and conclusions expressed here are those of the authorand do not necessarily reflect the views of funding agencies. The subleading constant term was already discussed in (4.21).
UANTIZATION VIA MIRROR SYMMETRY 43
References [AZ] M. Aldi and E. Zaslow,
Coisotropic Branes, Noncommutativity, and the Mirror Correspon-dence , JHEP (2005) 019, arXiv:hep-th/0501247.[A] P. Aspinwall,
Enhanced gauge symmetries and K3 surfaces , Phys. Lett.
B357 (1995) 329–334,arXiv:hep-th/9507012.[AB] M. Atiyah and R. Bott,
The Yang-Mills equations over Riemann surfaces , Phil. Trans. R.Soc. London A (1982) 523–615.[ASS] M. Atiyah and I. Singer,
The index of elliptic operators I , Ann. Math. (1968) 484–530;M. Atiyah and G. Segal, The index of elliptic operators II , Ann. Math. (1968) 531–545;M. Atiyah and I. Singer The index of elliptic operators III , Ann. Math. (1968) 546–609.[BFFLS] F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz and D. Sternheimer, DeformationTheory and Quantization. I. Deformations of Symplectic Structures , Ann. Physics (1978)61–110.[B] A. Beauville,
Vector bundles on curves and generalized theta functions: recent results andopen problems , MSRI Publications (1995) 17–33, arXiv:alg-geom/9404001.[BNR] A. Beauville, M.S. Narasimhan and S. Ramanan, Spectral curves and the generalised thetadivisor , J. Reine Angew. Math. (1989) 169–179.[BGV] N. Berline, E. Getzler and M. Vergne,
Heat kernels and Dirac operators , Springer-Verlag,Berlin-Heidelber-New York, 1992.[BJSV] M. Bershadsky, A. Johansen, V. Sadov and C. Vafa,
Topological reduction of 4-d SYM to2-d sigma models , Nucl. Phys. B (1995) 166, arXiv:hep-th/9501096.[BKR] T. Bridgeland, A. King, M. Reid,
The McKay correspondence as an equivalence of derivedcategories , J. Amer. Math. Soc. (2001) 535–554.[C] E. Calabi, M´etriques K¨ahl´eriennes et fibr´es holomorphes , Ann. Scient. ´Ecole Norm. Sup., (1979) 269–294.[CV] S. Cecotti and C. Vafa, Topological antitopological fusion , Nucl. Phys.
B367 (1991) 359–461.[CGLP] M. Cvetic, G. W. Gibbons, H. Lu and C. N. Pope,
Hyper-Kaehler Calabi metrics, L harmonic forms, resolved M2-branes, and AdS (4) /CF T (3) correspondence , Nucl. Phys.
B617 (2001) 151, arXiv:hep-th/0102185.[DM] M. Douglas and G. Moore,
D-branes, Quivers, and ALE Instantons , arXiv:hep-th/9603167.[D] B. Dubrovin,
Integrable systems in topological field theory , Nucl. Phys.
B379 (1992) 627–689.[FW] E. Frenkel and E. Witten,
Geometric Endoscopy and Mirror Symmetry , Communicationsin Number Theory and Physics, (2008) 113–283 (arXiv:0710.5939).[FG] E. Frenkel and S. Gukov, S-duality of branes and geometric Langlands correspondence , toappear.[Go] W. Goldman,
The modular group action on real SL (2) -characters of a one-holed torus ,Geom. Topol. (2003) 443–486.[Gu] S. Gukov, Surface Operators and Knot Homologies , proceedings of the International Con-gress on Mathematical Physics 2006, Fortsch. Phys. (2007) 473–490, arXiv:0706.2369.[GW] S. Gukov and E. Witten, D-branes and Quantization , Adv. Theor. Math. Phys. (2009)1–73, arXiv:0809.0305.[GW1] S. Gukov and E. Witten, Gauge theory, ramification, and the geometric Langlands pro-gram , Current Developments in Mathematics (2008) 35–180.[GW2] S. Gukov and E. Witten,
Rigid Surface Operators , Adv. Theor. Math. Phys. (2010)87–177, arXiv:0804.1561.[Ha] T. Hausel, Geometry of the moduli space of Higgs bundles , PhD thesis (1998),arXiv:math/0107040.[HT] T. Hausel and M. Thaddeus,
Mirror symmetry, Langlands duality, and the Hitchin system ,Invent. Math. (2003) 197–229, arXiv:math/0205236.[Hi] N. Hitchin,
The Self-Duality Equations On A Riemann Surface , Proc. London Math. Soc.(3) (1987) 59–126.[KV] M. Kapranov and E. Vasserot, Kleinian singularities, derived categories and Hall algebras ,Math. Ann. (2000) 565–576, arXiv:math/9812016.[KO] A. Kapustin and D. Orlov,
Remarks On A -Branes, Mirror Symmetry, And The FukayaCategory , J. Geom. Phys. (2003) 84–99, arXiv:hep-th/019098.[KW] A. Kapustin and E. Witten, Electric-magnetic Duality And The Geometric Langlands Pro-gram , Communications in Number Theory and Physics (2007) 1–236, arXiv:hep-th/0604151. [K] M. Kontsevich, Homological algebra of mirror symmetry , Proceedings of ICM, Zurich 1994,vol. I, Birkh¨auser (1995) 120–139, alg-geom/9411018.[K1] B. Kostant,
Quantization and unitary representations. I. Prequantization , Lecture Notes inMath. , Springer, Berlin (1970) 87–208.[K2] B. Kostant,
Line Bundles And The Prequantized Schrodinger Equation , Coll. Group Theo-retical Methods in Physics (Marseilles, 1972) 81.[M] S. Mukai,
Symplectic structure of the moduli of sheaves on an abelian or K3 surface , Invent.Math. (1984) 101; On the moduli space of bundles on K3 surfaces, I , in “Vector Bundleson Algebraic Varieties” Tata Inst. of Fund. Research.[S] J.-M. Souriau,
Quantification Geometrique , Commun. Math. Phys. (1966) 374.[SYZ] A. Strominger, S.-T. Yau and E. Zaslow, Mirror Symmetry Is T –duality , Nucl. Phys. B479 (1996) 243–259.[T] M. Thaddeus,
Topology of the moduli space of stable bundles over a compact Riemann surface ,1990.[V] E. Verlinde,
Fusion rules and modular transformations in 2d conformal field theory , Nucl.Phys.
B300 (1988) 360–376.[Ve] M. Verbitsky,
Hyperholomorphic sheaves and new examples of hyperk¨ahler manifolds , Math-ematical Physics , International Press, 1999, arXiv:alg-geom/9712012.[Vo] D. A. Vogan, Jr., Singular unitary representations , Lecture Notes in Math. , Springer,Berlin (1981) 506–535.[W] E. Witten,
Quantum Field Theory and the Jones polynomial , Commun. Math. Phys. (1989) 351.[W1] E. Witten,
On quantum gauge theories in two-dimensions , Commun. Math. Phys. (1991) 153–209.[W2] E. Witten,
Geometric Langlands And The Equations Of Nahm And Bogomolny ,arXiv:0905.4795.[W3] E. Witten,
A New Look At The Path Integral Of Quantum Mechanics , arXiv:1009.6032.
California Institute of Technology, Pasadena, CA 91125, USA,Max-Planck-Institut f¨ur Mathematik, Vivatsgasse 7, D-53111 Bonn, Germany.
E-mail address ::