• Tidak ada hasil yang ditemukan

Compactification To Two Dimensions

Dalam dokumen And The Geometric Langlands Program (Halaman 110-116)

7. Fluxes and S-Duality

7.2. Compactification To Two Dimensions

We can similarly decompose the character e that is dual to a. Clearly, we have e = e0 +e1, where e0 is a character of a0 ∈ Z and e1 is a character of a1 ∈ H1(C,Z).

Hence, using the selfduality of Z and Poincar´e duality, e0 ∈ Z=Z

e1 ∈H1(C,Z)=H1(C,Z). (7.23) In view of (7.9), the transformation under S-duality is

(e0,m0)→(−m0,e0)

(e1,m1)→(−m1,e1). (7.24) Our next goal is to interprete0,e1,m0, andm1in the effective two-dimensional sigma- model with target MH(Gad). We will consider both the case of a closed Riemann surface Σ, and the case when Σ has a boundary. The latter case will enable us to understand the implications of (7.24) for the geometric Langlands program.

Interpretation Of m0

The easiest to interpret ism0 =ξ(E)0,2. The target space MH(Gad, C) of our sigma- model is not connected. Its components are labeled by the topological type of the Gad- bundle E →C. But this is exactly what is measured by m0.

Interpretation Of e0

Consider in general a sigma-model of maps Φ : Σ → X, for some X. A flat B-field is an element b ∈ H2(X, U(1)). A flat B-field is incorporated in the sigma-model path integral as follows, for the case that Σ has no boundary. Given a map Φ : Σ → X, one pulls back b to Φ(b) ∈ H2(Σ, U(1)) = U(1), and then one includes in the path integral a factor of Φ(b). Thus, in this situation, incorporating the flat B-field has the effect of weighting by phases the different components of maps of Σ to X.

Now e0 determines a flat B-field in the sigma-model of maps Φ : Σ → MH(Gad, C).

Indeed, there is as we have explained a natural class ζ ∈ H2(MH(Gad, C),Z), which expresses the obstruction to lifting the universal Gad bundle Ead → MH(Gad, C) to a G-bundle. By composing ζ with e0 : Z → U(1), we get a flat B-field be0 = e0(ζ) ∈ H2(MH(Gad, C), U(1)). We claim that the role of e0 in the effective sigma-model of maps Σ → MH(Gad) is precisely to weight every map in the way that one would expect for a sigma-model with the flat B-field be0.

In fact, the definition of e0 is that the path integral includes a phase factor e0(a0) = e0(ξ(E)2,0). Butξ(E)2,0 = Φ(ζ); this follows upon composing (7.19) with the projection Σ ×C → Σ. So the phase factor induced in the path integral by e0 is e0(ζ)) = Φ(e0(ζ)) = Φ(be0). This justifies our claim that the effect of e0 in the sigma-model is to generate a flat B-field be0.

(7.24) therefore means that S-duality exchanges the topological class m0 of a flat Gad-bundle with the flat B-field determined by e0.

Incorporation Of Branes

For applications to the geometric Langlands program, the real payoff is to understand the implications of all this for branes.

So we take Σ to be a Riemann surface with boundary. M = Σ×C is therefore a four-manifold with boundary ∂M = ∂Σ×C. On ∂M, we place some supersymmetric boundary condition.

The effective two-dimensional description is a sigma-model of maps Φ : Σ→MH(Gad), with boundary condition corresponding to some brane. We would like to understand the role of e0 and m0 in this description.

There is little new to say about m0. It labels the components of MH(Gad), whether Σ has a boundary or not.

The role ofe0 is more subtle. As we have seen, the sigma-model with target MH(Gad) is endowed with a flat B-field e0(ζ). A flat B-field has a very interesting effect on branes [98]. In the absence of a B-field, a brane on MH(Gad) has a Chan-Paton bundle, which is a vector bundle over MH(Gad) (or more generally a sheaf, perhaps supported on a submanifold, that defines a K-theory class of MH(Gad)). However, in the presence of a flat B-field, associated with an element b∈ H2(MH(Gad), U(1)), the Chan-Paton bundle becomes a twisted vector bundle (or more generally a twisted sheaf related to an element of the twisted K-theory of MH(Gad)), twisted by b in the sense of eqn. (7.15). It is because of this that we introduced the concept of a twisted vector bundle.

So in short, for e0 6= 0, the Chan-Paton bundle of a brane is a twisted vector bundle, twisted by be0 = e0(ζ). Luckily, from the analysis at the end of section 7.1, we have a plentiful supply of such twisted vector bundles. If ̺ is any irreducible representation of G such that the character of the center of G defined by ̺ is equal to e0, then the universal bundle E̺ in the representation ̺ is an example of a twisted vector bundle for the flat B-fieldbe0.

Other important examples of twisted branes can be constructed by picking a sub- manifold Y ⊂ MH(Gad) such that be0 is trivial when restricted to Y. In this case, the definition of a brane supported on Y is independent of e0 (up to a not quite canonical isomorphism). For example, Y could be a point in MH(Gad). Certainly be0 is trivial when restricted to a point, so zerobranes exist for any e0. From such zerobranes, we can form electric eigenbranes, as we explain in section 8. A slightly more subtle example is a mag- netic eigenbrane, a brane of rank one supported on a fiberF of the Hitchin fibration. The Chan-Paton bundle of such a brane should be a flat34 twisted line bundle. In [21], it is shown that the flatb-fieldbe0 is trivial when restricted toF, as a result of which the space of flat twisted line bundles is isomorphic, but not canonically isomorphic, to the space of ordinary flat line bundles on F.

We can now deduce from S-duality a statement about branes. The duality trans- formation S maps (e0,m0) → (−m0,e0), so it exchanges the topology of the component of MH on which a brane is supported with the flat B-field be0 by which its Chan-Paton bundle is twisted. It also, of course, exchanges Gad with35 LGad, and (as we explain in the concluding remark of this section), exchanges MH(Gad) withMbH(LGad), which we define to be the universal cover of MH(LGad).

As a special case of this duality, a point on one side, contained in a fiber F of the Hitchin fibration, is mapped on the other side to a brane of rank one supported on the corresponding fiber LF = Ξ(F) of the Hitchin fibration, and endowed with a flat twisted line bundle. F is a union of complex tori, labeled by the characteristic classm0 =ξ(E) of the Higgs bundle. The choice of a component ofF on one side determines on the other side the discrete electric fielde0 and hence the twist. The choice of a point on F determines a flat twisted line bundle on LF (to which it maps under the duality transformation S). Of course, this relationship between F and LF is reciprocal. This twisted duality between F and LF is in fact one of the main results of Hausel and Thaddeus [21].

Interpretation Of e1 And m1

34 One can here interpret flatness to mean, just as for ordinary line bundles, that the transition functions are constants.

35 By LGad, we mean the adjoint form of the group LG. The statement we are describing here is best expressed in terms of adjoint bundles on both sides to allow all possible topologies.

Momentarily we indicate explicitly whether we want a given component of MH or its universal cover.

Finally, let us discuss the interpretation of e1 and m1 in two-dimensional terms. It will be helpful to begin by comparing the theories with gauge groups G and Gad.

MH(G) is simply-connected. It has a natural group of symmetries EC = H1(C,Z).

Indeed, EC parametrizes Z-bundles over C. A G Higgs bundle (E, ϕ) can be tensored with aZ-bundle to make a newGHiggs bundle. (Concretely, this operation multiplies the holonomies of E around noncontractible loops in C by elements of the center of G.) So this gives a group EC of symmetries of the sigma-model with targetMH(G). The Hilbert space of this sigma-model can be decomposed in characters of EC.

On the other hand, MH(Gad) has no such geometrical symmetries. But it has a fundamental group MC = H1(C,Z), which is isomorphic to EC. To understand where this fundamental group comes from, a shortcut is to note that one component MH(Gad)0 of MH(Gad), namely the component that parametrizes Higgs bundles that can be lifted to G, is simply MH(Gad)0 = MH(G)/EC. Dividing by EC eliminates the geometrical symmetries ofMH(G), but of course it creates a fundamental group.36 Soπ1(MH(Gad)0)∼= EC.

Actually, the fundamental group is the same for any component of MH(Gad).

MH(Gad) is defined by dividing the space of all solutions of Hitchin’s equations, for gauge groupGad, by the groupGad(C) of allGad-valued gauge transformations onC. If one were to divide only by the connected component of Gad(C), one would get the universal cover

b

MH(Gad, C). The fundamental group of MH(Gad) is therefore the group of components of Gad(C), and this is EC =H1(C,Z), for any component of MH(Gad).

In sum, strings moving onMH(Gad) have a discrete group of conserved winding num- bersMC1(MH(Gad)) =H1(C,Z). Likewise strings moving onMH(G) have a discrete group of conserved momenta EC =H1(C,Z).

Let us compare the symmetries EC and MC to what we can see in the underlying gauge theory. In Gad gauge theory on M = Σ×C = S1 ×Se1 ×C, the bundle E and other data determine a map Φ : S1 ×Se1 → MH(Gad). In analyzing the topology of this situation, we expanded the characteristic class ξ(E) as ξ(E) = a⊕m, where m is the restriction of ξ(E) to Se1 ×C (more precisely, to p×Se1×C, for a point p ∈ S1). In the

36 EC does not act freely, so MH(Gad) has orbifold singularities. (It also has more severe singularities from reducible Higgs bundles.) As is familiar in sigma models, the fundamental group of MH(Gad) must be understood in an orbifold sense. Alternatively, one can rely on the four-dimensional gauge theory instead of reducing to the sigma model with its singularities.

low energy sigma-model, a and m are invariants of Φ : Σ →MH(Gad). In particular, m only depends on the restriction Φ|Se1 of Φ to Se1.

In (7.21), we further expanded m = m0 ⊕m1 with m1 ∈ H1(Se1, H1(C,Z)). m1 is a topological invariant of Φ|Se1 that vanishes for constant maps of Se1 to MH(Gad). So it measures the homotopy class of the map Φ|Se1 in π1(MH(Gad)) =H1(C,Z).

Similarly, we could exchange the role of S1 and Se1. The restriction of ξ(E) to S1×C (that is, toS1×q×C, for a point q∈Se1), is in the above notationm0⊕a1. In particular, a1 ∈ H1(S1, H1(C,Z)) measures the winding of Φ in the S1 or “time” direction. The charactere1 which is dual toa1 therefore measures the conserved momentum of the strings.

An Example

For an important illustration of all this, consider the Langlands dual pair Gad and

LG. In Gad gauge theory, we set e = 0 and consider a sigma-model with target MH(Gad).

In this model, there are no conserved momenta, but there is a symmetry MC of string windings.

In the dual picture, the gauge group is LG, we set m= 0, and the sigma-model with target MH(G) has no conserved string windings, but a symmetry group EC of discrete conserved momenta.

S-duality or Montonen-Olive or Langlands duality exchanges the two pictures, ex- changing EC(LG) with MC(Gad). The fact that the duality exchanges the discrete con- served momenta and windings of strings is an aspect of its relation to T-duality in two dimensions.

A Concluding Comment

More generally, we can simply specify m0 and e0 as we please. Then we consider branes on a component of MH(Gad) labeled topologically bym0; the Chan-Paton bundles of the branes are twisted by e0.

We still have two ways to proceed with the quantization. If we divide by allGad-valued gauge transformations, then the target space is the component of MH(Gad, C) labeled by m0. In this case, there is a finite groupM1 that classifies the string winding numbers, but there is no group E1 of geometrical symmetries. If we divide by only the connected gauge transformations, then the target is the universal cover MbH(Gad, C). In this case, there is a group E1 of geometrical symmetries, but no group M1 of string winding numbers.

S-duality exchanges E1 and M1, so (in addition to exchanging G and LG) it exchanges the two methods of quantization.

In particular, when we apply S-duality to branes of specified e0 and m0, we must exchange the two methods of quantization in addition to exchanging the two adjoint groups Gad and LGad.

Dalam dokumen And The Geometric Langlands Program (Halaman 110-116)