• Tidak ada hasil yang ditemukan

Review

Dalam dokumen And The Geometric Langlands Program (Halaman 103-110)

7. Fluxes and S-Duality

7.1. Review

iǫ and Γ2367ǫ = ǫ. This system is physically sensible (if we interpret it with Lorentz signature); it is described by a real, positive-definite Hilbert space with a positive-definite, hermitian Hamiltonian H. The four unbroken supersymmetries are the two topological supercharges Qi, i = 1,2 described earlier, and their hermitian adjoints Qi. They obey a physical supersymmetry algebra {Qi, Qj} = 2δijH, {Qi, Qj} = 0 = {Qi, Qj}, for i, j = 1,2.

Group Center

SU(N) ZN

Spin(2n+ 1) Z2 Spin(4n+ 2) Z4

Spin(4n) Z2 ×Z2

Sp(n) Z2

E6 Z3

E7 Z2

Table 3. Simple and simply-connected Lie groups with non-trivial centers. F2, G4, and E8, which are omitted, have trivial center.

An important property is that the group Z is naturally selfdual. The dual of an abelian group B isB= Hom(B, U(1)). A finite abelian group is always isomorphic to its own dual, but not naturally. The centerZ of Galways has, however, a natural selfduality.

This duality is best expressed as a homomorphism

Υ :Z × Z →U(1), (7.1) which is symmetric, Υ(a, b) = Υ(b, a), and such that any homomorphism of Z to U(1) is a → Υ(a, b) for a unique b. Such a Υ is called a “perfect pairing.” Υ is constructed as follows. If G is simply-laced, then Z = Λ/Λ, where Λ is the root lattice and Λ is the weight lattice of G. The pairing Υ is then defined as

Υ(a, b) = exp(2πiha, bi), (7.2)

where h , i is the usual quadratic form on Λ (which is even and integral when restricted to Λ and takes rational values on Λ). If G is not simply-laced, its center is trivial or Z2, and so admits precisely one selfduality, which we call Υ.

A Gad-bundle E onM has a characteristic class ξ(E) that takes values inH2(M,Z).

For example, if Gad = SO(3), ξ(E) is the second Stieffel-Whitney class w2(E). One can define a partition function Zξ for every choice of ξ by restricting the path integral to bundles with givenξ. The Zξ transform in a unitary representation of theS-duality group [45,26], but this is something that we will not explain here.

Instead, we concentrate on the Hamiltonian description, and specialize toM =S1×W for a three-manifoldW. In this case, we haveH2(M,Z) =H1(W,Z)⊕H2(W,Z), soξ(E) can be decomposed asξ =a⊕m, with a∈H1(W,Z), m∈H2(W,Z).

Consider the pairing

Υ :b H2(M,Z)×H2(M,Z)→H4(M, U(1)) =U(1) (7.3) obtained by composing the cup productH2(M,Z)×H2(M,Z)→H4(M,Z × Z) with the map Υ : Z × Z → U(1). According to Poincar´e duality, Υ is a perfect pairing, makingb H2(M,Z) a selfdual abelian group. More generally, for a closed oriented manifold Y of dimension n, Poincar´e duality together with selfduality ofZ gives perfect pairings

Hd(Y,Z)×Hn−d(Y,Z)→U(1), 0≤d≤n. (7.4) In particular, we have the perfect pairing

Υ :H1(W,Z)×H2(W,Z)→U(1). (7.5) So these are dual abelian groups, and in particular

H2(W,Z) = Hom(H1(W,Z), U(1)). (7.6) Let us now recall how S-duality is implemented in a Hamiltonian framework. The partition function of the gauge theory with m specified has a natural Hilbert space in- terpretation, because m is part of the data at an initial time and is independent of time.

However, becausea cannot be expressed in terms of the data at a fixed time, the partition function with fixed a cannot be given a Hamiltonian interpretation. Instead one must introduce a character e of the finite abelian group H1(M,Z), that is, a homomorphism

e :H1(W,Z)→U(1). (7.7)

For each choice of e and m, the sum P

ae(a)Za,m can be interpreted in terms of a trace in a Hilbert space He,m. This sum has a Hilbert space interpretation because summing over a with the weight factor e(a) is compatible with cutting and pasting (or what in quantum field theory is usually called cluster decomposition) in the S1 direction. e and m are called respectively the discrete electric and magnetic flux [97]. Because of (7.6),e can be alternatively viewed as an element of H2(W,Z), just like m.

So we get a Hilbert space Hm,e for each choice of

e∈E= Hom(H1(W,Z), U(1))

m∈M=H2(W,Z). (7.8)

Happily, the finite groups E and M are isomorphic according to (7.6). This makes it possible to have S-duality symmetry. The duality transformation S exchanges E and M and maps (e,m)→(−m,e):

S :He,m → Hm,e. (7.9)

It also, of course, maps τ → −1/τ. (Eqn. (7.9) is a discrete analog of (2.23), with (m, ~~ n) replaced by their discrete counterparts (m,e).)

While it is possible to define all the Hilbert spacesHe,m, a given gauge theory construc- tion may not use all of them. If we do gauge theory onM =S1×W with simply-connected gauge group G, then we must set to zero the characteristic classes a and m that enter in the partition function Za,m(M). For the Hilbert space, this means thate, which arises by a Fourier transform with respect to a, is arbitrary, whilem= 0. Hence

H(G, W) =⊕eHe,0. (7.10)

Taking into account the assumed transformation law for e,m underS-duality, we see that the Hilbert space of the dual theory must be

mH0,m. (7.11)

How is this Hilbert space related to a path integral on M = S1 ×W? Setting e to zero means summing over a. So in terms of the variablesa,m, (7.11) means one must sum over all a,m with equal weight. In other words, in the dual theory we sum over LGad-bundles with all possible ξ. Thus the dual theory has gauge group LGad, as expected.

The Universal Bundle

We need another piece of background, which is the concept of the universal bundle.

Consider first the moduli space M(G, C) ofG-bundles over a Riemann surfaceC. For every point p∈M(G, C), there is a corresponding G-bundleEp →C parametrized byp; it is determined up to isomorphism. A universal bundle, if it exists, is aG-bundleE→M×C such that for any p ∈ M, E restricted to p×C is isomorphic to Ep. Gauge theory gives

[88] a very direct method to analyze the universal bundle, though we will not explain this here.

Naively, we just defineE so that its restriction top×C is “the” bundle overC that is determined byp. The reason that a problem arises is that, to begin with,ponly determines an isomorphism class of G-bundle over C, not an actual bundle. This causes no difficulty locally33; we just pick a bundle in the right isomorphism class at a particular p and then deform it in a small neighborhood. So for each small open set Ui ⊂ M, we can pick a universal bundle Ei→Ui×C, and moreover Ei is unique up to isomorphism.

The next step is to glue together the Ei over (Ui ∩Uj)×C. The bundles Ei and Ej are isomorphic over (Ui ∩Uj)×C, so we pick an isomorphism Θij : Ei ∼= Ej over this intersection, with Θji = Θ−1ij .

Now if on triple intersections (Ui∩Uj∩Uk)×C, the composition ΘkiΘjkΘij is equal to 1, then we can consistently glue together the Ei to get the desired universal bundle E.

If the gauge group is the adjoint group Gad, there is no problem in the gluing. A generic Gad bundle has no automorphisms (exceptions occur at singularities of M). So ΘkiΘjkΘij, just because it is an isomorphism of Ei (restricted to (Ui∩Uj ∩Uk)×C) to itself, is the identity at a generic point of the triple intersection and hence everywhere.

Thus, the universal Gad-bundle Ead does exist. Now let us repeat this discussion, taking the gauge group to be the simply-connected cover G (or more generally, any cover of Gad with a non-trivial center). We start in the same way. For each small open set Ui ∈ M, we pick a universal bundle Ei over Ui ×C, and try to glue to make a universal G-bundle E.

The reason that the result is different is that a generic G-bundle does have a non- trivial group of automorphisms; the centerZ is a group of automorphisms of anyG-bundle E.

Hence, in the above argument, the composite map ΘkiΘjkΘij over a triple intersection is not necessarily the identity; instead

ΘkiΘjkΘij =fijk, (7.12)

33 This statement holds near a smooth point inM. A more precise analysis than we will give shows that the class ζ(Ead) that obstructs the universal bundle can have a local contribution at singularities.

with fijk ∈ Z. The fijk combine to a two-cycle defining an element ζ ∈ H2(M,Z). The element ζ is known to be non-trivial; it is the obstruction to the existence of the universal G-bundle E.

There is another way to explain the existence ofζ. LetM(Gad, C)0 be the component ofM(Gad, C) that parametrizes topologically trivial bundles – the ones that when restricted top×C, forp∈M(Gad, C), can be lifted toG-bundles. Consider the universal bundleEad over M(Gad, C)0×C. If we could lift E to a G-bundle E, this (after being pulled back to M(G, C)×C, which is a finite cover ofM(Gad, C)0×C) would be the universalG-bundle.

In general, the obstruction to lifting aGad-bundleEadto aG-bundleEis the characteristic class ξ(Ead). So the obstruction to existence of the universalG-bundle isζ =ξ(Ead).

There is a similar story for Higgs bundles. The universal Higgs bundle is a pair (E,ϕ),b with E being a G-bundle over MH(G, C)×C, and ϕb∈H0(MH(G, C)×C,ad(E)⊗KC), obeying the following condition. For each p∈MH(G, C), the restriction of (E,ϕ) tob p×C should be isomorphic to the Hitchin pair (Ep, ϕp) parametrized by p.

The arguments that we have already given can be carried over with no essential change to show that for gauge group Gad, the universal Higgs bundle (Ead,ϕ) does exist. But forb the simply-connected gauge group G (or any nontrivial cover of Gad), the universal Higgs bundle does not exist. It is obstructed by the fact that a generic Hitchin pair (E, ϕ), with E aG-bundle, has the groupZ of automorphisms. The obstruction is just the obstruction to lifting the “bundle” part of the universal Gad Higgs bundle (Ead,ϕ) to ab G-bundle. It is therefore

ζ =ξ(Ead). (7.13)

The Universal Bundle As A Twisted Vector Bundle

Although for G not of adjoint type, the universalG-bundle does not exist as a vector bundle or as a principal bundle, it does exist as a twisted vector bundle. We will describe this concept in a very pedestrian way.

One way to construct an ordinary vector bundle V of rank N over a manifold X is to cover X with small open sets Ui, on each of which we pick a rank N trivial bundle Vi ∼= Ui ×CN. Then we glue Vi to Vj on the intersection Ui ∩Uj via a gluing map vij :Vi∩Vj →U(N), with vji=v−1ij . If on triple intersections we have

vkivjkvij = 1, (7.14)

then the Vi can be glued together consistently to make a rank N vector bundle V →X.

Now suppose that we are givenb∈H2(X, U(1)). Then bcan be represented explicitly by a U(1)-valued cocyclebijk defined on triple intersectionsUi∩Uj∩Uk. A twisted vector bundle is defined by the same sort of data vij : Vi ∩Vj → U(N). But now, instead of (7.14), we ask for

vkivjkvij =bijk. (7.15)

Thus, a b-twisted vector bundle V is not a vector bundle in the usual sense. However, the associated adjoint bundle ad(V) is an ordinary vector bundle, since the phase bijk in (7.15) disappears if we pass to the adjoint representation.

Now let us return to the problem of constructing a universal bundle. In this paper, we are generally a little imprecise about whether by a G-bundle, we mean a principal G-bundle, whose fiber is a copy of G, or an associated vector bundle in some faithful representation of G. Principal bundles make possible a uniform analysis good for any G, but for a group likeU(N) that has a convenient faithful representation (theN-dimensional representation) it is useful to think in terms of vector bundles.

In discussing the universal bundle, it is helpful to be more precise. We interpret the transition functions Θij of our discussion above as G-valued functions, with no particular choice of representation. Their projection to Gad gives us transition functions for a uni- versal principal Gad-bundle Ead, but we cannot lift this to a principal G-bundle, because of the relation

ΘkiΘjkΘij =fijk, (7.16)

where the fijk define the class ζ = ξ(Ead) ∈ H2(MH(Gad, C),Z). Now let us pick an irreducible representation ̺ of G. We write Θ̺ij for the transition functions Θij evaluated in the representation ̺. Likewise we write ̺(f) for f evaluated in the representation ̺.

In the irreducible representation ̺, the center of G acts by scalar multiplication, so ̺(f) takes values in U(1). We have

Θ̺kiΘ̺jkΘ̺ij =̺(fijk). (7.17) The quantities ̺(fijk) are a cocycle defining the element ̺(ζ)∈H2(MH(Gad, C), U(1)).

If ̺(f) = 1, the objects Θ̺ are transition functions that define a vector bundle E̺ → MH(G, C) that we may call the universal bundle in the representation ̺. In general this is not the case. However, comparing eqns. (7.16) and (7.15), we see that while E̺ may not exist as an ordinary vector bundle, it does always exist as a twisted vector bundle, twisted by ̺(ζ)∈H2(MH(Gad, C), U(1)).

Dalam dokumen And The Geometric Langlands Program (Halaman 103-110)