T-duality for Langlands dual groups (Q2257180)

This article addresses the following question: Is the relationship between a complex reductive group and its Langlands dual a form of \(T\)-duality? This question arises because a reductive group can be interpreted as a principal torus bundle for a maximal torus, and the maximal tori of Langlands dual groups are canonically \(T\)-dual. We can make the above question more precise. Suppose \(G\) is a complex reductive group. Choosing a maximal torus \(T \subset G\) allows one to define a root datum of \(G\) \[ R(G, T) = \{C \subset \Lambda, R \subset {\Lambda}^{\bot}\} \] where \(R\) and \(C\) are the sets of roots and coroots of \(G\), while \({\Lambda}^{\bot} = \Hom(T, \mathbb C^*)\) and \(\Lambda = \Hom(\mathbb C^*, T)\) are its character and co-character lattices, \(\mathbb C^* = \mathbb C -\{0\}\). A root datum determines the reductive group \(G\) up to isomorphism. Interchanging the roles of roots and coroots and of the character and cocharacter lattices results in a new root datum \[ R(G, T)^{\vee} = \{ R \subset {\Lambda}^{\bot}, C \subset \Lambda\}. \] The Langlands dual group of \(G\) is the complex reductive group \(G^{\vee}\) (unique up to isomorphism) determined by the dual root datum \(R(G, T)^{\vee}\). A root datum also implies a choice of maximal torus \(T \subset G\) via the canonical isomorphism \[ T \simeq \Hom({\Lambda}^{\bot}, \mathbb C^*), \] and likewise a natural choice of maximal torus for the Langlands dual group \(G^{\vee}\): \[ T^{\vee} \simeq \Hom({\Lambda}, \mathbb C^*) \subset G^{\vee}. \] By definition, the torus \(T^{\vee}\) is the \(T\)-dual of \(T\). This is the noncompact \(T\)-dual torus \(T^{\vee} \simeq (\mathbb C^*)^n\). To obtain a compact version, which is more common in the \(T\)-duality literature, one simply begins with a compact form of \(G\) instead of a complex form. The translation action of a maximal torus makes \(G\rightarrow G/T\) a principal torus bundle. Thus given that the maximal tori of Langlands dual groups are \(T\)-dual, it is natural to ask whether \(G\) viewed as a principal \(T\)-bundle, and \(G^{\vee}\) viewed as a principal \(T ^{\vee}\)-bundle, are \(T\)-dual to one another. A problem that immediately presents itself is that \(G\rightarrow G/T\) and \(G^{\vee}\rightarrow G^{\vee}/T^{\vee}\) are torus bundles over different spaces, whereas \(T\)-duality is a relationship between torus bundles over a common base space. However, if the simple factors of \(G\) are all of Dynkin type A, D, or E, then the manifolds \(G/T\) and \(G^{\vee}/T^{\vee}\) are isomorphic. If the base manifolds \(G/T\) and \(G^{\vee}/T^{\vee}\) are isomorphic, \(G\) and \(G^{\vee}\) only have a hope of being \(T\)-dual if we twist the duality by introducing NS-fluxes on each space. An NS-flux on a principal \(T\)-bundle is a closed, \(T\)-invariant, integral \(3\)-form on the total space of the bundle. Indeed, a pair of torus bundles without NS-flux are \(T\)-dual if and only if both torus bundles are flat, which is generally not the case for the bundles \(G\rightarrow G/T\) and \(G^{\vee}\rightarrow G^{\vee}/T^{\vee}\). The main theorem of this paper asserts (roughly) that when the flag manifolds \(G/T\) and \(G^{\vee}/T^{\vee}\) are isomorphic there is a natural choice of NS-flux on \(G\) and on \(G^{\vee}\) such that the two groups are \(T\)-dual.