Topological Twists and the Geometric–Langlands Interface

The geometric-Langlands interface in four-dimensional field theory is a protected statement built from the GL twist of $N=4$ Yang–Mills, a globally specified S-duality arrow, and categories of boundary conditions and line operators. Compactification on a Riemann surface turns the twist into a topological sigma model on Hitchin moduli space. The resulting categorical correspondence is deep evidence and a physical construction of geometric Langlands structures, but the twist deliberately discards most untwisted observables.

Required background. Use the general construction of topological and holomorphic twists and the globally complete duality groupoid and walls.

Helpful background. Six-dimensional duality frames explain the mapping-class-group origin, while Q-cohomology in supersymmetric quantum mechanics supplies the finite-dimensional model for the quotient.

The GL twist of N=4 Yang–Mills

The geometric-Langlands or GL twist identifies an $SU(2)$ factor of Euclidean rotations with part of the $N=4$ R-symmetry. It produces two scalar supercharges, and a projective family

$$Q_t=uQ_++vQ_-, \qquad t=\frac vu\in\mathbb{CP}^1.$$

For gauge-invariant operators, $Q_t^2$ is a gauge transformation and hence vanishes in cohomology. The stress tensor is Q-exact up to topological terms, so separated Q-cohomological observables are metric independent within the allowed class of backgrounds.

The pair $(t,\tau)$ is not the invariant parameter seen by the topological theory. With

$$\tau=\frac\theta{2\pi}+\frac{4\pi i}{g^2},$$

the canonical parameter is

$$\Psi =\frac\theta{2\pi} +\frac{4\pi i}{g^2} \frac{t-t^{-1}}{t+t^{-1}}.$$

Different pairs $(t,\tau)$ with the same $\Psi$ define equivalent topological dependence, subject to the global and boundary data Kapustin and Witten 2007, §§3–4. Singular-looking values such as $t=\pm i$ are handled by the projective family and correspond to $\Psi=\infty$ rather than to a failure of the twist.

S-duality of the canonical parameter

Let ${}^LG$ be the Langlands-dual group and let $n_G$ be the ratio of squared long- and short-root lengths. The basic S-duality arrow acts as

$$(G,\Psi) \longmapsto \left({}^LG,-\frac1{n_G\Psi}\right).$$

For simply laced groups, $n_G=1$. This formula is incomplete unless the global form, discrete theta angle, genuine line lattice, and boundary backgrounds are transformed with it. Langlands duality exchanges weight and coweight lattices, which is why Wilson and ‘t Hooft operators are exchanged.

At $\Psi=0$, the S transformation reaches the $\Psi=\infty$ theory. These are the two protected limits used in the geometric-Langlands construction. A generic $SL(2,\mathbb Z)$ element produces a quantum or twisted variant rather than the same classical correspondence.

Compactification and Hitchin moduli space

Place the theory on $\Sigma\times C$, with $C$ a compact Riemann surface, and take $C$ small compared with the scales on $\Sigma$. The low-energy two-dimensional theory is a topological sigma model whose target is the Hitchin moduli space

$$\mathcal M_H(G,C).$$

A point is represented by a connection $A$ and Higgs field $\phi$ obeying Hitchin’s equations,

$$F_A-[\phi,\phi]=0, \qquad d_A\phi=0, \qquad d_A^*\phi=0,$$

modulo gauge transformations, with conventional factors of $i$ depending on whether fields are taken Hermitian or anti-Hermitian. The space is hyperkähler. The GL twist selects an A- or B-model structure determined by $(t,\tau)$.

S-duality becomes mirror symmetry

$$\mathcal M_H(G,C) \longleftrightarrow \mathcal M_H({}^LG,C).$$

Fiberwise, this exchanges dual Abelian varieties in the Hitchin fibrations. Singular fibers, disconnected components, and global gerbe data require separate treatment; the smooth generic-fiber picture is not the whole equivalence.

Branes, D-modules, and Hecke operators

Boundary conditions of the four-dimensional topological theory become branes in the Hitchin sigma model. At the geometric-Langlands point, the canonical coisotropic A-brane generates a category related to twisted differential operators on $\operatorname{Bun}_G(C)$. Under mirror symmetry, B-branes supported at flat ${}^LG$ local systems produce Hecke eigensheaf data.

The line-operator map supplies the eigenvalue property:

$$\text{Wilson line for }{}^LG \quad\longleftrightarrow\quad \text{'t Hooft/Hecke operation for }G.$$

Moving a line to a boundary acts as an endofunctor on the boundary category. Junctions become natural transformations. Thus the appropriate protected object is categorical; equality of two partition functions would be far too weak to capture it.

What the physical construction establishes

The twisted field theory provides:

• a canonical parameter $\Psi$ and its duality action;
• a brane-category interpretation of boundary conditions;
• Wilson–‘t Hooft exchange and Hecke functors;
• mirror symmetry of Hitchin moduli spaces;
• ramified extensions from surface operators;
• a route from S-duality to geometric-Langlands categories.

It does not retain the generic untwisted spectrum, scattering amplitudes, non-BPS correlators, or a positive Hilbert-space structure. It also imports the nonperturbative S-duality of $N=4$ Yang–Mills as physical input. Consequently, the construction is not by itself a theorem proving every mathematical formulation of geometric Langlands for every curve, group, ramification, and derived enhancement.

Global and categorical checks

1. Verify that the scalar supercharge is globally defined and anomaly free.
2. State $G$, its global form, discrete theta data, and ${}^LG$.
3. Transform $\Psi$ and the genuine line lattice together.
4. Specify the A/B model, B-field, and brane category at each endpoint.
5. Include singular Hitchin fibers and gerbe data when relevant.
6. Check Wilson–‘t Hooft actions on boundary conditions and their fusion.
7. Separate field-theoretic predictions from mathematical results proved independently.

Exercises

For a simply laced group, apply S-duality twice to $\Psi$.

Solution

The first transformation gives $\Psi'=-1/\Psi$. Applying it again gives $\Psi''=-1/\Psi'=\Psi$. On complete theory data, the square can also act by charge conjugation, so equality of the numerical parameter is not the full groupoid statement.

References

• Kapustin, A., and E. Witten. “Electric–Magnetic Duality and the Geometric Langlands Program.” Communications in Number Theory and Physics 1 (2007): 1–236. DOI; Open PDF.