Geometric Langlands and Physics

⚡ Key Takeaways

The Geometric Langlands Correspondence is a conjectural equivalence between two categories attached to a complex algebraic curve $X$ and a reductive group $G$ : the derived category of D-modules on the moduli stack $\mathrm{Bun}_G(X)$ of $G$ -bundles, and the derived category of quasi-coherent sheaves on the moduli stack $\mathrm{LocSys}_{{}^L G}(X)$ of flat ${}^L G$ -connections (local systems). Physics realizes this as electric-magnetic (S-)duality of the topologically twisted $\mathcal{N}=4$ super Yang-Mills theory in four dimensions, compactified on $X$ . It applies in number theory, representation theory, and quantum field theory; the program crystallized through work of Drinfeld, Laumon, Beilinson-Drinfeld (1980s-90s) and was given its decisive physical interpretation by Kapustin-Witten (2006).

💡 The Core Insight

Compactify four-dimensional $\mathcal{N}=4$ super Yang-Mills (in a particular topological twist, the GL-twist) on a product $\Sigma \times X$ of two Riemann surfaces. Reducing along the small $X$ produces a two-dimensional sigma model on $\Sigma$ whose target is the Hitchin moduli space $\mathcal{M}_H(G,X)$ . The S-duality $G \leftrightarrow {}^L G$ , which acts on the 4d theory, becomes mirror symmetry of these Hitchin sigma models. Branes in the A-model on $\mathcal{M}_H({}^L G)$ map to branes in the B-model on $\mathcal{M}_H(G)$ ; translated mathematically, this is exactly the kernel/transform underlying geometric Langlands. The "electric eigenbrane" of a Wilson line becomes a "magnetic eigenbrane" of a 't Hooft line — the physical avatar of a Hecke eigensheaf.

DefinitionModuli Stacks of the Correspondence

Fix a smooth projective curve $X$ over $\mathbb{C}$ and a reductive group $G$ with Langlands dual ${}^L G$ .

$\mathrm{Bun}_G(X)$ — the moduli stack of principal $G$ -bundles on $X$ . It is a smooth Artin stack, generally non-separated, of dimension $(g-1)\dim G$ .
$\mathrm{LocSys}_{{}^L G}(X)$ — the moduli stack of ${}^L G$ -local systems, i.e. pairs $(P,\nabla)$ of a ${}^L G$ -bundle with a flat connection, equivalently ${}^L G$ -representations of $\pi_1(X)$ up to conjugacy.

The two sides carry the categories

1

\quad D\text{-}\mathrm{mod}\big(\mathrm{Bun}_G(X)\big) \;\xleftrightarrow{\;\sim\;}\; \mathrm{QCoh}\big(\mathrm{LocSys}_{{}^L G}(X)\big).

Derivation hint: D-modules encode "automorphic" data via the Riemann–Hilbert philosophy; the spectral side parametrizes Galois-type objects. The equivalence is the geometric analogue of "automorphic ↔ Galois."

DefinitionHecke Functors and Eigensheaves

For each point $x \in X$ and each cocharacter $\lambda$ of $G$ (a dominant weight of ${}^L G$ ), there is a Hecke modification correspondence

2

\quad \mathrm{Bun}_G \xleftarrow{\;h^{\leftarrow}\;} \mathrm{Hecke}_{x,\lambda} \xrightarrow{\;h^{\rightarrow}\;} \mathrm{Bun}_G,

inducing a Hecke functor $H_{x,\lambda}(\mathcal{F}) = h^{\rightarrow}_{*}\big(h^{\leftarrow *}\mathcal{F} \otimes \mathrm{IC}_\lambda\big)$ . A D-module $\mathcal{F}$ is a Hecke eigensheaf with eigenvalue the local system $E$ if

3

\quad H_{x,\lambda}(\mathcal{F}) \;\cong\; V_\lambda(E_x)\otimes \mathcal{F},\qquad \forall x,\lambda,

where $V_\lambda$ is the irreducible ${}^L G$ -representation of highest weight $\lambda$ and $E_x$ is the fiber at $x$ . Derivation hint: This generalizes Hecke operators in classical automorphic theory; the eigenvalue records, via the Satake correspondence, the conjugacy class of Frobenius/monodromy.

TheoremGeometric Satake Equivalence (Lusztig, Ginzburg, Mirković–Vilonen)

Let $\mathrm{Gr}_G$ be the affine Grassmannian of $G$ . There is an equivalence of tensor categories

4

\quad \mathrm{Perv}_{G(\mathcal{O})}\big(\mathrm{Gr}_G\big) \;\simeq\; \mathrm{Rep}\big({}^L G\big),

intertwining the convolution product with the tensor product of ${}^L G$ -representations. Derivation hint: The equivariant perverse sheaves on $\mathrm{Gr}_G$ form a symmetric monoidal abelian category; the Tannakian reconstruction theorem identifies it with representations of a reductive group, which turns out to be the Langlands dual ${}^L G$ . The cohomology functor is the fiber functor.

TheoremThe Best-Hope Categorical Equivalence (Beilinson–Drinfeld / Arinkin–Gaitsgory)

There is a conjectural equivalence of (derived, dg-)categories

5

\quad D\text{-}\mathrm{mod}\big(\mathrm{Bun}_G(X)\big) \;\simeq\; \mathrm{IndCoh}_{\mathcal{N}}\big(\mathrm{LocSys}_{{}^L G}(X)\big)

exchanging Hecke functors on the left with the action of tautological "Wilson" tensor functors on the right, and sending the structure sheaf of a point $\{E\}\in \mathrm{LocSys}_{{}^L G}$ to a Hecke eigensheaf with eigenvalue $E$ . Derivation hint: The naive $\mathrm{QCoh}$ must be refined to ind-coherent sheaves with nilpotent singular support $\mathcal N$ (the global nilpotent cone) to obtain a genuine equivalence; this correction was identified by Arinkin–Gaitsgory and is essential for the statement to hold.

ProofFrom S-duality to Geometric Langlands

Step 1 — The 4d theory and its twist. Start with $\mathcal{N}=4$ super Yang-Mills with gauge group $G$ on a 4-manifold $M$ . There is a family of topological twists parametrized by $t \in \mathbb{P}^1$ , the GL-twist (Geometric Langlands twist). The twist combines two of the four supercharges into a scalar BRST operator $Q_t$ .

Step 2 — Montonen–Olive duality. The theory enjoys S-duality acting on the complexified coupling

6

\quad \tau = \frac{\theta}{2\pi} + \frac{4\pi i}{g^2},\qquad S:\ \tau \mapsto -\frac{1}{n_{\mathfrak g}\,\tau},\quad G \leftrightarrow {}^L G,

where $n_{\mathfrak g}$ is the lacing number. S exchanges electric and magnetic charges, hence Wilson and 't Hooft operators. Derivation hint: The Witten effective Lagrangian is modular in $\tau$ ; the full duality group is $SL(2,\mathbb Z)$ (or a congruence subgroup), and $S$ is the order-4 generator interchanging the weight and coweight lattices, which is precisely Langlands duality of root data.

Step 3 — Compactify on $X$ . Take $M = \Sigma \times X$ , shrink $X$ . The low-energy theory on $\Sigma$ is a 2d sigma model into the Hitchin moduli space

7

\quad \mathcal{M}_H(G,X)=\Big\{(A,\phi):\ F_A - \phi\wedge\phi = 0,\ d_A\phi=0,\ d_A\!\star\!\phi=0\Big\}/\mathcal{G},

a hyperkähler manifold of real dimension $4(g-1)\dim G$ . Derivation hint: The BPS equations of the twisted theory on $X$ are the Hitchin equations; their moduli space is the target. Hyperkähler structure arises because $\mathcal N=4$ has $SU(4)_R$ R-symmetry.

Step 4 — Twist becomes A/B-model. The parameter $t$ selects a complex structure in the $\mathbb{P}^1$ of the hyperkähler structure. At one extreme the twisted theory is the B-model in complex structure $J$ (where $\mathcal{M}_H \cong \mathrm{LocSys}_{G_{\mathbb C}}$ via the non-abelian Hodge / Hitchin correspondence); at another it is the A-model in the symplectic structure $\omega_K$ .

Step 5 — Mirror symmetry of Hitchin systems. Hitchin's map $\mathcal{M}_H(G)\to \mathcal{B}$ and $\mathcal{M}_H({}^L G)\to \mathcal{B}$ have the same base $\mathcal{B}$ , with dual abelian-variety fibers (SYZ fibration / Donagi–Pantev). S-duality acts as fiberwise T-duality:

8

\quad \text{A-model}\big(\mathcal{M}_H({}^L G)\big)\ \simeq\ \text{B-model}\big(\mathcal{M}_H(G)\big).

Derivation hint: Generic Hitchin fibers are dual abelian varieties (Jacobians vs. their duals); SYZ mirror symmetry is fiberwise T-duality on such torus fibrations, and S-duality of the 4d theory descends to exactly this T-duality.

Step 6 — Branes to categories. The B-model branes form $D^b\mathrm{Coh}(\mathrm{LocSys}_{{}^L G})$ ; the A-model branes (including coisotropic branes) generate a category equivalent to $D\text{-}\mathrm{mod}(\mathrm{Bun}_G)$ via the canonical coisotropic brane $\mathcal{B}_{cc}$ . Mirror symmetry [8] then yields the equivalence [1]. Wilson lines (electric) act on B-branes by tensoring with tautological bundles; 't Hooft lines (magnetic) act on A-branes as Hecke modifications — so eigenbranes of 't Hooft operators are Hecke eigensheaves. $\blacksquare$

∎

EW

Edward Witten

b. 1951

Witten, with Anton Kapustin, transformed the Geometric Langlands program in their 2006 paper "Electric-Magnetic Duality and the Geometric Langlands Program." Witten's prior work on Donaldson theory, topological field theory, mirror symmetry, and Seiberg–Witten theory provided the entire toolkit. The key conceptual move was recognizing that the ramified, quantum, and "real" forms of the correspondence all sit inside the structure of $\mathcal{N}=4$ gauge theory: surface operators give ramification, $\theta$ -angle gives the quantum geometric Langlands, and the choice of real form of $\tau$ gives the relationship to conformal field theory. Witten later (with Gukov, Frenkel, and others) extended the dictionary to the quantum/deformed setting and to the relation with the Bethe ansatz and integrable systems.

🧠 Intuition

Think of the classical Fourier transform: functions on a vector space $V$ correspond to functions on the dual $V^*$ , and the operation of multiplication by a coordinate on one side corresponds to differentiation on the other. Geometric Langlands is a vast nonabelian, nonlinear categorification of this. The role of $V$ is played by $\mathrm{Bun}_G$ and the role of $V^*$ by $\mathrm{LocSys}_{{}^L G}$ . "Functions" are replaced by sheaves/D-modules. The operation that gets diagonalized — like the exponentials $e^{i\langle x,\xi\rangle}$ that are eigenfunctions of differentiation — is the Hecke operator, and the eigenobjects are Hecke eigensheaves. The abelian case ( $G=GL_1$ ) really is a Fourier–Mukai transform between $\mathrm{Bun}_{GL_1}=\mathrm{Pic}(X)$ and its dual, making this analogy literal.

Figure

The Hitchin fibration and S-duality as fiberwise T-duality

```svg

```

ExampleAbelian Case:

G = GL_1

and Fourier–Mukai

Given: $X$ a curve of genus $g$ , $G=GL_1$ , so ${}^L G = GL_1$ as well. Then $\mathrm{Bun}_{GL_1}=\mathrm{Pic}(X)$ , the Picard variety, and $\mathrm{LocSys}_{GL_1}(X)= H^1(X,\mathbb{C}^*)$ -torsor of rank-1 local systems.

Steps: 1. The relevant transform is the Fourier–Mukai transform with kernel the Poincaré bundle $\mathcal{P}$ on $\mathrm{Pic}(X)\times \mathrm{Pic}^0(X)$ . 2. A rank-1 local system $E$ has monodromy data; the associated Hecke eigensheaf $\mathcal{F}_E$ on $\mathrm{Pic}(X)$ is the character D-module obtained by integrating $E$ via the Abel–Jacobi map. 3. The Hecke functor at $x$ tensors by the line $\mathcal{O}(x)$ ; the eigenvalue equation [3] becomes $H_x(\mathcal F_E)=E_x\otimes\mathcal F_E$ , with $E_x\in\mathbb C^*$ a scalar. 4. Count dimensions: $\dim\mathrm{Pic}^0(X)=g$ and $\dim\mathrm{LocSys}^0_{GL_1}=2g$ (real) $= g$ complex — they match the dual abelian varieties.

Answer:

\boxed{\ \mathcal{F}_E = \text{(Fourier–Mukai dual of the skyscraper } \mathcal O_{\{E\}}),\quad H_x(\mathcal F_E)= E_x\otimes \mathcal F_E\ }

This is unconditionally a theorem (Laumon, Rothstein), confirming the program in the abelian case.

ExampleEigensheaf count for

G=GL_2

,

X=\mathbb{P}^1

with ramification

Given: $G=GL_2$ , base curve $X=\mathbb P^1$ with tame ramification (regular singular points) at $\{0,1,\infty\}$ . We want the dimension of the space of Hecke eigensheaves corresponding to a rigid local system, i.e. a 2-dimensional representation $\rho:\pi_1(\mathbb P^1\setminus\{0,1,\infty\})\to GL_2(\mathbb C)$ .

Steps: 1. $\pi_1(\mathbb P^1\setminus\{0,1,\infty\})$ is free on 2 generators with relation $\gamma_0\gamma_1\gamma_\infty=1$ . 2. Count moduli: dimension of the character variety is

\dim = (2g-2+n)\dim G - \dim Z + \sum_i \dim \mathcal C_i,

with $g=0,\ n=3,\ \dim GL_2=4$ . For generic semisimple conjugacy classes $\dim\mathcal C_i=2$ . 3. Compute: $(2\cdot 0 - 2 + 3)\cdot 4 = 4$ , minus $\dim Z=1$ for the center acting, plus boundary terms... For a rigid (zero-dimensional moduli) configuration the local data are fixed and the eigensheaf is essentially unique up to twist. 4. The corresponding D-module on $\mathrm{Bun}_{GL_2}(\mathbb P^1, \{0,1,\infty\})$ is the hypergeometric D-module.

Answer:

\boxed{\ \#\{\text{eigensheaves}\}=1\ \text{(rigid case): the hypergeometric/Kloosterman sheaf}\ }

This matches the classical fact that hypergeometric local systems are rigid (Katz), giving the explicit ramified geometric Langlands correspondence on $\mathbb P^1$ .

⚠️ QCoh is not enough — singular support

A frequent error is to state the correspondence as $D\text{-}\mathrm{mod}(\mathrm{Bun}_G)\simeq \mathrm{QCoh}(\mathrm{LocSys}_{{}^L G})$ with naive quasi-coherent sheaves. This is false as stated. The stack $\mathrm{LocSys}_{{}^L G}$ is not smooth — it is a derived/quasi-smooth stack — and the correct spectral side is $\mathrm{IndCoh}$ with nilpotent singular support along the global nilpotent cone $\mathcal N$ (Arinkin–Gaitsgory). The discrepancy between $\mathrm{QCoh}$ and $\mathrm{IndCoh}_{\mathcal N}$ is exactly what makes the equivalence delicate and is invisible in the smooth/abelian case.

⚠️ Eigensheaves are not eigenvectors

"Hecke eigensheaf" sounds like an eigenvector but is a categorical statement: equation [3] is an isomorphism of sheaves depending continuously on $x\in X$ , i.e. a compatible family forming a Hecke eigen-property, not a single numerical eigenvalue. Treating it as ordinary spectral theory loses the local-system structure of the eigenvalue.

CounterexampleNaive function-correspondence fails over

\mathbb{C}

Over a finite field $\mathbb F_q$ , the function-sheaf dictionary (Grothendieck's faisceaux–fonctions) lets one extract numerical automorphic forms from sheaves by taking traces of Frobenius. Over $\mathbb C$ there is no Frobenius, so one cannot reduce the categorical equivalence to a numerical correspondence of functions on $\mathrm{Bun}_G(\mathbb F_q)$ . Attempting to "evaluate" a D-module to a function over $\mathbb C$ has no meaning; the correspondence is irreducibly categorical. This is why the geometric theory must be phrased with categories rather than spaces of functions.

connection

Geometric Langlands sits at a crossroads of many fields. Conformal Field Theory enters via the Beilinson–Drinfeld construction of Hecke eigensheaves from conformal blocks of the affine Kac–Moody algebra at the critical level $\kappa=-h^\vee$ , where the center of the enveloping algebra (Feigin–Frenkel center) is identified with functions on ${}^L\mathfrak g$ -opers. Mirror Symmetry provides the mechanism (Step 5) translating S-duality into the equivalence of A- and B-branes on Hitchin moduli. Hitchin Integrable System is the classical limit, with its abelian fibration giving the SYZ structure. The Affine Grassmannian and Geometric Satake Equivalence supply the local building blocks and the very definition of ${}^L G$ . The arithmetic Langlands Program is the motivating shadow: replacing $\mathbb F_q$ -points and Galois groups by curves and fundamental groups. Finally Topological Quantum Field Theory organizes everything as a 4d TQFT whose boundary/defect data reproduce the categories.

🔬 Where this appears

⏳ History

- 1967–70 — Langlands formulates his conjectures linking automorphic forms and Galois representations; the dual group

{}^L G

is born.

- 1980s — Drinfeld proves the geometric Langlands correspondence for

GL_2

over function fields, introducing the idea of realizing automorphic forms as sheaves on moduli of shtukas/bundles.

- 1987 — Laumon formulates the geometric Langlands conjecture in sheaf-theoretic terms for

GL_n

and proves the abelian (Fourier–Mukai) case with Rothstein.

- 1991–95 — Beilinson and Drinfeld construct Hecke eigensheaves via the Hitchin system, opers, and conformal blocks at the critical level; the "quantization of the Hitchin system" appears.

- 1995 (Lusztig), 1998–2000 — Ginzburg, and Mirković–Vilonen establish the Geometric Satake equivalence, giving

{}^L G

a purely geometric definition.

- 2002 — Frenkel–Gaitsgory–Vilonen prove the geometric Langlands correspondence for

GL_n

in the unramified de Rham setting (vanishing conjecture).

- 2006 — Kapustin and Witten reinterpret geometric Langlands through S-duality of

\mathcal{N}=4

gauge theory and mirror symmetry of Hitchin systems.

- 2007–15 — Arinkin–Gaitsgory formulate the precise categorical statement with nilpotent singular support; Gaitsgory and collaborators develop the full derived-categorical framework.

- 2024 — Gaitsgory, Raskin, and collaborators announce a complete proof of the (de Rham) geometric Langlands conjecture in a series of papers, a landmark culmination of the program.

open-problem

Several deep issues remain at the frontier. (i) The quantum geometric Langlands correspondence deforms by the level/coupling $\kappa$ ( $\theta$ -angle physically), relating $\kappa$ -twisted D-modules for $G$ to $\check\kappa$ -twisted D-modules for ${}^L G$ with $\kappa\check\kappa=1$ ; this is far less understood than the de Rham (critical) case. (ii) The Betti version (Ben-Zvi–Nadler), using local systems with topological rather than algebraic structure, and its compatibility with the de Rham version through Riemann–Hilbert, is still being developed. (iii) Ramification — wild ramification and the analogue of the local Langlands correspondence at irregular singular points (Stokes data, isomonodromy) is only partially worked out. (iv) The analytic Langlands program (Etingof–Frenkel–Kazhdan) over $\mathbb C$ seeks genuine spectral theory of Hecke operators on $L^2(\mathrm{Bun}_G)$ — a Hilbert-space realization whose foundations are recent. (v) The precise dictionary between the recently announced de Rham proof and the physical Kapustin–Witten picture (making mirror symmetry of Hitchin systems mathematically rigorous, including singular fibers and gerbes) remains an active and only partly completed program.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also