Quantum geometric Langlands at every irrational level

← All digests

Family 069 · Algebraic and complex geometry
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-12

1. The question

Fourier transformation lets the same problem be described in two different languages. Functions become frequency data, while operations that are complicated on one side can become simple on the other. Quantum geometric Langlands predicts a much more elaborate kind of dual description: whole categories of geometric differential equations for one symmetry group should correspond to those for a dual group.

The parameter controlling the differential equations is called the level. Under the proposed duality it changes to a negative reciprocal, with a correction determined by the root system. The geometry on both sides lives over the same complex algebraic curve, but the bundles being parametrized belong to different groups.

This manuscript claims the global unramified correspondence at every irrational complex level, including non-real levels. The conclusion is an equivalence of full categories, with compatibility under the local operations that identify it as the intended Langlands correspondence.

2. Background and essential definitions

Bundles, their parameter space, and the dual group

Let $X$ be a smooth connected projective curve over $\mathbb C$. Analytically, it is a compact Riemann surface. A principal $G$-bundle is a geometric object locally modeled on $G$, with transition functions taking values in $G$. For matrix groups, familiar vector bundles with additional structure provide concrete examples.

The notation $\operatorname{Bun}_G(X)$ denotes the moduli stack of such bundles. A moduli space parametrizes objects; a stack also remembers their automorphisms. This distinction matters because different bundles can have different symmetry groups, and those symmetries affect differential equations on the parameter space.

A connected simple complex algebraic group has a root system encoding its infinitesimal symmetries. Its full root datum includes additional integer lattices describing characters and one-parameter subgroups. Exchanging roots with coroots and the corresponding lattices defines the Langlands dual $G^\vee$.

For example, $\mathrm{SL}_2$ and $\mathrm{PGL}_2$ are dual. They share closely related infinitesimal geometry, but one has a nontrivial finite center and the other has a different global topology. Keeping only the Lie algebra would discard information that the theorem explicitly retains.

Categories of differential equations

On a smooth space, a $D$-module is a sheaf on which functions and differential operators act compatibly. Locally, an equation such as $Pf=0$, with $P$ a differential operator, can be encoded by a module over the ring of differential operators. This makes systems of equations amenable to algebraic operations.

A category records both objects and maps between them. An equivalence of categories preserves all mapping information and reaches every target object up to isomorphism; it is much stronger than matching a list of dimensions or numerical invariants.

The manuscript uses derived differential graded, or DG, categories. Their objects and mapping spaces are organized by complexes, retaining higher cohomological information. “Full” means all twisted $D$-modules in the specified cocomplete category, rather than only a convenient finite or special subcategory. Cocompleteness allows constructions such as arbitrary direct sums and the limits of increasing diagrams called colimits.

The level and its normalization

A twisted $D$-module uses differential operators glued with an additional line-bundle-dependent correction. A complex exponent specifies that correction; an actual fractional root of the line bundle is not required.

For a principal bundle $E$, let $\mathfrak g_E$ be its associated adjoint vector bundle. The determinant line used here is

$$\mathcal L_G|_E= \det H^0(X,\mathfrak g_E)\otimes \det H^1(X,\mathfrak g_E)^{-1}.$$

The determinant of a vector space is its top exterior power. Thus this construction turns two cohomology groups into one line varying over the bundle stack.

Write $h^\vee$ for the dual Coxeter number, a positive integer attached to the root system. The manuscript’s shifted-level convention is

$$D_c(\operatorname{Bun}_G(X)) =D\!\left(\mathcal L_G^{(c-h^\vee)/(2h^\vee)}\right).$$

The formula defines the notation on the left. It is needed because an unshifted parameter would give a different-looking reciprocal rule.

3. Development history and previous partial results

Ordinary geometric Langlands relates geometric differential equations on a bundle stack to a spectral side built from local systems for the dual group. The quantum version deforms that picture and compares two twisted bundle-side categories.

An early inverse-parameter formulation appears in Stoyanovsky’s account, which attributes the quantum deformation to Drinfeld. Kapustin and Witten developed the gauge-theoretic interpretation through electric–magnetic duality. Gaitsgory’s quantum Langlands notes organized the mathematical correspondence using local categories, localization and Whittaker coefficients.

Important local ingredients were established separately. Feigin–Frenkel duality identifies principal $W$-algebras at inverse shifted levels. These algebras arise by reducing representations of loop Lie algebras, which describe symmetries depending on a formal local coordinate. Arakawa–Frenkel and Raskin developed representation-theoretic and categorical forms of that reduction. Campbell, Dhillon and Raskin proved fundamental local equivalences, supplying the pointwise comparisons used here in the irrational range.

The ordinary geometric Langlands theorem was proved in a five-paper collaboration, with the final multiplicity-one paper by Gaitsgory and Raskin. The present manuscript uses that body of work as an input for global cohomological control. It is not a replacement proof of the ordinary theorem.

Two recent predecessors are especially close. Bogdanova’s quantum Whittaker nonvanishing paper proves a detection result for cuspidal twisted objects for adjoint groups. Her quantum Betti Langlands paper constructs a functor in the topological, or Betti, setting with central compatibility. The current claim is a full global equivalence in the algebraic de Rham setting, not merely the construction of a candidate functor or a restatement of the Betti result.

The remaining bridge is substantial: pointwise local equivalences must fit into families, survive collisions of points and reconstruct all global objects. These are the manuscript’s two principal areas of claimed new work.

4. What the new manuscripts claim

Let $G$ be any connected simple complex algebraic group, with its given global form, and let $X$ be any smooth connected projective complex curve. Let $r$ be the lacing number: the ratio of squared long-root to short-root lengths, equal to $1$, $2$, or $3$.

For every $c\in\mathbb C\setminus\mathbb Q$, the manuscript claims an equivalence

$$D_c(\operatorname{Bun}_G(X)) \simeq D_{-1/(rc)}(\operatorname{Bun}_{G^\vee}(X)).$$

Each side uses its own group’s determinant line and dual Coxeter number in the convention above. When roots have one length, $r=1$; for $\mathrm{SL}_2$, the dual parameter is simply $-1/c$.

“Irrational” here means outside the rational numbers as a subset of $\mathbb C$. It includes every non-real complex number, and real algebraic irrational numbers as well as transcendental ones. The claim has no additional unspecified exceptional irrational levels.

The equivalence includes all connected components of the bundle stacks and the full twisted categories. It is compatible with localization, Whittaker coefficients, moving marked points, their collisions and insertion of the trivial local object. Fixed pinnings—choices normalizing root directions—and a theta characteristic, a square root of the canonical bundle of the curve, set the construction’s normalizations.

The global form is reflected in a concrete exchange. Source objects have a central character $\eta$, describing how the center acts, and a component label $p$ in the algebraic fundamental group. Under the root-datum identifications, the manuscript gives

$$(\eta,p)\longmapsto(p,-\eta).$$

It also claims compatibility between twisting bundles by central torsors and tensoring by the corresponding dual local systems. A torsor is a bundle with the specified group acting freely and transitively on each fiber. These compatibilities retain the finite central data that would disappear on replacing every group by its simply connected or adjoint form.

5. Why the advance matters

Importance: 5/5. A full global quantum geometric Langlands equivalence in this parameter range would resolve a central part of a major duality program linking algebraic geometry, representation theory and mathematical physics.

Its force comes from the combination of breadth and compatibility. It treats every connected simple complex group and every irrational complex level, and it identifies the functor through the operations used to build and analyze geometric objects. An abstract resemblance between categories would provide less usable information.

The rating reflects the significance of the stated result if correct. Rational levels, ramification and other extensions remain outside the headline, so the claim is not the completion of every form of quantum Langlands.

6. The main idea

Whittaker coefficients play the role of a transform

A Whittaker coefficient extracts information by averaging along unipotent symmetries with a chosen nontrivial character. Unipotent matrices have all eigenvalues equal to one; upper triangular matrices with ones on the diagonal provide an elementary model. The character supplies an oscillatory or exponential factor, making the operation resemble Fourier analysis.

A Poincaré construction is meant to reconstruct global objects from those coefficients. The manuscript claims an identity schematically of the form

$$J_sR_s\simeq\operatorname{Id}\qquad(s\notin\mathbb Q),$$

where $R_s$ takes Whittaker coefficients and $J_s$ is the suitably normalized reconstruction functor. This holds on the entire twisted category. At the ordinary level a residual subcategory complicates the analogous statement, so reconstruction at irrational level requires additional work.

Why irrationality matters

For a determinant exponent $s$, the local twisting connection has monodromy

$$q=\exp(2\pi\sqrt{-1}\,s).$$

Monodromy records how a local solution changes after going around a loop. When $s\notin\mathbb Q$, $q$ is not a root of unity. This separates the theorem’s range from the special torsion behavior of rational exponents.

The manuscript approximates its global kernel calculations by finite diagrams. For these finite pieces, quasi-unipotent monodromy makes relevant cohomology and map ranks constant away from roots of unity. But finite-piece constancy alone does not control an infinite colimit. The additional claimed mechanism combines logarithmic deformations near $q=1$ with a uniform cohomological weight bound, using ordinary geometric Langlands. It shows that each potential error is killed at a finite later stage, and transports that vanishing to non-torsion $q$.

A finite Whittaker test detects whether an error object remains. The resulting vanishing supplies the reconstruction identity. Irrational twisting also removes sufficiently unstable bundle directions, allowing the categorical duality arguments to be controlled on bounded geometric regions.

Make local comparisons survive collisions

Local representation theory gives a correspondence near one marked point of $X$. To globalize, one must allow any finite collection of points, move them and let them collide. Data defined only for distinct points can fail to extend coherently when two points meet.

The manuscript sends the local categories for the two dual groups into a common category of $W$-algebra modules. Low-degree cohomological estimates recover maps between objects from this common target. It then uses these recovered maps to identify the boundary data across collision divisors. Separate arguments retain compatibility when objects are inserted in both slots of the pairing.

Finally, localization turns loop-algebra representations into twisted $D$-modules on the bundle stack. The Ran space organizes all finite subsets of the curve at once; integration over it removes the auxiliary marking choices. A global unipotent-period identity makes the local pairings compatible with localization. The two Whittaker reconstruction identities, one for each group, provide inverse functors and hence the claimed equivalence.

7. Scope, limitations, and open questions

The theorem is unramified: the bundle categories carry no prescribed singular or parabolic data at a fixed set of marked points. Moving auxiliary points in the construction is not the same as proving a ramified Langlands theorem.

The level restriction is exact. Rational levels, including special values with additional representation-theoretic phenomena, are not obtained by continuity from the irrational result. The proof’s non-torsion monodromy mechanism is precisely what distinguishes the two regimes.

The headline concerns connected simple complex algebraic groups and de Rham twisted $D$-modules over complex projective curves. Reductive groups enter intermediate constructions, but the digest does not enlarge the final theorem to arbitrary reductive groups or to positive characteristic.

The categories contain much more information than their simplest objects. A categorical equivalence does not automatically provide an explicit elementary formula for the image of every differential equation, and the compatibility statements are part of what makes the proposed correspondence identifiable. Extending these constructions to rational or ramified settings would require additional arguments addressing their different local behavior.

8. Sources and further reading

For the program’s formulation, see Gaitsgory’s quantum Langlands notes. Kapustin–Witten explains the physical duality motivation. The fundamental local equivalences, ordinary geometric Langlands series, and Bogdanova’s Betti functor locate the main preceding results. The source below specifies the level convention, family comparisons and central data used in the claimed global equivalence.

Source manuscripts

• M1. OpenAI, Global quantum geometric Langlands at irrational level (October 4, 2026).

Sources consulted on 12 October 2026 at repository revision fd4aeeb. Download the manuscript citations.