Finite Langlands Correspondence

• Finite Langlands correspondence is a framework linking irreducible representations of finite reductive groups with enhanced Weil–Deligne parameters that are Frobenius-semisimple and special.
• It employs dual-group methods and Lusztig’s canonical quotient to form finite packets, with fibers parametrized by associated finite enhancement groups.
• The theory bridges finite-field character theory and local Langlands, integrating concepts like Jordan decomposition, theta correspondence, and finite-packet finiteness.

Finite Langlands correspondence denotes a family of closely related parametrization problems at the boundary between finite groups of Lie type, dual-group methods, and local Langlands theory. In one now explicit sense, it is a correspondence for connected reductive groups over a finite field $k=\mathbf F_q$, sending irreducible $\mathbf Q_\ell$-representations of $G(k)$ to special Frobenius-semisimple Weil–Deligne parameters $\varphi:\mathrm{WD}_k\to {}^LG$, with fibers parametrized by irreducible representations of a finite enhancement group $A_\varphi$. In a second, older sense, the phrase refers to the requirement that local Langlands fibers—the $L$-packets $\Pi_\phi$, or the fibers of semisimple parameter maps such as $\mathrm{CSS}$—should be finite and non-empty, so that the correspondence is finite on each parameter fiber (Imai, 20 Aug 2025, Harris, 2022).

1. Scope of the term and neighboring theories

The expression is potentially ambiguous because several adjacent theories involve finite fields without being a finite-field Langlands correspondence in the strict sense. The finite-field theory concerns the finite group of rational points $G(k)$ and parameters built from the finite-field Weil or Weil–Deligne group. By contrast, local Langlands over a local function field $F$ concerns smooth representations of $\mathbf Q_\ell$0, where $\mathbf Q_\ell$1 is a field such as $\mathbf Q_\ell$2, and parameters for the Weil group $\mathbf Q_\ell$3 or Weil–Deligne group $\mathbf Q_\ell$4. Li–Huerta’s work on $\mathbf Q_\ell$5 is of this local-field type: it constructs a map

$\mathbf Q_\ell$6

and proves that it is the usual local Langlands correspondence after forgetting monodromy (Li-Huerta, 2021).

A second neighboring theory is geometric Langlands over a finite field. In the explicit $\mathbf Q_\ell$7 case on $\mathbf Q_\ell$8, the objects are local systems, Hecke eigensheaves, and moduli stacks of parabolic bundles; the automorphic side is not the finite group $\mathbf Q_\ell$9, but a stack of bundles equipped with Hecke correspondences. This setting is therefore a geometric or function-field incarnation over $G(k)$0, not a correspondence for finite reductive groups in the sense of $G(k)$1 versus finite-field $G(k)$2-parameters (Bos, 2019).

Setting	Representation-theoretic side	Parameter side
Finite field $G(k)$3	$G(k)$4	$G(k)$5
Local function field $G(k)$6	Irreducible smooth representations of $G(k)$7	$G(k)$8- or $G(k)$9-parameters
Geometric Langlands over $\varphi:\mathrm{WD}_k\to {}^LG$0	Hecke eigensheaves, cusp forms on moduli	Local systems on punctured curves

This distinction matters conceptually. A finite Langlands correspondence in the strict sense is neither the usual local Langlands correspondence for $\varphi:\mathrm{WD}_k\to {}^LG$1 nor the sheaf-theoretic geometric Langlands correspondence over $\varphi:\mathrm{WD}_k\to {}^LG$2. A plausible implication is that the same dual-group vocabulary persists across all three settings, but the objects being parameterized and the meaning of packets differ substantially.

2. Finite-field parameters and enhancement data

For a finite field $\varphi:\mathrm{WD}_k\to {}^LG$3, the finite-field Weil and Weil–Deligne groups are defined by

$\varphi:\mathrm{WD}_k\to {}^LG$4

where conjugation by $\varphi:\mathrm{WD}_k\to {}^LG$5 acts on $\varphi:\mathrm{WD}_k\to {}^LG$6 by the $\varphi:\mathrm{WD}_k\to {}^LG$7-power map and on $\varphi:\mathrm{WD}_k\to {}^LG$8 by multiplication by $\varphi:\mathrm{WD}_k\to {}^LG$9 on the $A_\varphi$0-component. For a connected reductive group $A_\varphi$1 over $A_\varphi$2, an $A_\varphi$3-parameter of Weil–Deligne type is a morphism

$A_\varphi$4

compatible with projection to $A_\varphi$5. The parameters relevant to the finite correspondence are required to be Frobenius semisimple, meaning that $A_\varphi$6 is semisimple, and special, meaning that $A_\varphi$7 is a special unipotent element of $A_\varphi$8 (Imai, 20 Aug 2025).

A central structural feature is that equivalence is not naive $A_\varphi$9-conjugacy. Writing

$L$0

one first forms

$L$1

then Lusztig’s canonical quotient $L$2, and then an extension

$L$3

with

$L$4

The Frobenius element $L$5 determines an element

$L$6

and two parameters are equivalent only if, after conjugating $L$7, the induced bijection on the corresponding $L$8-groups carries $L$9 to the matching Frobenius class. The finite packet group is then

$\Pi_\phi$0

This refinement is decisive. The packet is not indexed merely by a component group of a centralizer, but by the centralizer of Frobenius inside a canonical quotient attached to $\Pi_\phi$1. A common oversimplification is to identify finite-field parameters with semisimple conjugacy classes alone; the finite correspondence instead retains inertial data, unipotent data, and Frobenius-centralizer enhancement simultaneously (Imai, 20 Aug 2025).

3. The finite correspondence for $\Pi_\phi$2

After fixing a Whittaker datum, the principal theorem asserts the existence of a natural map

$\Pi_\phi$3

and for every $\Pi_\phi$4 a natural bijection

$\Pi_\phi$5

Thus the finite Langlands correspondence is not a plain bijection between irreducible representations and bare parameters. It is a map to special parameters whose fibers are finite $\Pi_\phi$6-packets

$\Pi_\phi$7

This is the finite-field analogue of enhanced local Langlands parametrization rather than of the older bare-parameter formulation (Imai, 20 Aug 2025).

The construction is underwritten by a categorical decomposition. For each semisimple parameter block and two-sided cell $\Pi_\phi$8, Lusztig associates a finite group $\Pi_\phi$9, finite sets $\mathrm{CSS}$0, stabilizer groups $\mathrm{CSS}$1, and automorphisms $\mathrm{CSS}$2, leading to a natural equivalence

$\mathrm{CSS}$3

Here $\mathrm{CSS}$4 is the category of $\mathrm{CSS}$5-equivariant sheaves on the finite set $\mathrm{CSS}$6 for the $\mathrm{CSS}$7-twisted conjugation action. The categorical statement is the mechanism behind the packet theorem: irreducible objects in these summands are converted into dual-group data, first as $\mathrm{CSS}$8-type parameters and then, via the Jacobson–Morozov comparison,

$\mathrm{CSS}$9

The finite correspondence is built to recover and rigidify established finite-group character theory. The decomposition

$G(k)$0

recovers the organization by Lusztig series, while the theorem

$G(k)$1

gives a dual-group parametrization by $G(k)$2-stable special conjugacy classes in the dual group. The claim made for this formulation is not merely restatement but canonical rigidification, especially when the center is disconnected. The theory applies in particular to tori, where packets are singletons, and to $G(k)$3, where the finite correspondence is described as essentially due to Macdonald (Imai, 20 Aug 2025).

4. Jordan decomposition, canonicity, and finite relative duality

The finite correspondence sits inside the older framework of Deligne–Lusztig theory and Lusztig’s Jordan decomposition. For a finite classical group $G(k)$4, Lusztig’s partition

$G(k)$5

organizes irreducibles by semisimple conjugacy classes in the dual finite group $G(k)$6, and Jordan decomposition identifies each Lusztig series with unipotent representations of the finite centralizer $G(k)$7. The nontrivial point is that this decomposition is not canonically normalized in general. For symplectic and orthogonal groups, ambiguity remains because different irreducibles can share the same uniform projection and the defining Deligne–Lusztig character identities do not determine a unique correspondence (Wang, 2024).

For classical groups, a canonical choice $G(k)$8 has been constructed under the assumption that $G(k)$9 is large enough for Srinivasan’s result to apply. It is uniquely characterized by simultaneous compatibility with parabolic induction and with theta correspondence for both $F$0 and $F$1. In this normalization, theta correspondence becomes factorwise on Jordan parameters, and the refined combinatorics of Lusztig symbols, partition data, and sign characters become functorial rather than choice-dependent (Wang, 2024).

An important application is a finite-field instance of relative Langlands duality in the sense of Ben-Zvi–Sakellaridis–Venkatesh. The paper proves commutative diagrams relating theta correspondence, Alvis–Curtis duality, the canonical Jordan decomposition, and the finite Gan–Gross–Prasad problem. This does not produce a new finite Langlands correspondence in the full sense of local or global Langlands; rather, it shows that once a canonical finite Jordan decomposition is fixed, two different relative representation-theoretic constructions—theta lifting and GGP branching—become dual manifestations of the same spectral organization (Wang, 2024).

A persistent misconception is that finite Langlands over finite fields can be reduced to semisimple dual-group labels alone. The classical-group results show that this is inadequate: compatibility with parabolic induction, theta correspondence, and branching requires a canonical treatment of the unipotent-centralizer factor as well.

5. Finite packets in local Langlands

A second major use of the phrase concerns finiteness of local Langlands fibers. For a connected reductive group $F$2 over a non-archimedean local field $F$3, the conjectural local correspondence is a map

$F$4

with

$F$5

In this formulation, finiteness and non-emptiness are not ancillary properties but part of the expected canonical correspondence itself. Harris isolates the corresponding issue for semisimple parameterizations, such as Genestier–Lafforgue in positive characteristic and Fargues–Scholze over $F$6-adic fields, where one has only a canonical semisimple map

$F$7

and one does not know in general whether the fiber

$F$8

is finite or non-empty (Harris, 2022).

The semisimple/full distinction is essential. The semisimplification forgets the nilpotent operator $F$9 of the Weil–Deligne parameter, and Harris emphasizes that semisimple parameter maps do not by themselves solve the local Langlands correspondence in the classical sense. For $\mathbf Q_\ell$00, by contrast, the local correspondence is known, and in the function-field setting Li–Huerta construct a semisimple Weil parameter $\mathbf Q_\ell$01 and prove that it is the usual local Langlands correspondence after forgetting monodromy; in that case the full $\mathbf Q_\ell$02 theory supplies the benchmark against which more general groups are measured (Li-Huerta, 2021).

The first unconditional finiteness result in Harris’s program concerns pure unramified parameters in positive characteristic. Under hypotheses including $\mathbf Q_\ell$03 of positive characteristic, the Genestier–Lafforgue parameterization, purity, and in the corollary the split semisimple assumption with $\mathbf Q_\ell$04, Harris combines Theorem 8.1 and Fintzen’s exhaustion theorem to deduce that if $\mathbf Q_\ell$05 is pure irreducible with unramified semisimple parameter, then $\mathbf Q_\ell$06 is a constituent of an unramified principal series representation. This gives finiteness of the fiber in that special case, because the packet consists only of constituents of unramified principal series (Harris, 2022).

The general strategy remains conditional. It proceeds by reducing to irreducible semisimple parameters, globalizing via potential automorphy, passing up solvable towers so that the global parameter becomes everywhere unramified, invoking the conjecture that there are no pure incorrigible supercuspidal representations, and then descending by cyclic base change and endoscopy to reconstruct the original packet. This suggests a local notion of “finite Langlands correspondence” in which the central problem is not construction of a parameter map alone, but proof that its fibers are finite and non-empty.

6. Finite-field shadows of $\mathbf Q_\ell$07-adic local Langlands

The finite theory also appears as a reduction or bookkeeping device for $\mathbf Q_\ell$08-adic local Langlands. For $\mathbf Q_\ell$09, a conjecture formulated by Vogan is established in the form of a surjection

$\mathbf Q_\ell$10

from irreducible representations of the finite group $\mathbf Q_\ell$11 to inertia-equivalence classes of tame Langlands parameters for $\mathbf Q_\ell$12. The fibers are described by a finite abelian component group: $\mathbf Q_\ell$13 acts simply transitively on the fiber over $\mathbf Q_\ell$14. This is not a full finite-field Langlands correspondence for $\mathbf Q_\ell$15, but a finite-field reduction of the depth-zero/tame part of the $\mathbf Q_\ell$16-adic theory, with packet structure explicitly mirroring Vogan-style enhancement (Collacciani, 15 Jan 2025).

A deeper depth-zero construction uses finite reductive quotients directly. For a vertex $\mathbf Q_\ell$17 in the Bruhat–Tits building, one forms

$\mathbf Q_\ell$18

where the full quotient $\mathbf Q_\ell$19 may be disconnected. A depth-zero supercuspidal representation contains a finite representation

$\mathbf Q_\ell$20

lying in a Lusztig series $\mathbf Q_\ell$21 determined by the reduced toral character. Mishra uses a pinned Jordan decomposition for possibly disconnected finite reductive quotients,

$\mathbf Q_\ell$22

followed by pinned unipotent duality, to turn $\mathbf Q_\ell$23 into a cuspidal unipotent label on a dual centralizer. Combined with the toral LLC for maximally unramified elliptic tori and the Feng–Opdam–Solleveld correspondence, this yields a pinning-normalized bijection

$\mathbf Q_\ell$24

for depth-zero supercuspidal representations. In this setting the finite Jordan decomposition is reversible, so the finite quotient data are not auxiliary but the mechanism making the correspondence invertible (Mishra, 6 May 2026).

A third avatar appears modulo $\mathbf Q_\ell$25. Dat shows that the classical Jacquet–Langlands bijection does not behave well under reduction mod $\mathbf Q_\ell$26, but the Langlands–Jacquet transfer on Grothendieck groups does. There is a unique homomorphism

$\mathbf Q_\ell$27

compatible with reduction and characterized by Brauer character transfer on elliptic $\mathbf Q_\ell$28-regular elements. From this one extracts a bijection between irreducible $\mathbf Q_\ell$29-representations of $\mathbf Q_\ell$30 and super-Speh $\mathbf Q_\ell$31-representations of $\mathbf Q_\ell$32, with the Zelevinsky involution playing the normalizing role. This again is not a finite-field Langlands correspondence for $\mathbf Q_\ell$33, but a modular shadow of local Langlands/Jacquet–Langlands in which finite-type and packet phenomena are indispensable (Dat et al., 2010).

Taken together, these developments show that “finite Langlands correspondence” now has a precise finite-field meaning, but also a broader methodological meaning. In the strict sense it refers to a packet-valued correspondence for $\mathbf Q_\ell$34 using special Weil–Deligne parameters and enhancement groups. In the broader local sense it refers to the finiteness of local parameter fibers and to the finite reductive data through which depth-zero, tame, or modular local Langlands is organized. What remains open is substantial: the finite-field theory presently lands only in special parameters, and the local finite-packet problem is unresolved beyond special cases and conditional strategies (Imai, 20 Aug 2025, Harris, 2022).