Arthur parameters and geometric Langlands over finite fields

← All digests

Family 014 · Number theory
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-10

1. The question

Many problems in number theory ask how large an arithmetic eigenvalue can be. An eigenvalue measures how a function changes under an averaging operation; unusually large values signal extra structure. The Ramanujan conjectures predict sharp bounds for important classes of these functions. Arthur’s framework goes further: when the simplest bound fails, it predicts an organized algebraic reason for that failure.

Family 014 studies this picture over function fields, such as the field of rational functions over a finite field. Its eight manuscripts connect numerical eigenvalue bounds with a geometric correspondence between two kinds of objects: bundles on a curve and representations describing transport around that curve.

There are several questions here. Can the same algebraic pattern explain eigenvalues at every place? Can those local patterns be assembled into one global Arthur parameter? Does the geometric correspondence reach every allowed parameter? Can it produce well-behaved eigensheaves with specified ramification? The manuscripts address these questions in different, explicitly delimited settings.

2. Background and essential definitions

From a curve to arithmetic eigenvalues

Let $X$ be a smooth projective geometrically connected curve over the finite field $\mathbb F_q$, and let $F=\mathbb F_q(X)$ be its field of rational functions. A closed point $x$ plays the role of a prime; its residue field has size $q_x=q^{d_x}$, where $d_x$ is its degree.

A group $G$, such as $\mathrm{SL}_n$, specifies the symmetries under consideration. A principal bundle is a family locally modeled on this group, with compatible transition functions. Automorphic functions can be described as functions on isomorphism classes of bundles. A Hecke operator averages over prescribed modifications of a bundle at a point. Simultaneous eigenfunctions of these operators are the arithmetic counterpart of modes in Fourier analysis.

A cuspidal function has vanishing constant terms along every proper parabolic subgroup. Parabolic subgroups generalize block upper-triangular matrices; their constant terms detect contributions induced from smaller groups. Cuspidality removes these induced constant terms.

A level divisor $D$ records extra data at finitely many points. Full level means a trivialization on the entire finite subscheme $D$, including its multiplicities. A representation is unramified at a point when it has vectors fixed by the usual integral maximal compact subgroup there. At such a point, its Hecke eigenvalues are encoded by a Satake parameter, a semisimple conjugacy class in the Langlands dual group $\widehat G$. This dual group exchanges the root and coroot data of $G$.

Temperedness and the extra SL2

For a unitary local representation, temperedness means weak containment in the regular representation: its matrix coefficients have the decay expected from ordinary harmonic analysis. At an unramified point, the normalized Satake parameter is then conjugate into a compact subgroup of $\widehat G(\mathbb C)$.

Arthur’s extra ingredient is an algebraic homomorphism

$$\phi:\mathrm{SL}_2\longrightarrow\widehat G.$$

Its standard nilpotent matrix maps to a nilpotent element of the dual Lie algebra. A matrix is nilpotent when some power is zero; a nilpotent orbit groups such elements under conjugation. The homomorphism contributes a controlled size factor

$$\phi\!\begin{pmatrix}q_x^{1/2}&0\\0&q_x^{-1/2}\end{pmatrix}.$$

Thus an unramified parameter can have a compact part multiplied by an explicitly structured noncompact part. The zero nilpotent orbit corresponds to a trivial extra factor and hence to temperedness.

From functions to sheaves

A local system assigns vector spaces with consistent transport along paths, interpreted algebraically through the étale fundamental group. A dual-group local system assigns such a system to every representation of $\widehat G$, compatibly with tensor products. A sheaf on the space of bundles retains geometric information that a numerical function forgets. Over a finite field, taking traces of Frobenius on a suitable sheaf recovers a function.

A Hecke eigensheaf $\mathcal M$ with eigenvalue $\rho$ satisfies, schematically,

$$\mathsf H_V(\mathcal M)\simeq\mathcal M\boxtimes V_\rho.$$

Here $V$ is a dual-group representation, $V_\rho$ its local system, and $\boxtimes$ means exterior product. The identity says that modifying bundles has a prescribed effect determined by transport on the curve. With several moving modification points, these identities must remain compatible when points collide; this is fusion compatibility.

A perverse sheaf is a complex satisfying dimension-adjusted support and dual-support conditions. Local constructibility means finite-dimensional, stratified behavior on every finite-type chart. Neither phrase simply means “an arbitrary sheaf.” Singular support records cotangent directions where a sheaf fails to vary smoothly; a nilpotent singular-support condition restricts those directions through nilpotent Higgs fields, which are bundle-valued differential forms.

3. Development history and previous partial results

Drinfeld’s work for $\mathrm{GL}_2$, and Laurent Lafforgue’s work for $\mathrm{GL}_n$, established the function-field Langlands correspondence in those settings and the purity underlying Ramanujan bounds. Vincent Lafforgue later constructed global parameters for cuspidal representations of general reductive groups using excursion operators, which compare several fundamental-group elements simultaneously. These results are the starting point of M1, M3 and M4, rather than consequences first introduced here.

Arthur’s framework explains why non-tempered discrete automorphic representations should involve an additional $\mathrm{SL}_2$. Earlier function-field progress outside general linear groups required extra local assumptions. Ciubotaru and Harris, for example, obtained unramified temperedness from hypotheses including a tempered component and a generic unramified component. Their work also relates the possible spectral behavior to nilpotent classes. See their primary paper.

The geometric route developed through Drinfeld, Laumon, Frenkel–Gaitsgory–Vilonen and Gaitsgory’s constructions of eigensheaves, followed by categorical formulations for general groups. The characteristic-zero geometric Langlands theorem is an established input here. Gaitsgory and Raskin subsequently obtained a positive-characteristic restricted equivalence onto a union of spectral components; whether every component occurred remained a separate question. Their positive-characteristic paper explains this distinction.

Gaitsgory–Lafforgue–Raskin also formulated rational and $\ell$-adic Arthur decompositions of automorphic functions. Raskin’s ICM account describes related unramified progress, including global enhancement under its hypotheses. Accordingly, this family’s contribution should be understood through the particular extensions it states—full finite level, rational support, ramified temperedness and prescribed tame eigensheaves—not as the beginning of these subjects.

4. What the new manuscripts claim

M3 and M1: one orbit, then one global parameter

For split connected semisimple $G$, M3 claims a decomposition at every full finite level:

$$C_{D,\mathbb Q}=\bigoplus_{\mathcal O}C_{D,\mathcal O,\mathbb Q}.$$

Here $C_{D,\mathbb Q}$ is the finite-dimensional rational cusp space, and the sum runs over dual nilpotent orbits. The corresponding decomposition holds after extending coefficients to $\overline{\mathbb Q}_\ell$, for $\ell\ne\operatorname{char}\mathbb F_q$. Every eigenvalue in a summand has the same orbit factor at every unramified point, and its residual compactness holds under every complex embedding. Divisor multiplicities are unrestricted.

For split connected adjoint absolutely simple groups, M3 further claims that genericity at one unramified place implies temperedness at all unramified places. Genericity means having a nonzero Whittaker functional, a nondegenerate Fourier coefficient along a maximal unipotent subgroup. Global genericity supplies this local condition.

M1 takes the finite-level decomposition as an explicit input. For each occurring excursion parameter $\sigma$, it claims a global homomorphism $\phi$ and a commuting Weil-group representation $\tau$ such that, after one simultaneous conjugation,

$$\sigma(w)=\tau(w)\phi\!\begin{pmatrix}a^{\deg w}&0\\0&a^{-\deg w}\end{pmatrix}, \qquad a=q^{1/2}.$$

The square root is interpreted in the chosen $\ell$-adic coefficient field. The Weil group consists of fundamental-group elements with integral Frobenius degree. The formula holds for all its elements, including inertia. The representation $\tau$ lands in the full centralizer of $\phi(\mathrm{SL}_2)$, which can be disconnected, and its good Frobenius eigenvalues have weight zero: all complex absolute values equal one. This is stronger than choosing unrelated local factorizations.

M4: the ramified exceptional places

M4 claims that every local component of a globally generic cuspidal representation is tempered for split connected adjoint groups of types

$$G_2,\ F_4,\ E_6,\ E_7,\ E_8.$$

Its statement imposes no characteristic or ramification-depth restriction. It uses M3 for the unramified places and supplies a separate argument for the ramified ones, where monodromy information is essential.

M2 and M5: rationality and full spectral support

Several geometric statements share four characteristic hypotheses. In Lie-algebra language these require: a nondegenerate invariant symmetric form remaining nondegenerate on the center of every Levi Lie algebra; Chevalley restriction for every Levi; semisimple centralizers that are Levi subgroups; and, over every field extension, containment of each nilpotent element in the Lie algebra of a parabolic’s unipotent radical. A Levi is the block-diagonal-type reductive part of a parabolic. Chevalley restriction says that invariant polynomial functions are detected on a maximal torus. These conditions are part of the results, not optional conveniences.

Under these hypotheses, M2 claims rationality of every canonical unramified Arthur-filtration step on all finitely supported automorphic functions, including noncuspidal ones. If $A_E$ is this function space with $E=\overline{\mathbb Q}_\ell$, and $F_Y$ is the step associated with a closed invariant union $Y$ of nilpotent orbits, then

$$F_YA_E=E\otimes_{\mathbb Q}(A_{\mathbb Q}\cap F_YA_E).$$

The filtration therefore comes from rational functions and is independent of the coefficient prime in the stated sense. No Hecke-finiteness or rational-point assumption is imposed. On cusp spaces it agrees with the appropriate orbit sums from M3 in their common scope.

M5 claims that the restricted geometric Langlands equivalence reaches every spectral component. It treats connected reductive groups under the four hypotheses over an algebraic closure of a finite field. For an arbitrary algebraically closed base, it additionally assumes that the characteristic is very good for the group and does not divide its Weyl-group order. “Very good” is a root-system restriction; for example, type $A_r$ excludes primes dividing $r+1$, while type $E_8$ excludes $2,3,5$.

The resulting general eigenobjects are in an ind-constructible category; bounded constructibility and perversity do not follow automatically. M5 also has a rational-coefficient support statement for a specified rational spectral action. Its rational arithmetic trace consequence remains conditional on the corresponding primed trace comparison supplied as input.

M6–M8: actual tame perverse eigensheaves

These three papers impose genus at least two and dense parameters. Zariski density means the image is contained in no proper algebraic subgroup. Tame unipotent monodromy kills wild inertia and has the form $\exp(t_\ell N)$, where $N$ is nilpotent and $t_\ell$ is the tame $\ell$-adic inertia coordinate. Borel level specifies a flag at each marked point.

Paper	Group and marked points	Parameter and conclusion
M6	Simple simply connected group; one point; four characteristic hypotheses	Geometric dense dual local system with regular-unipotent monodromy; nonzero locally constructible perverse eigensheaf.
M7	$\mathrm{SL}_n$, characteristic $p>n$; at least two points	Geometric dense $\mathrm{PGL}_n$ local system; any unipotent Jordan types, including zero; the same kind of eigensheaf.
M8	$\mathrm{SL}_n$, $p>n$; one point	Arithmetic, geometrically dense $\mathrm{PGL}_n$ parameter with regular-unipotent monodromy; a perverse Weil eigensheaf after a finite extension of constants.

Regular means that the nilpotent centralizer has the smallest possible dimension, equal to the group’s rank. All three constructions include parabolic nilpotent singular support and the full tensor, permutation and fusion compatibilities. M8 additionally preserves Frobenius on the entire eigenstructure, not merely on its underlying sheaf.

5. Why the advance matters

Provisional importance: 5/5. Taken as stated, this is an exceptional collection of advances linking central spectral predictions to geometric realization. The common-orbit decomposition controls the possible departure from temperedness; global enhancement turns that control into one arithmetic object. The exceptional-group theorem reaches ramified components, where unramified eigenvalue information alone is insufficient.

On the geometric side, full support removes the possibility of allowed parameters having no corresponding automorphic category. The tame papers then produce more concrete sheaves with specified finiteness, perversity and compatibility properties. These are substantial improvements over existence of an unrestricted categorical object. The rating reflects this combined reach within function-field Langlands, not a claim to solve every Langlands conjecture.

6. The main idea

Spectral sizes become geometric degree bounds

M3 interprets an offending eigenvalue through cohomology of shtukas, geometric objects combining bundles, modifications and Frobenius. Two arguments bound the same cohomological degree. One bound grows with the eigenvalue’s size; the other is controlled by compatible nilpotent orbits. Scaling the modification forces a contradiction unless the size comes from an allowed $\mathrm{SL}_2$ factor. Purity then makes the orbit consistent across places.

M1 retains whole excursion tuples and a nilpotent operator in one global geometric support. Locally, the useful configurations form an open orbit. If every test missed it, products of suitable boundary tests would annihilate the global support. A cuspidal Hecke quotient simultaneously forces the same object to survive. The resulting contradiction produces a global enhancement, including simultaneous conjugacy.

M2 detects filtration membership through bilinear pairings of functions and measures on compact parameter sets. These are finite complex measures, not necessarily positive probability measures. Preventing cancellation from lower orbit boundaries is crucial. Once the detector uses only rational operations, coefficient descent gives the rational filtration.

Ramification and specialization require additional information

At a ramified place, a Weil–Deligne parameter includes a nilpotent monodromy operator. M4 compares two possible completions of the same semisimple parameter: one arising from global purity and one from local representation theory. Adjoint regularity places both monodromies in the same open orbit. Purity then contradicts the positive exponent that a non-tempered representation would require.

M5 supposes a spectral component is missing and transports a nonzero characteristic-zero sheaf toward positive characteristic. Two Hecke modifications and transverse slices detect a nonzero nearby-cycle contribution, contradicting the disappearance required by a missing component.

M6 and M7 use two specializations. A nodal degeneration first creates flag level in characteristic zero; mixed-characteristic specialization then reaches the desired finite-field curve. Separate nonvanishing arguments, perverse truncation and descent from enhanced flags preserve the required eigenstructure. M8 makes this construction Frobenius compatible, including the gluing at the node. A finite constant-field extension enters when an auxiliary point must be fixed by a power of Frobenius.

7. Scope, limitations, and open questions

These are function-field and geometric statements. They do not establish number-field Ramanujan or a complete Arthur packet classification. M1 does not provide an ellipticity assertion or multiplicity formula, and explicitly depends on the decomposition input. M3 alone concerns temperedness at unramified places; M4’s stated all-place theorem has the exceptional adjoint scope above.

The characteristic restrictions of M2, M5 and M6 must not be silently transferred away by M3’s unrestricted statement. Similarly, M7’s arbitrary unipotent types at several points do not turn M8 into an arithmetic theorem for arbitrary tame data. M8 permits extension of constants and concerns Weil structures; it does not promise realization over the original finite field without extension.

Broader ramification, nondense parameters, relaxed characteristic assumptions and stronger finiteness properties outside the stated eigensheaf cases remain separate questions. The eight papers fit together, but their hypotheses do not merge into one universal theorem.

8. Sources and further reading

For the numerical side, begin with M3, Theorem 1.1 and Corollary 1.2, then M1, Theorem 1.2, and M4, Theorem 1.1. For rationality and categorical support, read M2, Assumption 1.1 and Theorem 1.2, and M5, Hypotheses 1.1–1.3 and Theorems 1.4–1.5. M6–M8 each state their eigensheaf theorem as Theorem 1.1.

The historical papers linked in section 3 explain why parameters, spectral support and eigensheaves are different stages of the problem. M3’s introduction gives references to Drinfeld, both Lafforgues, Arthur and geometric Satake; M6–M8 explain the earlier geometric constructions and the role of gluing. Optional background on the Langlands program can help orient the terminology, but the claims here use the specific hypotheses above.

Source manuscripts

• M1. OpenAI, Global Arthur Enhancements of Cuspidal Excursion Parameters (October 5, 2026).
• M2. OpenAI, Rationality of the Canonical Unramified Arthur Filtration (September 24, 2026).
• M3. OpenAI, Ramanujan-Arthur Decompositions of Cuspidal Functions at Full Finite Level (September 24, 2026).
• M4. OpenAI, Temperedness at ramified places for globally generic exceptional groups (October 5, 2026).
• M5. OpenAI, The Restricted Geometric Langlands Equivalence in Positive Characteristic (September 24, 2026).
• M6. OpenAI, Constructible tame Hecke eigensheaves in positive characteristic (October 5, 2026).
• M7. OpenAI, Tame Hecke Eigensheaves with Several Marked Points (October 5, 2026).
• M8. OpenAI, Frobenius Structures on Tame Hecke Eigensheaves (October 5, 2026).

Sources consulted on 10 October 2026 at repository revision fd4aeeb. Download the manuscript citations.