VibeMathedMath problems solved with AI

Automorphy of elliptic Langlands parameters for split groups over global function fields

Let FF be a global function field of characteristic pp, GG a split connected reductive group over FF with dual group G^\widehat G, and ℓ≠p\ell\ne p. V. Lafforgue decomposed the cusp forms on GG by semisimple parameters σ:ΓF→G^(Q‾ℓ)\sigma:\Gamma_F\to\widehat G(\overline{\mathbb Q}_\ell), through excursion operators; his theorem does not say which parameters occur. Conversely: is every semisimple elliptic parameter (continuous, unramified at almost all places, with CG^(im σ)/Z(G^)C_{\widehat G}(\mathrm{im}\,\sigma)/Z(\widehat G) finite) the parameter of a nonzero cusp form on GG over FF itself? For GLnGL_n this is L. Lafforgue's theorem. Böckle, Harris, Khare and Thorne proved it for everywhere unramified, Zariski dense parameters only after a finite extension of FF, and wrote that one expects no extension to be needed.

Result
Proved(see note)
Status
Candidate (review pending)
AI contribution
AI co-developed
Method
Argument
Field
Langlands correspondence over function fields
Posed by
The reciprocity direction of the Langlands correspondence over function fields; expected explicitly by Böckle, Harris, Khare and Thorne (Acta Math. 2019) for Zariski dense unramified parameters
Year posed
—
Years open
—
Solved
2026-11-12
Model
GPT-6 Astra Pro; Claude Fable 5.2
Vendor
OpenAI; Anthropic
Collaborators
Boyang Hu
Verification
Unreviewed
Publication
Preprint
Significance
52 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What was actually shown

Theorem A (Part I): for GG split semisimple and p>b(G)p>b(G), every semisimple elliptic parameter σ\sigma has a nonzero cusp form on G(F)\G(AF)/KG(F)\backslash G(\mathbb A_F)/K, for some level KK, on which every excursion operator acts through σ\sigma. Here b=0b=0 for types A,B,C,DA,B,C,D (every characteristic, including 2), b=3b=3 for G2,F4G_2,F_4, 55 for E6E_6, 77 for E7,E8E_7,E_8. The field and its constants are not extended and σ\sigma may be ramified, with finite monodromy. Theorems B and C treat non-elliptic parameters and central lattices. Part II proves existence for simply connected groups when p∤∣W∣p\nmid|W|, by a geometric method; Part III treats classical groups by descent from GLnGL_n and theta correspondence. Not covered: exceptional types in the excluded small characteristics, multiplicities, packets, and local compatibility at ramified places.

What the AI did

From the 'Statement on the use of AI' printed in each paper and repeated on the Hexagon records: "The mathematical arguments of the three papers of this series were developed with GPT-6 Astra Pro. The papers were then organized and rewritten with Claude Fable 5.2, and in the course of this several arguments were replaced or expanded. The three papers were then reviewed in detail, line by line, by GPT and Claude models, in sessions which had not taken part in writing them, and the errors found in this way were corrected. Boyang Hu organized the work and is responsible for the manuscript." Hexagon labels all three 'Primarily AI-generated text', human understanding 'some parts'.

Verification

No independent mathematician has checked this yet. Read by this site on 19 November 2026: the three Hexagon records and PDFs (hexagon 2611.00398v1, 2611.00399v1, 2611.00402v1, about 225 pages), their AI statement, the introduction and Theorem A of Part I with its section on earlier work, and the existence theorems (Theorem 1.2) of Parts II and III. The expectation that such parameters are automorphic over FF itself was confirmed in the source of Böckle-Harris-Khare-Thorne (arXiv 1609.03491). None of the proofs was read or refereed, and the Python checks in the papers' ancillary folders were not run. Part I reduces everything to the two existence theorems; Part II rests on the Gaitsgory-Raskin restricted geometric Langlands equivalence in characteristic zero, and says it takes ideas, but no results, from three OpenAI preprints. By the papers' own account the arguments were developed by one model, rewritten by another and reviewed only by models; Hexagon is moderated but not peer reviewed. No expert reaction was found. Candidate until someone independent checks the argument in public.

Sources

Submitted by MellowGander562 on

Changelog2 changes

Discussion