VibeMathedMath problems solved with AI

The Period-Index Conjecture

For a Brauer class on a variety, the period-index conjecture bounds the index in terms of the period and the dimension. Disproved: for any uncountable algebraically closed field kk of characteristic 00 and any d3d \geq 3 there is a dd-dimensional variety over kk carrying a Brauer class that violates it, for Hodge-theoretic reasons. For d=3d = 3 the construction needs no uncountability, so the conjecture fails already over Q\overline{\mathbf{Q}}.

Result
Disproved(see note)
Status
Resolved
AI contribution
AI co-developed
Method
Construction
Field
Algebraic geometry
Posed by
Jean-Louis Coliot-Thèléne
Year posed
2001
Years open
25y
Solved
2026-08-04
Model
ChatGPT
Vendor
OpenAI
Collaborators
Alexander Perry
Verification
Unreviewed
Publication
Preprint
Significance
30 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

The authors conjecture the period-index conjecture holds true only with corrections at small primes, and to this end proved its analog in Hodge theory.

What the AI did

The paper's AI disclosure is unusually specific about a partial success. Prompted to find a counterexample by the Hodge-theoretic strategy, ChatGPT produced an example that was flawed, but whose shape survived into the final solution: a quotient of the same form, with an abelian surface, a genus two curve and the group Z/4. The author refined that into the working construction.

Verification

arXiv preprint, not yet peer-reviewed.

Sources

Changelog1 change
  • Matarisvanset Year posed to 2001, also Age footnote, What was actually shown, Posed by, More links

Discussion