VibeMathedMath problems solved with AI

Mahler's ZZ-number problem in its generalized form

Let p>q>1p>q>1 be coprime integers and let Zp/q(s,s+t)Z{p/q}(s,s+t) be the set of ξ>0\xi>0 whose fractional parts {ξ(p/q)n}\{\xi(p/q)^{n}\} all lie in [s,s+t)[s,s+t). The question for which parameters ZZ is empty remains open.

Result
Proved(see note)
Status
Partial result
AI contribution
AI-discovered
Method
Computation
Field
Distribution mod one; Mahler's Z-numbers
Posed by
Mahler
Year posed
1968
Years open
58y
Solved
2026-09-12
Model
Fable 5, Opus 5
Vendor
Anthropic
Collaborators
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
30 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

Theorem E gives Z3/2(27,57)=Z_{3/2}(\tfrac27,\tfrac57)=\emptyset: a window of length 37=0.428571\tfrac37=0.428571\ldots at base 3/23/2, beating the 31/8131/81 of [Dub19, Thm. 1.2], and with no assumption on the arithmetic nature of ξ\xi, where that result needs algebraicity. Mahler's problem itself is untouched: for which parameters ZZ is empty remains open, and this is a record window rather than a classification.

What the AI did

AI discovered theorems, formalized them and wrote a draft of the paper from the Lean.

Verification

Filed at Lean-checked, as submitted. Challenge2.lean is a trusted statement of record that imports nothing but Mathlib, redeclares verbatim every definition occurring in the certified theorems, and states the paper's lettered results with sorry proofs; comparator2.json names fourteen of them and, as configured, checks constant identity across the challenge and solution environments, restricts axioms to propext, Quot.sound and Classical.choice, and requires the Lean kernel to re-accept the solution from a fresh export with no olean loaded. That configuration reads correctly. It is not lifted to Lean-verified because no CI run was found on the repository and the comparator run was not reproduced here, so the machine check rests on the author's report - which is what Lean-checked means. PRIOR_ART.md is unusually candid and worth reading: it records that the search was targeted rather than exhaustive, that the whole of Section 3 was found in print with Theorems 3.5 and 3.6 a machine-checked fragment of [Bug04, Thm. 1], and concludes "Novelty is therefore recorded as unknown". Not peer reviewed and no independent expert endorsement.

Sources

Submitted by LucidKestrel185 on

Changelog2 changes

Discussion