VibeMathedMath problems solved with AI

Conjectures 2a and 2b of Kauers and Zeilberger

Conjectures 2a and 2b of Kauers and Zeilberger, on the asymptotics of a family of restricted lattice walks. Both are obtained from a local limit theorem for excursions of Markov-modulated random walks in cones.

Result
Proved
Status
Candidate (review pending)
AI contribution
AI-discovered
Method
Argument
Field
Lattice walk enumeration
Posed by
Manuel Kauers and Doron Zeilberger
Year posed
2020
Years open
6y
Solved
2026-08-11
Model
Claude
Vendor
Anthropic
Collaborators
Jaideep Sai Padhi
Verification
Unreviewed
Publication
Preprint
Significance
16 / 100
Disclosed cost
—
Wikipedia
No dedicated article

What the AI did

The paper's division of labour, in its own words: the model "generated the mathematical ideas, the proofs, and the verification code", while the author "selected the problems and determined when a line of attack should be abandoned; executed all computations, on hardware the system could not access; designed and enforced the verification protocol". It adds that neither role was passive: the system did not merely formalise ideas supplied to it, and the author did not merely execute instructions.

Verification

Checked by this site on 21 August 2026 against the paper (arXiv:2608.11290v2): the division of labour is verbatim as quoted, all five challenges are reported fully solved, and two further problems are reported only partial and are NOT entered here. The paper states a four-principle verification protocol that distinguishes proved, machine-verified and supported claims, and ships code and verification scripts. Those scripts were not re-run here. Days-old preprint, no independent review.

Sources

Changelog4 changes
  • GoldenWalrus805commented
  • Matarisvanset More links to “problem-record: Kauers & Zeilberger 2020 - Problem statement | https://arxiv.org/html/2006…”, also Year posed
  • Rasmus Lindahlcleared links
  • Rasmus Lindahladded this entry

Discussion1

GoldenWalrus80507 Nov 2026, 06:00 UTC

Thank you for making the manuscript, code, and verification materials available. While reviewing the local-limit argument, we found one point in D2(b)–(c) that may benefit from clarification.

The proof chooses η=ε2/3\eta=\varepsilon^{2/3}. The near-boundary contribution is written as

O(η/ε)=O(ε1/3),O(\eta/\varepsilon)=O(\varepsilon^{1/3}),

but direct substitution appears to give

ηε=ε2/3ε=ε−1/3,\frac{\eta}{\varepsilon} = \frac{\varepsilon^{2/3}}{\varepsilon} = \varepsilon^{-1/3},

which diverges as ε↓0\varepsilon\downarrow0.

The re-entry estimate uses

e−cη2/ε,e^{-c\eta^2/\varepsilon},

which, with the same choice of η\eta, appears to become

e−cε1/3,e^{-c\varepsilon^{1/3}},

rather than the displayed e−cε−1/3e^{-c\varepsilon^{-1/3}}. It therefore approaches 11, rather than 00.

Testing a general cutoff η=εa\eta=\varepsilon^a also seems problematic under the displayed estimates: making η/ε→0\eta/\varepsilon\to0 requires a>1a>1, while making η2/ε→∞\eta^2/\varepsilon\to\infty requires a<1/2a<1/2. We therefore could not find a single power-law cutoff that makes both error terms vanish.

This may be a repairable proof-interface issue rather than a problem with the final result. The exact counting objects, exponential growth bases, n−4n^{-4} behavior, and numerical evidence remain consistent with the claimed conclusions. Our concern is limited to whether the written D2 argument currently supplies the required rigorous link to the subsequent excursion assembly.

Could the authors please check whether there is an omitted compensating factor, a stronger boundary-layer estimate, a multiscale argument, or a different intended order of limits that resolves this conflict? Any clarification would be greatly appreciated.

1