VibeMathedMath problems solved with AI

The Lyons–White conjecture: rate-monotonicity of 2m\ell^{2m} distances for random walks on dihedral groups

Let DnD_n be the dihedral group of order 2n2n and run a continuous-time random walk on it driven by symmetric jump rates whose support generates the group. Call the pair (Dn,p)(D_n,p) rate-monotonic if, at every fixed time, the p\ell^p distance between the walk's distribution and the uniform distribution can only decrease when the rates are increased. Lyons and White (Ann. Probab. 51, 2023) proved this for p=2p=2 and p=p=\infty, found pairs (Dn,p)(D_n,p) that fail it for pp in [1,1.997][2.001,3.999][4.001,5.995][1,1.997]\cup[2.001,3.999]\cup[4.001,5.995], and asked whether any pair fails for p=4p=4 or p=6p=6.

Result
Proved(see note)
Status
Resolved
AI contribution
AI co-developed
Method
Argument
Field
Random walks on finite groups; mixing; harmonic analysis on groups
Posed by
Russell Lyons and Graham White, Monotonicity for continuous-time random walks
Year posed
2023
Years open
3y
Solved
2026-08-27
Model
AxiomProver
Vendor
Axiom Math
Collaborators
Colin Defant, Ken Ono
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
15 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

No such pair exists: for every positive integer mm and every nn, (Dn,2m)(D_n,2m) is rate-monotonic (Theorem A), and more generally so is every inversion extension of a finite abelian group by an involution, a family containing the generalized dihedral, dicyclic and generalized quaternion groups (Theorem B). The picture is completed in the other direction: for every real p1p\ge 1 that is not an even integer there is an nn with (Dn,p)(D_n,p) not rate-monotonic (Theorem C), so the even integers are exactly the exponents for which monotonicity holds on the dihedral groups.

What the AI did

The paper's own account: "The proofs in this paper were generated through human-AI collaboration. In dialogue with AI, the human authors developed and formalized [Theorems A, B and C] with AxiomProver, an AI system currently under development by Axiom Math. In particular, this resulted in a formal Lean certificate for these three theorems." Both authors are at Axiom Math. Co-developed rather than assisted because the proofs themselves, not only the formalization, are described as produced in dialogue with the system; not discovered, because the humans directed the work and no autonomous run is claimed.

Verification

Lean-checked, statement unaudited, as with the other AxiomProver entries here. The repository AxiomMath/LyonsWhite carries a Challenge/Basic.lean statement surface and a Comparator configuration, and its README says the development was verified locally against the challenge; the paper says the formalization "assumes standard facts from analysis and group theory" and lists none, so the certificate is conditional on those assumptions and this site has not enumerated them or rebuilt the development. A thirteen-page preprint ten days old, not peer reviewed, no independent reader on record.

Sources

Changelog1 change

Discussion