VibeMathedMath problems solved with AI

An Erdős–Kac law for base-bb palindromes and for reversed primes

For every base b2b\ge2, the number of prime factors of the λ\lambda-digit base-bb palindromes, and of the base-bb reversals of the λ\lambda-digit primes, obeys an Erdős–Kac law: counted with or without multiplicity, it is asymptotically normal with centring loglogbλ\log\log b^{\lambda} and scaling loglogbλ\sqrt{\log\log b^{\lambda}}. For reversed primes the law persists when the leading digit of the prime is prescribed. Erdős–Kac laws were already known for other digitally defined families — integers with a fixed digit sum, or with digits restricted to a fixed set — but for palindromes only the largest value of ω\omega had been studied, with nothing known about the typical value, and for reversed primes the level of distribution the argument needs became available only in 2025.

Result
Proved(see note)
Status
Resolved
AI contribution
AI co-developed
Method
Argument
Field
Analytic number theory
Posed by
Year posed
Years open
Solved
2026-08
Model
Claude Fable 5, Claude Opus 5
Vendor
Anthropic
Collaborators
Verification
Lean-checked, statement unaudited
Publication
Preprint
Significance
12 / 100
Disclosed cost
Wikipedia
No dedicated article

What was actually shown

New theorems, not a formalisation of previously known results. For every base b2b\ge2 the Erdős–Kac law is established for the λ\lambda-digit base-bb palindromes and for the base-bb reversals of the λ\lambda-digit primes, for ω\omega and Ω\Omega and for ωS,ΩS\omega_S,\Omega_S with any regular set SS of primes; with normal order loglogn\log\log n on both families, and, for ω\omega, all moments of order up to 12(loglogbλ)1/3\tfrac12(\log\log b^{\lambda})^{1/3} uniformly in the order. For reversed primes it also holds with the leading digit prescribed.

Not settled: the results rest on quoted inputs (Col for palindromes, the Bombieri–Vinogradov theorem of Dartyge–Rivat–Swaenepoel for reversed primes), and both families exclude the primes dividing b(b21)b(b^{2}-1). No rate of convergence is obtained. The question of Banks–Shparlinski on the *largest* value of ω\omega on palindromes is untouched: the trivial bound λ/logλ\ll\lambda/\log\lambda and their λo(1)\lambda^{o(1)} remain far apart.

What the AI did

Used at every stage: literature search, jointly working out the main arguments, and drafting the manuscript. Two contributions were decisive. The arithmetic input for palindromes — Col's theorem on the level of distribution of palindromes in arithmetic progressions — was located by the models. And the general Erdős–Kac criterion used here, a modification of the Granville–Soundararajan sieve moment estimate that allows a finite exceptional set of primes at which the local densities are arbitrary, was worked out jointly with them. The models also produced the Lean 4 formalisation. The author verified all statements, proofs and references, made the final decisions on content and presentation, and is responsible for any remaining errors.

Verification

Read independently here on 24 August 2026 at github.com/vibefrtz/vibemath, in addition to the submission's own detailed VERIFICATION.md, which this confirms rather than repeats. All 21 Lean files (about 7850 lines) carry no sorry, no admit and no native_decide; the sole textual match for "axiom" outside Cited.lean is a comment, not a declaration. axiom_audit.txt shows every one of the 27 theorems drawing only Lean's three standard axioms plus a subset of the eight declared in Cited.lean, matching the paper's citation structure theorem by theorem. Spot-checked Main.lean against the manuscript: pal_EK_omega and rev_EK_omega are Tendsto statements of the empirical distribution to Phi(t) with centring and scaling LL b lam and its square root, matching Theorems 1.1 and 1.2 as stated. Of the eight cited results, Col, Banks-Shparlinski, Dartyge-Rivat-Swaenepoel and Granville-Soundararajan were confirmed to exist with the stated venues; Dartyge-Rivat-Swaenepoel (arXiv:2506.21642) is from June 2025, corroborating the submission's claim that the reversed-prime argument's input became available only that year. This is a sampling audit, not the full informal-to-formal correspondence review the lean-verified tier requires, so the tier stays where the submission itself placed it.

Sources

Submitted by vibefrtz on

Changelog2 changes

Discussion