AI-built Lean proof reports Conway’s omnific refinement property

For cut-defined omnific integers a,b,c,d represented as numeric games, a Lean theorem says ab=cd implies omnific e,f,g,h with a=ef, b=gh, c=eg and d=fh, including zero cases. The theorem is stated at each Lean universe, rather than putting all surreal numbers into one type.

Sources for this record

What was already known

Conway posed this four-factor property in 1976; L’Innocente and Mantova still listed it as a conjecture in 2024. Abramov’s bounded literature search found no earlier proof.

AI involvement AI-led

Dan Abramov chose the problem, supplied sources and supervised checks. He credits ChatGPT and Claude with the mathematical argument and Lean proof.

Pinned provenance assigns all mathematical and Lean proof work to ChatGPT and Claude, with Abramov directing the task, sources and acceptance process.

Importance Notable

If the formal encoding matches Conway’s intended objects, the result resolves a longstanding factorization question.

Limitations

Lean kernel and statement-comparison checks are reported for the encoded theorem. No surreal-number specialist has independently confirmed the encoding’s match to the historical conjecture, and the owner says he does not understand the proof methods.

Technical verification Formal (Lean 4/Mathlib kernel and Palomar Comparator)

The repository reports Lean kernel checking of the full four-factor theorem and Palomar Comparator's exact Challenge/Solution type comparison with only standard logical axioms. The formal theorem is complete at its specified universe and cut-definition scope; a surreal-number specialist has not independently confirmed that this encoding captures the intended historical conjecture. This review did not rerun Lean.

Formal checking

kernel proof-term check, exact Challenge/Solution statement comparison and permitted-axiom audit

Checker
Lean kernel and Palomar Comparator, as reported in Abramov's pinned provenance
Scope
Four-factor refinement for cut-defined omnific integers represented by numeric games at each Lean universe, including zero cases
Basis
Reported by the linked checker
Evidence source

Checking a formal statement does not establish that it matches the original problem.

Human understanding Few

Abramov explicitly limits his understanding to the conjecture's statement rather than its proof methods. The proof guide's mathematical explanations are AI-written, and no named human explanation of the central mechanism was located. This is a documented gap, not merely a default.

Correction and publication history

2026-10-18 · publication

A distinct four-factor theorem with reported kernel checks and an explicit statement-fidelity gap.

Read this published revision