VibeMathedMath problems solved by AI

The Banks-Martin Conjecture on Primitive Sets

Banks and Martin conjectured in 2013 that for a primitive set AA and any set QQ of primes, the Erdos sum of the members of AA composed only of primes in QQ is at most the corresponding sum over QQ itself. The unrestricted form turned out to be false once QQ is allowed to contain 22; Lichtman proposed a revised form restricted to odd primes. That revised conjecture, long viewed as a unifying master theorem for the area, is proved here.

Result
Proved
Status
Resolved
AI contribution
AI co-developed
Method
Argument
Field
Number theory
Posed by
William D. Banks, Greg Martin; revised form proposed by Jared Duker Lichtman
Year posed
2013
Years open
13y
Solved
2026-05-01
Model
GPT-5.5 Pro (early version)
Vendor
OpenAI
Collaborators
Boris Alexeev, Kevin Barreto, Yanyang Li, Jared Duker Lichtman, Liam Price, Jibran Iqbal Shah, Quanyu Tang, Terence Tao
Verification
Unreviewed
Publication
Preprint
Significance
25 / 100
Disclosed cost
Wikipedia
No dedicated article

What the AI did

This paper discloses per theorem rather than in a blanket statement, and this theorem is one of the more modest entries: an early version of GPT-5.5 Pro was used to assist with the initial proof. Elsewhere in the same paper the model's role is larger, with the proof of the Erdos #1196 theorem generated by an autonomous GPT-5.4 Pro run whose transcript is public. Across all of it the authors state that the final proofs were generated and reviewed by them, using the AI-generated proofs as starting points where appropriate. The whole method, Markov chains with von Mangoldt weights, was itself suggested by model output.

Verification

arXiv preprint, not peer-reviewed. Other results in the same paper were formalized in Lean using Codex and Gauss, but this theorem was not among them.

Source

arXiv:2605.00301 - Primitive sets and von Mangoldt chains: Erdos Problem #1196 and beyond

Submitted by Curator34

Discussion