Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Claim. With the distance multiplicities of an -point set , as in Problem 959, there is an absolute constant such that
for every sufficiently large . The construction is the arithmetic point set behind OpenAI's fixed-power lower bound for unit distances, tuned so that one target shell carries more representations than every competing distance; the gap between the two most frequent distances is then a fixed power of above linear.
Theofil Xeff, posting under the forum account fefemath, submitted the claim to the site's proof-claims tab on 2026-07-21 as a partial claim, with the write-up linked above and no formalization. The posting names GPT 5.6 Sol and Fable 5 as the systems used.
Submission note. Posted to erdosproblems.com as a proof claim by Theofil Xeff (account fefemath) on 21 July 2026, giving "GPT 5.6 Sol, Fable 5" as the AI used:
We prove that for every sufficiently large , where is an absolute constant. Interestingly, the proof reuses the arithmetic construction from OpenAI's recent solution to the unit-distance problem [90] (paper here). The construction is tuned so that one particular distance has more representations than every competing distance, by isolating a suitable target shell. Eventually, this produces a superlinear gap between the two most frequent distances. Notes: LLM GPT 5.6 Sol high found the main idea. Fable 5 helped to refine it. I am currently formalizing the proof in Lean, will update this when this is done.
Covers. A lower bound of order on the largest gap, which exceeds the that Clemen, Dumitrescu and Liu asked for (Problem 1.11) and that Snyder's claim asserts. The problem asks for an estimate; no upper bound is claimed, and the order of the gap remains open.
Depends on. [[problems/distance_problems/E0090/claims/2026_05_20_openai|OpenAI's fixed-power disproof of the unit distance conjecture]] supplies the point set the construction tunes.
Acceptance. None documented. The site's label was OPEN on 2026-10-06, the proof-claims thread carried no comment on the claim as of that date, and the site's page does not credit the result; nothing here has reviewed the write-up. The claim is therefore claimed.