qedbot

Discrete geometry·borsuk:dimension

Smallest dimension in which Borsuk's conjecture is known to fail

record target minimiserecord held by a machine F0 no formal statement

No independent check recorded yet. A formal artifact, declaration or published object is attached, but no rebuild of it is recorded here.

Fidelity F0: Absent. No formal statement is attached to the result.

Borsuk asked in 1933 whether every bounded set in n-dimensional space can be split into n + 1 parts of smaller diameter. It is true in dimensions 2 and 3 and false in high dimensions; the record is the smallest dimension with a known counterexample.

Source

Record history

6 steps·dimension, lower is better
05001,0002000201020201325, Jeff Kahn and Gil Kalai, 1993298, Aicke Hinrichs and Christian Richter, 200365, Andriy Bondarenko, 2013-0564, Thomas Jenrich, 2013-0863, Max Grinsztajn, with GPT-5.5 Pro, 2026-05-2663, Yibo Ji, with GPT 5.6 Sol, 2026-08-12, did not move the record
Ink marks are steps by people and clay marks steps where AI took part; grey marks did not move the record, and hollow marks are candidates. How frontiers are drawn

Kahn and Kalai's 1993 counterexample works in dimension 1325; constructions from codes brought the dimension to 298 by 2003, and Bondarenko's two-distance set from the G2(4) graph led to 65 and then 64 in 2013. Several improvements between 1994 and 2002 are not yet drawn here. In May 2026 Max Grinsztajn posted a counterexample in dimension 63, obtained with assistance from GPT-5.5 Pro. In August an arXiv note reported the same dimension from a construction generated by GPT 5.6 Sol, and was withdrawn when its author found the earlier posting.

63

Yibo Ji, with GPT 5.6 Sol·12 Aug 2026·Source

The note said the example and proof were generated entirely by ChatGPT using GPT 5.6 Sol; its author withdrew it on 14 August on finding the earlier posting.

AI took partdid not move it
63

Max Grinsztajn, with GPT-5.5 Pro·26 May 2026·Source·Artifact

A 321-point set in which every subset of smaller diameter has at most five points, so at least 65 parts are needed. The repository says the construction and proof were obtained with assistance from GPT-5.5 Pro.

AI took partbest known
64

Thomas Jenrich·Aug 2013·Source

A 352-point subset of Bondarenko's set; published with Andries Brouwer in the Electronic Journal of Combinatorics in 2014.

moved the record
65

Andriy Bondarenko·May 2013·Source

A 416-point two-distance set from the G2(4) graph; published in Discrete & Computational Geometry in 2014.

moved the record
298

Aicke Hinrichs and Christian Richter·2003·Source

New sets with large Borsuk numbers.

moved the record
1325

Jeff Kahn and Gil Kalai·1993·Source

The first counterexample; Weißbach later showed the dimension-1325 case needs a corrected argument, which holds.

moved the record

AI activity

How grades work
GPT-5.5 Pro

26 May 2026·with Max Grinsztajn

A 321-point set in dimension 63 that cannot be split into 64 parts of smaller diameter, improving 64.

record A1 V2 F0
Reasoning and sources

Autonomy

The repository states the construction and proof were obtained with assistance from GPT-5.5 Pro.

GPT-5.6 Sol

12 Aug 2026·with Yibo Ji

The same dimension, 63, from a construction the author said was generated entirely by ChatGPT; withdrawn on 14 August when he found the earlier posting.

matched A3 V2 F0
Reasoning and sources

Autonomy

The note stated the example and proof were generated entirely by ChatGPT using GPT 5.6 Sol, with the author verifying them.

F0 no formal statement. Absent. No formal statement is attached to the result.

Follow and discuss

All discussion

Discussion and bounties for this problem load here.

Something wrong or missing here? Request a correction or add a claim, with its sources.

Sources

Cite this record

qed.bot, “Smallest dimension in which Borsuk's conjecture is known to fail”, https://qed.bot/t/borsuk-dimension, as of 30 Sep 2026.

This record as plain text, with each claim, its grades and its sources.