VibeMathedMath problems solved with AI
← All frontiers
FrontierGeometry & topologysignificance 40

Smallest dimension where Borsuk's conjecture fails

the smallest nn with a known counterexample to Borsuk's conjecture in Rn\mathbb R^n

Current best · lower is better
n=63n = 63
Max Grinsztajn, with GPT-5.5 Pro, 26 May 2026 · entry
steps
10
by AI
1
since
1993
2000201020201002005001k1993: n = 1325 (Jeff Kahn and Gil Kalai)1994: n = 946 (A. Nilli)1997: n = 561 (Andrei M. Raigorodskii)2000: n = 560 (Bernulf Weissbach)2002-01: n = 323 (Aicke Hinrichs)2002-02: n = 321 (Oleg Pikhurko)2003: n = 298 (Aicke Hinrichs and Christian Richter)2013-05: n = 65 (Andriy Bondarenko)2013-08: n = 64 (Thomas Jenrich and Andries E. Brouwer)2026-05-26: n = 63 (Max Grinsztajn, with GPT-5.5 Pro)

The line is the frontier over time, falling as the bound comes down: lower is better here. Filled dots are steps that moved it; muted dots are results that did not. Orange dots are catalog entries, results with AI in the loop. Hollow dots are candidates under review and never move the line.Grey dots along the top edge are results from before the quantity had a number, placed at the worst end because they have no value on this axis. Dots that would overlap are nudged sideways a few pixels. Hover a dot for its value and attribution.

About this frontier

Borsuk asked in 1933 whether every bounded set in Rn\mathbb R^n can be partitioned into n+1n+1 subsets of strictly smaller diameter. True for n3n \le 3; false in general, as Kahn and Kalai showed in 1993 with a counterexample in dimension 1325. Since then the question has been where it first fails, and the race has been to lower the dimension of an explicit counterexample: 946, 561, 560, 323, 321, 298, then Bondarenko's two-distance construction at 65 in 2013, Jenrich and Brouwer's 64, and 63 in 2026. The first failing dimension is open everywhere in 4n624 \le n \le 62, so the finish line is unknown; this frontier tracks the ceiling on it.

Every step, newest first

DateValueWhoModelStatusSource
26 May 2026n=63n = 63best
A 321-point subset of R^63 whose smaller-diameter subsets have at most 5 points, so at least 65 > 64 parts are needed: a 320-point rank-63 piece of the G2(4) set plus one scaled projected point. The repository's exact verifier was rerun here on 12 August. Found again independently by Nicholas Konz with Claude in August 2026. On Tao's ledger as the current best.
Max Grinsztajn, with GPT-5.5 ProGPT-5.5 ProAI step
site-confirmed
entry
Aug 2013n=64n = 64
A 352-point two-distance subset of Bondarenko's configuration, needing at least 71 parts, three months after his paper. Electronic Journal of Combinatorics 21 (2014); dated to the arXiv posting.
Thomas Jenrich and Andries E. Brouwerhistoricalsource ↗
May 2013n=65n = 65
The big drop: a 416-point two-distance set on the sphere in R^65, from the G2(4) strongly regular graph, that cannot be split into 83 parts of smaller diameter. Discrete and Computational Geometry 51 (2014); dated here to the May 2013 arXiv posting.
Andriy Bondarenkohistoricalsource ↗
2003n=298n = 298
Discrete Mathematics 270 (2003). The record for ten years.
Aicke Hinrichs and Christian Richterhistoricalsource ↗
Mar 2002n=321n = 321
Counterexamples in dimensions 321 and 322, one month after Hinrichs.
Oleg Pikhurkohistoricalsource ↗
Jan 2002n=323n = 323
Discrete Mathematics 243 (2002). A construction from spherical codes, which is the idea the next three steps refine.
Aicke Hinrichshistoricalsource ↗
2000n=560n = 560
Beitraege zur Algebra und Geometrie 41 (2000), 417-423, one dimension below Raigorodskii. The EMIS archive refused automated access from here; the value is as cited by Bondarenko and by Tao's ledger.
Bernulf Weissbachhistoricalsource ↗
1997n=561n = 561
Russian Mathematical Surveys 52 (1997), a two-page note.
Andrei M. Raigorodskiihistoricalsource ↗
1994n=946n = 946
Jerusalem Combinatorics '93, Contemporary Mathematics 178. A sharpening of the Kahn-Kalai construction; Nilli is a pseudonym of Noga Alon.
A. Nillihistoricalsource ↗
1993n=1325n = 1325
The disproof. Kahn and Kalai showed the Borsuk number grows at least like exp(c sqrt n), so the conjecture fails in every sufficiently large dimension, and exhibited the failure explicitly at n = 1325. Bulletin of the AMS 29 (1993). Everything below is the race to bring that number down.
Jeff Kahn and Gil Kalaihistoricalsource ↗

Every row is on Tao's optimisation-problems ledger (constants/28a.md), read here; each reference was then checked against the linked paper's own metadata, and Bondarenko's paper cites the 1994 to 2003 values as well. The two 2013 rows are dated to their arXiv postings, not their 2014 journal issues. Weissbach's paper is linked at the EMIS journal archive, which refused automated access from here; that row rests on Bondarenko's and Tao's citations. The 2026 row is published rather than a candidate because the entry is site-confirmed: its exact verifier was rerun here on 12 August.

Discussion