OpenAI proves the Falconer distance conjecture (Lean)
Awaiting review
Importance: major (4 of 5)The takeaway
Family 073 of OpenAI’s 6 Oct 2026 release proves that every compact set in ℝᵈ (d ≥ 2) of Hausdorff dimension greater than d/2 has a distance set of positive length. This is Falconer’s 1985 conjecture, the continuous analogue of Erdős’s distinct-distances problem; the best planar threshold before was 5/4. The 64-page proof comes with a Lean formalization whose statement uses Mathlib’s own Hausdorff dimension and Lebesgue measure, and the community comparator run accepted it. A first follow-up already builds on its method.
Status
- Claim
Awaiting review
- Our reporting
- Medium confidence
- Verification
- Lean-verified
- Importance
- Major (4 of 5)
- Last verified
- 29 October 2026
Your AI and this story
- GPT-6 Astra159 days after its cutoff
- Claude Opus 5.598 days after its cutoff
- Gemini 4 Argon189 days after its cutoff
- Grok 4.7128 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 98 days before it.
Key facts
- Abstract (verbatim): ‘We resolve the Falconer distance conjecture in every dimension. For every integer d ≥ 2, a compact subset of ℝᵈ with Hausdorff dimension greater than d/2 determines a set of Euclidean distances of positive Lebesgue measure.’ 64 pages, dated 23 Sep 2026
- Prior state (paper, §1.1): Falconer’s threshold (d+1)/2; Wolff’s planar 4/3; Erdoğan’s d/2 + 1/3 for d ≥ 3; Du, Guth, Ou, Wang, Wilson and Zhang 9/5 in dimension three; Du–Zhang d/2 + 1/4 + 1/(8d − 4); Guth, Iosevich, Ou and Wang dimH E > 5/4 in the plane; Du, Ou, Ren and Zhang d/2 + 1/4 − 1/(8d + 4) for d ≥ 3. Shmerkin and Wang handled sets whose Hausdorff and packing dimensions both equal d/2
- Method (paper, §1.2): a multiscale decomposition of measures into regular pieces with scale-dependent angular tests, building on Keleti–Shmerkin, Shmerkin–Wang and Orponen–Shmerkin; the retained incidence input is the Orponen–Shmerkin theorem; a ‘two-depth potential’ pays the costs of the recursion
- Lean: comparator challenge FalconerAllDimensions.lean, theorem
falconer_distance_conjecture: for every d ≥ 2 and compact E ⊆ EuclideanSpace ℝ (Fin d) with d/2 < dimH E,0 < volume {r : ℝ | ∃ x ∈ E, ∃ y ∈ E, dist x y = r}; plus PlanarFalconer.lean for the earlier planar case. Only Mathlib definitions are used, so the statement reads as the standard conjecture (our reading). Both were accepted in the oai-math-revised comparator run (25–26 Oct) - Follow-up: Guozheng Cheng, Xiang Fang, Menghan Li, Hongdou Qu and Chengbo Xiao, ‘A quantitative form of the planar Falconer distance theorem’ (arXiv 2610.22222, 15 Oct): ‘The proof develops the multiscale method of the September 2026 manuscript of OpenAI’; their comment line promises higher dimensions ‘building on the October 6 release of the OpenAI preprint’. Their AI note: an AI tool was used for language editing, LaTeX and ‘local mathematical reasoning’
- Our 19 Oct entry on the harmonic-analysis claims noted that 073 is among the few families in that area with a Lean link (see OpenAI claims 3D restriction and Bochner–Riesz conjectures); no press article or expert blog post on this family was found by 29 Oct
What happened
How big must a set be before the distances between its points fill out an interval of positive length? Kenneth Falconer showed in 1985 that Hausdorff dimension above (d+1)/2 is enough in ℝᵈ, and conjectured that d/2 suffices, which would be sharp. The problem is a continuous cousin of Erdős’s distinct-distances problem and became a test case for Fourier restriction and decoupling. Progress was slow: the planar threshold went from 4/3 (Wolff, 1999) to 5/4 (Guth, Iosevich, Ou and Wang, 2020).
OpenAI’s 64-page manuscript claims the full conjecture in every dimension. It works directly with the distance measure through a multiscale decomposition of the set’s measure, angular tests at each scale and a recursive potential argument, using known incidence and radial projection theorems as inputs. The main theorem is formalized in Lean against a statement written with Mathlib’s Hausdorff dimension and Lebesgue measure, and the comparator check passed.
Why it matters
Falconer’s conjecture is one of the central problems linking geometric measure theory and harmonic analysis. With a Lean proof of a faithful statement it is among the better-checked claims in the release, but human understanding of the argument is only beginning.
What is disputed or not yet verified
| Verification | Lean formalization of the main theorem; no published human verification yet |
|---|
Reactions
The first outside paper, by Guozheng Cheng, Xiang Fang, Menghan Li, Hongdou Qu and Chengbo Xiao, gives a quantitative planar version by developing the manuscript’s multiscale method and promises the higher-dimensional case. We found no blog post, talk or press piece assessing the proof.
Sources
6 sources from 2 sites. Numbers match the chips in the text.
6 sources: 5 primary, 1 reaction
Primary
- OpenAI: The Falconer distance conjecture in all dimensions (23 Sep)github.com, paper
- Lean scope note for family 073github.com, code
- Lean comparator challenge: FalconerAllDimensions.leangithub.com, code
- arXiv 2610.22222: Cheng, Fang, Li, Qu, Xiao, A quantitative form of the planar Falconer distance theoremarxiv.org, paper
- oai-math-revised: comparator run on all challenges (LEAN_CHECKS.md)github.com, code
Reactions
- oai-math-revised: community notes on the Falconer manuscriptgithub.com, discussion
Changes
- Filed (families 001–125; manuscript introduction, Lean statement, comparator results and the follow-up arXiv paper read)