Claimed AI-led

Erdős Problem 111: graphs that are not countably colorable need superlinear deletions to become bipartite

A Lean-checked proof that if every finite subgraph on mm vertices of a graph can be made bipartite by deleting at most CmCm edges, the graph is countably colorable, so hG(n)/n→∞h_G(n)/n \to \infty for every graph of chromatic number ℵ1\aleph_1, answering the second question of Erdős Problem 111; the proof was found by GPT-6 Astra and Claude Opus 5.5 in alternation.

Model
GPT-6 Astra (Codex) and Claude Opus 5.5 (Claude Code), alternating; Claude Opus 5.5 (Lean formalization)
Field
Mathematics
Date
2026-10-19
Human collaborators
Jacob Parish
Problem posed
1981 · open 45 yrs

What was found

For a graph GG, hG(n)h_G(n) is the least number such that every subgraph on nn vertices can be made bipartite by deleting at most hG(n)h_G(n) edges. Erdős, Hajnal and Szemerédi showed that hG(n)≫nh_G(n) \gg n for every graph of chromatic number ℵ1\aleph_1 and built such graphs with hG(n)≤2n3/2h_G(n) \le 2n^{3/2}, and Erdős asked whether hG(n)/n→∞h_G(n)/n \to \infty. The proof builds random sign assignments along a continuous chain of closed vertex sets, forces each vertex with an infinite trace to oppose its derived sign, passes to limits along an ultrafilter, and splits the resulting signs, as unit vectors in L2L^2, into countably many classes within which greedy coloring succeeds. Per the repository, Claude Opus 5.5 wrote the first complete proof and GPT-6 Astra checked and revised it, and Claude Opus 5.5 then wrote the Lean. The first, open-ended question of the problem, the exact growth of hG(n)h_G(n), is not addressed.

Novelty check

On 2026-10-20 erdosproblems.com listed Problem 111 as open with no proof claims. The repository records its own search of 17 October, of arXiv, the web and Formal Conjectures, which has no merged statement of the problem and two open pull requests that state it without proof. An Erdős Problem a Day report of 26 July 2026 proves the reduction of the limit to the finite-density form for graphs of chromatic number ℵ1\aleph_1, by the same countable-deletion and packing argument, but does not solve the problem; the repository says its argument does not rely on it.

Caveats and known objections

Graded claimed, not formal, although the Lean is Palomar-registered: the Challenge file defines the bipartization number, hGh_G and the chromatic cardinal itself rather than taking them from Mathlib or a merged Formal Conjectures statement. The repository says these agree with Formal Conjectures' definitions and that its theorem implies the statement in one of the two open pull requests, checked in Lean, but that statement is unmerged and nobody without a stake has audited either. One edge case deserves an outside look: the bipartization number of a subgraph with infinitely many edges is zero by convention, which the repository says the statements never use. The only review of the argument was by GPT-6 Astra, and no mathematician is on record as having read it. Autonomy is ai-led because Plasma AI chose the problem, ran the research loop and directed the formalization.

Machine-checked elsewhere

PalomarPALOMAR-2026-10-19-000007

plasma-ai/erdos-111 at de72caf0e388 · checked 2026-10-19

ProvedErdos111.countably_colorable_of_linearErdos111.tendsto_of_not_countably_colorableErdos111.erdos_111

A Lean proof that typechecks against the recorded statement at a pinned commit under a declared axiom set, checked mechanically rather than by human review. It does not certify that a result is new or of research interest: the only filter on that is a language-model screen, and Palomar states that it adds no human editorial step.

Permitted axioms are exactly propext, Quot.sound and Classical.choice. The definitions the three statements use are the authors' own, not Formal Conjectures'.

Nobody outside the lab has checked this yet.

Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.

Or on GitHub: submit a check challenge the grade send a correction or send a pull request

Entry history (1 event)
  1. AddedEntered the registry graded Claimed and AI-led.

Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.

Graded claimed for verification and ai-led for autonomy. What these mean.

Cite this entry

Plain text
whataifound.org. (2026). Erdős Problem 111: graphs that are not countably colorable need superlinear deletions to become bipartite. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-19-erdos-111-bipartization
BibTeX
@misc{whataifound-plasmaai-2026-bipartization,
  title        = {Erdős Problem 111: graphs that are not countably colorable need superlinear deletions to become bipartite},
  author       = {{whataifound.org}},
  year         = {2026},
  howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
  note         = {Result by Plasma AI. Verification: Claimed. Autonomy: AI-led.},
  url          = {https://whataifound.org/finding/2026-10-19-erdos-111-bipartization}
}

Related findings

← All mathematics findings in the registry