Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Claim. There is d0d_0 such that for every increasing sequence (σi)i≥1(\sigma_i)_{i\ge1} of positive even integers with σi+1≤exp⁡(σi1/10)\sigma_{i+1}\le\exp(\sigma_i^{1/10}) for all ii, every graph with average degree at least max⁡{d0,σ12}\max\{d_0,\sigma_1^2\} contains a cycle of length σi\sigma_i for some ii. This is Corollary 1.3 of H. Liu and R. Montgomery, A solution to Erdős and Hajnal's odd cycle problem, J. Amer. Math. Soc. 36 (2023), 1191--1234, first posted as arXiv:2010.15802 on 2020-10-29, where it is deduced from the even-cycle interval theorem, Theorem 1.1.

Since log⁡(2x)=o(x1/10)\log(2x)=o(x^{1/10}), the powers of two satisfy the growth condition from some index i0i_0 on. Taking A={2i:i≥i0}A=\{2^i: i\ge i_0\}, a set of density zero, and c=max⁡{d0,4i0}c=\max\{d_0,4^{i_0}\} answers the question of Problem 72 affirmatively, with no condition on the number of vertices. The same argument makes every increasing sequence of even integers with σi+1≤Cσi\sigma_{i+1}\le C\sigma_i unavoidable after finitely many initial terms are dropped. This settles the problem a second time, after Verstraëte's non-constructive proof, and contradicts Erdős's expectation that the powers of two are avoidable.

Acceptance. The paper is a refereed publication in the Journal of the American Mathematical Society (published online 2023-03-31), and the site's curator, Thomas Bloom, records it as proving the statement for the powers of two, which is the reviewed evidence listed. The corpus's source card reconstructs the proof from the arXiv v2 manuscript; that reconstruction is incomplete at one lemma and is author-recorded, not independently reviewed, which concerns the compilation and not the published result. The acceptance recorded here rests on the publication and the site's acceptance.

Depends on. Corollary 1.3, deduced on that page from Theorem 1.1; the compilation is author-recorded and incomplete at Lemma 3.13, which concerns the corpus's reconstruction and not the published result.

Formalization. The file src/latest/ErdosProblems/Erdos72.lean of Boris Alexeev's repository plby/lean-proofs, at the pinned commit, declares itself a Lean formalization of a solution to Problem 72, naming Verstraëte, Liu and Montgomery as its informal authors and, as its formal authors, Codex and GPT-5.6 Sol in its first header and OpenAI Codex in its second. It proves Erdos72.erdos_72, the existence of a set of natural density zero and a constant such that every sufficiently large graph of average degree at least that constant has a cycle with length in the set, by taking the powers of two as the set and deducing their unavoidability (powerTwoUnavoidable) from Liu and Montgomery's finite power-tail theorem, imported from the repository's development for Problem 63; the file's text declares no sorry and ends with a #print axioms command. The formal-conjectures statement file for the problem carries, since 2026-09-20, formal_proof attributes on erdos_72 and on its variant powers_of_two pointing to this file, and the community database records the problem as formalized. The formalization follows this paper's route, so it is a link on this page and not a claim of its own. This corpus has not built or audited the development, so no formalized evidence is listed.