Wiki
Wiki

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

Updated


Claim. Let RT(n,K4,m)\mathbf{RT}(n,K_4,m) be the largest number of edges of a K4K_4-free graph on nn vertices with independence number less than mm. Fox, Loh and Zhao prove that if m=ne−f(n)m=ne^{-f(n)} with f(n)=o((log⁡n/log⁡log⁡n)1/2)f(n)=o\bigl((\log n/\log\log n)^{1/2}\bigr), then

RT(n,K4,m)≥(18−o(1))n2,\mathbf{RT}(n,K_4,m)\ge\Bigl(\frac18-o(1)\Bigr)n^2,

by carrying out the quantitative analysis of the Bollobás--Erdős construction that earlier presentations of it left implicit. The paper presents the theorem as settling in the negative its Problem 1.4, the question Erdős, Hajnal, Simonovits, Sós and Szemerédi posed, which is the question of Problem 615 in Ramsey--Turán notation. The passage to the site's wording is one line, made on the problem page and on the library's result page and not spelled out in the paper: n/log⁡n=ne−log⁡log⁡nn/\log n=ne^{-\log\log n} and (log⁡log⁡n)3/2/(log⁡n)1/2→0(\log\log n)^{3/2}/(\log n)^{1/2}\to0, so m=n/log⁡nm=n/\log n lies in the theorem's range; hence for every c>0c>0 and all large nn some K4K_4-free graph on nn vertices has at least (1/8−c)n2(1/8-c)n^2 edges and independence number below n/log⁡nn/\log n, that is, no independent set of n/log⁡nn/\log n or more vertices, and no constant cc has the property the problem asks for. The theorem is paged at Theorem 1.10 of the library's source card, whose edition is the arXiv v3 (23 September 2014).

Scope. Full. The site's wording quantifies over all nn and the sources read it for all large nn; the answer is no in both readings, since the theorem denies the inequality for all large nn and the case n=2n=2 already fails it on its own. The base of the logarithm changes n/log⁡nn/\log n by a constant factor, which the theorem's range absorbs.

Depends on. Nothing in this wiki; the result rests on the cited paper alone.

Acceptance. Reviewed: the site's curator, Thomas Bloom, labels the problem DISPROVED (LEAN) and credits Fox, Loh and Zhao in the problem's commentary with the negative answer and with the quantitative lower bound behind it (page accessed 2026-09-18); the thread and the proof-claim tab are empty, so the commentary is the whole of the site's record. Refereed: Combinatorica 35 (2015), no. 4, 435--476, DOI 10.1007/s00493-014-3025-3, published online 22 October 2014 (the Crossref record and the arXiv listing's journal reference). The edition cited is the arXiv v3; the journal text is not held and was not compared.

Formalization. The file src/latest/ErdosProblems/Erdos615.lean of Boris Alexeev's repository plby/lean-proofs, linked above at the commit of 30 August 2026 that the formal-conjectures statement file for the problem names in its formal_proof attribute (the problem page's Formalization section describes that file), declares itself a formalization of this result: its header names Fox, Loh and Zhao as the informal authors, the formal-conjectures authors as the statement authors and, as formal authors, the automated systems Codex and GPT-5.6 Sol, with Lean 4 and Mathlib pinned at v4.33.0. It states

lean
theorem not_erdos_615 : ¬ ∃ c : ℝ, 0 < c ∧ ∀ᶠ (n : ℕ) in atTop, ∀ G : SimpleGraph (Fin n), (1 / 8 - c) * n ^ 2 ≤ G.edgeFinset.card → ¬ G.CliqueFree 4 ∨ (n : ℝ) / Real.log n ≤ G.indepNum

the negation of the formal-conjectures statement erdos_615, which adds "for all sufficiently large nn" to the site's wording and takes the natural logarithm; proves it from a lemma exists_counterexample, which gives for every c>0c>0 and every NN a K4K_4-free graph on some n≥Nn\ge N vertices with at least (1/8−c)n2(1/8-c)n^2 edges and independence number below n/log⁡nn/\log n, the construction living in a companion module ErdosProblems.Erdos615.Erdos615Construction; aliases erdos_615 to the negation; and ends with #print axioms not_erdos_615 without recording the output. The repository first added the file on 16 August 2026 (authored 15 August 2026) and edited it through 4 September 2026; the pinned commit does not touch it. At the pinned commit the proof file (21,379 bytes, 507 lines) contains no sorry and no axiom line, but the companion module is unexamined here, nothing was built or kernel-checked in this corpus and no statement-fidelity audit exists, so the file is a link and not formalized evidence. It is the artifact behind the (LEAN) suffix of the site's label; the community database lists the problem as disproved (Lean), with no URL, as of its last update of that field on 23 August 2026.

Read depth. Claims checked: Problem 1.4 (p. 3) and Theorem 1.10 with the paragraph before it (p. 4) of the arXiv v3; the proof (Section 8, the isoperimetric estimates on the sphere behind Theorem 8.1; the proof itself is on p. 28; Section 9's modified Bollobás--Erdős graph serves Theorems 1.7 and 1.9) was not read, and nothing is independently reviewed in this corpus. The elementary range check above warrants nothing beyond itself.

Postings. arXiv:1208.3276, v1 of 16 August 2012 (the first posting, which dates this page) and v3 of 23 September 2014, the edition cited; the journal article; the Lean file, first added to its repository on 16 August 2026; the site's problem page, whose thread and proof-claim tab were empty on 2026-09-18.