Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Conlon 2023 new bound brown erdos sos problem
conjecture_1_1: The Brown–Erdős–Sós conjecture in the special form the paper states: for every e >= 3, a 3-uniform hypergraph on n vertices with no e edges spanned by at most e + 3 vertices has o(n^2) edges; the paper does not prove it.
corollary_2: The r-uniform form of the paper's main theorem: for every 2 <= k < r and e >= 3, an r-uniform hypergraph on n vertices with no e edges spanned by at most (r-k)e + k + ceil(26 log e / log log e) - 2 vertices has o(n^k) edges.
proposition_1_2: The folklore reduction the paper proves: for 2 <= k < r, e >= 3 and d >= 1, f_r(n, (r-k)e + k + d, e) is at most an explicit factor of order n^(k-2) times f_3(n, e + 2 + d, e), so 3-uniform bounds give r-uniform ones.
theorem_1: Conlon, Gishboliner, Levanzov and Shapira's main theorem: for every e >= 3, a 3-uniform hypergraph on n vertices with no e edges spanned by at most e + ceil(26 log e / log log e) vertices has o(n^2) edges.
theorem_4: The paper's application to the Erdős–Gyárfás generalized Ramsey function: there is an absolute constant C such that colorings of K_n in which every K_p gets at least q colors need binom(n,2) - o(n^2) colors whenever p >= 4 and q >= q_quad(p) + C log p / log log p.
Conlon, David and Gishboliner, Lior and Levanzov, Yevgeny and Shapira, Asaf, A new bound for the {B}rown-{E}rdős-Sós problem. J. Combin. Theory Ser. B 158 (2023), 1--35. https://doi.org/10.1016/j.jctb.2022.08.005. The copy read for this card is arXiv:1912.08834v2 (9 June 2022). The arXiv record names arXiv's non-exclusive distribution license (arXiv:1912.08834), every other right reserved.
The Brown-Erdos-Sos conjecture (Conjecture 1.1, p. 2) asks whether f_3(n, e+3, e) = o(n^2) for every e >= 3, where f_3(n,v,e) is the largest number of edges in a 3-uniform hypergraph on n vertices with no e edges spanned by at most v vertices; the best previous bound, due to Sarkozy and Selkow, had error term 2 + floor(log_2 e). Theorem 1 (p. 2) improves this to f_3(n, e + ceil(26 log e / log log e), e) = o(n^2) for every e >= 3, the first general asymptotic improvement; the paper remarks, without giving the details, that sharper factorial estimates replace the constant 26 by 6 + o(1). The proof applies the r-uniform hypergraph removal lemma for all r, not just r = 3, which is what breaks the barrier in the Sarkozy-Selkow argument. Proposition 1.2 (p. 2) records the folklore reduction showing the general k, r form of the conjecture follows from the 3-uniform case, with an explicit inequality bounding f_r(n, (r-k)e + k + d, e) in terms of f_3(n, e+2+d, e). Combining the two, Corollary 2 (p. 3) gives, for every 2 <= k < r and e >= 3, f_r(n, (r-k)e + k + ceil(26 log e / log log e) - 2, e) = o(n^k). The paper is cited for Problem 1178 on the Brown-Erdos-Sos question: the case k = 2 of Corollary 2 bounds the problem's d_r(e) above by (r-2)e + ceil(26 log e / log log e), against the conjectured (r-2)e + 3. Section 5 applies Corollary 2 with r = 4 to the Erdos-Gyarfas generalized Ramsey function g(n,p,q): Theorem 4 (p. 27) improves Sarkozy and Selkow's range for g(n,p,q) = binom(n,2) - o(n^2) to q >= q_quad(p) + C log p / log log p.
Source: https://arxiv.org/abs/1912.08834.
Bears on.
- #1178: the case k = 2 of Corollary 2 gives d_r(e) <= (r-2)e + ceil(26 log e / log log e) for all r, e >= 3 (Theorem 1 is its case r = 3), against the conjectured d_r(e) = (r-2)e + 3; it is an upper bound above the conjectured value and settles no case. Conjecture 1.1 is the problem's upper half for r = 3, and Proposition 1.2 with k = 2, d = 1 carries it to every r; the paper does not prove the lower bound d_r(e) >= (r-2)e + 3.
Results.
- Conjecture 1.1 (p. 2): Brown-Erdos-Sos conjecture: f_3(n, e+3, e) = o(n^2) for every e >= 3; settled only for e = 3 (Ruzsa-Szemeredi (6,3)-theorem).
- Theorem 1 (p. 2): For every e >= 3, f_3(n, e + ceil(26 log e / log log e), e) = o(n^2); the paper remarks, without details, that the constant 26 can be taken as 6 + o(1).
- Proposition 1.2 (p. 2): For 2 <= k < r, e >= 3, d >= 1: f_r(n, (r-k)e + k + d, e) <= (binom(r-k+2,3)(e-1)+1) * binom(n,k-2)/binom(r,k-2) * f_3(n, e+2+d, e), reducing the general conjecture to the 3-uniform case.
- Corollary 2 (p. 3): For 2 <= k < r and e >= 3, f_r(n, (r-k)e + k + ceil(26 log e / log log e) - 2, e) = o(n^k).
- Theorem 4 (p. 27): There is an absolute constant C with g(n,p,q) = binom(n,2) - o(n^2) for every p >= 4 and q >= q_quad(p) + C log p / log log p, where g(n,p,q) is the least number of colors on the edges of K_n giving every K_p at least q colors (the Erdos-Gyarfas function).
Read status. Claims checked: the statements of the five results above were read clause by clause against the arXiv v2 print. The proofs of Proposition 1.2 and Proposition 5.1 and the derivation of Theorem 1 from Lemma 2.1 were read; the proofs of Lemmas 2.4 and 2.6 (Sections 3 and 4) were not checked step by step.
No file of this source is held: the license of the edition read permits no redistribution, and the card cites that edition. The Crossref record of the journal version names CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/) for the version of record, which was not read for this card.