Wiki
Wiki

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

Updated


Statement

Theorem 1.2 (p. 3, display (4)). For fixed s≥4s\ge4 and sufficiently large nn,

log⁡r3(s,n)≤(s−3(s−2)!+o(1))ns−2log⁡n.\log r_3(s,n)\le\left(\frac{s-3}{(s-2)!}+o(1)\right)n^{s-2}\log n.

Here r3(s,n)r_3(s,n) is the least NN such that every red-blue coloring of the triples of an NN-element set has a red set of size ss or a blue set of size nn, a set being red (blue) when all its triples are (p. 2); the logarithm is natural (p. 2). The abstract states the bound in the form r3(s,n)≤2ns−2log⁡nr_3(s,n)\le2^{n^{s-2}\log n} for fixed ss (p. 1).

The bound it improves is the 1952 Erdős--Rado inequality rk(s,n)≤2(rk−1(s−1,n−1)k−1)r_k(s,n)\le2^{\binom{r_{k-1}(s-1,n-1)}{k-1}} (display (3)), which with the graph bound (1) gives r3(s,n)≤2c n2s−4/log⁡2s−6nr_3(s,n)\le2^{c\,n^{2s-4}/\log^{2s-6}n} for fixed ss (p. 3). The paper adds that this result together with (3) gives a similar improvement for higher uniformity (p. 3).

Source. D. Conlon, J. Fox and B. Sudakov, Hypergraph Ramsey numbers, arXiv:0808.3760v1, Theorem 1.2 on p. 3; Theorem 2.1 on p. 6, Lemma 2.2 and Corollary 2.3 on p. 7 (J. Amer. Math. Soc. 23 (2010), 247--266, not compared). The edition read is identified on the source card.

Read depth. Claims checked: the statements of Theorem 1.2, Theorem 2.1, Lemma 2.2 and Corollary 2.3 were read clause by clause on the page images. The proofs were read for the outline below; their steps were not checked.

Proof pointer

Section 2, pp. 4--9. The paper refines the Erdős--Rado greedy argument (p. 4) with the vertex on-line Ramsey game (p. 5): vertices arrive one at a time, a builder chooses which edges back to earlier vertices to expose, and a painter colors each exposed edge at once. Theorem 2.1 (p. 6): if the builder can force a red Ks−1K_{s-1} or a blue Kn−1K_{n-1} using at most vv vertices, rr red edges and mm edges in total, then for every 0<α≤1/20<\alpha\le1/2, r3(s,n)≤(v+1)α−r(1−α)r−mr_3(s,n)\le(v+1)\alpha^{-r}(1-\alpha)^{r-m}. Lemma 2.2 (p. 7) gives a builder strategy forcing a red KsK_s or a blue KnK_n with at most (s+n−2s−1)\binom{s+n-2}{s-1} vertices, (s−2)(s+n−2s−1)+1(s-2)\binom{s+n-2}{s-1}+1 red edges and (s+n−4)(s+n−2s−1)+1(s+n-4)\binom{s+n-2}{s-1}+1 edges in all. Combining the two with α=r/m\alpha=r/m gives Corollary 2.3 (p. 7): for 4≤s≤n4\le s\le n,

r3(s,n)≤2s−3(s−2)!(s+n)s−2log⁡2(64n/s),r_3(s,n)\le2^{\frac{s-3}{(s-2)!}(s+n)^{s-2}\log_2(64n/s)},

which the paper says implies (4) (p. 7).

Dependencies

Theorem 2.1, Lemma 2.2 and Corollary 2.3 of the same paper; the greedy argument of Erdős and Rado (the paper's [15]).

Bears on

No problem page of this corpus asks for the off-diagonal numbers r3(s,n)r_3(s,n) with ss fixed; none is linked.