Wiki
Wiki

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

Updated

Bell 2021 note sunflowers

../

lemma_2: Bell, Chueluecha and Warnke's key lemma: there is a constant C >= 4 such that, with r(p,k) = Cp log k, every r(p,k)-spread family of at least r(p,k)^k sets of size k contains p disjoint sets, for all integers p, k >= 2.

lemma_4: Bell, Chueluecha and Warnke's Lemma 4, from a construction of Alweiss, Lovett, Wu and Zhang, showing Theorem 3 needs its spread parameter: for r at most 0.25 delta^{-1} log(k/eps) some r-spread family of r^k k-sets has a member inside X_delta with probability less than 1 - eps.

theorem_1: Bell, Chueluecha and Warnke's sunflower bound: there is a constant C >= 4 such that every family of at least (Cp log k)^k distinct k-element sets contains a sunflower with p petals, for all integers p, k >= 2.

theorem_3: The main technical estimate of Rao and of Tao as Bell, Chueluecha and Warnke state it: an r-spread family of at least r^k k-sets, with r >= B delta^{-1} log(k/eps), has a member inside the random delta-subset with probability more than 1 - eps.


Bell, T. and Chueluecha, S. and Warnke, L., Note on sunflowers. Discrete Math. 344 (2021), no. 7, 112367. doi:10.1016/j.disc.2021.112367.

Theorem 1 proves that there is a constant C≥4C\geq4 with Sun(p,k)≤(Cplog⁡k)k\mathrm{Sun}(p,k)\leq(Cp\log k)^k for all integers p,k≥2p,k\geq2, where Sun(p,k)\mathrm{Sun}(p,k) is the least ss such that any ss distinct kk-element sets contain a sunflower with pp petals; this removes the log⁡p\log p factor from Rao's bound (Cplog⁡(pk))k(Cp\log(pk))^k (p. 1). The improvement comes from Lemma 2: with r(p,k)=Cplog⁡kr(p,k)=Cp\log k, an r(p,k)r(p,k)-spread family of at least r(p,k)kr(p,k)^k sets of size kk contains pp disjoint sets (p. 1, proved p. 2). Theorem 1 follows by induction on kk with a case split on whether the family is spread (p. 1). The twist over Rao's and Tao's proofs is to partition the ground set at random into 2p2p classes rather than pp and to use linearity of expectation instead of a union bound, invoking their main technical spread estimate (Theorem 3, p. 2) only with error parameter 1/21/2. The appendix (p. 3) derives Theorem 3 from Rao's proof. Lemma 4 (p. 2) shows Theorem 3 is essentially optimal in its spread parameter, by a product construction the paper credits to Alweiss, Lovett, Wu and Zhang, building on Erdős and Rado. For problem 20, in the problem's notation (sets of size nn, sunflowers with kk petals), Theorem 1 gives f(n,k)≤(Cklog⁡n)nf(n,k)\leq(Ck\log n)^n; this is not the cknc_k^n bound the problem asks for, and the paper states that the Erdős–Rado conjecture remains open (p. 1).

Source: https://arxiv.org/abs/2009.09327. The copy read for this card is the arXiv preprint, version 2 (revised March 17, 2021), and the labels and page numbers below are its. The arXiv record names arXiv's non-exclusive distribution license (arXiv:2009.09327), every other right reserved.

Read status: claims checked for Theorem 1, Lemma 2, Theorem 3 and Lemma 4, read clause by clause on the page images; the proofs of Theorem 1, Lemma 2 and Lemma 4 and the appendix derivation of Theorem 3 followed; the arguments of Rao and Tao behind Theorem 3 were not read here. Nothing here is independently reviewed. Result pages: theorem_1, lemma_2, theorem_3 and lemma_4.

Bears on. #20: Theorem 1 (p. 1) gives f(n,k)≤(Cklog⁡n)nf(n,k)\leq(Ck\log n)^n for all n,k≥2n,k\geq2, with the problem's nn and kk as the paper's kk and pp; this is a bound of the form (log⁡n)n(1+o(1))(\log n)^{n(1+o(1))} for fixed kk, not the cknc_k^n bound the problem asks for, and it decides nothing about the problem.

Results.

  • Theorem 1 (p. 1): there is a constant C≥4C\geq4 with Sun(p,k)≤(Cplog⁡k)k\mathrm{Sun}(p,k)\leq(Cp\log k)^k for all integers p,k≥2p,k\geq2.
  • Lemma 2 (p. 1): there is a constant C≥4C\geq4 such that, with r(p,k)=Cplog⁡kr(p,k)=Cp\log k, for all integers p,k≥2p,k\geq2 every r(p,k)r(p,k)-spread family of at least r(p,k)kr(p,k)^k sets of size kk contains pp disjoint sets.
  • Theorem 3 (p. 2, the main technical estimate of Rao and Tao): there is a constant B≥1B\geq1 such that for every integer k≥2k\geq2, all reals 0<δ,ϵ≤1/20<\delta,\epsilon\leq1/2 and r≥Bδ−1log⁡(k/ϵ)r\geq B\delta^{-1}\log(k/\epsilon), if a family S\mathcal S of kk-element subsets of a finite set XX is rr-spread with ∣S∣≥rk|\mathcal S|\geq r^k, then with probability more than 1−ϵ1-\epsilon the random subset XδX_\delta contains a member of S\mathcal S.
  • Lemma 4 (p. 2): for all reals 0<δ,ϵ≤1/20<\delta,\epsilon\leq1/2 and integers k≥1k\geq1 and 1≤r≤0.25 δ−1log⁡(k/ϵ)1\leq r\leq0.25\,\delta^{-1}\log(k/\epsilon) there is an rr-spread family of rkr^k kk-subsets of {1,…,rk}\{1,\ldots,rk\} with P(∃S∈S:S⊆Xδ)<1−ϵ\mathbb P(\exists S\in\mathcal S:S\subseteq X_\delta)<1-\epsilon.

No file of this source is held: no license on record permits its redistribution, and the card cites the edition it names above.