Wiki
Wiki

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

Updated

Kolupaev 2023 erdos matching conjecture almost perfect matchings

../

theorem_1_2: Kolupaev and Kupavskii's Theorem 1.2: for integers s > k >= 5 with s > 101k^3 and (s+1)k <= n < (s+1)(k + 1/(100k)), every family of k-subsets of [n] with matching number at most s has at most as many members as the family of all k-subsets of [(s+1)k-1].


Kolupaev, Dmitriy and Kupavskii, Andrey, Erdős matching conjecture for almost perfect matchings. Discrete Math. 346 (2023), no. 4, Paper No. 113304, 9. The arXiv record names arXiv's non-exclusive distribution license (arXiv:2206.01526), every other right reserved.

The copy read for this card is arXiv:2206.01526v2, dated December 20, 2022, eleven pages; page numbers are its own. The journal version was not compared.

The Erdős matching conjecture (Conjecture 1.1, p. 1) asserts that for positive integers n,k,sn,k,s with n≥(s+1)kn\ge(s+1)k, every family F\mathcal F of kk-subsets of [n][n] with matching number at most ss satisfies ∣F∣≤max⁡{∣A∣,∣B∣}|\mathcal F|\le\max\{|\mathcal A|,|\mathcal B|\}, where A\mathcal A is the family of all kk-subsets of [(s+1)k−1][(s+1)k-1] and B\mathcal B the family of kk-subsets of [n][n] meeting [s][s]. The paper treats the range where A\mathcal A is extremal, that is where the forbidden matching is almost perfect, and improves Frankl's 2017 theorem, quoted as Theorem 1.1 (p. 2), which covers s>k≥2s>k\ge2 and (s+1)k≤n<(s+1)(k+k−2k−1/2)(s+1)k\le n<(s+1)(k+k^{-2k-1}/2). The main result, Theorem 1.2 (p. 2), proves ∣F∣≤∣A∣|\mathcal F|\le|\mathcal A| for s>k≥5s>k\ge5, s>101k3s>101k^3 and (s+1)k≤n<(s+1)(k+1100k)(s+1)k\le n<(s+1)(k+\frac1{100k}): the window in nn widens from k−2k−1/2k^{-2k-1}/2 to 1/(100k)1/(100k) at the cost of the lower bound on ss. The authors explain that Frankl's theorem needs no such bound because for s<2k2k+1s<2k^{2k+1} its range forces n=(s+1)kn=(s+1)k, the case proved by Kleitman. The proof (sections 2 and 3, pp. 2--10) follows Frankl's framework of shifted families and the trace on [(s+1)k−1][(s+1)k-1], reducing the bound to a weighted comparison over kk-subsets of [s][s]. The authors call it a very interesting question to replace 1/(100k)1/(100k) by some small constant (p. 2).

Results.

  • Theorem 1.2 (p. 2): for s>k≥5s>k\ge5, s>101k3s>101k^3 and (s+1)k≤n<(s+1)(k+1100k)(s+1)k\le n<(s+1)(k+\frac1{100k}), every F⊂([n]k)\mathcal F\subset\binom{[n]}k with matching number at most ss has ∣F∣≤((s+1)k−1k)|\mathcal F|\le\binom{(s+1)k-1}{k}.

Read status: claims checked for Conjecture 1.1 and Theorems 1.1 and 1.2, read clause by clause on the print; the proof was read for its structure only and no step of it was checked. Nothing here is independently reviewed.

Source: https://arxiv.org/abs/2206.01526.

Bears on. #1020: with the problem's rr for the paper's uniformity kk and the problem's kk for s+1s+1, Theorem 1.2 gives the conjectured value f(n;r,k)=(rk−1r)f(n;r,k)=\binom{rk-1}{r} for r≥5r\ge5, k−1>101r3k-1>101r^3 and rk≤n<k(r+1100r)rk\le n<k(r+\frac1{100r}). It covers that range near n=rkn=rk only.

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