Wiki
Wiki

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

Updated


Statement

Notation (pp. 1--2). A kk-uniform hypergraph G=(V,E)G=(V,E) has a vertex set V⊆NV\subseteq\mathbb N and a family EE of kk-element subsets of VV, its edges; v(G)=∣V∣v(G)=|V| and e(G)=∣E∣e(G)=|E|. A matching is a family of pairwise disjoint edges, and μ(G)\mu(G) is the size of the largest matching in EE. The number νk(n,s)\nu_k(n,s) is the largest number of edges of a kk-uniform hypergraph GG with v(G)=nv(G)=n and μ(G)=s\mu(G)=s, and Mk(n,s)\mathcal M_k(n,s) is the family of extremal hypergraphs: H∈Mk(n,s)H\in\mathcal M_k(n,s) when v(H)=nv(H)=n, μ(H)=s\mu(H)=s and e(H)=νk(n,s)e(H)=\nu_k(n,s). Covk(n,s)\mathrm{Cov}_k(n,s) is the family of hypergraphs on nn vertices whose edges are all kk-subsets meeting a given set S⊆VS\subseteq V with ∣S∣=s|S|=s; such a hypergraph has (nk)−(n−sk)\binom nk-\binom{n-s}k edges.

Theorem 1 (p. 2, quoted). "If k⩾3k\geqslant3 and n>2k2slog⁡kn>\frac{2k^2s}{\log k}, then Mk(n,s)=Covk(n,s)\mathcal M_k(n,s)=Cov_k(n,s)."

The inequality is the paper's display (2). The paper does not state the base of the logarithm. In words: in this range the largest number of edges of a kk-uniform hypergraph on nn vertices whose largest matching has exactly ss edges is νk(n,s)=(nk)−(n−sk)\nu_k(n,s)=\binom nk-\binom{n-s}k, and the hypergraphs attaining it are exactly the covers. The paper presents the theorem (p. 2) as confirming the statement Mk(n,s)=Covk(n,s)\mathcal M_k(n,s)=\mathrm{Cov}_k(n,s) for n≥g(k)sn\ge g(k)s with g(k)≥2k2/log⁡kg(k)\ge2k^2/\log k, against the earlier ranges g(k)≥2k3g(k)\ge2k^3 of Bollobás, Daykin and Erdős and g(k)≥3k2g(k)\ge3k^2 of Huang, Loh and Sudakov.

The abstract (p. 1) states the range as n>3k2s/2log⁡kn>3k^2s/2\log k and names the hypergraph HH where it introduced GG; the theorem on p. 2 and its proof use display (2), and this page follows the theorem.

Source. P. Frankl, T. Łuczak and K. Mieczkowska, On matchings in hypergraphs, Electron. J. Combin. 19(2) (2012), Paper 42, as identified on the source card: Theorem 1 on p. 2, with the definitions on pp. 1--2.

Read depth. Claims checked: the statement and the definitions it uses were read clause by clause on the print. The proof (pp. 2--4) was read for its structure only; nothing here is independently reviewed.

Proof pointer

Pages 2--4, by shifting. By the paper's Lemmas 2 and 3 (p. 2, stated as well known, with pointers to Frankl's shifting survey and to Łuczak and Mieczkowska), it suffices to treat a shifted hypergraph HH. Lemma 4 (p. 2) places every edge of a shifted GG on [n][n] with μ(G)=s\mu(G)=s in the union of the families Ai\mathcal A_i of kk-sets meeting {1,…,i(s+1)−1}\{1,\ldots,i(s+1)-1\} in at least ii elements, i=1,…,ki=1,\ldots,k. Lemma 5 (p. 3) deduces that for n≥k(s+1)−1n\ge k(s+1)-1 all but at most s(s+1)2(n−1k−2)\frac{s(s+1)}2\binom{n-1}{k-2} edges meet {1,…,s}\{1,\ldots,s\}. Claim 6 (p. 3) uses this count to show that for s≥2s\ge2 the edge {1,ks+2,…,ks+k}\{1,ks+2,\ldots,ks+k\} is present, and Claim 7 (p. 4) that every kk-set containing vertex 11 is then an edge, so deleting vertex 11 and its edges leaves a member of Mk(n−1,s−1)\mathcal M_k(n-1,s-1). Display (2) survives replacing n,sn,s by n−1,s−1n-1,s-1, and the case s=1s=1 is the Erdős–Ko–Rado theorem.

Bears on

  • Problem 1020: the problem asks whether, for r≥3r\ge3 and n≥krn\ge kr, the largest number f(n;r,k)f(n;r,k) of edges in an rr-uniform hypergraph on nn vertices with no kk independent edges is max⁡((rk−1r),(nr)−(n−k+1r))\max\left(\binom{rk-1}r,\binom nr-\binom{n-k+1}r\right). The paper's uniformity kk is the problem's rr and its matching number ss is the problem's k−1k-1. Take r≥3r\ge3, k≥2k\ge2 and n>2r2(k−1)/log⁡rn>2r^2(k-1)/\log r. A hypergraph with no kk independent edges has matching number some s′≤k−1s'\le k-1; for s′≥1s'\ge1 the range holds with s′s' in place of k−1k-1, so the theorem bounds its edges by (nr)−(n−s′r)≤(nr)−(n−k+1r)\binom nr-\binom{n-s'}r\le\binom nr-\binom{n-k+1}r. A cover on a (k−1)(k-1)-set attains the bound, and the clique on rk−1rk-1 vertices has no kk independent edges, so the value is the problem's maximum: the theorem gives f(n;r,k)=(nr)−(n−k+1r)f(n;r,k)=\binom nr-\binom{n-k+1}r for n>2r2(k−1)/log⁡rn>2r^2(k-1)/\log r. This deduction is this page's; the paper states only the theorem. It says nothing for smaller nn.