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. 25--26). An rr-graph G=(V,T)G=(V,T) has a finite vertex set VV and a set TT of rr-element subsets of VV, its rr-tuples or edges; edges are independent when they are pairwise disjoint, and the paper assumes r≥2r\ge2 throughout. For 0⩽k⩽n0\leqslant k\leqslant n, Er(n,k)E_r(n,k) is the rr-graph on nn vertices whose edges are all rr-sets meeting a fixed kk-set WW; it has

er(n,k)=(nr)−(n−kr)e_r(n,k)=\binom nr-\binom{n-k}r

edges and no k+1k+1 independent edges. The second configuration Fr(n,k)=(V1,T1)F_r(n,k)=(V_1,T_1) has ∣V1∣=n≥k+r|V_1|=n\ge k+r, disjoint sets W1,R⊂V1W_1,R\subset V_1 with ∣W1∣=k−1|W_1|=k-1 and ∣R∣=r|R|=r, and a vertex v∈V1−W1−Rv\in V_1-W_1-R; its edges are the rr-sets meeting W1W_1, the rr-sets containing vv and meeting RR, and RR itself. It has

fr(n,k)=(nr)−(n−kr)−(n−k−rr−1)+1=er(n,k)−(n−k−rr−1)+1f_r(n,k)=\binom nr-\binom{n-k}r-\binom{n-k-r}{r-1}+1=e_r(n,k)-\binom{n-k-r}{r-1}+1

edges; the paper notes that "If n⩾(k+1)n\geqslant(k+1) [sic]" it is a maximal rr-graph without k+1k+1 independent rr-tuples (p. 26). The printed condition is weaker than the requirement n⩾k+rn\geqslant k+r of the definition, which suggests a misprint; the print does not correct it.

Theorem 1 (p. 27, quoted). "Let G=(V,T)G=(V,T) be an rr-graph with r⩾2r\geqslant2, k⩾1k\geqslant1, ∣V∣=n>2r3k|V|=n>2r^3k and ∣T∣>fr(n,k)|T|>f_r(n,k). Suppose GG contains at most kk independent rr-tuples. Then G⊂Er(n,k)G\subset E_r(n,k); in other words there exists W⊂VW\subset V with ∣W∣=k|W|=k such that every rr-tuple of GG intersects WW."

The paper presents the theorem (p. 26) as extending the Hilton–Milner theorem, its case k=1k=1 (there for n≥2rn\ge2r), to every k≥1k\ge1, and as a sharper and more explicit form of Erdős's 1965 result that some constant crc_r makes er(n,k)+1e_r(n,k)+1 edges on n>crkn>c_rk vertices force k+1k+1 independent edges. The graph Fr(n,k)F_r(n,k) shows that the bound fr(n,k)f_r(n,k) on the number of edges cannot be lowered (p. 26).

Source. B. Bollobás, D. E. Daykin and P. Erdős, Sets of independent edges of a hypergraph, Quart. J. Math. Oxford Ser. (2) 27 (1976), 25--32, as identified on the source card: Theorem 1 on p. 27, with the definitions on pp. 25--26.

Read depth. Claims checked: the statement and the definitions it uses were read clause by clause on the print. The proof (pp. 27--28) was read for its structure only; nothing here is independently reviewed.

Proof pointer

Pages 27--28, by induction on kk, the case k=1k=1 being the Hilton–Milner theorem. If deleting some vertex uu leaves at most k−1k-1 independent edges, the remaining edges number more than fr(n−1,k−1)f_r(n-1,k-1) and the induction hypothesis applies to G−uG-u. Otherwise the paper's Lemma 1 (p. 27, an upper bound on the degree of such a vertex and a vertex of degree at least ∣T∣/(rp)|T|/(rp) when at most pp edges are independent) bounds ∣T∣|T| above by r2k2(n−2r−2)r^2k^2\binom{n-2}{r-2}, which the binomial inequalities (1) and (2) of p. 27 show to be incompatible with ∣T∣>fr(n,k)|T|>f_r(n,k) when n>2r3kn>2r^3k.

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 kk is the problem's k−1k-1. Applied with the paper's k=k′−1≥1k=k'-1\ge1, the theorem gives f(n;r,k′)=er(n,k′−1)=(nr)−(n−k′+1r)f(n;r,k')=e_r(n,k'-1)=\binom nr-\binom{n-k'+1}r for n>2r3(k′−1)n>2r^3(k'-1): a hypergraph with no k′k' independent edges has at most fr(n,k′−1)≤er(n,k′−1)f_r(n,k'-1)\le e_r(n,k'-1) edges or lies in Er(n,k′−1)E_r(n,k'-1), which attains the bound. The clique on rk′−1rk'-1 vertices has no k′k' independent edges, so (rk′−1r)≤er(n,k′−1)\binom{rk'-1}r\le e_r(n,k'-1) in this range and the value is the problem's maximum. This deduction is this page's; the paper states only the theorem and records the conjecture (p. 26). It settles the problem for n>2r3(k′−1)n>2r^3(k'-1) and says nothing for smaller nn.