Wiki
Wiki

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

Updated


Statement

Theorem 2 (p. 324, quoted). "Suppose that H=(V,E)H=(V,\mathcal E), ∣V∣=n≥5\lvert V\rvert=n\ge5 and any 4 points of VV span 0 or 2 edges. Then max⁡∣E∣\max\lvert\mathcal E\rvert is attained exactly for HH of the form HSH_S for some equipartition, i.e., $\lfloor n/6\rfloor\le\lvert V_i\rvert\le\lceil n/6\rceil$."

Here HSH_S is the blow-up of S(6)S(6) over a partition V=V1∪⋯∪V6V=V_1\cup\dots\cup V_6 (Example 1, p. 323), and a hypergraph has every vertex in some edge (p. 323). The argument on p. 327 adds that the Example 2 bound equals the Example 1 maximum only for n≤5n\le5, where "the two examples coincide"; so at n=5n=5 the extremal HSH_S is also of the circle form of Example 2.

Source. P. Frankl and Z. Füredi, An exact result for 3-graphs, Discrete Math. 50 (1984), 323--328, doi:10.1016/0012-365X(84)90058-X; Theorem 2 on p. 324, its deduction from Theorem 1 on p. 327. The edition is identified on the source card.

Read depth. Claims checked: the statement and the p. 327 comparison were read clause by clause on the page images. The count behind the comparison was not replayed, and the paper writes out no argument that the equipartition maximizes the edge count of HSH_S.

Proof pointer

P. 327. By Theorem 1 an extremal 3-graph is of the form of Example 1 or Example 2. For Example 2, with d1,…,d2k+1d_1,\ldots,d_{2k+1} points placed on the vertices of a regular (2k+1)(2k+1)-gon, the edge count is largest when the did_i are as equal as possible, which bounds it by (n/(2k+1))3(2k+1)(k+12)/3(n/(2k+1))^3(2k+1)\binom{k+1}2/3, less than n3/24n^3/24; the paper states that this never exceeds the largest 3-graph of Example 1, with equality only for n≤5n\le5.

Dependencies

Theorem 1; Example 1.

Bears on

  • Problem 794: a 3-graph whose four-point sets span zero or two edges has no four vertices spanning three edges, so it counts toward m(n,3,4,3)m(n,3,4,3), the quantity of the site commentary's reading. Theorem 2 shows that within this class nothing beats HSH_S; the larger iterated construction behind the lower bound of Theorem 3 leaves the class: a triple added inside V1V_1 together with a vertex of V2V_2 spans exactly one edge (an observation of this page). The problem page names Theorems 1--2 only as context, recording that they show HSH_S over an equipartition extremal for n≥5n\ge5 under the stricter local condition.