Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.
Updated
Source. Published p. 262, Theorem 1.7, and pp. 272–274, Section 4 (PDF). This is the Section 4 proof, separate from the later general counting theorem.
Statement. Given , there are and such that, for , a family with and an integer satisfy
Proof. First prove the following middle-layer form: for any there are such that, for large , of density and satisfy
Write and . For a -set , let count members with . The incidence graph is regular of degree on both sides. Lemma 4.1 gives at least choices of with .
Choose a small , set , and let count pairs with
For a fixed pair with intersection , its four atoms have sizes . There are exactly
choices of ; infeasible coefficients mean zero. As , the entropy estimate gives . If (2) failed, . Averaging over the at least popular choices of gives one with
Here ; choose first so its loss is small compared with , then smaller, then smaller, and finally large.
Delete all endpoints of these ordered pairs from the local family. At most members are lost, so at least remain. Each remaining member has an -set as its intersection with , and an -set in the complement. Let consist of inner -sets with residual fiber of size at least . Nonpopular fibers contribute at most members; each popular fiber has at most members. Thus .
Choose sufficiently small relative to . Theorem 1.1 on the -set gives distinct with . Their residual fibers each have size at least , so Theorem 1.4 on the complementary -set gives residual intersection . The required buffers are positive when and is small: is bounded away from both zero and . The reconstructed pair satisfies every condition defining , contradicting its deletion. This proves (2).
Now start with the family in (1). Choose a size level containing at least members. For any fixed , the entropy estimate forces once is small enough and is large. If , average containment over all -subsets to find a containing set on which the -set family has at least its original relative density. If , enlarge the ground set to size by unused points. In either case the new ambient size is within of , all members have size , and their relative density in that layer is at least
If , add two new points and adjoin to every member. The new ambient size and each member's size is ; the target intersection becomes . Otherwise set and . This operation is injective and preserves target pairs under the stated shift. It changes (5) only by an absolute factor.
Apply (2) with a sufficiently small output relative to . After its constants are fixed, choose small compared with , then and smaller still. Equations (5) and verify its hypotheses for large . The resulting pair count is at least
All these pairs were already in the original family, proving (1).
Source precision. A reference to “Theorem 1.2” in the inner-family step on p. 274 is to Theorem 1.1; there is no such numbered theorem. The fiber threshold is taken consistently as after deletion. The large- qualification used in the source proof is necessary in the statement: for fixed small and the full cube, the fraction of pairs with one prescribed intersection is less than one, whereas as . Decreasing the input tolerance cannot remove that obstruction. For example, at the full cube has target-pair proportion , less than for sufficiently small . The proof above includes its threshold.
Dependencies. lemma_4_1, theorem_1_1, theorem_1_4, entropy_estimates.