Wiki
Wiki

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

Updated


Source. Published pp. 284–285, Section 11 and Theorem 11.1 (PDF).

Let m(k)m(k) be the least size of a family of 2k2k-subsets of [4k][4k] such that every 2k2k-set GG has ∣F∩G∣=k|F\cap G|=k for some family member FF.

Statement. There is an absolute c>0c>0 such that m(k)≥ckm(k)\ge ck for every positive odd integer kk. Also m(k)≤2km(k)\le2k for every positive integer kk.

Proof. Let V⊆F24kV\subseteq\mathbb F_2^{4k} be the span of the characteristic vectors of a qualifying family F\mathcal F. When kk is odd, every weight-2k2k vector has nonzero inner product with at least one generator, so V⊥V^\perp contains no vector of weight 2k2k. Since it is a linear subspace, it contains no pair at Hamming distance 2k2k either: their difference would have that weight. Theorem 1.10 at alphabet size two, length 4k4k, and distance 2k2k gives ∣V⊥∣≤24ke−c04k|V^\perp|\le2^{4k}e^{-c_0 4k} for an absolute c0>0c_0>0. Taking base-two logarithms and using orthogonal-complement dimensions,

∣F∣≥dim⁡V=4k−dim⁡V⊥≥4c0log⁡2k.|\mathcal F|\ge\dim V=4k-\dim V^\perp \ge\frac{4c_0}{\log2}k.

If a theorem threshold was imposed before its finite-dimensional extension, decrease the final constant to include the finitely many smaller positive odd kk.

For the upper bound, cyclically order [4k][4k] and let FiF_i be the consecutive window of length 2k2k starting at ii, for 1≤i≤2k1\le i\le2k. Fix a 2k2k-set GG and also consider F2k+1=F1cF_{2k+1}=F_1^c. The integers ai=∣Fi∩G∣a_i=|F_i\cap G| satisfy ∣ai+1−ai∣≤1|a_{i+1}-a_i|\le1 and a2k+1=2k−a1a_{2k+1}=2k-a_1. If a1=ka_1=k we are done; otherwise the endpoints are on opposite sides of kk, so an integer intermediate value equals kk. If the last endpoint equals kk, the first does too. Thus one of the first 2k2k windows works for every GG, proving the upper bound. □\square

Source precision and historical scope. The printed interval [i,i+k−1][i,i+k-1] in the upper construction has the wrong size for a family of 2k2k-sets; the proof uses length 2k2k as above. The paper's later Conjecture 11.2 is explicitly reported as proved in its own added-in-proof note. No current open-problem or optimal-constant claim is inferred from the earlier conjecture paragraph.

Dependencies. theorem_1_10.