Wiki
Wiki

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

Updated


Source. Corollary 13, p. 10, of Ben Green, The Cameron-Erdős conjecture, Bull. London Math. Soc. 36 (2004), no. 6, 769--778, cited from the arXiv manuscript math/0304058v1 (4 April 2003) whose pages the labels below follow, as identified on the source card.

Statement

Sum-free means that no x,y,zx,y,z in the set satisfy x+y=zx+y=z, and [N]={1,…,N}[N]=\{1,\ldots,N\} (p. 1).

Corollary 13 (p. 10). "With o(2N/2)o(2^{N/2}) exceptions, all sum-free subsets of [N][N] consist entirely of odd numbers, or else are contained in {⌈(N+1)/3⌉,…,N}\{\lceil(N+1)/3\rceil,\ldots,N\}." (quoted)

So the number of sum-free A⊆[N]A\subseteq[N] that contain an even number and also an element of {1,…,⌈(N+1)/3⌉−1}\{1,\ldots,\lceil(N+1)/3\rceil-1\} is o(2N/2)o(2^{N/2}) as N→∞N\to\infty.

Proof pointer

Proof on pp. 10--11. Each sum-free AA lies in a member FF of the family of Proposition 6; since that family has 2o(N)2^{o(N)} members, the sets AA whose container has ∣F∣≤(12−1120)N|F|\le(\tfrac12-\tfrac1{120})N number o(2N/2)o(2^{N/2}). For larger FF, Proposition 7 (p. 8) puts FF, up to a small exceptional part, either inside a short interval or almost entirely among the odd numbers. The paper then bounds the sets AA that break the dichotomy by counting choices over disjoint pairs from which AA can take at most one element, except in the interval case when AA has at most 32ϵ1/8N32\epsilon^{1/8}N elements in [(1−1120)N,N][(1-\tfrac1{120})N,N]; there it counts AA by the bound 2N/2+o(N)2^{N/2+o(N)} applied to the sum-free set A∩[1,(1−1120)N]A\cap[1,(1-\tfrac1{120})N]. The proof on p. 11 cites "Theorem 12" for that bound, which the paper states as Proposition 12 (p. 10).

Dependencies

Proposition 6, Proposition 7 (p. 8) and Proposition 12 (p. 10). Read depth: claims checked; the statement was read on p. 10 and the proof for its structure only.

Bears on

  • Problem 748: the corollary is the structural step that reduces the count of all sum-free subsets of {1,…,n}\{1,\ldots,n\} to the sets of odd numbers and the sum-free subsets of the top interval, which with the Cameron-Erdős count of the latter gives Theorem 2; on its own it gives no count.