Continuous Ramsey Theory and Sidon Sets

TL;DR

This paper explores the relationship between symmetric subsets of real intervals and Sidon sets, establishing bounds for a function D(x) related to symmetry measures and for the size of B*[g] sets, revealing deep connections between continuous and discrete problems.

Contribution

It introduces new bounds for D(x) and R(g,n), demonstrating a surprising effectiveness of combining harmonic analysis with combinatorial and probabilistic methods.

Findings

D(x) = 2x - 1 for 11/16 < x < 1

0.59 x^2 < D(x) < 0.8 x^2 for 0 < x < 11/16

R(g,n) < 1.31 √(gn) and R(g,n) > 0.79 √(gn) for large g,n

Abstract

A symmetric subset of the reals is one that remains invariant under some reflection x --> c-x. Given 0 < x < 1, there exists a real number D(x) with the following property: if 0 < d < D(x), then every subset of [0,1] with measure x contains a symmetric subset with measure d, while if d > D(x), then there exists a subset of [0,1] with measure x that does not contain a symmetric subset with measure d. In this paper we establish upper and lower bounds for D(x) of the same order of magnitude: for example, we prove that D(x) = 2x - 1 for 11/16 < x < 1 and that 0.59 x^2 < D(x) < 0.8 x^2 for 0 < x < 11/16. This continuous problem is intimately connected with a corresponding discrete problem. A set S of integers is called a B*[g] set if for any given m there are at most g ordered pairs (s_1,s_2) \in S \times S with s_1+s_2=m; in the case g=2, these are better known as Sidon sets. We also establish upper and lower bounds of the same order of magnitude for the maximal possible size of a B*[g] set contained in {1,...,n}, which we denote by R(g,n). For example, we prove that R(g,n) < 1.31 \sqrt{gn} for all n > g > 1, while R(g,n) > 0.79 \sqrt{gn} for sufficiently large integers g and n. These two problems are so interconnected that both continuous and discrete tools can be applied to each problem with surprising effectiveness. The harmonic analysis methods and inequalities among various L^p norms we use to derive lower bounds for D(x) also provide uniform upper bounds for R(g,n), while the techniques from combinatorial and probabilistic number theory that we employ to obtain constructions of large B*[g] sets yield strong upper bounds for D(x).