Kneser's theorem for upper Banach density (Q873834)

Suppose \(A \subset \mathbb{N}\) has a certain density \(\alpha\) and \(A + A\) has density less than \(2\alpha\). One possible reason for this could be that for some \(g \in \mathbb{N}\), we have \(A \subset [0, \lceil \alpha g\rceil - 1] + g\mathbb{N}\), so that \(A + A \subset [0, 2\lceil \alpha g\rceil - 1] + g\mathbb{N}\). A theorem of Kneser essentially states that if one works with the lower density, then this is in fact the only reason. The present article gives an analogue for the upper Banach density, which is defined as \(\text{BD}(A) := \lim_n \sup_k| A \cap [k,k+n]| /(n+1)\). More precisely, the result is the following. Suppose that \(\text{BD}(A) = \alpha\) and \(\text{BD}(A+A) < 2\alpha\). Then there exists a \(g \in \mathbb{N}\) and a set \(W\) which is the union of \(\lceil 2\alpha g - 1\rceil\) arithmetic progressions of difference \(g\) such that \(A+A \subset W\). Moreover, suppose that \([a_n, b_n]\) are intervals whose lengths tend to \(\infty\) and in which the relative density of \(A\) approaches \(\alpha\). (Such intervals exist by definition of the Banach density.) Then there are slightly smaller intervals \([c_n,d_n] \subset [a_n, b_n]\) such that on \([2c_n,2d_n]\), \(A+A\) is exactly equal to \(W\). ``Slightly smaller'' means \((d_n-c_n)/(b_n-a_n) \to 1\). The proof uses nonstandard analysis to simplify asymptotic arguments (e.g. concerning the upper Banach density). The article contains a quick introduction to nonstandard analysis.