A note on Behrend sequences (Q2563787)

Let \(\mathcal A\) be a sequence of integers greater than 1 and \({\mathcal M}({\mathcal A})= \{n:\exists a|n\) with \(a\in{\mathcal A}\}\). If \({\mathcal M}({\mathcal A})\) has asymptotic density 1, then \(\mathcal A\) is said to be a Behrend sequence. \textit{P. Erdös} [Can. J. Math. 11, 161-167 (1959; Zbl 0092.04301)] showed in effect that \({\mathcal A}=\{pq: p<q\leq p^{1+\varepsilon_p}\}\), where \(p\), \(q\) denote primes, is a Behrend sequence if and only if \(\sum_p\min(1,\varepsilon_p)p^{-1}\) diverges. His result follows from the main result of this paper in which a necessary and sufficient condition for \(\mathcal A\) to be a Behrend sequence is established for the case when the elements of \(\mathcal A\) have exactly two prime factors (possibly equal). This condition requires a certain sum (too complicated to give here) to be divergent. The `obvious' generalization of Erdös' criterion, namely the divergence of \[ \sum_p\min(1, \lambda_p)p^{-1},\quad\text{where }\lambda_p= \sum_{\substack{ pq\in{\mathcal A}\\ q>p}} q^{-1}, \] is necessary but not sufficient for \(\mathcal A\) to be Behrend, as is demonstrated by an example. The paper ends with an arithmetical version of a probabilistic proof of Behrend's inequality, due to I. Z. Ruzsa.