When subset-sums do not cover all the residues modulo \(p\). (Q1427982)

If \({\mathcal A}\) is a subset of some group, \(S_{\mathcal A}\) denotes the sum of the elements of \({\mathcal A}\) and \({\mathcal A}^*= \{S_{\mathcal B},\,{\mathcal B}\subset{\mathcal A}\}\). In [Acta Arith. 9, 149--159 (1964; Zbl 0156.04801)], \textit{P. Erdős} and \textit{H. Heilbronn} proved that every subset \({\mathcal A}\subset\mathbb{Z}/p\mathbb{Z}\setminus \{0\}\) such that \(|{\mathcal A}|\geq 3\sqrt{6p}\) satisfies \({\mathcal A}^*= \mathbb{Z}/p\mathbb{Z}\). They conjectured that the conclusion still holds if we replace the coefficient \(3\sqrt{6}\) by \(2\). Olson proved it and finally, Dias da Silva and ould Hamidoune proved a best possible result of this form. The present paper looks for subsets \({\mathcal A}\subset \mathbb{Z}/p\mathbb{Z}\) such that \(|{\mathcal A}|\) is large and \({\mathcal A}^*\neq \mathbb{Z}/p\mathbb{Z}\). Observe that as soon as \(\sum_{a\in{\mathcal A}}\| a/p\|< 1-1/p\) \((\| z\|= \min_{n\in\mathbb{Z}}| z-n|)\), \({\mathcal A}^*\neq \mathbb{Z}/p\mathbb{Z}\) must hold. More generally, if for some non-zero modulo \(p\) integer \(s\), \(\sum_{a\in{\mathcal A}}\| sa/p\|< 1-1/p\), then \({\mathcal A}^*\neq \mathbb{Z}/p\mathbb{Z}\) must hold. The present paper looks for a kind of converse statement for this criterion. The following theorem is proved: Let \(c> \sqrt{2}\) and \(p\) be a prime. Let \({\mathcal A}\subset \mathbb{Z}/p\mathbb{Z}\), with \(|{\mathcal A}|\geq c\sqrt{p}\). If we assume \({\mathcal A}^*\neq \mathbb{Z}/p\mathbb{Z}\) then \(\sum_{a\in{\mathcal A}}\| as/p\|< 1+ O(p^{-1/4}\log p)\) for some integer \(s\) that is, non-zero modulo \(p\).