Inverse questions for the large sieve (Q484089)

Let \(\mathcal{A}\) be an infinite set, \(x\geq 1\) and \(\mathcal{A}(x)=\mathcal{A}\cap\{1,2,\dots,x\}\). If the set \(\mathcal{A}\) occupies at most \((p+1)/2\) residue classes modulo \(p\), for every sufficiently large prime number \(p\), then the large sieve inequality implies that \(\mathcal{A}(x)\ll x^{1/2}\). This estimate is sharp up to the value of the constant in \(\ll\), as shown by taking \(\mathcal{A}\) to be the set of squares or the set of integer values taken by any rational quadratic. According to the authors the simplest form of the inverse large sieve problem asks whether whether they are the only examples. The authors prove various results and formulate various conjectures related with the inverse large sieve problem. Below there are presented two typical results from the paper. Theorem A. Suppose that for each prime \(p\leq x^{1/2}\) one has a set \(S_p\subset\mathbb{Z}/p\mathbb{Z}\) of size at most \((p+1)/2\). Suppose there is some \(\delta>0\) such that, for each \(p\), \(S_p\) has at least \((\delta+1/16)\,p^3\) quadruples \((s_1,s_2,s_3,s_4)\) with property \(s_1+s_2=s_3+s_4\). If \(\mathcal{A}\pmod p \subset S_p\) for all \(p\), then \(\mathcal{A}(x)\leq x^{1/2-c\delta^2}\) with an absolute positive constant \(c\). Theorem B. Suppose that \(\mathcal{A}\) is a set of integers such that \(\mathcal{A}\pmod p\) lies in some interval \(I_p\) for each prime \(p\). {\parindent=6mm\begin{itemize}\item[{\(\bullet\)}] If \(|I_p|\leq(1-\varepsilon)\, p, \;\varepsilon\in(0,1),\) for at least a proportion \(\varepsilon\) of the primes in each dyadic interval \([z,2z]\), then \(\mathcal{A}(x)\ll_\varepsilon(\log\log x)^{c\log(1/\varepsilon)}\) with some positive absolute constant \(c\). \item[{\(\bullet\)}] If \(|I_p|\leq \,p/2\) for all \(p\), then \(|\mathcal{A}(x)|\leq(\log\log x)^{\gamma+o(1)}\), where \(\gamma=\log 18/\log(3/2)\).\item[{\(\bullet\)}] If \(I_p=[\alpha p,\beta p]\) for all primes \(p\) and some fixed \(0\leq\alpha<\beta<1\), then \(|\mathcal{A}|\ll_{\beta-\alpha}1\).\end{itemize}}