On problems of Erdős and Rudin. (Q1427633)

The still open Rudin conjecture on the set of squares implies that \((*)\) for all \(\varepsilon>0\) there exists a constant \(c_\varepsilon\) such that \((\int_\Pi | \sum_{j=1}^k e^{in_j^2 x}| ^4\,dx)^{1/4}<c_\varepsilon k^{{1\over2}+\varepsilon}\) for any \(k\) distinct integers \(n_1,\dots,n_k\), where \(\Pi\) denotes the usual circle group. The author proves that \((\int_\Pi | \sum_{j=1}^k e^{in_j^2 x}| ^4\,dx)^{1/4}<C_\varepsilon {k^{{3\over2}}\over (\log k)^{{1\over 48}-\varepsilon}}\) for any \(n_1,\dots,n_k\) and all \(\varepsilon>0\). The main ingredients of the proof is Gower's version of the Balog-Szemerédi theorem, Freiman's theorem and results on the density of squares in arithmetical progressions. Then in the paper a combinatorial interpretation of \((*)\) in the spirit of `sums and products along graphs' as considered by \textit{P. Erdős} and \textit{E. Szemerédi} [Studies in Pure Mathematics, to the memory of P. Turán, 213--218 (1983; Zbl 0526.10011)] is given. More precisely, if \(A=\{a_i\in{\mathbb Z}\mid a_i<a_j\text{ if }i<j\}\) is a set of \(n\) distinct integers and \(G\subset\{(i,j)\mid i,j\in{\mathbb Z},1\leq i,j\leq n\}\) is a graph then \(\text{Sum}_GA=\{a_i+a_j\mid (i,j)\in G\}\) and similarly \(\text{Diff}_GA\) and \(\text{Prod}_GA\) for the difference and product, resp. can be defined. In the paper it is proved that \((*)\) is equivalent to the statement: \(| \text{Sum}_GA| \cdot| \text{Diff}_GA| \cdot| \text{Prod}_GA| >c_\varepsilon| G| ^{2-\varepsilon}\) for all \(\varepsilon>0\), and \(A\subset{\mathbb Z}\). Then it is proved that \((*)\) follows from one of the Erdős conjectures: If \(| G| >cn\), then for \(a_a,\dots,a_k\in{\mathbb Z}\) we have \(| \text{Sum}_GA| +| \text{Prod}_GA| >c_\varepsilon | G| ^{1-\varepsilon}\).