Quasi-diagonal behaviour and smooth Weyl sums (Q1573762)

An integer \(n\) is called an \(R\)-smooth number if every prime divisor of \(n\) does not exceed \(R\). Let \({\mathcal A}(P,R)\) be the set of \(R\)-smooth numbers up to \(P\), \(k\) be a fixed integer \(>2\), and define \[ f(\alpha;P,R)=\sum_{x\in{\mathcal A}(P,R)}e(\alpha x^k),\qquad U_s(P,R)=\int_0^1|f(\alpha;P,R)|^s d\alpha, \] where \(e(z)=e^{2\pi iz}\) and \(s\) denotes a positive real number. The importance of the latter mean value has been widely recognised. A real number \(\delta_s=\delta_{s,k}\) is called an \textit{associated exponent}, if for each \(\varepsilon>0\), there exists \(\eta=\eta(\varepsilon, s, k)>0\) such that whenever \(R\leq P^{\eta}\), one has \[ U_s(P,R)\ll_{\varepsilon,s,k}P^{s/2+\delta_s+\varepsilon}. \] When \(R>P^{\eta_1}\) for a fixed \(\eta_1>0\), the lower bound \(U_s(P,R)\gg P^{s/2}\) is known for any \(s>0\). Also the upper bound \(U_s(P,R)\ll P^{s/2}\) is known for \(0<s\leq 4\), and it is conjectured that the latter upper bound is valid for \(4<s\leq 2k\) as well. The theme of this article is to establish upper bounds for \(U_s(P,R)\) when \(s\) is greater than and close to 4, in other words, to discuss associated exponents \(\delta_{4+\sigma}\) for small positive \(\sigma\). The author first shows that \(\delta_{4+\sigma}=(k\sigma/24)^{1+(k+1)/(2e)}\) is an associated exponent when \(k\geq 60\) and \(0<\sigma\leq 8e/(k+1)\). This improves substantially the previously available value \(\delta_{4+\sigma}=e\delta_6\sigma^{(\log(k+1))/\log 2}\). Two more theorems are presented to record the conclusions obtained by his method when \(\sigma>8e/(k+1)\), and when \(k\) is smaller. All three theorems arise from the author's recent breaking of ``classical convexity'' in Waring's problem [Invent. Math. 122, 421-451 (1995; Zbl 0851.11055)]. While this method leads to particularly sharp upper bounds for \(U_s(P,R)\), its flexibility causes serious difficulty to find the optimal choice of parameters in practice. The paper under review not only provides the currently sharpest bounds for the mean values \(U_s(P,R)\), but also offers heuristic guidance on sizes of the optimal parameters in applications of his ``breaking convexity'' method.