On generating functions in additive number theory. II: Lower-order terms and applications to PDEs (Q2225612)

This paper considers the Weyl sum \[ f_k(\alpha,\beta)=\sum_{n\le P}e(\alpha n+\beta n^k), \] for a fixed integer \(k\ge 2\). Let \(\beta=b/q+\theta\) with \((b,q)=1\) and \(\theta\) small, and let \(a\) be an integer with \(|\alpha-a/q|\le 1/2q\). One might expect that the difference \(f_k(\alpha,\beta)\) is well approximated by \(q^{-1}S(q;a,b)I(\alpha-a/q,\beta-b/q)\), where \[ S(q;a,b)=\sum_{n\le q}e((an+bn^k)/q))\;\;\;\mbox{and}\;\;\; I(\phi,\theta)=\int_0^Pe(\phi x+\theta x^k)dx. \] In particular, when \(\alpha=0\) it was shown by \textit{R. C. Vaughan} (Theorem 4.1 of [The Hardy-Littlewood method. 2nd ed. Cambridge: Cambridge University Press (1997; Zbl 0868.11046)]) that \[ f_k(0,\beta)=q^{-1}S(q;0,b)I(0,\theta) +O(q^{1/2+\varepsilon}(1+|\theta|P^k)^{1/2}) \] for any fixed \(\varepsilon>0\). For general \(\alpha\) it was shown by \textit{J. Brüdern} and \textit{O. Robert} [Nagoya Math. J. 218, 51--100 (2015; Zbl 1371.11139)] that one gets a similar error term except that the exponent of \(q\) is \(1-1/k+\varepsilon\) instead of \(1/2+\varepsilon\). The principal result of the paper shows that one can (essentially) achieve the error term that one has for \(\alpha=0\) but that one needs to modify the main term. For any real \(x\) let \([[x]]\) denote an integer closest to \(x\). Then it is shown that \[ f_k(\alpha,\beta)=q^{-1} \sum_d S(q;d[[a/d]],b)I(\alpha-d[[a/d]]/q,\theta) +O(q^{1/2+\varepsilon}(1+|\theta|P^k)^{1/2}\log P), \] where the sum over \(d\) runs over divisors of \(q\) for which \([[a/d]]\) is coprime to \(q/d\). The error term may be replaced by \(O(q^{1/2+\varepsilon})\) when \(|\theta|\le(4kqP^{k-1})^{-1}\). Moreover it is said that the factor \(\log P\) in the error term may be removed via a more careful analysis. The paper gives an application to the fractal dimension of solutions to the Schrödinger and Airy equations, which have solutions of the form \[ q(t,x)=\sum_n a_n e^{itn^k+xn} \] for \(k=2\) and \(k=3\) respectively.