Incomplete higher-order Gauss sums (Q1399334)

This paper explores the fascinating world of the exponential sum \[ S_m(B)=\sum_{j=0}^B \exp(2\pi ij^m/N) \] where \(m\) is a fixed integer, \(N, B\) are integers with \(0\leq B< N-1\) and \(N, B\rightarrow\infty\). The graphs of the sums in the complex plane, obtained by joining the points \(S_m(B)\) for \(B=1,2,\ldots\), provide the motivation for the work. These graphs contain spiral condensation points, apparently first observed by \textit{D. H. Lehmer} [Mathematika 23, 125-135 (1976; Zbl 0346.10020)] in the quadratic case \(m=2\) where the graphs turn out to be periodic. For \(m>2\), the patterns are more chaotic, but contain a primary condensation point at \(S_m(B_1)\) and a sequence of progressively smaller condensation points \(S_m(B_k), k=2,3,\dots\), with \(B_k\sim (kN/(2m))^{1/(m-1)}\). The authors give an asymptotic estimate for \(S_m(B)\) near \(B_1\) and locate \(S_m(B_k)\) for \(k=2,3,\dots\). The estimates can be made very precise in the case \(m=2\). Lehmer's work was led by the geometry of the spirals. More precise asymptotics arose from work in optics (where the spirals are the classical Cornu spirals), using Euler-McLaurin and Poisson summation. This is the basic analytic machinery used in the present paper, with the estimates being carried through to a higher order of accuracy than in earlier work. The analysis is detailed, but very clearly and carefully described. The detail also produces interesting connections with the so-called Springer and Euler numbers. Finally, the authors return to the geometry in an appendix and explain the intuitive approach to the spirals.