Kummer's conjecture for cubic Gauss sums (Q5932012)

In J. Reine Angew. Math. 310, 111--130 (1979; Zbl 0412.10028) the author and the reviewer proved that cubic Gauss sums defined over primes in the ring \(\mathbb Z[\omega]\), \(\omega^2+ \omega+1=0\) were uniformly distributed; this result showed in particular that conjecture (or rather suggestion) about the distribution of the values of cubic Gauss sums was not tenable. The proof consisted of two parts. The first was an application of methods going back to T. Kubota to determine the behaviour of such Gauss sums over arithmetic progressions. The second was the application of R. C. Vaughan's version of Vinogradov's sieve method to deduce the behaviour over primes. In this paper the author substantially improves the size of the error term by means of two improvements. The first is in the Tauberian method used in the first part of the argument, which leads to a small but subtle and significant improvement that is affected through a mean-square estimate for the Dirichlet series involved. The much more substantial improvement is in the second part. To achieve this he proves a very sharp (and probably not far from optimal) mean square estimate (over \(m\in\mathbb Z [\omega])\) for sums of the form \(\Sigma_nc_n ({n\over m})_3\) where \(({n\over m})_3\) denotes the cubic residue symbol. This is done by means of a modification of an argument of the author's paper [Acta Arith. 72, 235--275 (1995; Zbl 0828.11040)], which, apart from the law of cubic reciprocity, is based on Fourier-theoretic methods. The order of magnitude of the error obtained is at a level that one may expect to be essentially optimal; at least the reviewer has ventured a conjecture on the asymptotic behaviour of the Gauss sums that is just within the order of magnitude of this new estimate [see \textit{S. J. Patterson}, J. Reine Angew. Math. 303--304, 126--143 (1978; Zbl 0384.10017)].