Incomplete Kloosterman sums and a divisor problem. Appendix: On some exponential sums by Bryan J. Birch and Enrico Bombieri (Q1062092): Difference between revisions

The authors obtain estimates of two kinds of averages of incomplete Kloosterman sums, by the profitable use of modified ideas of Burgess in connection with the estimation of (multiplicative) character sums based on Weil's proof of the RH for curves. Let \(\psi\) be a primitive additive character mod q, \(q\in {\mathbb{N}}\), i.e. \(\psi (r)=\exp (2\pi i\mu r/q)\) with \(\mu\in {\mathbb{Z}}\), \((\mu,q)=1\). Then the first sum considered is \[ S(M,N)=\sum_{n\leq N}\quad | \quad \sum_{M_ 1<n\leq M_ 1+M}\psi (\bar m \bar n)|, \] where \({\mathbb{N}}\ni M,N<q\), \(M_ 1\) is an arbitrary integer, and \(\bar n\) is the multiplicative inverse of n mod q, and the second one is \[ S(H,M,N)=\sum_{1\leq h\leq H, (h,q)=1}\quad \sum_{1\leq m\leq M, 1\leq n\leq N}\psi (h\overline{m n}). \] Weil's fundamental estimate for the Kloosterman sum gives \((\epsilon >0\) is arbitrary, also in the sequel) \[ S(M,N)\ll_{\epsilon}\quad N q^{1/2+\epsilon}\quad and\quad S(H,M,N)\ll_{\epsilon}\quad H \min \{MN, q^{1+\epsilon}\}. \] The distinguishing feature of the authors' results is that they take the incompleteness of the sums into consideration and bring the effect of the average into relief. The estimate for S(M,N) given in Theorem 3 reads for q prime: \[ S(M,N)\ll_{\epsilon}\quad q^{\epsilon}(q^{1/4} M^{7/12} N^{5/6}+M^{3/4} N), \] and the estimate for S(H,M,N) is given in Theorem 4. The proof of Theorem 3 depends on \textit{C. Hooley}'s result on complete Kloosterman sums [Journées arithmétiques, Exeter 1980, Lond. Math. Soc. Lect. Note Ser. 56, 92-122 (1982; Zbl 0488.10041)] and the proof of Theorem 4 rests on (a special case of) a theorem of Birch and Bombieri in the appendix. Both of them in turn draw on Deligne's fundamental work on the RH for algebraic varieties defined over finite fields. As a modest application of their powerful theorem the authors consider the sum \[ A_ 3(X;a,q):=\sum_{n\leq X, n\equiv a (mod q)}d_ 3(n)\quad with\quad (a,q)=1 \] and obtain an error estimate valid even for \(q<X^{58/115}\). Although the obtained error estimate is weaker than those of \textit{R. A. Smith} [Math. Ann. 260, 255-268 (1982; Zbl 0467.10034)] and \textit{K. Matsumoto} [Nagoya Math. J. 98, 37-42 (1985; Zbl 0545.10032)] for \(q\leq X^{242/487}\), the essential advantage lies in the wide range of uniformity in q. Thus, after \textit{C. Hooley}'s very wide range of uniformity in q \((q\leq X^{2/3})\) for \(A_ 2(X;a,q)\) [Proc. Lond. Math. Soc., III. Ser. 7, 396-413 (1957; Zbl 0079.273)] the authors for the first time have succeeded in superseding the limit 1/2 for \(A_ 3(X;a,q)\) by 1/230.