Primes in arithmetic progressions to large moduli. II (Q579311): Difference between revisions

The principal theorem of this paper is the estimate \[ \sum_{Q<q\leq 2Q;\quad (a,q)=1}| \psi (x;q,a)-x/\phi (q)| \quad \ll_ a\quad x (\frac{\log y}{\log x})^ 2 (\log \log x)^ B \] where \(a\neq 0\), \(3\leq y\leq x\), \(Q^ 2\leq xy\), and B is an absolute constant. One obtains the asymptotic formula \[ \pi (x;q,a) \sim li(x)/\phi (q)\quad (x\to \infty) \] for fixed a, for almost all \(q\leq x^{1/2} \exp (\log x/(\log \log x)^{B'})\). In contrast the Bombieri-Vinogradov theorem allows one to vary a, but is restricted to \(q\leq x^{1/2}(\log x)^{-B''}.\) The proof builds on earlier work, by \textit{E. Fouvry} [Acta Math. 152, 219-244 (1984; Zbl 0552.10024)] and the authors [ibid. 156, 203-251 (1986; Zbl 0588.10042)], for example. These papers depend fundamentally on the theory of averages of Kloosterman sums, as developed by \textit{J. M. Deshouillers} and \textit{H. Iwaniec} [Invent. Math. 70, 219-288 (1982; Zbl 0502.10021)]. The proof decomposes \(\psi\) (x;q,a) using products of Dirichlet polynomials, and handles most of the resulting convolutions asymptotically; for these terms one could extend the range to \(Q\leq x^{+\delta}.\) There remain a few terms corresponding to highly restricted convolutions of 4, 5 and 6 factors, which can only be estimated by upper bounds. These terms account for the relative weakness of the result as Q grows beyond \(x^{1/2}\).