New equidistribution estimates of Zhang type (Q486421): Difference between revisions

One of the most important problems in analytic number theory is to extend the range of the Bombieri-Vinogradov theorem. One version of the theorem states that for any \(A>0\) one has \[ \sum_{q\leq x^{\theta}}\max_{(a,q)=1}\max_{y\leq x}\left|\psi(x;a,q) -\frac{x}{\varphi(q)}\right|\ll_{A,\theta}x(\log x)^{-A} \] provided that \(\theta<1/2\). In fact this is equivalent to the corresponding statement in which one omits the maximum over \(y\leq x\), and sets \(y=x\). The Elliott-Halberstam Conjecture is the corresponding statement in which any \(\theta<1\) is allowed. A number of important variants allowing larger moduli \(q\) have been established, in which one imposes restrictions on \(a\), or on \(q\), or replaces the moduli signs by weights \(\lambda(q)\). Thus, for example, \textit{E. Bombieri} et al. [Acta Math. 156, 203--251 (1986; Zbl 0588.10042)] showed that if \(a\not=0\) is fixed, then \[ \sum_{q\leq x^{\theta}, (q,a)=1}\lambda(q)\left(\psi(x;a,q) -\frac{x}{\varphi(q)}\right)\ll_{a,A,\theta}x(\log x)^{-A} \] provided that \(\theta<4/7\), whenever the coefficients \(\lambda(q)\) are ``well-factorable of level \(x^{\theta}\)''. This directly enables one to improve the upper bound for the number of twin primes up to \(x\), by using \(a=2\). However it does not lead to an improvement for the corresponding problem of representations of a large even integer \(N\) as a sum of two primes, since one would want to choose \(a=N\), which is not fixed. Interest in such results was heightened by the work of \textit{D. A. Goldston} et al. [Ann. Math. (2) 170, No. 2, 819--862 (2009; Zbl 1207.11096)] who showed that any improvement on the Bombieri-Vinogradov theorem would suffice to show the existence of infinitely many bounded gaps among the primes. This was followed by the breakthrough of \textit{Y. Zhang} [Ann. Math. (2) 179, No. 3, 1121--1174 (2014; Zbl 1290.11128)] who established a suitable extension, allowing \(\theta<1/2+1/584\), for appropriate residue classes \(a\), and suitably smooth moduli \(q\). As ones sees by comparison with the result of Bombieri, Friedlander and Iwaniec, the two key points here are firstly that one allows ``appropriate residue classes \(a\)'', and secondly that it suffices to use ``suitably smooth moduli \(q\)''. (This latter possibility had been observed by \textit{Y. Motohashi} and \textit{J. Pintz} [Bull. Lond. Math. Soc. 40, No. 2, 298--310 (2008; Zbl 1278.11090)].) The present paper sharpens and simplifies Zhang's result. Suppose one has a sequence of integers \((a_p)\) indexed by the primes, and such that \(p\nmid a_p\) for every \(p\). One then chooses \(a_q\) so that \(a_q\equiv a_p\pmod{p}\) for every prime divisor \(p\) of \(q\). Then for any \(A,\delta>0\) it is shown that \[ \sum_{q\leq x^{\theta}}\left|\psi(x;a,q) -\frac{x}{\varphi(q)}\right|\ll_{A,\theta,\delta}x(\log x)^{-A}, \] provided that \[ \theta<\frac{1}{2}+\frac{7}{300}, \] where the sum over \(q\) is restricted to square-free \(x^{\delta}\)-smooth integers. It should be emphasized that the implied constant is independent of the sequence \((a_p)\). There are more refined versions in which the smoothness assumption is replaced by a requirement that the moduli \(q\) should have factors in suitable ranges. The application of these results to gaps between primes has been overtaken by the work of \textit{J. Maynard} [Ann. Math. (2) 181, No. 1, 383--413 (2015; Zbl 1306.11073)] (and of Tao, unpublished). The proof follows the same general strategy as is used by Zhang. The prime number sum is decomposed into standard Type I and Type II sums, along with the Type III (trilinear) sums introduced by Zhang. These then lead to incomplete exponential sums. One important source of additional saving comes from the observation that these exponential sums have relatively well factorable moduli. This enables one to use the \(q\)-analogue of van der Corput's method, as in the works of \textit{S. W. Graham} and \textit{C. J. Ringrose} [Prog. Math. 85, 269--309 (1990; Zbl 0719.11006)] and the reviewer [Proc. Lond. Math. Soc. (3) 82, No. 3, 554--596 (2001; Zbl 1023.11048)]. However there are many other sources of saving used in the paper.