Small differences between prime numbers (Q1119687)

Let \(E_ r=\liminf_{n\to \infty}((p_{n+r}-p_ n)/\log p_ n)\), where \(p_ n\) is the n-th prime. It is conjectured that \(E_ r=0\) for all values of r, and it is immediate from the prime number theorem that \(E_ r\leq r\). For \(r=1\), which is perhaps the most interesting case, \textit{P. Erdős} [Duke Math. J. 6, 438-441 (1940; Zbl 0023.29801)] showed that \(E_ 1<1\). \textit{E. Bombieri} and \textit{H. Davenport} [Proc. R. Soc. Lond., Ser. A 293, 1-18 (1966; Zbl 0151.042)] obtained \(E_ r\leq r-,\) and \textit{M. N. Huxley} [Mathematika 24, 142-152 (1978; Zbl 0367.10038)] showed that \[ E_ r\leq \frac{2r-1}{16r}(4r+(4r-1)\frac{\vartheta_ r}{\sin \vartheta_ r}), \] where \(\vartheta_ r+\sin \vartheta_ r=\pi /4r\) (whence \(E_ 1\leq 0.4425\) and \(E_ 2\leq 1.4105).\) The present paper combines the methods of Bombieri, Davenport and Huxley with a technique introduced by the author [Adv. Math. 39, 257-269 (1981; Zbl 0457.10023)]. The key idea here is that the primes are denser than average by a factor \(e^{\gamma}\) (\(\gamma\) is Euler's constant) in the arithmetic progressions s(mod P(z)), where P(z) is the product of all primes below z, and \(s\leq z(\log z)^ 3\). One therefore replaces an exponential sum over all primes, which occurs in the earlier works, by one restricted to the above congruence classes. The end result is that Huxley's estimate is improved by a factor \(e^{\gamma}\), giving \(E_ 1\leq 0.248...\) and \(E_ 2\leq 0.79...,\) in particular. The technical arguments needed to push the analysis through are hard work. For example, one requires a version of the Bombieri-Vinogradov Theorem in which the moduli are restricted to multiples of P(z). Overall, the paper is a considerable achievement, and the result a most significant advance on what was known previously.