Pair correlation of zeros, primes in short intervals and exponential sums over primes (Q1590269)

Let \(S(\alpha) = \sum_{n\leq X}\Lambda(n)e(n\alpha)\), where \(e(x)=e^{2\pi ix}\), and \(\Lambda(n)\) is the von Mangoldt function defined to be \(\log p\) if \(n=p^m\), \(p\) a prime, \(m\geq 1\), and zero otherwise. This paper is concerned with the short \(L^2\) mean \[ S(X,\xi) = \int_{-\xi}^\xi |S(\alpha)|^2 d\alpha , \] with \(0\leq \xi \leq {1\over 2}\). The function \(S(X,\xi)\) is useful in studying the existence of Goldbach numbers in short intervals, an idea that goes back to \textit{Yu. V. Linnik} [Izv. Akad. Nauk SSSR, Ser. Mat. 16, 503-520 (1952; Zbl 0049.03104)], however Linnik used short \(L^2\) estimates for the related exponential sum \(\tilde S(\alpha) = \sum_{n=1}^\infty \Lambda(n)e^{-{n\over X}}e(n\alpha)\). For applications of \(S(X,\xi)\) to the Goldbach problem, see the papers of \textit{D. A. Goldston} [Glasg. Math. J. 32, No. 3, 285-297 (1990; Zbl 0719.11065)] and \textit{A. Languasco} and \textit{A. Perelli} [Ann. Inst. Fourier 44, No. 2, 307-322 (1994; Zbl 0799.11040)]. To study \(S(X,\xi)\) more closely, one removes the expected main term and defines \[ R(X,\xi) = \int_{-\xi}^\xi |S(\alpha) -T(\alpha)|^2 d\alpha , \] where \(T(\alpha)=\sum_{n\leq X}e(n\alpha)\). In this paper the authors show that there is an equivalence between an asymptotic formula for \(R(X,\xi)\) and an asymptotic formula for the second moment for primes in short intervals. By work of \textit{D. A. Goldston} and \textit{H. L. Montgomery} [Prog. Math. 70, 183-203 (1987; Zbl 0629.10032)] this in turn is equivalent to Montgomery's strong form of the pair correlation conjecture for zeros of the Riemann zeta-function. Let \[ J(X,h) = \int_0^X |\psi(x+h)-\psi(x)-h|^2 dx, \] and \[ F(X,T)= \sum_{0<\gamma, \gamma'\leq T}X^{i(\gamma -\gamma ')}w(\gamma -\gamma ') \] where \(\gamma\) and \(\gamma '\) are the imaginary part of zeros of the Riemann zeta-function, and \(w(x) = 4/(4 +x^2)\). The main result of the paper is that, assuming the Riemann Hypothesis and letting \(X\to \infty\), then the following are equivalent: (i) for every \(\varepsilon>0\), \(R(X,\xi) \sim 2X\xi \log X\xi\), uniformly for \(X^{-1/2 +\varepsilon}\leq \xi \leq {1\over 2}\), (ii) for every \(\varepsilon >0\), \(J(x,h) \sim h X \log {X\over h}\), uniformly for \(1\leq h \leq X^{{1\over 2}-\varepsilon}\), (iii) for every \(\varepsilon >0\) and \(A\geq 1\), \(F(X,T)\sim {T\over 2\pi}\log \min(X,T)\), uniformly for \(X^{{1\over 2}+\varepsilon}\leq T \leq X^A\). The second and third results above actually are equivalent in the wider range \(1\leq h \leq X^{1-\varepsilon}\) and \(X^{\varepsilon}\leq T \leq X^A\), and the authors point out that for this wider range the equivalence in (i) is not with \(R(X,\xi)\) but rather with an appropriately modified asymptotic formula for a short \(L^2\) mean of \(\widetilde S(\alpha)\) in the range \(X^{-1+\varepsilon}\leq \xi \leq 1/2\). Returning to \(S(X,\xi)\), from (i) above one obtains the asymptotic formula \[ S(X,\xi)\sim X(1 +2\xi \log X\xi) \] uniformly for \({\log^{10} X\over X} \leq \xi \leq {1\over 2}\). (It is known that this asymptotic formula holds unconditionally for \({\log^{10} X\over X} \leq \xi \leq x^{-1/6 -\varepsilon}\).) The authors also prove unconditionally that the asymptotic formula for \(R(X,\xi)\) in (i) holds unconditionally in the range \({F(X)\over \log X} \leq \xi \leq {1\over 2}\), where \(F(X)\) is a function that increases arbitrarily slowly to infinity, with corresponding results for \(J(X,h)\) and (on the Riemann hypothesis) for \(F(X,T)\). The proofs depend on a generalization of Gallagher's lemma together with Abelian and Tauberian type arguments.