Large gaps between the zeros of the Riemann zeta function (Q2474297)

Assuming the Riemann Hypothesis (RH) one can label the non-trivial zeros in the upper half-plane as \({1\over 2}+ i\gamma_n\) with \(\gamma_1\leq \gamma_2\leq\cdots\). The differences \(\gamma_{n+1}- \gamma_n\) are then about \(2\pi/(\log\gamma_n)\) on average. However there is considerable interest in establishing the existence of differences which are much smaller, or much larger, than this average. Prior to the present paper the best known results on large gaps were that of \textit{R. R. Hall} [Mathematika 52, No. 1--2, 101--113 (2005; Zbl 1119.11050)] who showed that if \[ \lambda:= \limsup{(\gamma_{n+1}- \gamma_n)\log\gamma_n\over 2\pi}, \] then \(\lambda\geq 2.63067\dots\), and of \textit{J. B. Conrey}, \textit{A. Ghosh} and \textit{S. M. Gonek} [Mathematika 33, 212--238 (1986; Zbl 0615.10048)] who showed (assuming GRH) that \(\lambda> 2.68\). It is now shown (subject only to RH) that \(\lambda> 3\), which represents a significant improvement on these results. The method develops further that used by \textit{J. B. Conrey}, \textit{A. Ghosh} and \textit{S. M. Gonek} (loc. cit.), which itself derives from the technique introduced by \textit{J. Mueller} [J. Number Theory 14, 327--331 (1982; Zbl 0483.10035)]. Suppose that \(|H({1\over 2}+ it)|\) is a continuous function, and let \[ m(T;\alpha):= \sum_{T<\gamma\leq 2T}|H(\textstyle{{1\over 2}}+ i(\gamma+ \alpha))|^2, \] where \(\gamma\) runs over the ordinates of the zeros of \(\zeta(s)\). Set \[ M_1(T):= \int^{2T}_T |H(\textstyle{{1\over 2}}+ it)|^2\,dt\quad\text{and}\quad M_2(T;\eta):= \int^\eta_{-\eta} m(t; \alpha)\,d\alpha. \] Then if \(M_2(T; \eta)< M_1(T)\) the range \([T, 2T]\) must contain an interval of length \(2\eta\) free of points \(\gamma\). In general it is the evaluation of \(m(T;\alpha)\) which is hardest, and which restricts most severely the feasible choices of \(H(s)\). Mueller applied the above principle with \(H(s)=\zeta(s)\), while Conrey, Ghosh and Gonek used \(H(s)=\zeta(s)A(s)\), where \(A(s)=\sum_{n\leq y} n^{-2}\). In this case one is still able to estimate \(m(t;\alpha)\) if \(y\) is slightly smaller than \(T^{1/2}\). In general, the choice for \(H(s)\) can be motivated by the work of \textit{C. P. Hughes} [J. Phys. A, Math. Gen. 36, No. 12, 2907--2917 (2003; Zbl 1074.11047)], who showed, subject to suitable conjectures, that one can achieve \(\lambda=\infty\) using \(H(s)= \zeta(s)^k\) for arbitrarily large \(k\). In an attempt to mimic \(H(s)= \zeta(s)^{1+r}\) the present paper takes \[ A(s)= \sum_{n\leq y} d_r(n) P_y(n) n^{-s}, \] where \(d_r(n)\) is the generalized divisor function, and \(P_y(n)\) is a weight function, taken to be polynomial in \(\log n\). Again it is possible to handle \(m(T;\alpha)\) when \(y\) is slightly below \(T^{1/2}\). It turns out that the choice \(r= 2\) and \(P_y(n)= (\log n)^{30}\) is enough to get \(\lambda> 4\). The paper does not aim to find the optimal choices of \(r\) and \(P_y\), but it would appear that any further improvements are likely to be small.