Omega result for the mean square of the Riemann zeta function (Q2487940): Difference between revisions

Let \[ E(t) := \int_{0}^{t} \biggl| \zeta\biggl({1\over 2}+iu \biggr)\biggr| ^{2}\, du - t\log \biggl({t\over 2\pi}\biggr) - (2\gamma -1)t , \qquad (t>0). \] The main result of this paper is \[ E(t) = \Omega((t\log t)^{{1\over 4}}(\log\log t)^{{3\over 4}(2^{{4\over 3}}-1)}(\log\log\log t)^{-{5\over 8}}). \] This improves the result of \textit{J. L. Hafner} and \textit{A. Ivić} [J. Number Theory 32, 151--191 (1989; Zbl 0668.10045)] who had \((3+\log 4)/4\) as the power of \(\log\log t\). The method depends on \textit{K. Soundararajan}'s [Int. Math. Res. Not. 2003, No. 36, 1987--1998 (2003; Zbl 1130.11329)] approach for his \(\Omega\)-result for \(\Delta(x)\), the error term in the divisor problem. However, since the Voronoi-type Atkinson series for \(E(t)\) has an oscillating factor \((-1)^{n}\) which is not present in the Voronoi series for \(\Delta(x)\), Soundararajan's method is not directly applicable to the problem at hand. \textit{M. Jutila} [Ark. Mat. 21, 76--96 (1983; Zbl 0513.10040)] observed that \(\Delta^{\ast}(x) = -\Delta(x)+2\Delta(2x)-{1\over 2}\Delta(4x)\) is the right approximation to \((2\pi)^{-1}E(2\pi x)\). The authors use the combination for \(\sqrt{x}\Delta^{\ast}(x^2)\) in order to establish a relation between the Voronoi series for their problem which has the oscillating factor \((-1)^{n}\) and Voronoi series without this oscillating factor to which Soundararajan's method can be applied. Furthermore, finding that \(\Delta^{\ast}(x)\) does not provide sufficient approximation to \((2\pi)^{-1}E(2\pi x)\), the authors pass to a smoothing of \(E_{1}(x) := (2x)^{-{1\over 2}}E(2\pi x^{2})\) by taking convolutions with the Fejer kernel. Soundararajan's heuristic arguments for believing that his \(\Omega\)-result for \(\Delta(x)\) is best possible suggest to the authors that their \(\Omega\)-result is also best possible within a factor of \((\log\log t)^{o(1)}\). As in Soundararajan's work, the presence of the phase angle \(-{\pi\over 4}\) in the cosines of the summands of the Voronoi series prevents obtaining \(\Omega_{+}\) or \(\Omega_{-}\)-results. It is trivial that an upper-bound for \(\Delta(x)\) implies an upper-bound for \(\Delta^{\ast}(x)\). Here the authors also prove the converse statement that \(\Delta^{\ast}(x) \ll x^{\sigma}\) for some \(\sigma \leq {1\over 3}\) implies \(\Delta(x) \ll x^{\sigma}\).