On the integral of Hardy's function (Q1881019): Difference between revisions

Let \(\zeta(s)\) denote the Riemann zeta function, \(\chi(s)=2^s\pi^{s-1}\sin(\frac{\pi s}{2})\Gamma(1-s)\), and denote Hardy's function by \(Z(t)=\chi^{-\frac12}(\frac12 +it)\zeta(\frac12+it)\). The author proves: \[ \int_0^T Z(t)\,dt=O_\varepsilon(T^{\frac14+\varepsilon}). \tag{1} \] In the proof the author uses a smoothed variant of the approximate functional equation with a sharp error term instead of the standard approximate functional equation for \(\zeta(s)\) (this is the Riemann-Siegel formula) that has the error term \(O(t^{-1/4)}\), which is not sufficiently good to produce the bound above. The variant was proved in the author's book [The mean values of the Riemann zeta-function. Berlin: Springer-Verlag (1991; Zbl 0758.11036)]. Therefore it suffices to prove the following: \[ \int_T^{2T}Z(t)\,dt = 2\sum_{n\leq 2\sqrt{T/\pi}}n^{-{1\over 2}}\text{Re}\,\left\{\int_{T_ {1}}^{2T}\rho\left({n\over \tau}\right)e^{iF(t)}\,dt \right\}+ O\left(T^{{1\over 4}}\right), \] where \(T_1=\max(T,2\pi ({n\over 2})^2)\) and \( F(t) = t\log({\tau\over n})-{t\over 2}-{\pi\over 8}\). The summation is then split into five parts which are bounded either by an elementary estimate for exponential integrals or by applying the saddle-point method. Finally the author gives some remarks on the fact that ``the true order of the integral of \(Z(t)\) remains elusive. In particular, it would be of interest to find an omega result for this quantity. Is it true that perhaps \[ \int_0^T Z(t)\,dt = \Omega(T^{1/4})\quad (= \Omega_{\pm}(T^{1/4})) ? \tag{2} \] If yes, then the result of the Theorem would be (up to the factor ``\(\varepsilon\)) best possible. The reason that (2) seems plausible is that \(T^{1/4}\) is the order of the terms coming from the saddle points, and in the evaluation of exponential integrals one usually expects the saddle points to produce the largest contribution.''