Sums of the error term function in the mean square for \(\zeta (s)\) (Q956670)

In 2005 \textit{J. Furuya} [J. Number Theory 115, No. 1, 1--26 (2005; Zbl 1089.11055)] investigated the classical error term in the Dirichlet divisor problem, namely sums of the form \(\Delta^k (n)\) \((n\in\mathbb N)\) when \(k = 2\) or \(k = 3\), where \[ \Delta (x) =\sum_{n\leq x} d(n) - x(\log x + 2\gamma - 1), \] \(d(n)\) is the number of divisors of \(n\), and \(\gamma\) is Euler's constant. In this paper the authors provide a natural generalization of Furuya's results to sums of the form \(\sum_{n\leq x} E^k (n)\) (\(k \in \mathbb N\) is fixed), where \[ E(T) =\int_0^T |\zeta(1/2 + it)|^2 \,dt - T\left(\log\frac{T}{2\pi} + 2\gamma - 1\right) \] is the error term in the mean square formula for \(|\zeta(1/2 + it)|\), and \(\zeta(s) = \sum_{n=1}^{\infty} n^{-s}\) \((\text{Re}\,s>1)\) is the Riemann zeta-function. The emphasis is on the case \(k=1\), which turns out to be more intricate than the corresponding problem for sums of \(\Delta (n)\). The starting point is the formula \[ \sum_{n\leq x} E(n) =\pi x + \int_0^x \psi(t) |\zeta(1/2 + it)|^2 \,dt + G(x) + O(x^{1/3}), \] where \[ G(x) :=\int_0^x (E(u) - \pi) \,du, \quad G(x) = O(x^{3/4}), \quad G(x) = \Omega_{\pm} (x^{3/4}). \] To bound the integral \(\int_0^x \psi(t) |\zeta(1/2 + it)|^2 \,dt\), the authors first consider subintegrals over the intervals of the form \([T,2T]\) and employ the approximate functional equation for \(\zeta^2 (s)\), followed by asymptotic analysis by the saddle-point method which shows that the integral is asymptotic to sums of the type (\(C>0\) is a generic constant) \[ C \sum_{m \leq (1/(2 \pi) + \varepsilon) \log T} \frac{e^{\pi m}}{m} \sum_{n \asymp T e^{-2 \pi m}} d(n) \exp(-2 \pi i n e^{2 \pi m}), \] so that the problem is reduced to the estimation of the inner sum over \(n\). Then use is made of several well-known results due to Wilton. The first result is the functional equation \[ D(x,\eta) =\eta^{-1} D(\eta^2 x, -\eta^{-1}) + O(x^{1/2} \log x), \quad D(x,\eta) =\sum_{n\leq x} d(n) e^{2\pi i\eta n}, \] where \(\eta^2 x\gg 1\) and \(0< \eta \leq 1\). The second result is the bound, for irrational \(\eta\), \(D(x,\eta) =o(x \log x)\) \((x \to \infty)\), which holds uniformly in \(\eta\). Wilton's functional equation is employed with \(x = CT e^{2\pi m}\) and \(\eta = \{e^{2\pi m}\}\), where \(\{y\} = y - [y]\) is the fractional part of the real number \(y\). Since all the numbers \(e^{2 \pi m}\) are transcendental, Wilton's bound applies in this case. Wilton also proved the following: Let \[ \frac{1}{\eta} = a_0 + \frac{1}{a_1+} \frac{1}{a_2+}\cdots \quad (0 < \eta \leq 1), \] be the expansion of \(1/\eta\) as a simple continued fraction. If \(\eta\) is irrational and the partial quotients \(a_n\) satisfy \(a_n \ll n^{1 + K}\) \((K \geq 0)\), then \(\sum_{n\leq x} d(n) \sin(2 \pi n \eta) \ll x^{1/2} \log^{2 + K} x.\) Moreover, if \(a_n \ll e^{K n}\) \((K > 0)\), then \(\sum_{n \leq x} d(n) \exp(2 \pi i n \eta) \ll x^H \log x\) where \(H = (4L + \log 2)/(4K + 2\log 2)\). The analysis for the results on \(\sum_{n\leq x} E(n)\) requires bounds for the irrationality measure of \(e^{2 \pi m}\) and for the partial quotients in its continued fraction expansion. Lemma 1 gives an explicit bound for the partial quotients of \(e^{\pi m}\) in its continued fraction expansion \(e^{\pi m} = [a_0(m);a_1(m),\dots]\), where \(m\) is any non-zero integer. It states that there is an absolute positive constant \(c\) such that the inequality \[ \log\log a_n(m) <c(n + \log |m|) \log(n + \log |m|) \] holds for any non-zero integer \(m\) and any non-negative integer \(n\). This allows the authors to derive a non-trivial bound which sharpens Wilton's bound by a quantity that is less than \((\log x)^{\varepsilon}\) for any given \(\varepsilon > 0\). Specifically, Theorem 1 states that if \[ D(x,\eta(m)) :=\sum_{n\leq x} d(n) \exp(2 \pi i \eta(m)) \quad (\eta(m) = e^{-2 \pi m}), \] then for some \(1 \leq m \leq \log_2 x/ \log_3 x\) and some constant \(C > 0\) \[ D(x,\eta(m)) \ll x \log x \exp\left(-C\frac{\log_2 x}{\log_3 x}\right). \] Theorem 2 follows from Theorem 1, and it states that \[ \sum_{n\leq x} E(n) =\pi x + H(x), \] where, for some \(C > 0\), unconditionally \[ H(x) \ll x \log x \exp\left(-C \frac{\log_2 x}{\log_3 x}\right). \] Moreover, if the bound \[ \sum_{n\leq x} d(n) \sin(2 \pi n \eta) \ll x^{1/2} \log^{2 + K} x \] holds with \(\eta = e^{2 \pi m}\) uniformly for \(1\leq m \ll \log x\), then \[ H(x) =O(x^{3/4}). \] Finally, in general, for \(k \in \mathbb N\) fixed, \[ \sum_{n\leq x} E^k (n) =\int_0^x E^k(u) \,du =\int_0^x E^k(u) \,du - \int_0^x E^k(u) \,d\psi(u). \] There are two cases here: \(k = 2, 4\) and \(k = 3, 5, 7, 9\). The results for \(\sum_{n \leq x} E^k (n)\) for \(k > 1\) are summarized in Theorem 3, which states that \[ \begin{aligned} &\sum_{n \leq x} E^2 (n) =C_2 x^{3/2} + O(x^{5/4} \log^2 x) \quad (C_2 > 0), \\ &\sum_{n \leq x} E^3 (n) = C_3 x^{7/4} + O_{\varepsilon}(x^{7/4 - 83/393 + \varepsilon}) \quad (C_3 > 0), \\ &\sum_{n \leq x} E^4 (n) =C_4 x^2 + O_{\varepsilon}(x^{2 - 3/28 + \varepsilon}) \quad (C_4 > 0), \end{aligned} \] and for \(k = 5, 7, 9,\) \[ \sum_{n \leq x} E^k (n) =C_k x^{1 + k/4} + O(x^{1 + k/4 + \delta(k)}) + O(x^{k/3} \log x) \quad (C_k >0), \] where \(\delta(k)\) is a positive constant which may be explicitly evaluated. Larger values of \(k\) are not considered in the paper, as the authors do not have precise formulas for the integrals of \(E^k\).