Fast Algorithms for Multiple Evaluations of the Riemann Zeta Function

Authors:
  • AOA. M. Odlyzko
    A. M. Odlyzko
  • ASA. Schonhage
    A. Schonhage

Abstract
The best previously known algorithm for evaluating the Riemann zeta function, <!-- MATH $\zeta (\sigma + it)$ --> , with bounded and large to moderate accuracy (within <!-- MATH $\pm {t^{ - c}}$ --> for some 0$">, say) was based on the Riemann-Siegel formula and required on the order of operations for each value that was computed. New algorithms are presented in this paper which enable one to compute any single value of <!-- MATH $\zeta (\sigma + it)$ --> with fixed and <!-- MATH $T \leqslant t \leqslant T + {T^{1/2}}$ --> to within <!-- MATH $\pm {t^{ - c}}$ --> in <!-- MATH $O({t^\varepsilon })$ --> operations on numbers of bits for any <!-- MATH $\varepsilon > 0$ --> 0$">, for example, provided a precomputation involving <!-- MATH $O({T^{1/2 + \varepsilon }})$ --> operations and <!-- MATH $O({T^{1/2 + \varepsilon }})$ --> bits of storage is carried out beforehand. These algorithms lead to methods for numerically verifying the Riemann hypothesis for the first zeros in what is expected to be <!-- MATH $O({n^{1 + \varepsilon }})$ --> operations (as opposed to about operations for the previous method), as well as improved algorithms for the computation of various arithmetic functions, such as . The new zeta function algorithms use the fast Fourier transform and a new method for the evaluation of certain rational functions. They can also be applied to the evaluation of -functions, Epstein zeta functions, and other Dirichlet series.

This publication has 13 references indexed in Scilit: