Möbius convolutions and the Riemann hypothesis (Q2368483)

\textit{M. Riesz} [Acta Math. 40, 185--190 (1916; JFM 46.0497.03)] proved that the Riemann hypothesis RH is equivalent to the estimate \[ \sum_{n=1}^\infty {{(-1)^{n+1} x^n}\over {(n-1)! \; \zeta(2n)}} \ll x^{{1\over 4}+\epsilon}, \hbox{ as } x\to\infty. \] In [Acta Math. 41, 119--196 (1917; JFM 46.0498.01)], there is a modification of this result due to \textit{G. H. Hardy} and \textit{J. E. Littlewood}. The author's aim is to embed these results into a general theorem for a class of entire functions. The Mellin transform of a complex--valued function \(f\) is \[ f^\wedge(s) = \int_0^\infty t^{-s-1} f(t) dt, \] when the integral converges absolutely. Define for real \(a\) the norm \(N_a(f) = \int_0^\infty t^{-a-1} | f(t) | dt\). The function \(\Phi\) is \textit{proper}, if \(N_\sigma (\Phi) < \infty\) at least for \( -{1\over 2} < \sigma \leq 0\), and it is \textit{Mellin--proper}, if \(\Phi\) is proper and \(\Phi^\wedge(s) \not=0 \) in \(-{1\over 2} < \sigma < 0\). Using the sum--function \(\displaystyle g(x) = \sum_{n\leq x} {1\over n} \cdot \mu(n)\) with the Möbius--function \(\mu\), the convolution operator \textbf{G}\ is defined by \[ \hbox{{\mathbf G}}\Phi(x) = \int_0^\infty g(xt) \Phi\left({1\over t} \right) {{dt} \over t}. \] The author proves: If \(\Phi\) is Mellin--proper, then RH is equivalent with the condition \textbf{G}\(\Phi(x) \ll x^{-{1\over 2} + \epsilon}, \) and the ``entire function RH--criterion'': For an entire, Mellin--proper function \(\Phi(z) = a_1 z + a_2 z^2 + \dots\) with the associated entire function \(\displaystyle \Phi^\ast(z) = \sum_{n=1}^\infty {{a_n z^n} \over {n \zeta(n+1)}}\; \) the equivalence \[ \hbox{RH} \iff \Phi^\ast(x) \ll x^{- {1\over 2} + \epsilon} \] is true. Concerning the Riemann hypothesis with only simple zeros (RHS), the author proves, that the relation \textbf{G}\(\Phi(x) \ll x^{-{1\over 2}}\), as \(x\to\infty\), implies RHS, if \(\Phi\) is Mellin--proper and if \(\Phi^\wedge\) [holomorphic in the strip \(-{1\over 2} < \sigma < 0\) and continuous on \(-{1\over 2} \leq \sigma \leq 0\)] does not vanish on the line \(\sigma = - {1\over 2}\). Finally the author shows that \textbf{G}\(\Phi(x) \not= o\left(x^{-{1\over 2}}\right)\), if \(\Phi\) is Mellin--proper, and --- if \(\Phi\) is real--valued in addition --- there is a \(\beta, \, 0 \leq \beta < {1\over 2}\) so that \[ \liminf_{x\to\infty} x^{\beta + \epsilon}\hbox{{\mathbf G}} \Phi(x) = - \infty, \quad \limsup_{x\to\infty} x^{\beta + \epsilon}\hbox{{\mathbf G}} \Phi(x) = \infty \] for all \(\epsilon > 0\).