A modified Möbius \(\mu \)-function (Q660457)

The function in the title is a modification of the classical Möbius \(\mu\)-function (generated by \(1/\zeta(s)\)). This is the arithmetic function \(\mu_\infty(n)\), which is a multiplicative function defined (\(p\) is a generic prime) as \[ \mu_\infty(p^\nu) = (-1)^{|B(\nu)|}, \] where \(|B(\nu)|\) is the number of non-zero terms in the binary expansion of the non-negative integer \(\nu\). This function was introduced by \textit{G. Cohen} and \textit{P. Hagis jun.} [Int. J. Math. Math. Sci 16, 373--383 (1993; Zbl 0774.11006)]. The main results of the paper are as follows. It is first shown that the generating function of \(\mu_\infty(n)\) is \(\prod_{j=0}^\infty\zeta(2^js)^{-1}\), valid for \(\text{Re}\, s > 1\). If one sets \({\mathcal M}(x) := \sum\limits_{n\leq x}\mu_\infty(n)\), then it is shown that \[ {\mathcal M}(x) \ll x\exp\Bigl(-c(\log x)^{3/5}(\log\log x)^{-1/5}\Bigr), \] where \(c\) is a suitable positive constant. This is analogous to the best-known error term in the prime number theorem (or the upper bound for \(\sum\limits_{n\leq x}\mu(n)\)). It is shown that the famous Riemann hypothesis (that all complex zeros of the Riemann zeta-function \(\zeta(s)\) have zeros whose real parts are 1/2) is equivalent to the bound \({\mathcal M}(x) \ll_\varepsilon x^{1/2+\varepsilon}\). If the Riemann hypothesis holds, then an explicit formula for \({\mathcal M}(x)\) (involving complex zeros \(\rho\) of \(\zeta(s)\)) is given. It is also shown that (unconditionally) \({\mathcal M}(x) = o(\sqrt{x}\,)\) as \(x\to\infty\) cannot hold. The main tool in the proofs is Perron's inversion formula for Dirichlet series, coupled with the best known zero-free region for \(\zeta(s)\).