On a generalization of the Euler totient function (Q5917922): Difference between revisions

A polynomial Euler product is a function \(F(s) = \prod_p \prod_{j=1}^d (1-\alpha_j(p)p^{-s})^{-1}\), which is absolutely convergent for \(s= \sigma + it\) with \(\sigma > 1\). Here \(p\) runs over the primes and the complex numbers \(\alpha_j(p)\) have norm at most \(1\) for all \(p\) and \(1 \leq j \leq d\). The associated Euler totient function is defined as \[ \varphi(n , F) = n \prod_{p|n} \prod _{j=1}^d(1- \alpha_j(p)p^{-1}). \] Thus for \(d=1\) and \(\alpha_j(p) =1, F(s)\) is the Riemann zeta-function \(\zeta(s)\) and \(\varphi(n, \zeta)\) is the usual Euler totient function. The first result proved in this paper is the asymptotic formula: for all \(x \geq 1\), \[ \sum_{n \leq x} \varphi(n, F) = C(F) x^2 + O(x (\log 2x)^d), \] where \(C(F)\) is a constant depending on \(F\). Let \[ E(x, F) = \sum_{n \leq x} \varphi(n, F) - C(F)x^2 \] denote the above \(O\)-term. Suppose further that \(F(s)\) belongs to the Selberg class and it has a zero-free region of the form \[ \sigma > 1 - C_0(F)(\log(|t| +10))^{-1} \] for some positive constant \(C_0(F)\). Then it is proved that \[ \int_1^x |E(u, F)|^2\;du = \beta(F)x^3 + O( x^3 \exp(-c \sqrt{\log x})), \] for a certain \(c > 0\) and an explicit constant \(\beta (F)\) depending on \(F\). For the particular case that \(F = \zeta\), this improves on a 1932 result of Chowla, namely \[ \int_1^x |E(u, \zeta)|^2\, du = \frac{1}{6 \pi^2} x^3 +O(\frac{x^3}{\log^4x}). \] A remarkable result proved in this paper is that \(\beta (F) \geq \beta (\zeta)\) for all the \(F\) under consideration, and \(\zeta(s)\) can be characterized as the only polynomial Euler product from the Selberg class with the minimal square mean of \(E(u, F)\).