On some sums involving small arithmetic functions (Q6640974)

The article under review belongs to the papers dealing with sums of the shape \[S_f(x) := \sum_{n \leqslant x} f \left \lfloor x/n \right \rfloor\] a topic which has been introduced in [\textit{O. Bordellès} et al., J. Number Theory 202, 278--297 (2019; Zbl 1454.11009)], in which it is proven that, assuming \(f(n) \ll n^{\alpha} (\log n)^{\theta}\) and \(0 \leqslant \alpha < 1\) \[S_f(x) = x \sum_{n \geqslant 1} \frac{f(n)}{n(n+1)} + E_f(x)\] where \(E_f(x) \ll x^{\frac{1+\alpha}{2}} (\log x)^{\theta}\). In this work, assuming that \(f\) satisfies the Ramanujan hypothesis \(f(n) \ll n^{\varepsilon}\), the main aim of this paper is twofold: firstly, the author investigates the mean square of \(E_f(x)\), showing that \[\int_T^{2T} \left| E_f(x)\right|^{2} \, \mathrm{d}x \ll T^{9/5+\varepsilon} \quad \left( T \geqslant 10 \right).\] Secondly, the author proves that, if \(k \geqslant 2\) and \(\ell \geqslant 1\) are two fixed integers and if \(f \in \{ \tau_k, \mu_\ell,\Lambda_\ell,\omega_\ell\}\), then \(E_f(x) \ll x^{8/17+\varepsilon}\), where the notation \(g_\ell\) means \[g_\ell(n) := \sum_{n=n_1 \dotsb n_\ell} g(n_1) \dotsb g(n_\ell)\] for any arithmetic function \(g\). As a corollary, the author gets \[\sum_{\substack{n \leqslant x \\\N\lfloor x/n \rfloor \: \mathrm{prime} }} 1 = x \sum_{p} \frac{1}{p(p+1)} + O \left( x^{8/17+\varepsilon} \right) \] improving the result obtained in [\textit{R. Heyman}, Integers 22, Paper A59, 10 p. (2022; Zbl 1504.11095)] and in [\textit{R. Ma} and \textit{J. Wu}, Bull. Aust. Math. Soc. 108, No. 2, 236--243 (2023; Zbl 1537.11124)]. The proof rests on some usual techniques in analytic number theory, such as Vaaler's approximation of the first Bernoulli function, exponential sums techniques involving some spacing problems, and also two precise estimates of some specific sums of types I and II.