The fractional sum of small arithmetic functions (Q2162810)

Asymptotic formulas for the summatory function of a composite functions has been studied in 1960's, cf., e.g., [\textit{A. G. Postnikov}, Introduction to analytic number theory. Transl. from the Russian by G. A. Kandall. Ed. by Ben Silver. Appendix by P. D. T. A. Elliott. Providence, RI: American Mathematical Society (1988; Zbl 0641.10001)]. Let \([x]\) denote the greatest-integer function, i.e., \([x]=n\) if \(n-1<x\le n\), \(n\) being an integer. For an arithmetic function \(f\), an asymptotic formula has been studied for the summatory function of the composite functions (\(x>0\)) \[ S_f(x)=\sum_{n\le x}f\left( \left[ \frac{x}{n} \right] \right). \] In the case where \(f(n)=\sum_{d|n}g(d)\) and \(\sum_{d\le x}|g(d)\ll x^\alpha{(\log x)}^\theta\), \(0\le \alpha<1\), Theorem 2 gives \[ S_f(x)=C_fx+O(x^{\frac{1+\alpha}{3-\alpha}}{(\log x)}^\theta).\tag{1} \] The sums \[ \mathcal{T}_\delta(A,B)=\sum_{A<m\le x/B}\psi \left( \frac{x}{m+\delta}\right), \quad \delta=0,1 \tag{2} \] are the main error terms, where \(\psi(y)\) is the saw-tooth Fourier series. For the estimation of \(\mathcal{T}_\delta(A,B)\), the van der Corput exponent pair method is applied in an ingenious way. The more recent paper [\textit{J. Li} and \textit{J. Ma}, J. Number Theory 247, 35--45 (2023; Zbl 07662014)] studies the more general case \(f(n)n^{-k}\), \(f\) being a convolution as above.