On the behaviour of a certain error-term (Q1821146)

For any fixed positive integer k let \(\delta _ k(n)\) denote the greatest divisor of n which is prime to k and define the error term \(E_ k\) by \[ \sum _{n\leq x}\delta _ k(n)=\frac{k}{2 \sigma (k)}x^ 2+E_ k(x). \] \textit{D. Suryanarayana} [Math. Stud. 37, 147-157 (1969; Zbl 0206.338)] obtained the estimate \(E_ k(x)=O(x \log ^ 2x)\) which was sharpened by \textit{V. S. Joshi} and \textit{A. M. Vaidya} [Topics in classical number theory, Budapest, Colloq. Math. Soc. János Bolyai 34, 791-806 (1984; Zbl 0546.10038)] to \(E_ k(x)=O(x)\). For primes p they investigated the error-term \(E_ p(x)\) in more detail. Among other things they showed \(E_ p(x)=\Omega _{\pm}(x)\). Using other methods the present authors obtain similar results on \(E_ k\) for all \(k\in {\mathbb{N}}\) and in addition they give asymptotic formulas for various means of the error-function such as \[ \sum _{n\leq x}E_ k(n)=\frac{k}{4 \sigma (k)}x^ 2+O(x (\log x)^{\omega (k)}). \]