Über eine zum Kreisproblem verwandte Summe. (On a sum related to the circle problem) (Q1060242)

Let \(\Psi_ k( )\) denote the 1-periodic Bernoulli polynomial of order \(k\in {\mathbb{N}}\). An equivalent form of the classical circle problem is the unproved conjecture that for \(x\to \infty\) the estimate \[ (*)\quad \sum_{n\leq \sqrt{x/2}}\Psi_ 1((x-n^ 2)^{1/2})=O(x^{1/4+\epsilon}) \] still holds for arbitrary \(\epsilon >0\). In this note the corresponding conjecture with \(\Psi_ k( )\) (k\(\geq 2)\) instead of \(\Psi_ 1( )\) in (*) is settled in an affirmative way (even with \(\epsilon =0)\). It is worth noticing that the multiplicative analogue of (*), a special case of the so-called 'Conjecture of Chowla and Walum': \[ \sum_{n\leq \sqrt{x}}\Psi_ k(x/n)=O(x^{1/4+\epsilon}) \] is still unproved today for every \(k\in {\mathbb{N}}\); see the author's paper [Arch. Math. 44, 348-354 (1985; Zbl 0556.10032)]. Moreover, the generalized sum \(\sum_{n\leq \sqrt{x/2}}\Psi_ k((x- n^{\beta})^{\gamma})\), \(\beta\),\(\gamma\in {\mathbb{R}}\) with \(\beta >1\), \(0<\gamma <1\) and \(k\geq 2\), is estimated by means of an analogue to a classical result due to \textit{J. G. van der Corput} [Math. Z. 17, 250-259 (1923)]. The O-results are too complicated for citing here.