Binary Egyptian fractions (Q1587419)

A number of papers deal with the question of estimating the number of \(n \leq x\) such that for fixed \(a\) the equation \[ a/n=1/{m_1} + \cdots + 1/m_{k} \tag{1} \] has no solution in positive integers. This paper takes a different point of view. Let \(A_k(n)\) denote the number of positive integers \(a\) such that (1) admits a solution in positive integers. The authors work with \(A_k^*(n)\), where the \(*\) indicates that only those \(a\) with \((a,n)=1\) are counted. Results on \(A^*_k(n)\) imply results on the number \(A_k(n)\) by means of \(A_k(n)=\sum_{d \mid n} A^*_k(d)\). It is proved that \[ x ( \log x)^3 \ll \sum_{n \leq x} A_2^{*}(n) \ll x (\log x)^3. \] For \(k>2\) only an upper bound is shown: \[ A_k^{*}(n) \ll n^{\alpha_k + \varepsilon}, \tag{2} \] where \(\alpha_2=0\) and \(\alpha_k=1- \frac{1- \alpha_{k-1}}{2+\alpha_{k-1}}\). This implies that \( \sum_{n \leq x} A_k^{*}(n) \ll x^{1+\alpha_k + \varepsilon}\). It is conjectured that (2) also holds with \(\alpha_k=0\). For \(k=2\) a better result is obtained for the related problem where \(B^*(n)\) counts the number of solutions as \(a\) varies subject to \((a,n)=1\). It is proved that \[ \sum_{n \leq x} B^*(n) = \frac{1}{8} \prod_p \left(1- \frac{1}{p}\right)^2 \left(1+\frac{2}{p}\right) x \log^3 x + O\left( \frac{x \log^3 x}{\log \log x}\right). \] As the authors remark, a proof based on contour integration would sharpen the error term. The reviewer expects that a slight change of the argument (at the bottom of page 75) might sharpen the error term, in a very elementary way, to \(O\left( x \log^2 x \log \log x \right)\): one has \(\log \tau (ja) \ll \Omega(ja)\) so that \[ x \sum_{a \leq A} \frac{\varphi(a)}{a} \sum_{j \leq x/a} \frac{\log \tau(ja)}{ja} \ll x \log^2 x \log \log x \] holds. The paper uses elementary methods from multiplicative number theory. The key idea is to use a bijection between the solutions of \(a/n=1/m_{1}+1/m_{2}\) and the pairs \((u_1,u_2) \in \mathbb{N}^2\) with \((u_1,u_2)=1\), \(u_1u_2 \mid n\) and \(a \mid u_1+u_2\). W. Sierpiński attributes this classification of sums of two unit fractions to A. Schinzel [see \textit{W. Sierpiński}'s monograph ``On the decomposition of rational numbers into unit fractions'' (Polish), Pánstwowe Wydawnictwo Naukowe, Warsaw (1957; Zbl 0078.03502)].