Sums of numbers with many divisors (Q1283035)

For a fixed integer \(k\geq 2\), there is the interesting problem of representing \(n\) as \(m_1+\cdots+m_k\), with each \(m_i\) having, in some sense, many divisors. The divisor function \(d(n)\) has a maximum size which is essentially \(D(n)=\exp(\log 2\log n/\log\log n)\), and one may interpret the problem by prescribing the condition that \(d(m_i)>D(n)^\alpha\) for \(1\leq i\leq k\), with a fixed positive real parameter \(\alpha\leq 1\). The authors prove that \(| \{n\leq x:d(n)>D(x)^\alpha\}| =x^{1-\alpha+o(1)}\) as \(x\to\infty\), uniformly in \(0\leq\alpha\leq 1\). For the problem, it follows almost immediately that if \(\alpha>1-1/k\) then there are arbitrarily large values of \(n\) which are not representable with the prescribed condition. To complement this, they also prove that if \(\alpha<1-1/k\) then all large \(n\) are representable. The authors remark that, by applying the circle method as in Waring's problem, one could obtain an asymptotic formula for the number of representations, provided that \(\alpha\) is sufficiently small, but there seems to be no hope of obtaining such an asymptotic estimate when one assumes only that \(\alpha<1-1/k\). The paper is a model of clarity, precision and economy in style and presentation.