The distribution functions of \(\sigma(n)/n\) and \(n/\varphi(n)\). II (Q455801)

Let \(\phi(n)\) denote Euler's totient function, and set \[ B(t) := \lim_{N \rightarrow \infty} \frac{1}{N} \#\{n \leq N : n/\phi(n) \geq t\} . \] This limit is known to exist, and the author's main result in this paper asserts that for any \(m \geq 2\), \[ B(t) = \exp\left\{-e^{te^{-\gamma}}(1 + \sum_{j=2}^{m} \frac{a_j}{t^j} + O_{m}(\frac{1}{t^{m+1}}) )\right\} , \] where \(\gamma\) is Euler's constant and where the \(a_j\) are certain other constants all of whose values can be extracted from the proof. The same is true if, in the definition of \(B(t)\), one replaces \(n/\phi(n)\) by \(\sigma(n)/n\), where \(\sigma(n)\) denotes the sum of divisors of \(n\). The author [Proc. Am. Math. Soc. 135, No. 9, 2677--2681 (2007; Zbl 1169.11042)] has previously proved the weaker result that \[ B(t) = \exp\{-e^{te^{-\gamma}}(1 + O(\frac{1}{t^{2}}) )\} , \] which improves an estimate of Erdős. This was done by using Rankin's method in the form \[ B(t) \leq \frac{1}{t^s} \lim_{N \rightarrow \infty} \frac{1}{N} \sum_{n \leq N} \left(\frac{n}{\phi(n)}\right)^s , \;\;\; \mathrm{Re}(s) > 0 \] to obtain an upper bound for \(B(t)\), and using a special construction of numbers \(n\) (with relatively many small prime factors) to obtain a lower bound. In the present paper, it is shown that if one chooses \(s\) optimally in Rankin's method (i.e., to minimise the resulting upper bound), then in fact one has a very good estimate for the true size of \(B(t)\). This kind of behaviour and approach may remind the reader of the use of the saddle-point method when studying smooth/friable integers. In the present case, it is comparatively much easier to show that the Rankin upper bound is close to the truth (see Lemma 2 of the paper), both because one is only seeking a logarithmic asymptotic rather than a true one, and also because the function \(B(t)\) decays very rapidly with \(t\) (so there is a very close correspondence between \(B(t)\) and the Mellin transform of \(B(\cdot)\) evaluated at the optimal \(s=s(t)\)). Having neatly established this, the bulk of the paper is devoted to fiddly calculations of the Rankin upper bound to obtain the claimed expression \(\sum_{j=2}^{m} \frac{a_j}{t^j}\) in the exponential.