On differences of two harmonic numbers (Q7028813)

\textit{P. Erdős} and \textit{R. L. Graham} [Old and new problems and results in combinatorial number theory. L'Enseignement Mathématique, Université de Genève, Genève (1980; Zbl 0434.10001)] raised the following question concerning randomness in the harmonic numbers. Writing \(t=t(n)\) for the least integer with \(\displaystyle\varepsilon_n=\sum_{k=n}^t\frac{1}{k}-1\ge 0\) we have \(t=\bigl(\mathrm{e}+o(1)\bigr)n\) and when \(\displaystyle\sum_{k=n}^t\frac{1}{k}\) exceeds \(1\) it does so by no more than \(\bigl(1/\mathrm{e}+o(1)\bigr)/n\). There being no reason to imagine that this does not behave like a uniformly distributed random variable in the range \([0,1/(\mathrm{e}n)]\), Erdős and Graham asked the following: `How small can \(\varepsilon_n\) be? As far as we know this has not been looked at. It should be true that \(\displaystyle \liminf_{n\to\infty}n^2\varepsilon_n=0\) but perhaps \(n^{2+\delta}\varepsilon_n\to\infty\) as \(n\to\infty\) for every \(\delta>0\). \N\NThe first result proved here -- using elementary but ingenious arguments starting from the continued fraction expansion of \(\mathrm{e}\) -- shows that there is a constant \(C>0\) for which there are infinitely many pairs \((m,n)\in\mathbb{N}^2\) (constructed in concrete terms) with \(\displaystyle 1\le \sum_{\ell=n}^{m}\frac{1}{\ell} \le 1+\frac{C\log\log n}{n^2\log n}\), resolving the first question. By leveraging much deeper techniques due to \textit{I. Danicic} [Mathematika 5, 30--37 (1958; Zbl 0085.03304)], \textit{G. Harman} [Glasg. Math. J. 38, No. 3, 299--308 (1996; Zbl 0885.11046)], \textit{H. Heilbronn} [Q. J. Math., Oxf. Ser. 19, 249--256 (1948; Zbl 0031.20502)], \textit{C. Hooley} [Glasg. Math. J. 32, No. 3, 309--316 (1990; Zbl 0715.11033)], \textit{A. Zaharescu} [Invent. Math. 121, No. 2, 379--388 (1995; Zbl 0827.11040)] aimed at understanding how well real numbers can be approximated by rationals with square denominator a considerably more refined result is proved, showing that for any \(\varepsilon>0\) there are infinitely many pairs \((m,n)\in\mathbb{N}^2\) with \(\displaystyle\bigl\vert\sum_{\ell=n}^{m}\frac{1}{\ell}-1\bigr\vert \le \frac{1}{n^2(\log n)^{(5/4)-\varepsilon}}\).