On the convergence of thinned harmonic series (Q2635176)

It is an amusing exercise to show that the series \(\sum_{l\in A} 1/l\) converges where \(A\) denotes the set of natural numbers whose decimal expansion does not contain the digit \(9\); the first person to do this was \textit{A. J. Kempner} [``A curious converging series'', Am. Math. Mon. 21, No. 2, 48--50 (1914; \url{doi:10.2307/2972074})]. One might wonder about series of the form \(\sum_{l\in A} 1/l^\alpha\). A special case of the main result of this paper is that \(\sum_{l\in A} 1/l^\alpha \) converges if and only if \(\alpha> \log_{10}9\). The setup for the main result is this. One looks at \(d\)-adic expansions, selects a set of \(k<d\) digits \(\mathbf{a}=\{a_1,\dots,a_k\}\) in \(\{1,\dots,d-1\}\) and a set of nonnegative integers \(\mathbf{n}=\{n_1,\dots, n_k\}\) and defines \(A(\mathbf{a}, \mathbf{n})\) to be the set of all natural numbers that contain \(a_j\) precisely \(n_j\) times in their \(d\)-adic expansion. (The above case is \(d=10\), \(k=1\), \(a_1=9\), \(n_1=0\).) Then the series \(\sum_{l\in A(\mathbf{a}, \mathbf{n})} 1/l^\alpha\) converges if and only if \(\alpha>\log_d(d-k)\). In the case of convergence the author also gives an upper bound for the series.