A central limit theorem for the variation of the sum of digits (Q6596231)

The sum-of-digits function \(s\) in integer base \(b\ge2\) is studied, in particular, the differences\N\[\N\Delta^{(r)}(n):=s(n+r)-s(n),\N\]\Nfor integers \(r,n\ge0\), and asymptotic densities\N\[\N\mu^{(r)}(d):=\lim_{N\rightarrow\infty}\frac 1N\bigl\lvert\bigl\{n<N:\Delta^{(r)}(n)=d\bigr\}\bigr\rvert.\N\]\NThe probability measures \(\mu^{(r)}\) on \(\mathbb Z\) form the basis for the paper under review. Up to finitely many exceptional points, the function \(\Delta^{(r)}\) can be extended to the group \(\mathbb X\) of \(b\)-adic integers (endowed with the Haar measure \(\mathbb P\)). Proposition~2.2 states in particular that the values \(\mu^{(r)}(d)\) are recovered when the asymptotic density is replaced by \(\mathbb P(\{x\in \mathbb X:\Delta^{(r)}(x)=d\})\). Of course, the variance\N\[\N\mathrm{Var}(\mu^{(r)})=\sum_{d\in\mathbb Z}d^2\mu^{(r)}(d)\N\]\N(for which there exists an explicit recurrence relation, see Lemma~3.3) is of great interest for the paper; lower and upper bounds are needed, and studied in detail. The main result, Theorem~1.3, states that the renormalized measures \(\tilde\mu^{(r)}\) converge, in distribution, to the standard normal law, as soon as the number of \textit{blocks} (defined in Definition~1.1) in the base-\(b\) expansion tends to \(\infty\). More generally an estimate for the speed of convergence is given (Theorem~1.4). This should be compared to results by \textit{J. Emme} and \textit{P. Hubert} [Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 19, No. 2, 757--780 (2019; Zbl 1451.60031)], and by \textit{M. Wallner} and the reviewer [Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 24, No. 1, 1--31 (2023; Zbl 1521.11008)].\N\NIn the proofs, the well-known \textit{odometer} \(T\) on the set \(\mathbb X\), and \textit{Rokhlin towers} are employed. Clearly, the difference \(\Delta^{(r)}\) can be written (almost surely) as a sum of differences \(\Delta^{(1)}\) along subsequent iterations of the odometer; this small remark proves of great use. Theorem~1.4 is then proved by employing, among other things, a theorem by Sunklodas on the speed of convergence in the central limit theorem for \(\phi\)-mixing sequences [\textit{J. K. Sunklodas}, Lith. Math. J. 51, No. 2, 260--273 (2011; Zbl 1227.60031)].