Decomposing the sum-of-digits correlation measure
Bartosz Sobolewski, Lukas Spiegelhofer
TL;DR
This work develops a structural decomposition for the binary sum-of-digits correlation function $\gamma_t(\vartheta)$ by embedding the problem in an extended binary-expansion framework and encoding carries via matrix products. The main result expresses the extended $\gamma_{\mathbf t}$ as a finite sum of components with universal weights, clarifying how block lengths and their arrangement govern the correlations, and enabling explicit computations of the induced densities $c_{\mathbf t}$. By specializing to cases with at most two $1$-blocks and by analyzing Thue–Morse correlations, the paper derives concrete lower bounds for Cusick's conjecture and provides exact expressions for key densities in canonical configurations. The results shed light on the nonlocal carry effects that drive the central tendencies in the distribution of $\mathsf s(n+t)-\mathsf s(n)$ and propose concrete conjectures for extremal configurations and block-append operations, suggesting new avenues toward the unresolved Cusick conjecture.
Abstract
Let $s(n)$ denote the number of ones in the binary expansion of the nonnegative integer $n$. How does $s$ behave under addition of a constant $t$? In order to study the differences \[s(n+t)-s(n),\] for all $n\ge0$, we consider the associated characteristic function $γ_t$. Our main theorem is a structural result on the decomposition of $γ_t$ into a sum of \emph{components}. We also study in detail the case that $t$ contains at most two blocks of consecutive $1$s. The results in this paper are motivated by \emph{Cusick's conjecture} on the sum-of-digits function. This conjecture is concerned with the \emph{central tendency} of the corresponding probability distributions, and is still unsolved.
