Functional limit theorems for additive arithmetic functions on intervals (Q1897874)

An additive function \(h\) can be used to generate a stochastic process via simulating time by a truncation, that is, by putting \[ H_x (m,t) = \beta (x)^{-1} \sum_{p^k |m, p \leq z(t)} h(p^k) - \alpha \bigl( z(t), x \bigr) \] with suitable functions \(\alpha, \beta, z\). Here \(m\) is a typical element of the \(x\)th probability space; in the classical form it is the set \(\{1, \ldots, x\}\) with the uniform measure. In this paper this is modified to \(\{x - y, \ldots, x\}\), where \(y\) is a function of \(x\). For a wide class of functions a necessary and sufficient condition is found for these processes to converge to a process with independent increments (the result is too complicated to quote in detail). If \(\log y \sim \log x\) and the limit is the standard Brownian motion, then the condition reduces to \[ \sum_{p \leq x, |h(p) |\geq \varepsilon \beta (x)} p^{-1} \bigl |h(p)/ \beta (x) \bigr |^* \to 0, \] where \(u^* = \min (1, |u |) \text{sgn} u\).