Uniform distribution on divisors and Behrend sequences (Q1920468)

This paper's main objectives are stated in its first sentence: ``to give a consistent, largely self contained account on the theory of uniform distribution on divisors, and to establish effective estimates with immediate applications to the construction of Behrend sequences.'' A strictly increasing sequence \({\mathcal A}\) of integers greater than 1 is said to be Behrend if the set \(\{ma: a\in {\mathcal A}\), \(m \in\mathbb{N}\}\) of multiples has asymptotic density 1. For \(f\) an arithmetic function, let \[ \Delta (n,f) = \sup_{0 \leq u\leq v\leq 1} \Bigl|\sum_{d \mid n,\;u< \langle f(d) \rangle \leq v} 1-(v-u) \tau (n) \Bigr|, \] where \(\langle u\rangle\) denotes the fractional part of \(u\) and \(\tau(n)\) the divisor function. Then \(f\) is uniformly distributed on the divisors (abbreviated to erd) if \(\Delta (n,f) = o(\tau (n))\) on a sequence of asymptotic density 1. This paper gives a comprehensive account of results concerning these concepts, obtained during the last twenty years or so by various authors. Most of the fifteen theorems and eleven corollaries establish a set of conditions under which an arithmetic function \(f\) is erd and/or a specified upper bound for \(\Delta (n,f)\) holds on a sequence of logarithmic density 1, and in some cases this leads to a construction of a Behrend sequence. Functions of moderate and of excessive growth are considered, and section 5 is concerned with additive functions. To illustrate the range of the paper, we single out a few results that are simple to quote. Corollaries 2 and 3 (due to G. Tenenbaum, R. R. Hall, respectively) show that \(f(d) = (\log d)^\alpha\) is erd if and only if \(\alpha >0\), \(f(d) = (\log_2d)^\beta\) is erd if and only if \(\beta >1\). Corollary 5 states that the sequence \(\{n \geq 2: \langle n^\alpha \rangle \leq (\log n)^{-\delta}\}\) is Behrend when \(\alpha >0\), \(\alpha \notin \mathbb{N}\) and \(\delta < {1\over 2} \log {12 \over 11}\). In section 4, the function \(f(d) = \theta d\), \(\theta\) irrational, is shown to be erd. From corollary 11 (due to R. R. Hall, I. Kâtai), \(\Delta (n,\log) < (\tau (n))^\alpha\) holds on a sequence of logarithmic density 1 when \(\alpha > (\log {4 \over \pi})/ \log 2\).