Sums of Reciprocals of Fractional Parts II
Reynold Fregoli
TL;DR
The paper develops a general lattice-point counting framework for non-convex regions arising in Diophantine approximation by introducing φ-multiplicatively-badly-approximable matrices and proving a sharp upper bound for the lattice intersection count M(L,ε,R,T) that depends only on the geometric mean of the T-components. This geometric-mean bound, combined with tessellation techniques and a careful analysis of successive minima under diagonal maps, yields unconditional almost-sure asymptotics for sums of reciprocals of fractional parts: S(α,T) ≍ log T̄ · log T1 · ... · log Tn and S*(α,T) ≍ (log T)^{n+1} for a.e. α. The results settle conjectures by Beresnevich–Haynes–Velani and extend lower-bound frameworks to sharp upper bounds in higher dimensions, including non-averaged sums with general φ. Methodologically, the work fuses geometry-of-numbers tessellations with Schmidt–Cassels-type probabilistic arguments, producing a versatile approach to non-convex lattice-point problems with Diophantine applications.
Abstract
We prove an estimate for the number of lattice points lying in certain non-convex Euclidean domains of interest in Diophantine approximation. As an application, we generalise a result of Kruse (1964) concerning the almost sure order of magnitude of sums of reciprocals of fractional parts and solve a conjecture posed by Beresnevich, Haynes, and Velani. The methods are based both on the geometry of numbers and on probability theory.
