The paper tackles a fundamental question in mathematical analysis: how can we measure the "roughness" or variation of a function defined on an abstract geometric space, using only comparisons between pairs of points rather than traditional calculus? Classical tools like derivatives and integrals work well on flat Euclidean space, but many modern applications involve curved spaces, fractals, or networks where standard calculus breaks down. The authors work in a very general setting called a metric measure space, which just means a space with a notion of distance and a way to measure sizes of regions, satisfying some mild regularity conditions.
The central results are called Poincare inequalities, which are estimates that control how much a function can oscillate over a region in terms of a local measure of its average rate of change. The new twist here is that the "rate of change" is defined in a nonlocal and nonconvex way: instead of looking at derivatives, one sums up contributions from pairs of points where the function jumps by more than a threshold relative to their distance. The paper identifies precisely which versions of these nonlocal energy functionals lead to valid Poincare inequalities, and shows that the allowable parameter ranges are sharp, meaning they cannot be improved without losing the inequality in some spaces. Two different families of functionals are treated, one based on distances between points and one based on volumes of balls.
The paper also connects these nonlocal functionals to classical energy quantities like Sobolev energies and total variation, showing that the classical energies can be recovered as limiting cases when the nonlocal parameter is pushed to an extreme. The proofs rely on two key technical tools: a characterization of the geometric spaces via curves connecting points, and a clever adaptive subdivision argument using median values instead of averages, which behaves better when functions are irregular. The results unify and sharpen earlier work done in Euclidean settings, extending the theory to a broad class of spaces relevant in geometry, probability, and data analysis.