Best finite approximations of Benford's law (Q2312784)

The set \(\mathcal{P}\) of all Borel probability measures on a given real compact interval can be organized as a metric space. The paper refers to the probability metrics on \(\mathcal{P}\) due to Lévy, Kantorovich, and Kolmogorov. For each of these distances, the best finitely supported approximation (with arbitrary given number of atoms) of any \(\mu\in\mathcal{P}\) is identified. In particular, best (constrained or unconstrained) approximations are highlighted for Benford's law (logarithmic distribution of significands). In this way, certain known results in the literature are extended. The paper also provides new rigorous tools to evaluate the empirical observations on Benford's law. This fine study is complemented by appropriate remarks and interesting numerical examples.