Marshall lemma in discrete convex estimation (Q2344880)

This short note proves a Marshall-type inequality in a discrete setting. Given a convex probability mass function \(p\) defined on \(\mathbb{N}\), let \(F\) be the corresponding distribution function, and suppose we have empirical distribution function \(\mathbb{F}_n\). Letting \(\widehat{p}_n\) be the discrete convex least-squares estimator of \(p\) and \(\widehat{F}_n\) be the corresponding distribution function, the authors prove that \[ \|\widehat{F}_n-F\|_\infty\leq 2 \|\mathbb{F}_n-F\|_\infty, \] where \(\|\cdot\|_\infty\) is the usual supremum norm.