On Popoviciu's concept of convexity for functions of $d$ variables
Andrzej Komisarski, Teresa Rajba
TL;DR
The paper develops a multidimensional generalization of Popoviciu's higher-order convexity by introducing box-$\mathbf n$-convex functions for $\mathbf n=(n_1,\dots,n_d)$ and arbitrary open boxes. It derives integral representations that decompose box-$\mathbf n$-convex functions into pseudo-polynomials plus kernel-based measures, enabling differential methods via $\mathbf n$-regularity and leading to Raşa, Jensen, and Hermite-Hadamard inequalities in this setting. It also establishes stochastic box-$\mathbf n$-convex orders, with characterizations in terms of moments and convolutions, and discusses probabilistic representations and uniqueness of the representing measures. Overall, the work unifies and extends prior two-variable differentiable results to the general $d$-dimensional, non-regular regime, providing a robust framework for functional inequalities in higher-order convexity.
Abstract
We establish an integral representation for Popoviciu's convex functions of $d$ variables. This representation serves as a~foundation for deriving several functional inequalities, analogous to those well-known for usual convex functions. Our results generalize and extend the results obtained by S.~Gal, C.~Niculescu, B.~Gavrea, T.~Popoviciu, and others, who considered only differentiable functions of two variables. In contrast to other authors, we do not impose any additional regularity assumptions on the studied functions.
