A complete characterization of normal cones and extreme points for p-boxes

Probability boxes, also called p-boxes, correspond to sets of probability distributions bounded by a pair distribution functions. They belong the class models known as imprecise probabilities. One central issues related probabilities is intervals values corresponding expectations random variables, and in particular interval bounds. In general, these are reached at extreme points credal sets, which denote convex compatible probabilistic models. The goal this paper characterize identify p-boxes on finite domains. To achieve this, we use concept normal cones. context probabilities, variables whose common point. Our main results include characterization all possible cones their relation points, identification an adjacency structure collection that closely set points.