Stable limit theorems for empirical processes under conditional neighborhood dependence (Q1740523)

The authors introduce a new structure modelling dependence between random variables, called conditional neighborhood dependence (CND). Given a set of random variables and a set of \(\sigma\)-algebras, each indexed by the same set equipped with a neighborhood structure (that is, a notion of adjacent elements), the CND property states that any pair of non-adjacent sets are conditionally independent, given the collection of \(\sigma\)-algebras in the neighborhood of one of the two sets. This notion generalizes both dependency graphs and certain Markov random fields, allowing for a general and complex dependence structure. The main results of this paper are two limit theorems for random variables with the CND property. The first is a Berry-Esseen bound for sums of such random variables. The second is a Donsker-type stable limit theorem for an empirical process (in which the underlying random variables are CND) indexed by a class of functions. In each of these cases, problems caused by large maximal degrees in the underlying neighborhood are resolved by conditioning on the random variables associated with high degree vertices.