Convergence in law of random elements and random sets

Let \(T\) be a fixed topological space (not necessarily Hausdorff). Denote by \((\Omega,{\mathcal F}, P)\) and \((\Gamma,{\mathcal A},Q)\) two fixed probability spaces, and denote by \((E,E^{\ast},E_{\ast})\) the \(P\)-expectation, the upper and the lower \(P\)-expectation. Similarly, denote by \(({\mathbb E},{\mathbb E}^{\ast},{\mathbb E}_{\ast})\) the \(Q\)-expectation, the upper and the lower \(Q\)-expectation. Let \(Z\) be a \(T\)-valued random element on \((\Gamma,{\mathcal A},Q)\) and \((X_\pi\mid\pi\in\Pi)\) be a \(\Pi\)-net of \(T\)-valued random elements on \((\Omega,{\mathcal F},P)\). \(T\)-valued random element on \((\Omega,{\mathcal F}, P)\) or \((\Gamma,{\mathcal A},Q)\) means simply a map from \(\Omega\) or \(\Gamma\) into \(T\). Denote by \(C(T)\) the set of all bounded continuous functions \(\varphi : T\to{\mathbb{R}}\). We say that \((X_\pi)\) is convergent in Baire law to \(Z\) if \(Z\) is Baire \(Q\)-measurable and \(\lim_{\pi\uparrow\Pi}E^\ast\varphi(X_\pi)={\mathbb{E}}\varphi(Z)\) \(\forall\varphi\in C(T)\). Denote by \(\text{Usc}(T)\) the set of all upper semicontinuous functions \(f\) from \(T\) into the extended real line \(\overline{\mathbf{R}}:=[-\infty,\infty]\) such that \(\sup_{t\in T}f(t)<\infty\). We also say that \((X_\pi)\) converges in Borel law to \(Z\) if \(\limsup_{\pi\uparrow\Pi}E^{\ast}f(X_\pi)\leq{\mathbb E}_{\ast}f(Z) \quad\forall f\in\text{Usc}(T).\) If \(T\) is metrizable or more generally if \(T\) is perfectly normal, then the notions of convergence in Baire and Borel law are equivalent and behave nicely, and they satisfy most of the calculus of the classical convergence in law for sequences of measurable random variables with values in a separable metric space. However, when \(T\) is a general topological space, there exist some peculiarities of convergence in law, which seem to have overlooked in the past. The object of this paper is to study the notions of convergence in Baire and Borel law for nets of random elements in general and nets of random sets in particular. In Section 2, the reader will find the notation used throughout this paper. Section 3 contains a careful study of the so-called probability extents, which plays a cardinal role in the theory of convergence in law. Section 4 is devoted to a study to what extent the classical calculus for convergence in law of measurable random variables with values in a separable metric space carries over to Baire and Borel convergence for nets of random elements with values in an arbitrary topological space. In Section 5, the general theory of Section 4 is applied to the study of convergence of statistical estimators in particular. Before proceeding with the general theory, the author gives an example illustrating some of the peculiarities of convergence in law. For the entire collection see [Zbl 0883.00024].

Recommendations

Cited in

(8)

This page was built for publication: Convergence in law of random elements and random sets

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4387143)