Approximation of distributions by bounded sets (Q996788)

Let \((S,d)\) be a separable metric space, mostly supposed to be locally compact. Let \(\mathcal{A}\) be a collection of subsets of \(S\). \(\mathcal{A}\) is called \(K\)-bounded if the diameters of \(A\in \mathcal{A}\) do not exceed \(K\). Hence \(\mathcal{A}^K_k\) denotes the collection of \(k\)-unions \(\bigcup_1^k A_j\) of sets \(\mathcal{A}_j\) belonging to the \(K\)-bounded collection \(\mathcal{A}\). The authors investigate approximations of probabilities \(P\) on \(S\) by sets of \(\mathcal{A}^K_k\) w.r.t. a decreasing discrepancy function \(\varphi\) with corresponding loss function \(A \mapsto W(A, P) = \int_S \varphi(d(x,A)) P(dx)\). An optimal approximative set for \(P\) is a set \(A^*\) for which the infimum of \(W(A,P)\) is attained. The space \(2^S\) of subsets of \(S\) is endowed with the Hausdorff topology. Let \(P_n \to P\) (weakly) and assume \(S\) to be locally compact. The authors investigate the limit behaviour w.r.t. the Hausdorff topology of optimal approximative sets for \(P_n\) and \(P\), respectively.