No label defined (Q4354029)

This mostly expository paper first of all treats, from the unified point of view of coupling constructions, several examples of dominance relations among probability laws of random elements of abstract spaces. In each example, such relations among collections of laws \( \{ \mu_{\alpha}\), \(\alpha \in {\mathcal A}\} \) can be seen to be equivalent to almost-sure relations on some probability space between random elements \( \{ X_{\alpha}\), \(\alpha \in {\mathcal A} \}\); examples of this sort are (a) the equivalence between stochastic dominance \(\mu_1 < \mu_2 \) of probability laws on the real line and a.s. inequality \( X_1 \leq X_2 \) between suitably constructed r.v.'s with these laws, and (b) the Skorokhod-Dudley-Wichura equivalence between weak convergence of laws \( \mu_n \) on a separable metric space and the a.s. convergence of a `coupled' sequence of r.v.'s \( X_n \) with the corresponding laws. The paper gives other examples, and collects proofs and references concerning them in a very useful Appendix. In particular, the author gives a new relation among laws of random sets which he shows equivalent to a.s. inclusion of coupled random sets with the given laws. The second topic, and ultimate aim, of the paper is to show how a metric (the `Wasserstein distance') for inner-regular probability laws on a general metric space can be used to define an `ergodic coefficient' for the rate at which a Markov process with values in the metric space approaches equilibrium, i.e., its stationary distribution from arbitrary starting points. The author uses these ideas to discuss the accuracy of statistical results derived from simulations of Markov processes with general metric state-spaces. The topic of sensitivity analysis of simulations of Markov processes is discussed also in terms of derivatives of Markov kernels. For the entire collection see [Zbl 0869.00058].