On effective computation of expectations in large or infinite dimension (Q917260)

The important problem of the computation of expectations of functionals of stochastic processes is considered. The advantages of a method based on the implementation of the Bernoulli shift operator by pointers is presented. It is shown that in the case of a process whose functional is studied, it is itself mixing or simply ergodic. The shift method can be applied to the process itself, instead of the underlying coordinates. It is proved that for the Bernoulli shift method the speed of convergence can be arbitrarily near 0(1/n). The author announces that the shift method can be slightly improved by using the two-sided Bernoulli shift, which allows to never throw a box away. But the benefit is negligible in high dimensions.