Substituting Independent Processes

Abstract

It is shown by constructing Rohlins canonical measures that for a strictly stationary, d-dimensional vector-valued process X there exists another strictly stationary d-dimensional process U with uniform one-dimensional marginals and with the same mixing properties as X, such that X is a finitary factor of U of coding length 1, and such that the projection map is order preserving in each coordinate. As an application this extends the a.s. approximation of the empirical distribution function of weakly dependent random vectors with continuous distribution function in [1] and [3] to the general case.