A probabilistic ergodic decomposition result (Q985338)

From the different versions of ergodic decompositions existing in literature, in the present paper a probabilistic approach is chosen based on the notion of conditional probability, i.e., for a basic probability space \((X,\Sigma,\mu)\) a sub-\(\sigma\) algebra \(T\) of \(\Sigma\) is said to decompose the probability measure in an ergodic way, if one (and thus all) regular conditional probability \(P^T(x,B)\) with respect to \(T\) for \(\mu\)-almost all \(x\in X\) is 0-1-valued for every \(B\in T\). In this situation an integral decomposition in \(T\)-ergodic parts is given by \(\mu(\cdot)= \int P^T(x,\cdot)\,d\mu\). \textit{H. Shimomura} [Publ. Res. Inst. Math. Sci. 26, No.~5, 861--865 (1990; Zbl 0716.28005); ibid. 14, 359--381 (1978; Zbl 0391.60004)] proved, that the intersection of a decreasing sequence of separable sub-\(\sigma\)-algebras of \(\Sigma\) decomposes \(\mu\) in an ergodic way. He also gives an example of a suitable sub-\(\sigma\)-algebra for which the above decomposition is not ergodic. The main result of the paper deals with the preservation of ergodic decomposition under intersections of arbitrary families of sub-\(\sigma\)-algebras which improves H. Shimomura's result considerably. For ensuring existence of regular conditional probabilities, all results are restricted to probability spaces with a separable \(\sigma\)-algebra, containing an approximating compact class. There are given applications for probability measures being quasi-invariant under invertible bi-measurable transformations.