Stochastic processes and applications to countably additive restrictions of group-valued finitely additive measures (Q1915570): Difference between revisions

The paper deals with the following problem: given a continuous finitely additive measure \(m\) on the power set of a set \(\Omega\), are there countably additive restrictions of \(m\) to suitable sub-algebras, preserving continuity, and with the same range \(R\)? Here, continuity means that \(\Omega\) can be splitted into finitely many subsets, each having measure as small as desired. This problem is stated for topological group-valued finitely additive measures, and it is solved in the affirmative, provided some conditions are fulfilled: one condition is that \(m\) is controllable, i.e. there exists a real-valued non-negative finitely additive measure \(\nu\), such that \(m \ll \nu\); the other conditions are that \(\text{card} (\Omega) \geq c\), and that \(\text{card} (R) \leq \text{card} (\Omega)\). Although the result could be achieved by techniques similar to those used by \textit{K. P. S. Bhaskara Rao} and the authors [J. Math. Anal. Appl. 177, No. 1, 166-169 (1993; Zbl 0786.28005)], here a new proof is given, which looks more direct, and is related with the existence of nice distributions for some kinds of stochastic processes; more precisely the result used reads as follows: Given a finitely additive probability space \((S, \Sigma,P)\) and a stochastic process \(X = (X_t)_{t \in T}\) on it, if the marginal distribution functions \(F_t\) are right-continuous at every point, and satisfy \(\lim_{x \to - \infty} F_t (x) = 0 = 1 - \lim_{x \to + \infty} F_t (x)\) for all \(t\), then there exists a countably additive probability \(Q\) on the Borel product \((\mathbb{R}^T, {\mathcal B}^{(T)})\), which agrees with \(P_X\) on all the cylinders of the form \(]a_1, b_1] \times \cdots \times ]a_n,b_n] \times \mathbb{R}^T\).