On the equivalence of probability spaces (Q2412505): Difference between revisions

The authors consider an arbitrary sigma-finite measure space \((M, \mathcal{F},\sigma)\) and show the existence of a zero-mean Gaussian process \(\left(W^{(\sigma)}_A\right)_{A\in\mathcal{F}}\), indexed by sets of the sigma-algebra \(\mathcal{F}\), such that \(\mathbb{E}\left[ W^{(\sigma)}_AW^{(\sigma)}_B\right]=\sigma(A\cap B)\) for all \(A,B\in \mathcal{F}\). The authors discuss properties of \(W^{(\sigma)}\) and of the corresponding Wiener integrals, show that, if \(\sigma\) is refinable (Def. 3.20), the measure \(\sigma\) may be recovered as the quadratic variation of \(W^{(\sigma)}\), establish (a generalization of) the Ito formula. It is also shown that \(W^{(\sigma)}\) may be realized in the canonical infinite Cartesian product measure space; (generalizations of) the Karhunen-Loève formula and the Cameron-Martin-Maruyama theorem are obtained. Further, the authors consider two different realizations of Gaussian processes: in \(\mathcal{S}'(\mathbb{R})\) (via Gelfand triples \(\sim\) white noise analysis) and in \(C(\mathbb{R})\) (via the Kolmogorov consistency theorem) and show the equivalence of these two realizations by presenting an explicit formula which transforms one approach into the other.