On complex operator-valued O-U processes, T-positivity, and innovations of Okabe and Masani (Q910824)

Let X(t), \(t\in R\), be an infinite-dimensional Ornstein-Uhlenbeck process; i.e. X(t), \(t\in R\), is a continuous function, whose values are isometries between two Hilbert spaces H and K, with the covariance \(X(s)^*X(t)=K(t-s)\) depending only on t-s and such that for every \(t>0\) and \(x\in H\), the orthogonal projection of X(t)x onto the past, \(\overline{sp}\{X(s)y\), \(s\leq 0\), \(y\in H\}\) is equal to the orthogonal projection of X(t)x onto the present \(\overline{X(0)H}.\) It is shown that every infinite-dimensional Ornstein-Uhlenbeck process satisfies an appropriate Langevin type equation and that the regular part of the process is the only regular stationary solution to this equation. This gives a nontrivial generalization of the results by \textit{Y. Okabe} [J. Fac. Sci., Univ. Tokyo, Sect. IA 26, 115-165 (1979; Zbl 0433.60039)] and \textit{K. Itô} [Stochastic Analysis, Proc. Taniguchi Int. Symp., Katata \& Kyoto/Jap. 1982, North-Holland Math. Libr. 32, 197-224 (1984; Zbl 0564.60056)] to the case of processes with nonsymmetric covariances. It also provides the link between the classical concept of innovation for semigroups of isometries [cf. \textit{P. Masani}, Bull. Amer. Math. Soc. 68, 624-632 (1962; Zbl 0117.342)] and the construction of innovation for Ornstein-Uhlenbeck processes.