Nonlinear filtering of semi-Dirichlet processes (Q1041054)

The optimal filtering problem for a time-homogeneous Markov process \(\{X_t\}_{t\geq0}\) taking values in a general state space \(E\) consists of computing the conditional distributions of \(\{X_t\}_{t\geq0}\) based on observations of \(Y_t=\int_0^t h(X_s)ds+W_t\), where \(h:E\to\mathbb R^d\) is measurable an \(\{W_t\}_{t\geq0}\) is the standard \(d\)-dimensional Brownian motion independent of \(\{X_t\}_{t\geq0}\). In the case where \(\{X_t\}_{t\geq0}\) is a solution of the classical martingale problem or satisfies a stochastic differential equation, the well-known Fujisaki-Kallianpur-Kunita (also known as Kushner-Stratonovich) and Duncan-Mortensen-Zakai equations and the related results may be used to derive the filtering processes. Unfortunately, these fundamental results cannot be applied directly to the Markov processes that are not characterized by solutions of classical martingale problems or stochastic differential equations which arise naturally in many applications. The authors of this article investigate the nonlinear filtering problem for general right continuous Markov processes, which are assumed to be associated with a Dirichlet form, or more generally, a semi-Dirichlet form. Using the theory of Dirichlet forms provides a powerful tool for analysis of Markov processes associated with singular generators taking values in infinite-dimensional state spaces. (For examples of Dirichlet forms and their one-to-one correspondence with Markov processes see the monographs by [\textit{Z.-M. Ma} and \textit{M. Röckner}, Universitext. Berlin: Springer-Verlag. (1992; Zbl 0826.31001)] and by [\textit{M. Fukushima, Y. Oshima} and \textit{M. Takeda.} de Gruyter Studies in Mathematics. 19. Berlin: Walter de Gruyter. (1994; Zbl 0838.31001)].) The authors derive the filtering equations in the semi-Dirichlet form setting. The existence and uniqueness of solutions of the filtering equations is studied in the case where the signal processes are not characterized by solutions of the classical martingale problems or stochastic differential equations. The uniqueness of solutions of the filtering equations is proved for very general observation functions \(h(\cdot)\) via the Wiener chaos expansions. The absolute continuity of the filtering processes with respect to the reference measures is investigated and the density equations for the filtering processes are derived. The results obtained in this paper can be regarded as the starting point for many other important problems for filtering of singular signal processes. These problems include particle filters, Markov approximations to solutions of the conditional density equations, exponential stability of the nonlinear filters, etc.