Discrete time Galerkin approximations to the nonlinear filtering solution (Q1075039)

The paper concerns discrete time approximation to the solution of the filtering problem for diffusion. The signal is a Markov diffusion process and the observation is supposed to evolve according to \(''signal+white\) noise'' model. Conditions are specified for which the dynamic part of the filter can be represented by means of the Zakai formula for the unnormalized conditional density of the diffusion. Two families of schemes approximating the unnormalized conditional density, respectively in an ''average'' and in a ''pathwise'' sense are presented. The introduced Galerkin schemes have a finite element discretization in the space domain and an implicit Runge-Kutta form in the time domain discretization. In order to achieve better rate of convergence the second order powers of the observation process are incomporated. Convergence theorems are proved and \(L^ 2\) error estimates of the discrete time Galerkin methods are presented. It is shown that discrete time numerical approximations can be produced in such a way that 1) they converge in an ''average'' sense with linear rate in the time increment to the solution of the Zakai equation, 2) they converge in a ''pathwise'' sense with a linear rate in the modulus of continuity of the sample path to the solution of the ''pathwise'' version of the Zakai equation. The rate of convergence of the proposed schemes to the equations governing the filter for a diffusion absorbed by the boundaries of a given domain is under the hypotheses made declared to be optimal.