Real time solution of nonlinear filtering problem without memory. I (Q5929470)

The authors consider a nonlinear filtering problem for the estimation of a stochastic process \(x={x_t}\) (called the state process) according to observations of a process \(y={y_t}\). Let \(\rho(t,x)\) denote the conditional probability density of the state process according to the observations \({y(s): 0 \leq s\leq t}\). Then \(\rho(t,x)\) is given by normalizing a function, \(\sigma(t,x)\), which satisfies the Duncan-Mortensen-Zakai (DMZ) parabolic partial differential equation. The authors describe their algorithm that permits to solve a robust DMZ equation in real time and in a memoryless manner for any filtering system with arbitrary initial distribution. They solve the DMZ equation with ``freezed'' observation and demonstrate that the solution to the robust ``freezed'' equation approximates very well the solution to the original robust equation both pointwise and in an \(L^2\)-sense. Unlike the classical Kalman-Bucy case, which requires the real time solution of a system of ordinary differential equations, the authors' solution to the general nonlinear filtering problem does not need any on-line computation and requires only the knowledge of the observation at time \(t\).