No label defined (Q3973619)

[For the entire collection see Zbl 0724.00018.] The author considers the following one-dimensional filtering problem: \[ dx_ t = f(x_ t)dt + dw_ t,\qquad dy_ t = g(x_ t)dt + dv_ t,\tag{S} \] where \(w\) and \(v\) are independent one-dimensional Brownian motions. Under suitable assumptions on \(f,\;g\text{ and }p_ 0\) (the density of \(x_ 0)\), it is known that there exists a maximum a priori estimator (M.A.P.) defined as \(\hat \phi = \arg \max J(\phi)\), where \(J\) is a functional intrinsically associated to the system (S). This M.A.P. is related to non-linear stochastic partial differential equation. In this paper, finite dimensionality results for the M.A.P. are studied. The author uses a local version of Ocone's technique in order to prove that the only finite-dimensional realizations of the M.A.P. equations are those for which the filtering problem possesses a finite-dimensional solution.