Smoothness properties of the conditional expectation in finitely additive white noise filtering (Q1110191)

The nonlinear filtering model \((y=\xi +e)\) is considered, just as it was thoroughly studied by \textit{G. Kallianpur} and \textit{R. L. Karandikar} [Ann. Probab. 13, 1033-1107 (1985; Zbl 0584.60055)]. In this model the observation (y), the signal (\(\xi)\) and the noise (e) are all vectors in some Hilbert space H and the distribution of the noise is given by the canonical cylinder Gaussian measure on H. The main result is that the nonlinear filter E(g\(| Q y)\) is an infinitely many times Fréchet differentiable function on Q y (Q is a finitely dimensional orthoprojector in H and g is a nonlinear functional of \(\xi)\). This result has its analogue in the usual (i.e. countably additive) filtering model, in which the filter is proved to be \(C^{\infty}\) in Malliavin's sense. The authors' results, however, are very transparent and clear and the proofs are simple and elegant.