Asymptotic expansion in nonlinear filtering with homogenization (Q1568194)

The author considers the nonlinear filtering problem with the \(n\)-dimensional signal process \(x^\varepsilon(t)\) and \(d\)-dimensional observation \(z^\varepsilon(t)\): \[ dx^\varepsilon(t)= g\Biggl({x^\varepsilon(t)\over \varepsilon}\Biggr) x^\varepsilon(t) dt+ \sigma\Biggl({x^\varepsilon(t)\over \varepsilon}\Biggr) dw(t),\quad x^\varepsilon(0)= \xi, \] \[ dx^\varepsilon(t)= h\Biggl({x^\varepsilon(t)\over \varepsilon}\Biggr) x^\varepsilon(t) dt+ dv(t),\quad z^\varepsilon(0)= 0. \] Here \(w\) and \(v\) are independent standard Wiener processes which are independent of \(\xi\), \(\xi\) is a Gaussian variable, \(g\), \(\sigma\), and \(h\) are periodic with period 1 in all directions. The author obtains an asymptotic expansion for the conditional density of the estimate of \(x^\varepsilon(t)\), given the \(\sigma\)-algebra of observations \(Z^\varepsilon_t= \sigma(z^\varepsilon(s), 0\leq s\leq t)\).