Quenched large deviations for one dimensional nonlinear filtering
Consider the standard, one dimensional, nonlinear filtering problem for a diffusion processe Ξt observed in small additive white noise. Denote by q1ε(⋅) the density of the law of Ξ1 conditioned on σ(Ytε:0≤t≤1). We provide "quenched" large deviation estimates for the random family of measures q1ε(x)dx: there exists a continuous, explicit mapping Jˉ:R2→R such that for almost all B⋅,V⋅, Jˉ(⋅,X1) is a good rate function and for any measurable G⊂R, −infx∈GoJˉ(x,X1)≤liminfεlog∫Gq1ε(x)dx≤limsupεlog∫Gq1ε(x)dx≤−infx∈GˉJˉ(x,X1).