1: \begin{abstract}

2: Filter stability is a classical and extensively studied problem for partially observed Markov processes. For such a process, an incorrectly initialized non-linear filter is said to be stable if the filter eventually corrects itself with the arrival of new measurement information. This paper has three main contributions: (i) we introduce a general explicit definition of observability for stochastic non-linear dynamical systems which involves a functional characterization that is testable with little effort. Under this observability definition we establish stability of one-step predictors under weak convergence. (ii) We establish new implications between various criteria for filter stability, notably in arriving from weak convergence of predictors to that of filters through weak Feller continuity of filter transitions, from weak merging to total variation merging for both the predictor and filters, and finally in establishing the equivalence of total variation merging and relative entropy merging. (iii) We then strengthen the stability results to involve weak merging and total variation merging, both in expectation and in an almost sure sense, as well as in relative entropy (which also has implications in the context of relative entropy convergence of Markov chains to invariant measures). Our conditions, and the examples we study, complement and generalize the existing results on filter stability and a detailed comparison is reported. Numerous examples are presented. 

3: \end{abstract}

4: