On the mean field theory of Ensemble Kalman filters for SPDEs

TL;DR

This paper provides a rigorous mathematical analysis of continuous-time Ensemble Kalman Filters for infinite-dimensional SPDEs, establishing well-posedness, convergence rates, and connections to stochastic filtering methods.

Contribution

It introduces the well-posedness and convergence analysis of EnKBFs in infinite dimensions and links them to the Feedback Particle Filter for SPDEs.

Findings

Proved well-posedness of EnKBF and mean field limit.

Established quantitative convergence rates.

Connected EnKBF to Feedback Particle Filter in infinite dimensions.

Abstract

This paper is concerned with the mathematical analysis of continuous time Ensemble Kalman Filters (EnKBFs) and their mean field limit in an infinite dimensional setting. The signal is determined by a nonlinear Stochastic Partial Differential Equation (SPDE), which is posed in the standard variational setting. Assuming global one-sided Lipschitz conditions and finite dimensional observations we first prove the well posedness of both the EnKBF and its corresponding mean field version. We then investigate the quantitative convergence of the EnKBF towards its mean field limit, recovering the rates suggested by the law of large numbers for bounded observation functions. The main tool hereby are exponential moment estimates for the empirical covariance of the EnKBF, which may be of independent interest. In the appendix of the paper we investigate the connection of the mean field EnKBF and the Stochastic Filtering Problem. In particular we derive the Feedback Particle Filter for infinite dimensional signals and show that the mean field EnKBF can viewed as its constant gain approximation.