Fluctuations and Long-Time Stability of Multivariate Ensemble Kalman Filters

We develop a self contained stochastic perturbation theory for discrete generation and multivariate Ensemble Kalman filters. Unlike their continuous-time counterparts, discrete EnKF algorithms are defined through a two steps prediction update mechanism and exhibit non Gaussian fluctuations, even in linear settings. In the multivariate case, these fluctuations take the form of non central Wishart type perturbations, which significantly complicate the mathematical analysis. We establish non asymptotic, time-uniform stability and error estimates for the ensemble covariance matrix processes under minimal structural assumptions on the signal observation model, allowing for possibly unstable dynamics. Our results quantify the impact of ensemble size, dimension, and observation noise, and provide explicit bounds on the propagation of stochastic errors over long time horizons. The analysis relies on a detailed study of stochastic Riccati difference equations driven by matrix-valued noncentral Wishart fluctuations. Beyond their relevance to data assimilation, these results contribute to the probabilistic understanding of ensemble-based filtering methods in high dimension and offer new tools for the analysis of interacting particle systems with matrix-valued dynamics.

💡 Research Summary

This paper develops a comprehensive stochastic perturbation framework for discrete‑time, multivariate Ensemble Kalman Filters (EnKFs). Unlike continuous‑time formulations, discrete EnKFs operate via a two‑step prediction‑update cycle, which destroys the Gaussian structure of the ensemble statistics even for linear dynamics. In the multivariate setting the resulting fluctuations of the sample covariance matrix are shown to be non‑central Wishart‑type random matrices, a fact that dramatically complicates the analysis.

The authors first derive the exact distributional properties of these non‑central Wishart perturbations. By writing the covariance recursion as a stochastic Riccati difference equation
(P_{n+1}= \Phi(P_n)+\Delta_n),
they identify (\Phi) as the deterministic Riccati map and (\Delta_n) as a matrix‑valued martingale increment whose mean and variance are computed explicitly in terms of the ensemble size (N), the state dimension (d), and the model/observation covariances. This yields a precise second‑order description of the sampling noise that goes beyond the usual central Wishart approximations.

A central contribution is the establishment of time‑uniform, non‑asymptotic error bounds for the ensemble covariance. Using matrix martingale inequalities, the Ando‑Hemm inequality, and a refined Fréchet‑calculus for matrix‑valued maps, the paper proves that for any integer (r\ge1)
\