Existence and Uniqueness For Variational Data Assimilation in Continuous Time

TL;DR

This paper investigates the existence and uniqueness of solutions in a nonstandard variational data assimilation problem involving stochastic integrals, extending optimal control theory to irregular, noise-affected dynamical estimation scenarios.

Contribution

It introduces a framework for pathwise existence and maximum principles in data assimilation with irregular performance indices, broadening the scope of optimal control methods.

Findings

Established pathwise existence of minimisers.

Derived a maximum principle for the problem.

Extended results on the maximum a posteriori estimator.

Abstract

A variant of the optimal control problem is considered which is nonstandard in that the performance index contains "stochastic" integrals, that is, integrals against very irregular functions. The motivation for considering such performance indices comes from dynamical estimation problems where observed time series need to be "fitted" with trajectories of dynamical models. The observations may be contaminated with white noise, which gives rise to the nonstandard performance indices. Problems of this kind appear in engineering, physics, and the geosciences where this is referred to as data assimilation. Pathwise existence of minimisers is obtained, along with a maximum principle as well as preliminary results in dynamic programming. The results extend previous results on the maximum aposteriori estimator of trajectories of diffusion processes. To obtain these results, classical concepts from optimal control need to be substantially modified due to the nonstandard nature of the performance index, as well the fact that typical models in the geosciences do not satisfy linear growth nor monotonicity conditions.