A continuous data assimilation method for a variant of Oberbeck-Boussinesq system with randomly perturbed data

Technical Summary: Continuous Data Assimilation for a Stochastic Oberbeck–Boussinesq System

Problem Formulation
This paper addresses the problem of continuous data assimilation for a variant of the Oberbeck–Boussinesq (OB) system, which models the time evolution of velocity $u$ and temperature fluctuations $\theta$ in a bounded domain $Q \subset \mathbb{R}^d$ ($d=2,3$). The system is derived as a singular limit of a mildly stratified Navier–Stokes–Fourier system. The authors consider a scenario where the "observed" solution (the reference state) is subject to random perturbations in the data, necessitating a stochastic framework.

The primary mathematical challenge is to synchronize a "synchronized" solution (a prediction model) with the observed solution using only approximate measurements (interpolated data) over a sampling period $(0, T)$. The goal is to recover the exact state of the observed system during a subsequent prediction period $(T, T^+)$, even when the well-posedness of the observed system in three dimensions ($d=3$) remains an open problem. The authors work under the hypothesis that the reference solution is bounded in $L^\infty$, a condition satisfied by global weak solutions in $d=2$ and assumed for physically relevant cases in $d=3$.

Methodology
The methodology relies on the relative energy inequality technique adapted for stochastic partial differential equations (PDEs). The approach proceeds through the following steps:

1. Reformulation: The original OB system with non-local boundary conditions is reformulated into a Modified Oberbeck–Boussinesq (MOB) system with homogeneous Dirichlet boundary conditions. This involves introducing a new temperature variable $\Theta$ and a bounded, self-adjoint operator $M$.
2. Stochastic Synchronized System: A synchronized system is constructed by adding "nudging" terms to the momentum and heat equations. These terms force the synchronized solution $(\tilde{u}, \tilde{\Theta})$ toward the observed data. Crucially, the measurements are modeled as containing both deterministic errors and random errors represented by cylindrical Wiener processes ($W_u, W_\Theta$).
3. Weak Martingale Solutions: Since strong solutions for the synchronized system in $d=3$ are not guaranteed, the authors define and utilize the concept of weak martingale solutions. This allows for the existence of solutions in a probabilistic sense without requiring the high regularity needed for pathwise uniqueness.
4. Relative Energy Analysis: The core analytical tool is the derivation of a relative energy inequality. By treating the smooth, bounded observed solution as a test function for the weak martingale solution of the synchronized system, the authors derive an inequality governing the evolution of the distance (in $L^2$ norm) between the two solutions.
5. Interpolation and Nudging: The analysis incorporates interpolation operators $I_\delta$ to model the discrete nature of measurements. The authors establish that for sufficiently large nudging parameters ($\Lambda$) and sufficiently fine interpolation ($\delta$), the error terms can be controlled.

Key Contributions and Results
The main result, stated in Theorem 4.1, establishes the convergence of the continuous data assimilation method. Specifically, the paper proves that the expected value of the squared $L^2$-distance between the observed solution $(u, \Theta)$ and the synchronized solution $(\tilde{u}, \tilde{\Theta})$ decays exponentially over time, modulo the magnitude of the observation errors.

The error bound is given by:
$$\mathbb{E}[\|u - \tilde{u}\|^2 + \|\Theta - \tilde{\Theta}\|^2] \lesssim e^{-\gamma T} (\text{initial error}) + \Lambda^2 (\text{deterministic error}) + \Lambda^2 (\text{stochastic noise variance})$$

Key technical findings include:

• Dimensional Independence: The method is shown to work for both $d=2$ and $d=3$. This is significant because the well-posedness of the 3D OB system is an outstanding open problem; the data assimilation method provides a way to synchronize solutions even when the uniqueness of the reference solution is not fully established.
• Handling Stochasticity: The paper successfully extends the relative energy method to stochastic settings, accounting for cylindrical Wiener processes in the forcing terms.
• Error Dependence: The convergence rate depends on the magnitude of the nudging parameter $\Lambda$ and the interpolation resolution $\delta$. While increasing $\Lambda$ accelerates convergence, it also amplifies the impact of measurement errors (both deterministic and stochastic), requiring a balance in parameter selection.

Significance and Claims
The paper claims that its primary significance lies in providing a rigorous mathematical framework for data assimilation in fluid dynamics models where the underlying system may not be fully well-posed (specifically in 3D). By relying on the boundedness of the reference solution rather than its uniqueness or full regularity, the method is applicable to physically relevant scenarios in meteorology where such boundedness is a reasonable hypothesis.

The authors emphasize that the work is motivated by the seminal data assimilation methods of Azouani, Olson, and Titi, extending them to stochastic Oberbeck–Boussinesq systems. The results suggest that continuous data assimilation can effectively recover the state of a complex fluid system from sparse, noisy, and randomly perturbed data, provided the system's energy remains bounded. The paper does not propose new experimental setups or future applications beyond the mathematical convergence proof but positions the result as a theoretical advancement for stochastic PDEs in fluid mechanics.