Data assimilation for the barotropic Navier-Stokes system

1. Problem Statement and Context

The paper addresses the mathematical foundation of continuous data assimilation for the compressible (barotropic) Navier–Stokes system. The goal is to recover an "observed" physical solution (representing the true state of a fluid) using limited observation data collected over a specific time interval (the assimilation period) and to ensure that a "synchronized" solution (generated by a numerical model) converges to this observed solution.

Key Challenges:

• 3D Compressibility: Unlike previous works focusing on 2D incompressible flows, this study tackles the 3D compressible case, which lacks global well-posedness results for weak solutions.
• Solution Regularity: The observed solution is assumed to be a strong (regular) solution, while the synchronized solution (often obtained via numerical simulation) is treated as a dissipative solution. Dissipative solutions are a generalized class of weak solutions that allow for oscillations and concentrations (Reynolds stress), making them suitable for analyzing numerical approximations.
• Prediction vs. Assimilation: The method must ensure that the solutions synchronize during the data assimilation period $[0, T]$ so that they remain close during the subsequent prediction (forecast) period $(T, T^+)$.

2. Mathematical Formulation

The Observed System

The observed solution $(r, U)$ satisfies the standard barotropic Navier–Stokes equations:
$$\begin{aligned} \partial_t r + \text{div}_x(rU) &= 0 \\ \partial_t(rU) + \text{div}_x(rU \otimes U) + \nabla_x p(r) &= \text{div}_x S(D_x U) + rg \end{aligned}$$
Assumptions on the Observed Solution:

1. Boundedness: The density $r$ and velocity $U$ are uniformly bounded over the entire observation period $(T^-, T^+)$.
2. Regularity: The solution is smooth (regular) at the initial time $T^-$, which, combined with boundedness, implies it is a strong solution throughout the interval.

The Synchronized System

The synchronized system $(\varrho, u)$ is the same Navier–Stokes system perturbed by nudging terms (feedback control) active only during the assimilation period $(0, T)$:
$$\begin{aligned} \partial_t \varrho + \text{div}_x(\varrho u) &= -\Lambda_\varrho (\varrho - I_\delta[r]) \mathbb{1}_{(0,T)} \\ \partial_t(\varrho u) + \text{div}_x(\varrho u \otimes u) + \nabla_x p(\varrho) &= \text{div}_x S(D_x u) + \varrho g - \Lambda_u (1+\varrho)(u - I_\delta[U]) \mathbb{1}_{(0,T)} \end{aligned}$$

• Interpolation Operators ($I_\delta$): These operators map the continuous observed data to the grid points of the synchronized system based on discrete observations at specific space-time points.
• Nudging Parameters: $\Lambda_\varrho$ and $\Lambda_u$ are positive relaxation parameters that force the synchronized solution toward the observed data.
• Dissipative Solutions: The synchronized solution is defined in the sense of dissipative solutions (introduced by Feireisl et al.), which include a Reynolds stress defect measure $R$ to account for potential lack of strong convergence in numerical approximations.

3. Methodology

The core of the proof relies on the Relative Energy Method (also known as the relative entropy method).

1. Relative Energy Functional:
   The authors define a generalized distance (Bregmann divergence) between the synchronized solution $(\varrho, u)$ and the observed solution $(r, U)$:
   $$\mathcal{E}(\varrho, u | r, U) = \int_\Omega \left( \frac{1}{2}\varrho |u - U|^2 + P(\varrho) - P'(r)(\varrho - r) - P(r) \right) dx$$
   where $P$ is the pressure potential.

2. Conditional Regularity:
   Using results from Sun, Wang, and Zhang, the authors establish that the boundedness and initial regularity of the observed solution imply that it belongs to high-order Sobolev spaces ($W^{1,q}$ for density, $W^{2,q}$ for velocity). This provides the necessary bounds on derivatives to control error terms.

3. Relative Energy Inequality:
   By testing the momentum and continuity equations of the synchronized system against the observed solution (and vice versa), the authors derive a differential inequality for the relative energy. This inequality includes:

   • Dissipation terms: Viscous dissipation and the nudging terms (which provide exponential decay).
   • Error terms: Terms involving the interpolation error ($I_\delta[r] - r$) and the Reynolds stress defect.

4. Gronwall-type Estimates:
   The differential inequality is manipulated to show that the relative energy decays exponentially during the assimilation period, provided the nudging parameters are chosen correctly.

4. Key Contributions and Results

Main Theorem (Theorem 2.2)

The paper proves that complete synchronization occurs on any compact prediction interval $(T, T^+)$ if the nudging parameters and observation density are chosen appropriately. Specifically, the relative energy at time $T^+$ can be made arbitrarily small ($\varepsilon$) if:

1. Nudging Strength: $\Lambda_u$ is sufficiently large relative to $\Lambda_\varrho$ (specifically $\Lambda_u \geq \Gamma(\|data\|)\Lambda_\varrho$).
2. Observation Density: The mesh size $\delta$ of the observation points is small enough relative to the nudging strength ($\delta \Lambda_u \Gamma(\|data\|) \leq 1$).
3. Assimilation Time: The assimilation period $T$ is long enough to allow the exponential decay term $e^{-\Lambda_\varrho T}$ to become negligible.

Technical Innovations

• Handling Dissipative Solutions: The work extends data assimilation theory to the class of dissipative solutions, which is crucial for validating numerical schemes where strong convergence is not guaranteed.
• Interpolation on Observed Data Only: The nudging terms act on the difference between the synchronized variable and the interpolated observed variable (e.g., $u - I_\delta[U]$), rather than interpolating the synchronized variable itself. This avoids regularity issues with the numerical solution.
• Pressure Constraints: The result holds for the adiabatic exponent $\gamma > 6/5$. The authors note that if the observed solution is more regular, this can be relaxed to $\gamma > 1$.

5. Significance

• Theoretical Foundation for 3D Compressible Flow: This is a significant step forward in applying data assimilation to realistic 3D compressible fluid models, which are standard in meteorology and aerodynamics but mathematically difficult due to the lack of global regularity for weak solutions.
• Robustness of Numerical Methods: By utilizing dissipative solutions, the results validate that numerical approximations (which naturally converge to dissipative solutions) can be successfully synchronized with real-world data, even if the numerical solution does not converge strongly to a classical weak solution.
• Practical Applicability: The conditions derived (bounds on nudging parameters and observation density) provide concrete guidelines for implementing data assimilation algorithms in weather forecasting and fluid dynamics simulations. The paper demonstrates that reliable forecasts can be generated from short assimilation windows if the control parameters are tuned correctly.

In summary, Feireisl provides a rigorous mathematical proof that a numerical model of a compressible fluid, when driven by observational data via nudging, will converge to the true physical state, provided the observations are sufficiently dense and the feedback control is strong enough. This bridges the gap between the theoretical limitations of 3D Navier-Stokes well-posedness and the practical success of data assimilation in meteorology.