Quantitative approximation of the Vlasov(-Fokker-Planck)-Navier-Stokes system by stochastic particle systems

Abstract: This paper is concerned with a fluid-particle system given by the incompressible Navier-Stokes equations coupled with the Vlasov(-Fokker-Planck) equation through a drag force. Such a model arises naturally in the study of aerosols, sprays, and more generally two-phase flows. In dimensions $d\in {2,3}$, we establish a rate of convergence for a system of $N$ interacting stochastic particles coupled with a fluid, towards the Vlasov(-Fokker-Planck)-Navier-Stokes system, as $N\to \infty$. The case of particles with a noise that vanishes as $N\to \infty$ is considered and leads specifically to the Vlasov-Navier-Stokes system. More precisely, we prove that the empirical measure associated with the particle system converges to the Vlasov(-Fokker-Planck) component, while the fluid velocity converges to the Navier-Stokes component of the coupled system. The proofs combine stochastic calculus and PDE techniques to establish energy estimates and commutator estimates for both the discrete and continuous systems.

• The paper establishes explicit convergence rates linking stochastic particle systems with the coupled Vlasov(-Fokker-Planck)-Navier-Stokes equations.
• It employs advanced stochastic calculus and Sobolev space techniques to derive sharp error bounds in both diffusive (σ > 0) and non-diffusive (σ = 0) regimes.
• The results validate propagation of chaos and provide practical guidelines for particle-based numerical simulations in multiphase flow analysis.

Quantitative Mean-Field Convergence for Coupled Vlasov(-Fokker-Planck)-Navier-Stokes Systems

---

Introduction and Context

The coupling of kinetic and fluid equations is a central topic in the analysis of multiphase flows, such as aerosols, sprays, and suspensions. The system composed of the Vlasov(-Fokker-Planck) equation for the particle phase and the incompressible Navier-Stokes equation for the fluid phase—interacting via a Stokes drag term—has become a canonical model in this context.

This paper (2604.18871) investigates the quantitative convergence of an interacting stochastic particle system toward the coupled Vlasov(-Fokker-Planck)-Navier-Stokes (VFP-NS) system in dimensions two and three. The stochastic particle system is designed so that its empirical measure converges, in the mean-field limit, to the solution of the kinetic-fluid PDE system. The analysis yields explicit rates, with careful attention paid to both the regular regime (with diffusion, $\sigma > 0$) and the non-diffusive regime ($\sigma = 0$).

---

Main Problem and Microscopic-Macroscopic Modeling

The system of interest is the VFP-NS system: $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$ where $u$ is the fluid velocity, $F$ is the phase-space density of particles, and the particles and fluid are coupled by the Stokes drag term.

To relate this macroscopic PDE system to a microscopic description, the authors introduce a stochastic particle system: each particle evolves according to its own velocity, is subject to drag from the fluid, and possibly experiences diffusion (Brownian noise). The fluid, in turn, is forced at discrete positions by the Stokes drag from the ensemble of particles, represented via mollified Dirac masses. The empirical measure of the particle system—mollified appropriately in position and velocity—is the microscopic analogue of the kinetic density field.

The goal is to establish explicit rates of convergence for the empirical measure of the particle system and the associated fluid velocity toward solutions of the VFP-NS system as the number of particles $N\to\infty$, considering possibly vanishing noise (the Vlasov-Navier-Stokes limit).

---

Mathematical Contributions

1. Mean-Field Convergence in Weighted Norms

The authors provide a rigorous derivation of the VFP-NS system from the particle system by proving that, as $N\to\infty$:

• The empirical measure $F^N$ converges to $F$ in a weighted $L^2$-type norm,
• The fluid velocity $\sigma = 0$0 converges to $\sigma = 0$1 in Bessel (fractional Sobolev) spaces $\sigma = 0$2.

The main theorem establishes that, under suitable initial data and technical assumptions, for any $\sigma = 0$3: $\sigma = 0$4 where $\sigma = 0$5 is an explicit rate term depending algebraically on $\sigma = 0$6 (if $\sigma = 0$7) and logarithmically on $\sigma = 0$8 (if $\sigma = 0$9).

Notably, for $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$0, the rate of convergence is algebraic in $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$1, with dominant terms due to:

• The smoothing mismatch/incoherence from mollification,
• Discretization error in the empirical measure,
• Martingale fluctuations due to noise.

For $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$2, a logarithmic rate is obtained, reflecting the inherent challenges of controlling stochastic particle systems with degenerate diffusion.

2. Propagation of Chaos and Pathwise Quantitative Bounds

The analysis yields a quantitative version of propagation of chaos: as $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$3, the trajectories of individual particles become statistically independent and follow the mean-field flow generated by $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$4. Moreover, the paper provides explicit pathwise bounds: the deviation between the finite-particle trajectory and its mean-field analogue is controlled by the sum of error in the fluid field and any difference in the noise scaling.

3. Auxiliary PDE Analysis and PDE-ODE Comparison

A key technical ingredient is the construction and a priori analysis of an auxiliary smoothed PDE system with cut-offs and regularized noise. This enables precise estimates comparing the PDE fields and empirical measures, and helps manage terms that are delicate in the absence of uniform $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$5 bounds.

Further, the paper provides novel regularity results for weak solutions to coupled systems under fractional Sobolev assumptions on the initial fluid velocity, carefully exploiting the inherent regularization of the heat semigroup and mollifier structures.

4. Strong Numerical and Theoretical Bounds

Quantitative error bounds are provided for both Bessel and energy norms. For example, in the regime $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$6, the error in the fluid field (in $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$7) and the kinetic density (in weighted $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$8) is controlled by a combination of the initial data error and explicit negative powers of $$\begin{cases} \partial_t u - \Delta u + (u \cdot \nabla)u + \nabla p + \int (u-v) F\, dv = 0,\ \nabla \cdot u = 0, \ \partial_t F + v \cdot \nabla_x F + \nabla_v \cdot ((u-v) F) = \frac{\sigma^2}{2}\Delta_v F, \end{cases}$$9, up to constants depending on model parameters.

---

Implications and Theoretical Impact

The results are a substantial strengthening of previous works which, though they established qualitative mean-field limits for kinetic-fluid systems, did not provide quantitative convergence rates or treat such a general regime including both the Fokker-Planck case ($u$0) and the (deterministic) Vlasov case ($u$1), as well as both two and three spatial dimensions.

The uniform quantitative control has several implications:

• Direct guidance for particle-based numerical simulations: One can use the explicit bounds to estimate the required particle number $u$2 for a target accuracy in representing kinetic-fluid coupled systems, depending on the noise level.
• The analytical framework can be transferred to other kinetic-fluid models, including those with more singular interactions or different forms of drag and coupling.
• Propagation of chaos with explicit rates provides a theoretical underpinning for stochastic Lagrangian methods in computational fluid mechanics.

Moreover, the distinction between algebraic and logarithmic convergence rates as a function of the noise—a delicate probabilistic vs. deterministic dichotomy in the well-posedness and regularization of the mean-field limit—sharpens understanding of the interplay between stochastic regularization and mean-field behavior.

---

Discussion of Methods and Generalizations

The analysis combines:

• Stochastic calculus for the interacting particle system, tracking Itô corrections, martingale terms, and regularized empirical measures,
• Sharp commutator and energy estimates for both the discrete and continuous systems,
• Sobolev/Bessel space interpolation and embedding techniques to precisely manage regularity transmission from the fluid to the particle field,
• Gronwall-type convolution inequalities to propagate and close the error bounds.

The cut-off and mollification procedures, used to handle both mathematical and physical singularities (such as the Dirac interaction between particles and the fluid), are carefully tracked at each step, ensuring robustness of the rate results.

Potential generalizations include extending the methods to models with more complex physical interactions (e.g., volume-fraction effects in thick sprays), and kinetic-fluid systems with more intricate drift, boundary, or collision mechanisms.

---

Conclusion

This paper presents a comprehensive quantitative theory for the approximation of the coupled Vlasov(-Fokker-Planck)-Navier-Stokes system via stochastic particle systems. Explicit and rigorous convergence rates are established for both kinetic and fluid components, in both strong and weak diffusion regimes, and in both two and three dimensions. The technical core blends stochastic analysis and PDE theory, furnishing robust tools and results applicable to future investigations of mean-field and multiphase models in kinetic theory and fluid dynamics (2604.18871).