On the diffusive-mean field limit of a kinetic weakly interacting particle system

Technical Summary: On the Diffusive-Mean Field Limit of a Kinetic Weakly Interacting Particle System

Problem Statement
This paper investigates the joint diffusive-mean field limit for a system of $N$ weakly interacting particles governed by kinetic Langevin dynamics. The system is defined on a $d$-dimensional torus $\mathbb{T}^d$ with positions $q_i$ and momenta $p_i$, subject to a confining potential $V$, an interaction potential $W$, friction $\gamma$, and thermal noise at inverse temperature $\beta$. The study focuses on the behavior of the system as the number of particles $N \to \infty$ (mean-field limit) and the time-scale parameter $\varepsilon \to 0$ (diffusive limit).

A central challenge addressed is the potential non-commutativity of these two limits. While previous work [20] established that for overdamped (first-order) dynamics, the limits commute away from phase transitions but may not commute at low temperatures due to the existence of multiple stationary states, this paper extends the analysis to the kinetic (second-order, hypoelliptic) setting. The authors specifically examine whether the order of limits ($N \to \infty$ then $\varepsilon \to 0$, versus $\varepsilon \to 0$ then $N \to \infty$) affects the covariance matrix of the limiting Brownian motion when phase transitions occur.

Methodology
The authors employ a combination of modern hypocoercivity techniques, linearization arguments, and homogenization theory:

1. Hypocoercivity and Convergence to Equilibrium: To analyze the long-time behavior of the nonlinear McKean-Vlasov-Fokker-Planck equation, the authors utilize recently developed $L^2$-hypocoercivity techniques [23, 24, 74]. They treat the generator of the nonlinear process as a perturbation of a linearized operator around a stationary state $\mu_\infty$. By constructing a modified weighted $L^2$ norm and utilizing Poincaré inequalities for the position marginal, they prove exponential convergence to a local minimizer of the free energy, provided the initial distribution is sufficiently close to this minimizer.
2. Linearization and Invariance Principles: The diffusive limit is derived by viewing the nonlinear McKean SDE as a perturbation of a linear diffusion process with a fixed potential determined by the stationary measure. Using Itô's lemma and the Green-Kubo/Kipnis-Varadhan formula, the authors establish an invariance principle, showing that the rescaled particle positions converge to a Brownian motion. The covariance matrix is expressed via a Poisson equation associated with the linearized generator.
3. Analysis of the O(2) Model: To demonstrate the theoretical findings, the paper provides a detailed analysis of the kinetic O(2) model (noisy Kuramoto model) in a magnetic field. This involves characterizing the number and stability of stationary states at low temperatures and analyzing the bifurcation of the free energy landscape.
4. Numerical Validation: The theoretical predictions are validated through numerical simulations using the BAOAB integration scheme. The authors compute self-diffusion coefficients for both the $N$-particle system and the linearized mean-field dynamics starting from different stationary states.

Key Contributions and Results

• Extension to Kinetic Dynamics: The paper extends the results of [20] from overdamped to kinetic Langevin dynamics. It establishes that the generator of the kinetic system is hypoelliptic and hypocoercive, requiring specialized techniques for proving convergence to equilibrium.
• Commutativity of Limits:
  • High Temperatures (Unique Stationary State): If the free energy admits a unique global minimizer (typically at high temperatures or for H-stable potentials), the diffusive and mean-field limits commute. The limiting Brownian motion has a covariance matrix determined by the unique stationary state.
  • Low Temperatures (Phase Transitions): In the presence of phase transitions (multiple stationary states), the limits may not commute. The effective diffusion coefficient depends on the specific stationary state $\mu_\infty$ reached by the system.
  • Dependence on Initial Conditions: If the system starts near a local minimizer of the free energy that is not the global minimizer, the diffusive limit of the mean-field dynamics yields a covariance matrix associated with that local state. However, the joint limit of the particle system (starting from the same initial condition) converges to the diffusion coefficient associated with the global minimizer.
• Theoretical Characterization of the O(2) Model: The authors prove that for the O(2) model in a magnetic field, there exists a critical temperature $\beta_c$. Below this temperature, there are exactly three stationary states: one global minimizer (stable) and two unstable critical points (saddle points).
• Explicit Covariance Calculation: The paper derives the explicit form of the limiting covariance matrix using the solution to a Poisson equation involving the linearized generator around the specific stationary measure.

Significance and Claims
The paper claims to provide a rigorous framework for understanding the hydrodynamic behavior of weakly interacting kinetic systems in the presence of phase transitions. The primary significance lies in demonstrating that the transport coefficients (specifically the diffusion matrix) in kinetic systems can depend discontinuously on the microscopic data (initial conditions and temperature) when phase transitions are present.

The authors emphasize that their analysis relies on the systematic use of hypocoercivity to handle the non-self-adjoint nature of the kinetic generator and the nonlinearity of the mean-field interaction. They assert that their results generalize the overdamped findings of [20] to the hypoelliptic case, showing that the non-commutativity of limits is a robust feature of interacting particle systems exhibiting phase transitions, regardless of whether the dynamics are overdamped or underdamped. The numerical results for the O(2) model confirm that the particle system's long-time diffusive behavior is governed by the global free energy minimizer, even when initialized near an unstable or local stationary state, thereby illustrating the non-commutativity of the limits.