[Paper Review] Functional central limit theorem for Brownian particles in domains with Robin boundary condition

Zhen-Qing Chen, Wai-Tong Louis Fan|arXiv (Cornell University)|Apr 5, 2014

Advanced Mathematical Modeling in Engineering21 references3 citations

TL;DR

This paper establishes a functional central limit theorem for particle densities in bounded Lipschitz domains with Robin boundary conditions, using a Dirichlet form framework to analyze non-equilibrium fluctuations of reflected diffusions killed by a time-dependent potential proportional to boundary local time. The key result is the convergence of fluctuation fields to a generalized Gaussian process governed by a stochastic PDE, valid for any symmetric reflected diffusion, dimension $d \geq 1$, and bounded Lipschitz domain.

ABSTRACT

We rigorously derive non-equilibrium space-time fluctuation for the particle density of a system of reflected diffusions in bounded Lipschitz domains in $\mathbb R^d$. The particles are independent and are killed by a time-dependent potential which is asymptotically proportional to the boundary local time. We generalize the functional analytic framework introduced by Kotelenez [19, 20] to deal with time-dependent perturbations. Our proof relies on Dirichlet form method rather than the machineries derived from Kotelenez's sub-martingale inequality. Our result holds for any symmetric reflected diffusion, for any bounded Lipschitz domain and for any dimension $d\geq 1$.

Motivation & Objective

• To develop a rigorous framework for analyzing non-equilibrium space-time fluctuations in systems of reflected diffusions in bounded Lipschitz domains.
• To extend the functional analytic machinery of Kotelenez to handle time-dependent perturbations in fluctuation theory.
• To establish the convergence of fluctuation fields to a Gaussian process in the hydrodynamic limit, even under singular, boundary-localized killing potentials.
• To generalize existing results on particle density fluctuations to include Robin-type boundary conditions and time-dependent killing mechanisms.
• To provide a Dirichlet form-based proof that avoids reliance on Kotelenez’s submartingale inequality, enhancing generality and robustness.

Proposed method

• Formulate the particle system as i.i.d. reflected diffusions in a bounded Lipschitz domain $D \subset \mathbb{R}^d$, killed by a time-dependent potential $q_N(t,x) = \delta_N^{-1} \mathbf{1}_{D^{\delta_N}}(x) q(t,x)$, which concentrates near the boundary.
• Use the Dirichlet form method to analyze the generator and semigroup of the underlying diffusion process, leveraging the bilinear form $\mathcal{E}(f,g) = \frac{1}{2} \int_D \mathbf{a}(x) \nabla f \cdot \nabla g \, \rho(x) \, dx$.
• Define the fluctuation field $\mathcal{Y}^N_t(\phi) = N^{-1/2} \sum_{i=1}^N \left( \mathbf{1}_{\{t < \zeta_i^{(N)}\}} \phi(X_i(t)) - \mathbb{E}[\phi(X_i(t))] \right)$, where $\zeta_i^{(N)}$ is the killing time.
• Establish tightness and convergence of the fluctuation field $\mathcal{Y}^N$ via martingale techniques and the characterization of the limiting process as a solution to a stochastic PDE.
• Employ Hilbert space $\mathcal{H}_{-\alpha}$ for $\alpha > d \vee (d/2 + 1)$ to handle the regularity and integrability of the limiting Gaussian process.
• Use the evolution semigroup $Q_{(s,t)}$ and the associated stochastic integral $\int_0^t U_{(t,\theta)} dM_\theta$ to represent the limiting fluctuation field as a centered Gaussian process with specified covariance structure.

Experimental results

Research questions

• RQ1 How do non-equilibrium space-time fluctuations of particle densities behave in bounded domains with Robin boundary conditions?
• RQ2 Can the functional central limit theorem be extended to systems with time-dependent, boundary-localized killing potentials?
• RQ3 What is the limiting Gaussian process that describes the fluctuation field in the hydrodynamic limit?
• RQ4 How can the Dirichlet form method be used to replace submartingale inequalities in fluctuation analysis?
• RQ5 To what extent do the results hold for general symmetric reflected diffusions and arbitrary dimensions $d \geq 1$?

Key findings

• The fluctuation field $\mathcal{Y}^N$ converges weakly in $C([0,T]; \mathcal{H}_{-\alpha})$ to a centered Gaussian process $\mathcal{Y}$ for any $\alpha > d \vee (d/2 + 1)$.
• The limiting process $\mathcal{Y}$ solves a stochastic PDE driven by a space-time white noise and is characterized by a specific covariance structure involving the evolution semigroup $Q_{(s,t)}$.
• The convergence holds for any bounded Lipschitz domain $D \subset \mathbb{R}^d$, any dimension $d \geq 1$, and any symmetric reflected diffusion with generator $\mathcal{A} = \frac{1}{2\rho} \nabla \cdot (\rho \mathbf{a} \nabla)$.
• The proof relies on the Dirichlet form method rather than Kotelenez’s submartingale inequality, enabling broader applicability.
• The stochastic integral $\int_0^t U_{(t,\theta)} dM_\theta$ is well-defined and square-integrable in $\mathcal{H}_{-\alpha}$, ensuring the existence of the limiting Gaussian process.
• The result remains valid when the killing potential is replaced by a local time integral, i.e., $2 \int_0^t q(s,X_i(s)) dL^i_s$, as shown via Lemma 5.6 and modifications to the martingale structure.