[Paper Review] One Brownian Stochastic Flow

A. A. Dorogovtsev|ArXiv.org|Nov 24, 2006

Stochastic processes and financial applicationsEconomics, Econometrics and Finance1 references21 citations

TL;DR

This paper studies the weak limit of measure-valued processes generated by a flow of interacting Brownian motions with smooth, compactly supported kernels. As the kernel width $\varepsilon \to 0^+$, the system converges to a stochastic flow where particles perform independent Brownian motion until they meet, after which they coalesce and move together. The key result is the weak convergence of the measure-valued processes to a unique limit process characterized by coalescing Brownian motions.

ABSTRACT

The weak limits of the measure-valued processes organized as a mass carried by the interacting Brownian particles are described. As a limiting flow the Arrattia flow is obtained.

Motivation & Objective

• To analyze the weak limit of measure-valued processes induced by a flow of interacting Brownian particles with smooth, localized interaction kernels.
• To resolve the apparent contradiction between the non-independence of particle paths and the expectation that they should behave independently in the limit as $\varepsilon \to 0^+$.
• To formalize the intuitive idea of coalescing Brownian motions by studying the mass flow rather than individual particles.
• To establish weak compactness and uniqueness of the limit process in the space of continuous measure-valued processes.

Proposed method

• Define a stochastic flow $x_\varepsilon(u,t)$ via an SDE driven by a Wiener sheet, with $\varphi_\varepsilon$ as a mollifier kernel.
• Construct the measure-valued process $\mu^\varepsilon_t = \mu_0 \circ x_\varepsilon(\cdot,t)^{-1}$, representing the mass distribution at time $t$.
• Use the Wasserstein-type distance $\gamma_n$ on spaces $\mathfrak{M}_n$ of probability measures with finite $n$th moment to metrize weak convergence.
• Apply a criterion for weak relative compactness of measure-valued processes (from [4]) based on tightness and uniform equicontinuity in test functions.
• Prove that the limit process has the property that particles perform independent Brownian motion until they meet, then coalesce.
• Establish uniqueness of the limit via convergence of finite-dimensional distributions and dominated convergence.

Experimental results

Research questions

• RQ1 What is the weak limit of the measure-valued process $\mu^\varepsilon_t$ as $\varepsilon \to 0^+$, when the particle interactions are smoothed by $\varphi_\varepsilon$?
• RQ2 How can the apparent contradiction between non-independent particle paths and the expectation of independent motion in the limit be resolved?
• RQ3 What is the structure of the limiting stochastic flow in terms of particle coalescence and motion?
• RQ4 Does the family $\{\mu^\varepsilon\}$ admit a unique weak limit under $\varepsilon \to 0^+$?
• RQ5 Can the limiting process be characterized by a pathwise property such as coalescing Brownian motion?

Key findings

• The family $\{\mu^\varepsilon\}$ is weakly compact in $C([0,1], \mathfrak{M}_n)$ for initial measures $\mu_0 \in \mathfrak{M}_n$ with $n > 2$, ensuring existence of limit points.
• The limit process is uniquely characterized by the property that particles perform independent Brownian motion until they meet, after which they move together.
• For any finite set of initial positions $u_1 < \cdots < u_n$, the joint law of the particle paths in the limit coincides with that of coalescing Brownian motions.
• The limit measure-valued process $\{\nu_t\}$ satisfies $\mathbb{E} \prod_{k=1}^d \int \varphi_k(u) \nu_{t_k}(du) = \int \cdots \int \lim_{n \to 0^+} \mathbb{E} \left[ \prod_{k=1}^d \varphi_k(x_{\varepsilon_n}(u_k, t_k)) \right] \mu_0(du_1) \cdots \mu_0(du_d)$, proving uniqueness of the limit.
• The limiting distribution is supported on paths that remain in the closure of the set $\mathcal{G} = \overline{\bigcup_{\delta>0} \mathcal{G}_\delta}$, where $\mathcal{G}_\delta$ consists of paths that do not collide before time 1.
• The limit process is a well-defined, unique stochastic flow of continuous processes with coalescing paths, confirming the heuristic picture of mass flow with coagulation.