🔢 mathematics

Mean-field limit of particle systems with absorption

This paper establishes the mean-field limit and propagation of chaos for a system of one-dimensional particles interacting via a bounded kernel and subject to absorption at a barrier, proving the existence of weak solutions for the particle system and the strong well-posedness of the limiting nonlinear Fokker-Planck equation.

Original authors: Gaoyue Guo, Maxime Latypov, Milica Tomasevic

Published 2026-04-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Gaoyue Guo, Maxime Latypov, Milica Tomasevic

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a crowded dance floor where thousands of dancers (particles) are moving around. In a normal dance, they might bump into each other or move in sync with the crowd. But in this specific mathematical story, there's a dangerous edge to the floor—a "cliff" at zero.

If a dancer steps off the edge, they don't just fall; they vanish from the party forever. They stop dancing, stop talking to others, and are gone. This is what mathematicians call an absorbing barrier.

The paper by Gaoyue Guo, Maxime Latypov, and Milica Tomašević is about understanding what happens when you have a huge number of these dancers, and their movements are influenced by two things:

The Crowd: They listen to the average mood of everyone still on the floor (Mean-Field interaction).
The Cliff: If they get too close to the edge, they might get pushed off or fall off.

Here is the breakdown of their discovery using simple analogies:

1. The Problem: The "Vanishing Act"

Usually, when mathematicians study huge groups of interacting things, they assume everyone stays in the game. But in real life—like in financial markets (where a company goes bankrupt) or biology (where a species dies out)—things disappear.

The authors are looking at a system where particles (dancers) interact, but once they hit zero (the cliff), they are removed. The tricky part is that the act of them disappearing changes the behavior of the survivors. If 50% of the dancers fall off, the remaining 50% might panic and move differently. This creates a "singularity"—a point where the math gets messy and breaks down.

2. The Goal: Predicting the Future of the Crowd

The authors wanted to answer two big questions:

Can we predict the behavior of the whole crowd as the number of dancers goes to infinity? (This is the "Mean-Field Limit").
Is the prediction unique? (If we run the simulation twice, do we get the same result, or is it chaotic?)

3. The Solution: Two New Tools

The authors developed two clever ways to solve this puzzle.

Tool A: The "Survival Probability" Map (The PDE Approach)

Imagine trying to track the density of dancers on the floor. Most of them are dancing in the middle, but some are huddled near the edge, about to fall.

The Innovation: Instead of just tracking the dancers, the authors tracked the probability of survival. They created a mathematical map (a Partial Differential Equation) that shows exactly how many dancers are likely to be left at any given time.
The Result: They proved that this map is smooth and well-behaved. It's like having a crystal ball that tells you exactly how the "survival rate" changes over time, even as people keep falling off the cliff. This allowed them to prove that the system has a unique, predictable solution.

Tool B: The "Partial Girsanov" Trick (The Probabilistic Approach)

To prove that the huge crowd of $N$ dancers actually behaves like the smooth map they predicted, they needed a way to compare the two.

The Problem: You can't easily turn a complex, interacting crowd into a simple group of independent dancers because the "cost" of doing the math gets too high as the crowd gets bigger.
The Innovation: They used a technique called Partial Girsanov Transforms. Think of this as a magic lens. Instead of trying to transform the entire crowd at once (which is too heavy), they transformed just one dancer at a time.
The Analogy: Imagine you want to see if a noisy party sounds like a quiet library. Instead of silencing everyone at once, you put headphones on one person, then another, then another. By doing this one by one, you can prove that, statistically, the party does sound like the library, without the math exploding. This allowed them to prove that the "chaos" of the individual dancers averages out into a predictable pattern.

4. The Big Picture: Why Does This Matter?

This isn't just about abstract math; it's about real-world systems where things can "die" or "fail."

Financial Risk: Imagine a bank's capital. If it hits zero, the bank fails (absorbs). If many banks fail, the remaining ones might become more cautious or risky. This paper helps model how a cascade of failures happens.
Neuroscience: Neurons fire and then reset. If they hit a certain threshold, they "reset." This model helps understand how large networks of neurons fire together.
Epidemics: If a person gets sick and is removed from the population, how does that change the spread of the disease?

Summary

The authors took a messy, dangerous system where participants can vanish at any moment and proved that:

It works: The system has a clear, unique solution.
It's predictable: As the group gets huge, the randomness smooths out into a clear pattern.
We have the tools: They invented new mathematical "lenses" (PDE analysis and partial transforms) to look at these disappearing acts without getting lost in the math.

In short, they figured out how to predict the fate of a crowd when the floor is falling out from under them.

1. Problem Statement

The paper investigates the mean-field limit of a system of $N$ interacting one-dimensional particles that are subject to absorption at a boundary (specifically, the origin).

The System: The particles $X^{N,i}_t$ evolve on $(0, \infty)$ according to a stochastic differential equation (SDE) with a bounded mean-field drift and additive noise. Once a particle hits the barrier at $0$, it is absorbed (removed from the active system) and ceases to interact.
$dX^{N,i}_t = \iota(X^{N,i}_t) \left( \frac{1}{N} \sum_{j=1}^N b(t, X^{N,i}_t, X^{N,j}_t) \iota(X^{N,j}_t) \right) dt + \iota(X^{N,i}_t) \sqrt{2} dW^i_t$
where $\iota(x) = \mathbb{1}_{\{x>0\}}$ and $b$ is a bounded measurable interaction kernel.
The Singularity: Unlike previous works focusing on singularities arising from strong attractive/repulsive forces (blow-up), the singularity here arises from hitting times. The system experiences a "macroscopic mass loss" as particles are absorbed, and the cumulative loss feeds back into the dynamics of the surviving particles.
The Goal: To establish the well-posedness of the particle system, prove the propagation of chaos (convergence of the empirical measure to a limiting law), and characterize the limit via a nonlinear Fokker-Planck equation.

2. Methodology

The authors employ a hybrid approach combining probabilistic techniques with partial differential equation (PDE) analysis.

A. Probabilistic Tools

Partial Girsanov Transforms: A full Girsanov transform from the interacting system to independent killed Brownian motions is not uniform in $N$ . The authors introduce Partial Girsanov transforms, modifying only a finite number of coordinates (e.g., $r$ particles) at a time. This allows them to control the Radon–Nikodym derivatives uniformly as $N \to \infty$ , treating individual particles as stopped Brownian motions under a change of measure.
Tightness and Martingale Problems: They prove the tightness of the sequence of empirical measures and identify the limit points as weak solutions to the McKean-Vlasov SDE using a martingale problem approach.
Regularization: To handle the discontinuity of the indicator function $\iota(\cdot)$ at the boundary, they introduce a smooth regularization $f_\eta$ (equal to 0 on $(-\infty, 0]$ and 1 on $[\eta, \infty)$ ) to derive quantitative estimates, later removing the regularization via boundary-layer analysis.

B. PDE Analysis

Nonlinear Fokker-Planck Equation: The law of the limiting process $\mu_t$ is shown to decompose into a sub-probability density $p(t, x)$ on $(0, \infty)$ and a Dirac mass at the origin representing absorbed particles:
$\mu_t(dx) = \left(1 - \int_0^\infty p(t, y) dy\right) \delta_0(dx) + p(t, x) dx$
The density $p$ satisfies a nonlinear Fokker-Planck equation with homogeneous Dirichlet boundary conditions at $x=0$ .
Fixed-Point Arguments:
- Existence: Proven using Schauder's fixed-point theorem on a product space of probability flows and survival probabilities.
- Uniqueness: Established via a contraction mapping argument on the space of sub-probability densities. This involves analyzing the parametric dependence of the PDE solutions on the coefficients and using parametrix expansions (Aronson-type estimates) to bound the differences between solutions.

3. Key Contributions and Results

A. Well-posedness of the Particle System

Theorem 2.3: Under the assumption that the drift $b$ is bounded and measurable, the $N$ -particle system admits a unique strong solution. The proof relies on "survival-interval gluing," constructing the solution piecewise between absorption events.

B. Propagation of Chaos and Convergence

Theorem 2.4: The sequence of empirical measures $\mu^N$ is tight. Any limit point is a weak solution to the McKean-Vlasov SDE. If the drift $b$ satisfies Hölder continuity conditions, the limit is unique, implying propagation of chaos (convergence in probability of the empirical measure to the law of the limit process).
Theorem 2.6 (Quantitative Rate): Under a Lipschitz condition on $b$ , the authors establish a strong propagation of chaos with an optimal convergence rate:
$\mathbb{E}\left[ \sup_{t \in [0,T]} W_2(\mu^N_t, \mu_t)^2 \right] \leq \frac{C}{N}$
where $W_2$ is the 2-Wasserstein distance.

C. Well-posedness of the Mean-Field Limit

Theorem 2.5: Under regularity assumptions (Hölder continuity of $b$ and its spatial derivative) and a condition on the initial density near zero ( $\int_0^1 \rho(x)/x^2 dx < \infty$ ), the limiting McKean-Vlasov SDE admits a unique weak solution.
Classical Solution: The authors prove that the sub-probability density $p(t, x)$ is a classical solution to the associated nonlinear Fokker-Planck equation. This provides regularity results for the survival probability $t \mapsto \mathbb{P}(X_t > 0)$ .

D. Methodological Novelty

PDE-based Uniqueness: Unlike previous works that relied solely on probabilistic uniqueness, this paper uses a fixed-point argument on the space of sub-probability densities to prove uniqueness for the McKean-Vlasov equation.
Handling Degeneracy: The paper addresses the degeneracy at the boundary (where the diffusion coefficient vanishes for absorbed particles) by carefully analyzing the boundary layer behavior and using Mosco convergence of Dirichlet forms to handle the transition from regularized to unregularized systems.

4. Significance

Systemic Risk Modeling: The model is directly applicable to systemic risk in finance, where the "particles" represent the capital of financial institutions. The absorption at zero models default, and the mean-field interaction models contagion (the loss of one institution affects the drift of others).
Complement to Existing Literature: While previous works (e.g., Olivera et al. [29]) focused on singularities caused by drift coefficients (blow-up), this paper focuses on singularities caused by hitting times (absorption). It complements existing literature by providing a classical PDE description of the limit rather than just a distribution-valued one.
Quantitative Rigor: The derivation of the optimal $O(1/N)$ convergence rate in the Wasserstein metric is a significant advancement, providing precise error bounds for numerical approximations of such systems.
Generalizability: The methodology, particularly the use of partial Girsanov transforms and the PDE fixed-point strategy, offers a robust framework for analyzing other interacting particle systems with boundary interactions, including higher-dimensional domains and multiple barriers.

In summary, the paper provides a comprehensive theoretical framework for singular particle systems with absorption, bridging stochastic analysis and PDE theory to establish existence, uniqueness, and quantitative convergence rates for the mean-field limit.