Mean-field limit for collective behavior models with sharp sensitivity regions

José A. Carrillo, Yeji Choi, Maxime Hauray, Samir SalemJournal of the European Mathematical Society 2018 · DOI · arXiv

Abstract

We rigorously show the mean-field limit for a large class of swarming individual based models with local sharp sensitivity regions. For instance, these models include nonlocal repulsive-attractive forces locally averaged over sharp vision cones and Cucker–Smale interactions with discontinuous communication weights. We define global-in-time solutions through a differential inclusion system corresponding to the particle descriptions. We estimate the error between the solutions to the differential inclusion system and weak solutions to the expected limiting kinetic equation by employing tools from optimal transport theory. Quantitative bounds on the expansion of the 1-Wasserstein distance along flows based on a weak-strong stability estimate are obtained. We also provide various examples of realistic sensitivity sets satisfying the assumptions of our main results.

Introduction: The Challenge of Discontinuous Interactions

In the study of collective behavior—ranging from bird flocks to fish schools—Individual-Based Models (IBMs) are the primary tool for simulating how local interactions lead to global patterns. A key theoretical challenge is the "mean-field limit": proving that as the number of agents $N$ tends to infinity, the discrete particle system converges to a continuous "fluid" described by a kinetic partial differential equation.

While this limit is well-established for smooth interactions, biological reality is often "sharp." Animals typically have specific vision cones or sensitivity regions, meaning they are influenced by neighbors within a certain field of view but completely ignore those outside it. Mathematically, this introduces discontinuities into the interaction kernels. Standard calculus and mean-field proofs rely on Lipschitz continuity, which fails when forces "jump" at the boundary of a vision cone.

Visual representation of sharp sensitivity regions, such as a vision cone (left), which define the local interaction neighborhood for an agent.

Methodology: Differential Inclusions and Optimal Transport

To overcome the lack of regularity, this paper utilizes two powerful mathematical frameworks:

  1. Filippov’s Theory of Differential Inclusions: Because the force field is discontinuous, standard ordinary differential equations (ODEs) may not have well-defined solutions. By redefining the system as a differential inclusion, the authors allow the velocity to take a range of values at points of discontinuity, ensuring the existence of global-in-time solutions for the particle system.
  2. Optimal Transport and the $d_1$ Metric: To compare the discrete particle distribution to the continuous kinetic density, the authors employ the 1-Wasserstein distance ($d_1$). This metric is uniquely suited for this problem because it allows for stable estimates of how the distance between two distributions evolves over time, even when the underlying velocity fields are not perfectly smooth.

Geometric Regularity and the Averaged Force Field

A central insight of the work is that the averaged force field can remain regular even if the underlying interaction kernel is discontinuous. If the agent density $f$ is bounded ($L^\infty$), the force field $F(f)$ can be shown to be locally Lipschitz continuous, provided the sensitivity region satisfies specific geometric constraints (labeled H1 and H2).

Illustration of the geometric regularity conditions (H1 & H2) on the boundary K of a sensitivity region, ensuring the averaged force field remains mathematically tractable.

These constraints require that the boundary of the sensitivity region (e.g., the edge of a vision cone) is not too irregular. Specifically, the volume of an $\epsilon$-neighborhood around the boundary must be proportional to $\epsilon$, ensuring that the "jump" in the force field doesn't destabilize the system's large-scale behavior.

Key Results and Quantitative Bounds

The authors successfully derive the Vlasov-type kinetic equation for a wide class of models, including Cucker–Smale alignment and attractive-repulsive dynamics. They provide a rigorous quantitative bound on the error between the particle system and the kinetic model:

$$d_1(f(t), \mu^N(t)) \le e^{Ct} d_1(f(0), \mu^N(0))$$

This result demonstrates that the error grows at most exponentially over time. If the initial particle distribution is a good approximation of the continuous density, it will remain a valid approximation as the system evolves. This provides a solid mathematical foundation for using kinetic equations to study biologically realistic swarming models.

Conclusion and Future Directions

This research bridges the gap between discrete biological simulations and continuous mathematical analysis for models with sharp perception boundaries. While the results are currently limited to the $d_1$ distance and do not yet account for singular kernels (where forces become infinite at zero distance), they represent a significant step toward more realistic mathematical models of collective behavior. Future work may explore extending these stability estimates to other metrics or incorporating more complex, singular interaction forces.

Summary updated

Abstract

We rigorously show the mean-field limit for a large class of swarming individual based models with local sharp sensitivity regions. For instance, these models include nonlocal repulsive-attractive forces locally averaged over sharp vision cones and Cucker–Smale interactions with discontinuous communication weights.We define global-in-time solutions through a differential inclusion system corresponding to the particle descriptions. We estimate the error between the solutions to the differential inclusion system and weak solutions to the expected limiting kinetic equation by employing tools from optimal transport theory. Quantitative bounds on the expansion of the 1-Wasserstein distance along flows based on a weak-strong stability estimate are obtained. We also provide various examples of realistic sensitivity sets satisfying the assumptions of our main results.
BibTeX
@article{carrillo2018mean,
  author = {José A. Carrillo and Yeji Choi and Maxime Hauray and Samir Salem},
  title = {{Mean-field limit for collective behavior models with sharp sensitivity regions}},
  journal = {Journal of the European Mathematical Society},
  year = {2018},
  doi = {10.4171/jems/832},
}
Figure