Mean-Field limit of the non-exchangeable Cucker-Dong model

Abstract: In this article, we examine the mean-field limit of the non-exchangeable Cucker--Dong model. This model corresponds to a biologically more realistic version of the classic Cucker-Smale model, which is used to describe the alignment phenomenon in large animal groups. In addition to alignment forces, the non-exchangeable Cucker--Dong model integrates attraction/repulsion forces and network-structured interactions. In order to enable convergence towards a flocking profile, the attraction/repulsion forces are weighted by a second-order coefficient called the alignment measure, which is smaller when individuals are more aligned overall. Deriving the mean-field limit of this model relies on a new stability result that is in agreement with with both the second-order nature of the alignment measure and the non-exchangeability induced by the graph-dependent interactions.

• The paper rigorously establishes the mean-field limit for a non-exchangeable Cucker-Dong model incorporating network interactions and alignment-modulated attraction-repulsion.
• It constructs a kinetic Vlasov-type PDE framework with fibered Wasserstein spaces to prove existence and stability in W2 metrics.
• The analysis highlights the necessity of W2 stability over W1 for capturing the nonlinear dynamics, guiding future work on heterogeneous agent systems.

Mean-Field Limit for the Non-Exchangeable Cucker-Dong Model

Introduction and Context

The paper investigates the rigorous mean-field limit of the non-exchangeable Cucker-Dong model, a significant generalization of flocking-type particle systems used to study collective biological dynamics. Unlike its ancestor, the Cucker-Smale model, which captures velocity alignment, the Cucker-Dong model integrates both attraction-repulsion and velocity alignment, with the added complexity of a dynamically weighted coefficient (the alignment measure) for pairwise interactions. This work considers non-exchangeability, representing agents with heterogeneous, network-structured interactions modeled via graphons. This framework closely reflects systems with social or hierarchical structures, such as herds or swarms with leader-follower relations.

Model Formulation and Mathematical Framework

The study formulates a particle system incorporating network-structured alignments and spatial attraction-repulsion by the following ODE system:

$$\begin{cases} \dot{x}_i = v_i,\ \dot{v}_i = \dfrac{1}{N}\sum_{j=1}^N A_{ij}^N K(x_j - x_i, v_j - v_i) + \dfrac{\sigma(v)^{2\alpha-1}}{N}\sum_{j=1}^N \phi(\|x_i-x_j\|)(x_i-x_j), \end{cases}$$

where $A^N$ encodes the interaction graph, $K$ is a generic communication kernel, and $\phi$ describes collision avoidance.

A critical novelty is the network non-exchangeability: the interaction matrix $A^N$ is generally heterogeneous, precluding the usual symmetrizations leveraged in exchangeable particle systems. The incorporation of attraction/repulsion forces, weighted by a second-order alignment measure, reflects a biologically plausible feedback—when velocities align, attractive/repulsive influence fades.

For large systems, the evolution is captured by a kinetic Vlasov-type PDE for a “fibered” family of probability measures $\mu_t^\xi(x, v)$, where the parameter $\xi$ indexes agent identity (or network node):

$$\partial_t \mu_t^\xi(x, v) + v \cdot \nabla_x \mu_t^\xi + \nabla_v \cdot (F_a[\mu_t](\xi, x, v) \mu_t^\xi) = 0$$

with $F_a[\mu_t]$ incorporating both network structure and state-dependent interaction scaling.

The limit of $A^N$ as $A^N$0 is formalized in terms of graphon convergence under the cut distance—a necessity for treating networks.

Main Analytical Contributions

The primary technical contributions can be distilled as follows:

• Well-posedness in Fibered Wasserstein Spaces: The paper formalizes solution concepts in the space of fibered probability measures $A^N$1, with metrics parameterized by agent-identity. The equations admit global solutions for sufficiently regular kernels and initial data, using classical fixed-point arguments in Wasserstein spaces but generalized for non-exchangeable, parameterized measures.
• Stability in Wasserstein-2 (Not Wasserstein-1):

The nonlinearity arising from the second-order (alignment) coefficient rules out stability estimates in $A^N$2, as illustrated by explicit counterexamples. The work establishes that the kinetic equation is stable in $A^N$3, both with respect to perturbations of initial data and interaction graphon, with stability constants depending on network regularity, interaction regularity, and initial support.

• Rigorous Mean-Field Limit: Leveraging the $A^N$4 stability, the authors derive the mean-field limit for empirical measures constructed from the agent dynamics. If the sequence of interaction graphs $A^N$5 converges (up to relabeling) in cut distance to a symmetric, bounded graphon $A^N$6, then the associated empirical distributions converge in $A^N$7 to the solution of the limiting Vlasov-type PDE parameterized by $A^N$8.
• Sharp Structural Assumptions: The well-posedness and stability require the interaction graphons to be symmetric and bounded away from zero (for $A^N$9), and the collision avoidance function $K$0 and communication kernel $K$1 must be locally Lipschitz. The authors carefully outline cases in which graphon convergence is automatic, e.g., uniform boundedness or Lusin-type approximations, ensuring theoretical robustness.

Numerical and Theoretical Claims

A key qualitative result is the demonstration that $K$2 stability fails for any exponent $K$3, meaning the nonlinearity introduced by the alignment measure fundamentally changes the analytical structure compared to canonical alignment models. The findings clarify that the previously claimed mean-field limits for the Cucker-Dong model in $K$4 are not generally valid. Instead, the mean-field limit here is established in $K$5, aligning the metric structure with the underlying second-order nonlinearities.

No explicit computational or numerical experiments are presented; all results are mathematical and structural.

Implications and Directions for Future Work

The results have both practical and theoretical implications:

• Modeling Realism: Allowing for non-exchangeability and including alignment-modulated attraction/repulsion improves the biological fidelity of the model. Many biological groups and engineered networks exhibit highly structured interactions incompatible with exchangeability.
• Analysis of Non-Equilibrium Dynamics: Rigorous mean-field limits in this generality enable analytic study of large, heterogeneous multi-agent systems without reducing to mean-interaction or homogeneous approximations.
• Foundation for Collision Avoidance: Although the present framework does not extend to the regime of strongly singular interaction forces (required for rigorous collision avoidance), it provides the necessary foundation for future work. The authors note that achieving mean-field limits under truly singular forces (removing cutoffs) for repulsive collision avoidance, especially without network exchangeability, remains an open technical challenge.

Potential avenues include:

• Extending existence/uniqueness theory to singular kernels and repulsive regimes relevant for true collision avoidance.
• Analysis of convergence rates as a function of network heterogeneity (cut distance) and attraction/repulsion strength.
• Application to control and coordination design in engineered swarms or hierarchical agent networks.

Conclusion

This work advances the theory of multi-agent dynamical systems by establishing a rigorous, network-aware mean-field limit for the Cucker-Dong model with non-exchangeable interactions and second-order modulation. The stability analysis in Wasserstein-2 is both necessary and sufficient due to the alignment-modulated coupling. This framework bridges biologically realistic agent-based modeling and the rigorous kinetic theory of large populations within arbitrary networked interaction topologies, establishing new mathematical tools for future exploration of network-induced phenomena in self-organization.