Singular mean-field limits via a multiscale mollification metric

Abstract: We consider a general class of first order ODE systems for the evolution of $N$ interacting particles (in Euclidean space $\mathbb{R}<sup>d$) in a mean-field regime. The class of interactions treated includes singular interactions of inverse power type up to power $d+1$, attractive or repulsive, and not necessarily deriving from a potential -- unlike, for instance, the modulated energy method. We introduce a new method to prove quantitative convergence of the discrete system to solutions of the mean-field equation. It relies on studying the evolution of a metric encoding a multiscale control of the difference between the empirical measure and its limit, via mollification by heat kernels. We prove that the desired convergence holds (i) up to the maximal time of existence of the smooth solution to the limiting equation if the singularity is sub-coulombic in any dimension, or coulombic in dimensions 1 and 2 (where, to do so, we introduce a notion of weak solution to the ODE system), or (ii) for short time in the case of Coulomb singularity in dimension 3 and above and (iii) up to a short $N$-dependent timescale for super-coulombic interactions in all dimensions. The latter two results are demonstrated to be optimal as we prove that collisions occur within the same timescale for a class of attractive interactions.

• The paper introduces a novel multiscale mollification metric that rigorously quantifies mean-field convergence even with highly singular interactions.
• It establishes sharp convergence rates and thresholds for sub-Coulomb, Coulomb, and super-Coulomb regimes, linking discrete dynamics to continuum limits.
• The analysis proves propagation of chaos and offers actionable insights for designing particle-based numerical schemes to handle singular forces.

Singular Mean-Field Limits via a Multiscale Mollification Metric

Overview and Motivation

This paper addresses the rigorous derivation of mean-field limits for first-order systems of $N$ particles in $\mathbb{R}^d$ interacting through singular kernels, significantly extending the class of interactions for which quantitative convergence to the associated nonlinear PDE can be established. The focus is on force fields $F(x, y)$ that may be highly singular, including power-law singularities $\sim |x-y|^{-s}$ with $s < d+1$, encompassing sub-Coulomb, Coulombic, and super-Coulombic cases. In contrast to previous approaches that often rely on symmetry, positive definiteness, or potential structure, this work establishes convergence under minimal structural constraints, using a new metric based on multiscale mollification via heat kernels.

Main Contributions

1. Novel Multiscale Mollification Metric

The paper introduces a multiscale metric $M$, defined by mollifying the empirical particle measure with heat kernels at various scales $\eta \geq N^{-1/d}$. By measuring the weighted $L^\infty$ distance between the mollified empirical measure and the limiting PDE solution $\mu$, this approach effectively captures discrepancies between discrete and continuum dynamics at all relevant microscales. This development crucially leverages the semi-group property of the heat kernel, allowing refined error propagation across scales and optimal utilization of available regularity.

The primary metric is given by

$$M(\mathbf{X}_N, f, \eta) = \sup_{x \in \mathbb{R}^d} \langle x \rangle^{\gamma} \left| \frac{1}{N} \sum_{i=1}^N \Phi_{x_i}^\eta(x) - f(x) \right|,$$

summed across dyadic scales with appropriate weights to penalize discrepancies more severely at coarser resolutions.

2. General Treatment of Force Fields

Unlike classical results, which either require smoothness (Lipschitz continuity) or structural properties linked to derivation from a potential, this work permits force fields $\mathbb{R}^d$0 with strong singularities, non-symmetry, and significant freedom in their long-range behavior. Only a minimal cancellation property is required to make the convolution with $\mathbb{R}^d$1 well-defined in the mean-field PDE.

3. Sharp Regimes and Matching Lower Bounds

The authors provide a complete analysis spanning the regimes:

• Sub-Coulomb ($\mathbb{R}^d$2): Global-in-time classical solutions, uniform convergence, and quantitative chaos propagation are established.
• Coulomb ($\mathbb{R}^d$3): Full-time convergence in dimensions $\mathbb{R}^d$4 (with weak solutions after collisions); only short-time convergence in higher dimensions due to possible collisions.
• Super-Coulomb ($\mathbb{R}^d$5): Convergence is proven only up to a critical short time $\mathbb{R}^d$6, with matching lower bounds showing that collisions develop macroscopically precisely at this timescale.

The lower bounds are demonstrated via explicit analysis of well-prepared initial conditions, showing that the discrete particle system undergoes collapse due to attractive interactions at the same rate as the breakdown of validity of the mean-field limit.

4. Quantitative Convergence and Propagation of Chaos

The convergence rates in the new metric essentially achieve the microscale-optimal rate: for mollification at scale $\mathbb{R}^d$7, the bound $\mathbb{R}^d$8 holds, up to logarithmic factors cascading from the multiscale argument. Importantly, this convergence holds for the empirical measure itself, not only for finite-dimensional marginals or weak distances.

These results rigorously imply propagation of chaos: for i.i.d. initial data, all $\mathbb{R}^d$9-particle marginals are quantitatively well-approximated by tensor products of the mean-field solution.

5. Construction of Classical Solutions and Weak Continuations

The approach not only proves convergence but also constructs global-in-time classical solutions for singular particle systems in the sub-Coulomb regime, and provides weak solution concepts for continuation past collisions in low-dimensional Coulomb systems. This extends existence theory for deterministic singular ODE systems by direct mean-field analysis.

Technical Core and Analytical Methods

Multiscale Analysis via Heat Kernel

Central to the proof is the use of the heat kernel, enabling scale-wise error control and recursive estimates. The key is to analyze the evolution (in time) of the multiscale metric $F(x, y)$0, showing that while discrepancies at the finest scale may deteriorate, higher (coarser) scales can absorb errors via their dyadic weighting. The argument carefully tracks the contribution of particle interactions at distances $F(x, y)$1, requiring a coupled control of close-pair interactions via a secondary small-scale energy $F(x, y)$2 (measuring pair collisions). A Gronwall-type cascade then closes the estimates without cumulative loss.

Handling Singular Kernels

Precise kernel bounds, including pointwise, integral, and symmetrized truncations, are derived for the mollified and truncated force fields. These are used to control error terms arising in the comparison between discrete and continuum dynamics, and are checked against the cancellation conditions imposed on $F(x, y)$3. The analysis distinguishes sharply between sub-Coulomb, Coulomb, and super-Coulomb cases.

Microlocal and Dual Estimates

To supplement the $F(x, y)$4-based multiscale metric, the authors provide dual space bounds (in $F(x, y)$5) that permit propagation of weak convergence, crucially using backward transport equations and commutator estimates familiar from the study of rough transport PDEs.

Strong Numerical and Qualitative Results

Regime	Solution Existence	Quantitative Convergence	Chaos Propagation	Time Scale	Collisions
Sub-Coulomb	Global classical	$F(x, y)$6, up to log factors	Uniform in time	$F(x, y)$7	Absent
Coulomb 1,2D	Weak solutions	$F(x, y)$8, up to log factors	Uniform, weak sense	$F(x, y)$9	Handled via weak
Coulomb $\sim |x-y|^{-s}$0	Classical, short-time	$\sim |x-y|^{-s}$1, up to log factors	Up to collision	$\sim |x-y|^{-s}$2 possible, then breakdown	Present
Super-Coulomb	Classical, very short-time	$\sim |x-y|^{-s}$3, up to log factors	Up to $\sim |x-y|^{-s}$4	$\sim |x-y|^{-s}$5 sharp	Inevitable

Implications and Theoretical Outlook

This work shifts the landscape for mean-field theory, making precise the sharp singularity thresholds for the validity of deterministic mean-field equations for first-order systems. The new multiscale mollification metric bypasses the need for structural restrictions on $\sim |x-y|^{-s}$6 and yields strong, direct control over the empirical measures. The results clarify the precise influence of particle collisions and singular attractive forces on the breakdown of the mean-field approximation, connecting PDE well-posedness with discrete dynamics in a nontrivial fashion.

Practically, the techniques provide guidance for designing and interpreting particle-based numerical schemes for singular flows (including those in vortex dynamics, plasma physics, and sociophysical models with power-law interactions), particularly regarding the necessity of mollification and scale selection to avoid spurious aggregation.

From a theoretical perspective, this approach opens avenues to treat broader classes of interactions (including those without energy structure or with additional external fields), and prompts refined questions about propagation of chaos, fluctuations, and large deviations at and beyond the singularity threshold.

Future Directions

• Extension to second-order systems (e.g., Vlasov dynamics with inertia) and stochastic perturbations.
• Analysis of hypersingular regimes ($\sim |x-y|^{-s}$7) where standard mean-field limits break down and hydrodynamic limits emerge.
• Systematic study of noise regularization and its criticality (including links to rough SDE theory and stochastic regularization).
• Investigation into universality and limit theorems for macroscopic blowup and collision phenomena in discrete systems with singular forces.

Conclusion

This paper provides a rigorous, quantitative framework for mean-field convergence in the presence of singular interactions, establishing precise thresholds for well-posedness, convergence, and the onset of collisions. The multiscale mollification approach represents a substantial sharpening of existing techniques and paves the way for further advances in deterministic and probabilistic mean-field theory for complex systems.

Reference: "Singular mean-field limits via a multiscale mollification metric" (2607.10686)