A blob method for the aggregation equation

Katy Craig, Andrea L. Bertozzi

TL;DR

This paper introduces a novel "blob method" for numerically solving the aggregation equation, a fundamental model for phenomena like swarming and self-assembly. By regularizing the interaction kernel, this method achieves arbitrarily high polynomial convergence rates to exact solutions, outperforming traditional particle methods and offering greater flexibility in kernel choices.

What Problem Does This Paper Address?

The aggregation equation describes how a density distribution evolves under a velocity field derived from pairwise interactions. This equation is crucial for modeling diverse systems, from biological swarming to molecular self-assembly. A key challenge in numerically solving this equation, especially with singular interaction kernels (like the Newtonian potential), is the potential for blow-up in finite time and the difficulty in achieving high accuracy. Existing particle methods often struggle with these issues, leading to limitations in their convergence rates and applicability. This paper aims to develop a more robust and accurate numerical method that can handle a wider range of interaction kernels and achieve higher orders of convergence.

Methods

The core of this paper is the development of a "blob method" for the aggregation equation. This method is inspired by classical vortex blob methods used for the Euler equations. The key innovation lies in regularizing the singular interaction kernel by convolving it with a smooth, rapidly decreasing "blob function" (a mollifier). This regularization smooths out the velocity field, preventing singularities and enabling higher-order convergence.

The method involves:

  • Particle Approximation: Representing the density as a collection of particles, each carrying a portion of the total mass.
  • Kernel Regularization: Convolving the interaction kernel $K$ with a mollifier $\psi_\delta$ to create a regularized kernel $K_\delta = K * \psi_\delta$. This ensures that the velocity field and its derivatives are well-behaved.
  • Lagrangian Formulation: Evolving the particles according to ordinary differential equations (ODEs) derived from the regularized aggregation equation.
  • Convergence Proofs: Rigorously proving that the numerical solutions generated by the blob method converge to the exact classical solutions of the aggregation equation. The convergence rate depends on the accuracy of the chosen mollifier.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and numerical simulations using synthetic initial data.

Key Insights and Ideas

  • Regularization as a Path to High Accuracy: The paper demonstrates that regularizing the interaction kernel, rather than directly handling singularities, is a powerful strategy for achieving arbitrarily high polynomial convergence rates in numerical methods for the aggregation equation.
  • Blob Method's Versatility: The blob method is shown to be effective for a broad class of kernels, including singular ones like the Newtonian potential, as well as repulsive-attractive potentials (Morse and power-law).
  • Connection to Gradient Flows: The blob method preserves the underlying Wasserstein gradient flow structure of the aggregation equation, and the regularized energy functional is shown to decrease along particle trajectories. This provides a theoretical foundation for the method's stability.
  • Trade-off in Regularization: While regularization improves convergence and stability, it can also lead to phenomena like "bending" of particle trajectories to avoid collisions, which is an artifact of the smoothed interaction. This is observed in the numerical examples.

Key Results

  • Arbitrarily High Polynomial Convergence: The authors prove that their blob method achieves arbitrarily high polynomial rates of convergence to classical solutions, with the rate dependent on the accuracy of the mollifier. This is a significant improvement over existing particle methods, which typically have lower convergence rates (e.g., $O(h^{2-\epsilon})$ for Euler equations).
  • Conservation Laws: The method conserves mass and preserves the energy decreasing property of the regularized system.
  • Numerical Validation: Extensive numerical examples in one and two dimensions validate the theoretical convergence rates for various kernels and initial data types (regular and discontinuous). The results show that the blob method's accuracy is indeed dependent on the mollifier's order of accuracy.

Limitations and Future Work

  • Focus on Classical Solutions: The current convergence proofs are for classical solutions. Extending these results to weak measure solutions is a significant direction for future work, building on existing research for specific kernels.
  • Kernel Differentiation: While the method handles various kernels, the numerical computation of convolutions can still be computationally intensive, especially for high-dimensional problems or complex kernels.
  • Error Measurement: The paper notes that for certain problems (e.g., steady states), the $L^\infty$ norm might not be the most sensitive measure of error, and exploring other norms could be beneficial.
  • Adapting to Kernel Types: The authors suggest that future work could leverage the distinct dynamics of purely repulsive, purely attractive, or repulsive-attractive kernels to further improve numerical methods. For instance, methods that take advantage of steady-state formation in repulsive kernels could be explored.
  • Extension to Keller-Segel Equation: The blob method could potentially be extended to the Keller-Segel equation by incorporating diffusion, drawing inspiration from methods used for Navier-Stokes equations.

Summary updated

Abstract

: Motivated by classical vortex blob methods for the Euler equations, we develop a numerical blob method for the aggregation equation. This provides a counterpoint to existing literature on particle methods. By regularizing the velocity field with a mollifier or “blob function”, the blob method has a faster rate of convergence and allows a wider range of admissible kernels. In fact, we prove arbitrarily high polynomial rates of convergence to classical solutions, depending on the choice of mollifier. The blob method conserves mass and the corresponding particle system is energy decreasing for a regularized free energy functional and preserves the Wasserstein gradient flow structure. We consider numerical examples that validate our predicted rate of convergence and illustrate qualitative properties of the method.
BibTeX
@inproceedings{craig2016blob,
  author = {Katy Craig and Andrea L. Bertozzi},
  title = {{A blob method for the aggregation equation}},
  booktitle = {Math. Comput},
  year = {2016},
  url = {https://openreview.net/forum?id=JDPC2Q72CQ},
}
Figure