Positive and free energy satisfying schemes for diffusion with interaction potentials

Hailiang Liu, Wumaier Maimaitiyiming

TL;DR

This paper introduces new numerical schemes for solving diffusion equations with interaction potentials. These schemes are designed to be "positive and free energy satisfying," meaning they accurately preserve crucial physical properties like mass conservation, solution positivity, and energy dissipation, even for complex scenarios and without strict time step limitations. This advancement is crucial for reliably simulating a wide range of physical phenomena.

What Problem Does This Paper Address?

Many real-world phenomena, from the movement of polymers to the aggregation of biological cells, can be modeled by diffusion equations that include interaction potentials. These models often possess fundamental physical properties: the solution (representing density, for example) must remain non-negative, the total mass must be conserved, and the system's free energy should only decrease over time (dissipate). Developing numerical methods that faithfully preserve these properties is a significant challenge, especially for complex, nonlinear, and nonlocal equations. Existing methods often struggle with one or more of these properties, or impose restrictive time step limits, hindering their practical application for long-term simulations or complex systems. This paper aims to bridge this gap by developing schemes that robustly maintain these essential physical characteristics.

Methods

The authors propose novel numerical schemes based on a reformulation of the original diffusion equation. This reformulation, using a non-logarithmic Landau form, simplifies the discretization process. They develop both semi-discrete (spatial discretization only) and fully discrete (spatial and temporal discretization) schemes.

  • Semi-discrete scheme: A finite volume method is employed for spatial discretization, ensuring mass conservation, positivity, and free energy dissipation.
  • Fully discrete schemes:
    • First-order in time: An implicit-explicit approach is used, which is shown to preserve the desired properties without strict time step restrictions.
  • Second-order in time and space: This more advanced scheme utilizes a local scaling limiter to restore solution positivity when it might be lost. Crucially, they rigorously prove that this limiter does not compromise the second-order accuracy of the scheme.

The schemes are designed to be easy to implement and efficient for long-term simulations.

Datasets

Not discussed in the paper. The paper focuses on developing and analyzing numerical methods for a general class of equations, with numerical examples provided to demonstrate their performance.

Key Insights and Ideas

  • Reformulation is Key: The non-logarithmic Landau reformulation of the diffusion equation significantly simplifies the design of structure-preserving numerical schemes.
  • Positivity Limiter: The introduction and rigorous analysis of a local scaling limiter for restoring solution positivity in the second-order scheme is a major contribution. Proving that this limiter doesn't degrade accuracy is a critical validation.
  • Unconditional Stability (for first-order): The first-order fully discrete scheme achieves mass conservation, positivity, and free energy dissipation without any time step restrictions, a significant practical advantage.
  • Generalizability: The proposed methods are shown to be applicable to a broader class of nonlinear nonlocal equations, not just the specific model studied.

Key Results

  • First-order scheme: The fully discrete, first-order scheme (in time) is proven to satisfy mass conservation, positivity, and free energy dissipation for both one and two-dimensional problems with nonuniform meshes, without time step restrictions.
  • Second-order scheme: The fully discrete, second-order scheme (in time and space) also preserves these properties. The local scaling limiter effectively restores positivity without sacrificing second-order accuracy.
  • Numerical Examples: The paper presents numerical examples in one and two dimensions that demonstrate the accuracy and effectiveness of the proposed schemes, confirming the theoretical findings regarding mass conservation, positivity, and energy dissipation. For instance, accuracy tests show expected orders of convergence for both first and second-order schemes.

Limitations and Future Work

  • Positivity Limiter for Second-Order Scheme: While the second-order scheme is generally robust, the positivity-preserving limiter might be required for large time steps to guarantee non-negativity. The paper provides a theoretical condition for the limiter's effectiveness and an algorithm for its application.
  • Explicit Scheme CFL Condition: A remark mentions that an explicit Euler forward time discretization would preserve positivity but would require a CFL condition like $\tau < \gamma h^2$. The implicit-explicit and second-order schemes avoid such strict restrictions.
  • Future Work: The authors suggest that exploring explicit-implicit strategies for Discontinuous Galerkin (DG) schemes could be an interesting avenue for future research. They also note that for certain problems, like the aggregation equation, particle methods might be an alternative, though they can require a large number of particles. The paper also hints at the potential for extending these methods to more general nonlinear nonlocal equations.

Summary updated

Abstract

In this paper, we design and analyze second order positive and free energy satisfying schemes for solving diffusion equations with interaction potentials. The semi-discrete scheme is shown to conserve mass, preserve solution positivity, and satisfy a discrete free energy dissipation law for nonuniform meshes. These properties for the fully-discrete scheme (first order in time) remain preserved without a strict restriction on time steps. For the fully second order (in both time and space) scheme, we use a local scaling limiter to restore solution positivity when necessary. It is proved that such limiter does not destroy the second order accuracy. In addition, these schemes are easy to implement, and efficient in simulations over long time. Both one and two dimensional numerical examples are presented to demonstrate the performance of these schemes.
BibTeX
@article{liu2020positive,
  author = {Hailiang Liu and Wumaier Maimaitiyiming},
  title = {{Positive and free energy satisfying schemes for diffusion with interaction potentials}},
  journal = {Journal of Computational Physics},
  year = {2020},
  doi = {10.1016/j.jcp.2020.109483},
}
Figure