Stability and convergence analysis of high-order numerical schemes with DtN-type absorbing boundary conditions for nonlocal wave equations

Jihong Wang, Jerry Zhijian Yang, Jiwei Zhang

Introduction

Nonlocal models have emerged as a powerful mathematical framework for describing physical phenomena where the state of a point depends on a neighborhood—defined by a "horizon"—rather than just its immediate vicinity. These models are increasingly essential in fields such as peridynamics for fracture mechanics, image processing, and biological systems. However, a significant hurdle in the practical application of nonlocal wave equations is their simulation on unbounded domains. While local wave equations can often be truncated using simple boundary conditions, nonlocal interactions create a unique challenge: points near the edge of a computational window still interact with the region outside that window.

This paper addresses this challenge by developing a rigorous framework for high-order numerical schemes equipped with Dirichlet-to-Neumann (DtN) absorbing boundary conditions (ABCs). The authors provide the first formal stability and convergence analysis for these schemes, ensuring that waves can exit the computational domain without artificial reflections.

The Challenge of Nonlocal Boundaries

In standard local physics, the interaction between points is infinitesimal. When simulating waves in an infinite space, one can truncate the domain and apply boundary conditions that approximate the behavior of the wave as it moves toward infinity. In nonlocal models, the interaction "horizon" means that the domain truncation is inherently lossy. If the domain is simply cut off, the missing interactions from the "outside" manifest as non-physical wave reflections that propagate back into the simulation, destroying the accuracy of the results.

To solve this, researchers use Absorbing Boundary Conditions (ABCs). The goal is to create a boundary that acts as if the domain were still infinite. While previous research explored Dirichlet-to-Dirichlet (DtD) mappings, these were notoriously difficult to analyze for mathematical stability. This paper shifts the focus to DtN mappings, which relate the solution's value on the boundary to its "nonlocal flux."

Methodology: The DtN Mapping Framework

The authors employ a multi-step strategy to bridge the gap between infinite-domain theory and finite-domain computation:

  1. High-Order Discretization: The spatial nonlocal operator is discretized using a quadrature-based finite difference scheme. This approach is versatile, allowing for high-order Lagrange interpolations (quadratic, cubic, etc.) to achieve superior spatial precision.
  2. The Discrete Green’s Identity: The core innovation lies in using a discrete nonlocal Green’s first identity. This mathematical tool allows the authors to transform the exact but analytically stubborn DtD mappings into DtN mappings. This transformation is the key to unlocking stability proofs.
  3. Energy-Based Stability Analysis: To ensure the numerical scheme does not "blow up" over time, the authors utilize an energy method. They establish stability and convergence under a specific Courant–Friedrichs–Lewy (CFL) condition. This condition is specifically tailored to the nonlocal interaction kernel, providing a necessary roadmap for setting time-step restrictions in practical simulations.

Numerical Validation

The theoretical framework was tested through a series of synthetic numerical experiments in both 1D and 2D environments:

  • 1D Wave Propagation: The scheme was tested against three distinct kernel types: constant, non-integrable, and fractional Laplacian. These tests confirmed that the DtN boundary conditions effectively "vanish" waves at the boundary.
  • 2D Wave Propagation: Using a Gaussian kernel, the authors demonstrated the scheme's effectiveness in higher dimensions, showing that the waves pass through the artificial boundaries perfectly with no visible "ghost" reflections.
  • Convergence Verification: The experiments confirmed an optimal convergence order of $O(\tau^2 + h^q)$, where $\tau$ is the time step and $h^q$ represents the spatial discretization order. Specifically, linear interpolation yielded 2nd-order convergence, while quadratic and cubic interpolations achieved 4th-order convergence, matching the theoretical predictions.

Insights and Future Directions

The most significant contribution of this work is the "stability bridge" created by the DtN mapping. By proving that these high-order schemes are stable and convergent, the authors provide a robust foundation for future engineering applications in peridynamics and beyond.

However, some limitations remain. While spatial accuracy can be tuned to high orders, the temporal discretization remains limited to second-order accuracy. The authors suggest that "modified equation techniques" could be employed in future iterations to improve temporal precision. Additionally, the computational cost of evaluating convolution kernels at the boundary remains high. Future research will likely focus on adapting fast algorithms to these complex nonlocal kernels to reduce the computational overhead of these high-precision simulations.

Summary updated

BibTeX
@inproceedings{wang2023stability,
  author = {Jihong Wang and Jerry Zhijian Yang and Jiwei Zhang},
  title = {{Stability and convergence analysis of high-order numerical schemes with DtN-type absorbing boundary conditions for nonlocal wave equations}},
  booktitle = {IMA},
  year = {2023},
  doi = {10.1093/imanum/drad016},
}
Figure