High order matched interface and boundary method for elliptic equations with discontinuous coefficients and singular sources

Guo-Wei Wei, Yongcheng Zhou, Shan Zhao, Michael Feig, Guo-Wei Wei 0001Journal of Computational Physics 2006 · DOI

TL;DR

This paper introduces a novel "Matched Interface and Boundary" (MIB) method for accurately solving complex elliptic equations on simple Cartesian grids, even when coefficients change abruptly or sources are singular. The MIB method achieves high-order accuracy by cleverly using fictitious points and iteratively enforcing interface conditions, offering a robust and efficient alternative to existing techniques.

What Problem Does This Paper Address?

Many real-world physical phenomena, such as fluid dynamics, heat transfer, and electromagnetics, are described by elliptic partial differential equations (PDEs). Often, these phenomena involve interfaces where material properties (like diffusion coefficients) change drastically, or where singular sources exist. Solving these problems accurately on standard, regular grids (like Cartesian grids) is challenging because traditional numerical methods struggle with these discontinuities. Existing methods often require complex grid generation (body-fitting grids) or introduce inaccuracies near interfaces. This paper aims to provide a high-order accurate and efficient numerical method that can handle these interface problems directly on simple Cartesian grids.

Methods

The core of this paper is the development of the Matched Interface and Boundary (MIB) method. Here's how it works:

  • Disassociation of PDE and Interface: Unlike some methods that embed interface conditions directly into the PDE discretization, MIB separates these two tasks. The PDE is discretized using standard high-order finite difference (FD) schemes across the entire domain.
  • Fictitious Points (FPs): When the standard FD stencil for a grid point "looks" across an interface (i.e., it needs values from both sides), MIB introduces "fictitious points" on the other side. These FPs are not actual grid points but are introduced to allow the use of the same high-order FD stencil everywhere.
  • Iterative Enforcement of Jump Conditions: The crucial part is determining the values at these FPs. MIB uses the known physical jump conditions (e.g., continuity of the solution and flux across the interface) and applies them iteratively. By repeatedly enforcing the lowest-order jump conditions, MIB can construct arbitrarily high-order accurate schemes. This avoids the complexity of dealing with high-order derivatives that arise in direct high-order jump condition enforcement.
  • Regular and Irregular Interfaces: The method is first formulated for simple, straight interfaces (regular) and then extended to handle complex, curved interfaces (irregular). For irregular interfaces, auxiliary lines and points are used to manage the local topology.

Datasets

Not discussed in the paper. The paper focuses on numerical methods and uses analytical solutions for validation.

Key Insights and Ideas

  • Iterative Lowest-Order Conditions for High Order: The most significant insight is that by repeatedly applying the simplest (lowest-order) interface jump conditions, one can achieve arbitrarily high-order accuracy for the overall solution. This bypasses the significant challenge of deriving and implementing high-order jump conditions directly.
  • Systematic Approach for High Order: The MIB method offers a systematic way to construct high-order schemes, making it adaptable to different orders of accuracy by adjusting parameters.
  • Robustness to Jump Magnitude: The method demonstrates robustness even when the jump in coefficients or solution across the interface is very large, suggesting its accuracy is driven by the oscillatory nature of the solution rather than the interface itself.
  • Generalizability: The MIB approach is designed to be largely independent of the specific PDE, making it applicable to a wide range of elliptic problems.

Key Results

  • High-Order Accuracy: The MIB method successfully recovers the designed high-order accuracy of the underlying finite difference schemes, even in the presence of interfaces. For regular interfaces, schemes up to 16th-order were constructed, and for irregular interfaces, up to 6th-order were demonstrated.
  • Efficiency: The MIB method shows significant computational efficiency gains compared to lower-order methods. For example, a 12th-order MIB scheme achieved a target accuracy in a 3D problem with a mesh size that would require a second-order method to be orders of magnitude larger and slower.
  • Comparison with IIM and GFM: The paper compares MIB to the Immersed Interface Method (IIM) and Ghost Fluid Method (GFM). MIB is presented as a generalization, offering advantages in terms of systematic high-order scheme construction and robustness.
  • Numerical Validation: Extensive numerical experiments in 1D, 2D, and 3D, for both regular and irregular interfaces, with varying contrasts in diffusion coefficients, validate the method's accuracy and efficiency.

Limitations and Future Work

  • Irregular Interface Complexity: While MIB handles irregular interfaces, it still requires careful treatment of local topology at irregular grid points, which can be intricate.
  • Linear System Properties: The resulting linear systems for MIB are noted to be non-symmetric and not necessarily diagonally dominant, which can affect the choice and performance of iterative solvers compared to some IIM formulations.
  • Future Work: The authors mention ongoing investigations into applying MIB to Lipschitz continuous interfaces, moving interfaces, and for electromagnetic scattering problems. They also suggest exploring its use with other high-order methods for hyperbolic conservation laws.

Summary updated

BibTeX
@article{wei2006high,
  author = {Guo-Wei Wei and Yongcheng Zhou and Shan Zhao and Michael Feig and Guo-Wei Wei},
  title = {{High order matched interface and boundary method for elliptic equations with discontinuous coefficients and singular sources}},
  journal = {Journal of Computational Physics},
  year = {2006},
  doi = {10.1016/j.jcp.2005.07.022},
}
Figure