High order solution of Poisson problems with piecewise constant coefficients and interface jumps

Alexandre Noll Marques, Jean-Christophe Nave, Rodolfo Ruben RosalesJournal of Computational Physics 2017 · DOI

TL;DR

This paper introduces a novel, high-order accurate algorithm for solving Poisson's equation in complex geometries, particularly those with interfaces where the solution or its derivatives can jump. The method combines the Correction Function Method (CFM) with Boundary Integral Methods (BIM) to efficiently and accurately handle these challenging problems, achieving convergence rates of 3rd or 4th order.

What Problem Does This Paper Address?

Solving Poisson's equation is fundamental in many scientific and engineering fields, from fluid dynamics to electromagnetics. However, when the problem involves complex geometries or interfaces where the solution's properties change abruptly (e.g., different material properties or fluid interfaces), standard numerical methods like finite differences struggle. These discontinuities break the assumptions of smoothness that these methods rely on, leading to inaccurate results or requiring extremely fine grids. Existing methods for handling such interfaces often have limitations in terms of accuracy, computational cost, or the types of interface conditions they can handle. This paper specifically addresses the challenge of solving Poisson problems with piecewise constant coefficients and jumps in the solution and its normal derivative across interfaces, aiming for high accuracy and efficiency.

Methods

The core of the proposed algorithm is a clever decomposition of the problem into sub-problems that can be solved efficiently. The solution u is split into two components: u = v + w.

  • Component v (Correction Function Method - CFM): This component handles the "standard" parts of the problem, including the source terms and the jumps in the solution itself across interfaces. The CFM is a technique that allows finite difference methods to work accurately across interfaces by introducing "correction functions." These functions effectively smooth out the discontinuities for the finite difference stencils, enabling high-order accuracy on a regular Cartesian grid. This part of the problem is solved within a rectangular box that encloses the domain, allowing for the use of fast Poisson solvers based on the Fast Fourier Transform (FFT).
  • Component w (Boundary Integral Formulation - BIM): This component addresses the "deficit" left by the CFM, specifically the jumps in the normal derivative across interfaces and boundary conditions that the CFM cannot directly handle. The authors employ a boundary integral formulation, which is well-suited for solving Laplace-type equations. However, to avoid the computational cost of evaluating the solution everywhere, they use the BIM to compute potential densities at the interfaces and boundaries. These densities are then used to define a new problem for w within the rectangular box, which can again be solved using the CFM and FFT-based solvers.

This combination allows the algorithm to leverage the strengths of both methods: the geometric flexibility and high-order accuracy of the CFM on a Cartesian grid, and the efficient handling of interface conditions by the BIM.

Datasets

Not discussed in the paper. The paper focuses on developing and testing a numerical algorithm, and the "datasets" are generated by defining specific mathematical problems with known exact solutions for validation.

Key Insights and Ideas

  • Problem Decomposition: The key insight is splitting the solution into two components (v and w) that can be handled by different, specialized numerical techniques (CFM and BIM). This avoids the limitations of using a single method for all aspects of the problem.
  • CFM for Interface Jumps: The paper demonstrates how the CFM can be extended to handle not only solution jumps but also the more complex scenario of jumps in normal derivatives, by reformulating the problem.
  • BIM for "Deficit" Problems: Using BIM to compute potential densities and then solving a "deficit" Laplace problem with CFM is an efficient way to incorporate challenging interface conditions without the full cost of a traditional BIM solution everywhere.
  • Handling Large Coefficient Ratios: The algorithm is shown to be robust even when the coefficients across an interface differ by many orders of magnitude (e.g., 10⁶), a common scenario in multiphase flows. This is achieved through a specific constraint that mitigates poor conditioning.
  • High-Order Accuracy on Cartesian Grids: The combination of CFM and FFT-based solvers allows for high-order accuracy (3rd or 4th order) on a simple, regular Cartesian grid, simplifying implementation and potentially improving efficiency compared to methods requiring complex mesh generation.

Key Results

The paper presents several 2-D examples demonstrating the algorithm's performance:

  • Accuracy: The algorithm consistently achieves high-order convergence. For interior Dirichlet problems without interfaces, 4th-order accuracy is observed. For problems with Neumann boundary conditions or discontinuity interfaces, 3rd-order accuracy is achieved, which is attributed to the need for computing derivatives.
  • Robustness: The algorithm performs well even with very large ratios of coefficients across interfaces (up to 10⁶). A specific constraint is introduced to handle poorly conditioned problems arising from these large ratios, leading to accurate solutions where unconstrained methods fail.
  • Efficiency: The use of FFT-based solvers for the CFM components and efficient BIM techniques for potential densities results in a fast algorithm. The number of iterations for Krylov solvers (GMRES/CG) used to solve the integral equations remains relatively small and independent of problem size.
  • Versatility: The algorithm successfully handles interior Dirichlet and Neumann problems, open space problems with interfaces, and combinations thereof.

Limitations and Future Work

  • 3D Extension: The authors acknowledge that extending the algorithm to 3-D is a future goal. A bottleneck in 3-D is the evaluation of integrals in the boundary integral formulation. They suggest that recent work on Kernel-Free Boundary Integral Methods (KFBIM) could provide a path to address this efficiency issue.
  • Smoothness Assumptions: While the CFM can be extended to higher orders, its accuracy depends on the smoothness of the input data (source terms, jump functions). The paper focuses on smooth interfaces for simplicity, but mentions that extensions to handle corners in interfaces are possible with appropriate BIM choices.
  • Neumann Boundary Conditions: For Neumann problems, the solution is defined only up to an arbitrary constant. The algorithm uses GMRES to pick a constant, and for consistency, a shift is applied to match a known value (e.g., at a specific point). The paper also discusses how to manage poorly conditioned problems arising in the context of Neumann conditions.

Summary updated

BibTeX
@article{marques2017high,
  author = {Alexandre Noll Marques and Jean-Christophe Nave and Rodolfo Ruben Rosales},
  title = {{High order solution of Poisson problems with piecewise constant coefficients and interface jumps}},
  journal = {Journal of Computational Physics},
  year = {2017},
  doi = {10.1016/j.jcp.2017.01.029},
}
Figure