TL;DR
This paper introduces a novel technique to simplify the imposition of jump conditions for the Correction Function Method (CFM) when dealing with interfaces that don't align with the computational grid. By reformulating the integration of jump conditions over interface sections using local coordinate transformations, the method significantly simplifies implementation, especially in 3D, while maintaining high accuracy and robustness.
What Problem Does This Paper Address?
Solving partial differential equations (PDEs) like Poisson's equation in domains with interfaces (e.g., different materials, phases) is a common challenge in science and engineering. These interfaces often lead to discontinuities in the solution or its derivatives, requiring special treatment. The Correction Function Method (CFM) is a powerful technique for handling these problems with high accuracy and compact stencils. However, a key challenge in CFM, particularly in 3D, is efficiently computing integrals over interface sections when the interface is implicitly represented (e.g., by a level set function). This paper addresses this computational bottleneck by developing a new approach to impose jump conditions that simplifies these integral evaluations.
Methods
The core of this work is a new technique for imposing jump conditions within the CFM's least-squares framework. The authors leverage two key observations:
- Flexible Integration Domains: Jump conditions can be enforced over a collection of surfaces (specifically, L∞ balls on the interface) whose union approximates the area of a diagonal plane of the correction function patch.
- Local Coordinate Transformations: The level set function is used to define local coordinate transformations that map these L∞ balls on the interface to squares. This allows for the use of standard numerical quadrature for integration.
The paper introduces a local coordinate transformation ($T_q$) that maps L∞ balls on the interface to squares. It also presents an efficient algorithm to approximate the inverse of this transformation ($T_q^{-1}$), avoiding the need to solve nonlinear systems. This new approach modifies the CFM's minimization functional to incorporate integrals over these transformed interface sections, leading to simpler and more robust computations, especially in 3D. The overall CFM implementation uses a standard 9-point stencil in 2D and a 19-point stencil in 3D for discretizing Poisson's equation, and Hermite cubic polynomials for representing the correction function, achieving fourth-order accuracy.
Datasets
Not discussed in the paper. The paper uses synthetic problems defined by analytic expressions for the interfaces and solutions to demonstrate the method's performance.
Key Insights and Ideas
- Simplifying 3D Integrals: The most significant insight is that by transforming interface sections into squares, the challenging 3D surface integrals required for jump conditions can be handled with standard, efficient numerical quadrature. This bypasses the need for complex geometric computations or specialized integration techniques for implicitly defined surfaces.
- Leveraging Level Sets for Geometry: The level set function, already used to represent the interface, is cleverly employed not just for geometry but also to define coordinate transformations that simplify integration.
- Robustness through Local Transformations: The local nature of the coordinate transformations makes the CFM more robust to the arbitrary way an interface can intersect the computational grid.
- Decoupling Interface and Solution Grids: The paper highlights that the grid used for the level set function (Gr) can be distinct from the grid used for solving the Poisson equation (Gp), offering flexibility in mesh refinement strategies.
Key Results
The paper demonstrates that the new technique for imposing jump conditions leads to an efficient, accurate, and robust implementation of the CFM. Numerical experiments in 2D and 3D show:
- Fourth-order accuracy in the solution for Poisson's equation with constant diffusion coefficients, consistent with the CFM's theoretical capabilities.
- Third-order accuracy in the gradient of the solution in several examples.
- Comparable errors to previous CFM implementations, but with significantly simplified implementation, especially in 3D.
- Robustness in handling complex interface configurations, including touching interfaces (circles and spheres) and multiple interfaces close together, without requiring special case handling.
Limitations and Future Work
- Constant Diffusion Coefficient: The current work focuses on the case where the diffusion coefficient ($\beta$) is constant. The authors acknowledge that extending this approach to problems with discontinuous $\beta$ is the subject of ongoing research.
- Grid Resolution: The accuracy of the local coordinate transformation and its inverse approximation depends on the level set grid being fine enough to "resolve" the interface. The paper mentions heuristics to check for adequate grid refinement.
- Future Work: The authors explicitly state that an extension of this approach to problems involving discontinuous $\beta$ is the subject of current work.
Summary updated