Abstract
This paper provides the first rigorous mathematical proof for the accuracy of the vertex-centered "MACH-like" finite volume method (FVM) for solving diffusion problems with sharp material interfaces. While the MACH method has been a staple in engineering for decades, its global convergence properties remained unproven. By introducing novel analytical techniques, including dimension reduction via discrete sine transforms, the authors demonstrate that this method achieves a near-second-order global error estimate of $O(h^2 |\ln h|)$ in the maximum norm on fixed Eulerian grids. These findings bridge a significant gap between engineering practice and mathematical theory.
Introduction
In computational physics, simulating diffusion across material interfaces is critical for fields such as plasma physics and fluid-solid mechanics. Researchers typically choose between two grid paradigms: Lagrangian grids, which move with the material but risk mesh tangling, and Eulerian grids, which remain fixed. While Eulerian grids offer stability, they present challenges in accurately capturing the physics within cells intersected by a material interface.
The "MACH" (Multiblock Arbitrary Coordinate Hydromagnetic) method was developed in the late 1980s to address these challenges. Despite its empirical success, a strict mathematical guarantee of its global error estimate—how the error diminishes as the grid is refined—was previously unavailable. This research provides that missing theoretical link.
Methodology: Vertex-Centered Discretization
The authors analyze a vertex-centered MACH-like FVM. Unlike standard cell-centered methods, this approach focuses on the vertices (corners) of the grid to define control volumes.
As shown in Figure 1, the control volumes ($Q_k$) are constructed around each vertex $(i, j)$. When an interface passes through these cells (Figure 1b), the method must account for the discontinuity in material properties. A key technical innovation is the use of a diagonal "$\\times$" stencil rather than the traditional "+" stencil used in most finite volume schemes.
This "$\\times$" stencil (Figure 3a) utilizes diagonal neighbors, which introduces unique mathematical challenges for error estimation. To handle the material interface, the authors incorporate physical "jump conditions"—the specific ways the solution and its flux change across the boundary—directly into Taylor series expansions. Furthermore, they highlight that using a harmonic average for the diffusion coefficient ($\kappa$) within multi-material cells is essential for maintaining high-order accuracy.
Theoretical Error Analysis
The core contribution of this work is the derivation of local and global error estimates.
- Local Truncation Error: The authors prove that near the interface, the local error is generally $O(h)$. However, under "Assumption I"—a condition where the second derivative of the solution along the interface tangent is zero—this improves to $O(h^2)$.
- Global Convergence: By applying discrete sine transforms to reduce the 2D error equations into 1D versions, the authors prove that the global error in the maximum norm is $O(h^2 |\ln h|)$. This confirms that the method is nearly second-order accurate even in the presence of strong discontinuities.
Numerical Validation
The theoretical proofs were verified using synthetic numerical benchmarks. In a "stress test" where ideal assumptions were not met, the method still maintained a robust performance floor. In cases satisfying the authors' optimal conditions, numerical experiments confirmed the $O(h^2 |\ln h|)$ convergence rate: when the grid size was halved, the error decreased by a factor of nearly four, matching the theoretical prediction.
Conclusion
This research validates the MACH-like finite volume method as a high-precision tool for scientific computing. By proving its near-second-order convergence on Eulerian grids, the authors have provided the mathematical foundation necessary for its continued use in complex multi-material simulations. Future work will look to extend these rigorous estimates to 3D geometries, curved interfaces, and time-dependent diffusion problems.
Summary updated