Universal AMG accelerated embedded boundary method without small cell stiffness (Q6057148)

The document titled ``Universal AMG accelerated embedded boundary method without small cell stiffness'' by Zhichao Peng, Daniel Appelö, and Shuang Liu presents a significant advancement in numerical methods for solving partial differential equations (PDEs) with complex geometries. This paper addresses a critical challenge in the field of scientific computing: efficiently solving wave, heat, and Poisson equations in domains with intricate boundaries without encountering the common issue of small cell stiffness. The authors develop an embedded boundary finite difference method that yields a symmetric positive definite linear system. This advancement is crucial as it facilitates the use of the conjugate gradient method accelerated by algebraic multigrid techniques, ensuring efficient and stable computations across various applications, including shape optimization problems and simulations involving moving boundaries like the Stefan problem, Navier-Stokes equations, and Grad-Shafranov equations. The method is based on the principle of modifying the stencils near the boundary to incorporate boundary conditions directly, avoiding the generation of small cells that lead to numerical stiffness. This approach is differentiated from others by interpolating to interior boundary points instead of exterior ghost points, which significantly mitigates the small-cell stiffness problem. For convex and non-convex geometries where traditional line-by-line interpolation fails, the authors propose a novel combination of polynomial and radial basis function interpolation that maintains the symmetry and positive definiteness of the system matrix. Through extensive numerical experiments, including comparisons with existing finite element methods, the paper demonstrates the accuracy, stability, and efficiency of the proposed method. The method's universality is highlighted by its applicability to a broad range of PDEs and complex geometrical domains, making it a valuable tool for scientific and engineering simulations. This research contributes significantly to the field of scientific computing by providing a robust, efficient, and universally applicable method for solving PDEs in complex domains. Its ability to avoid small cell stiffness and maintain the system's symmetric positive definiteness opens new possibilities for fast and stable simulations across various applications, marking a notable advancement in numerical methods for scientific computing.