Sparse Approximate Multifrontal Factorization with Butterfly Compression for High Frequency Wave Equations

Yang Liu,Pieter Ghysels,Lisa Claus,Xiaoye Sherry Li
DOI: https://doi.org/10.1137/20M1349667
2021-10-19
Abstract:We present a fast and approximate multifrontal solver for large-scale sparse linear systems arising from finite-difference, finite-volume or finite-element discretization of high-frequency wave equations. The proposed solver leverages the butterfly algorithm and its hierarchical matrix extension for compressing and factorizing large frontal matrices via graph-distance guided entry evaluation or randomized matrix-vector multiplication-based schemes. Complexity analysis and numerical experiments demonstrate $\mathcal{O}(N\log^2 N)$ computation and $\mathcal{O}(N)$ memory complexity when applied to an $N\times N$ sparse system arising from 3D high-frequency Helmholtz and Maxwell problems.
Mathematical Software,Computational Engineering, Finance, and Science
What problem does this paper attempt to address?