CF-UAMG: A Coarse/Fine Splitting-Based Parallel Unsmoothed Aggregation Algebraic Multigrid Method
Year: 2026
Type: preprint
Abstract: In this paper, we propose a coarse/fine splitting-based parallel unsmoothed aggregation algebraic multigrid (AMG) method, referred to as CF-UAMG. The proposed method utilizes the strong-connection structure identified by classical AMG (C-AMG) coarse/fine splitting to guide the selection of aggregate roots and the formation of aggregates, aiming to improve the quality of the aggregation-based coarse space and thus the convergence of unsmoothed aggregation AMG (UA-AMG). We also prove uniform two-grid convergence of CF-UAMG with a convergence bound independent of the matrix size for symmetric positive definite M-matrices with nonnegative row sums. Numerical experiments on three classes of three-dimensional problems, including Poisson-type equations, radiation diffusion equations, and incompressible Navier-Stokes equations, demonstrate the robustness both in convergence and parallelism of CF-UAMG. Compared with UA-AMG methods employing four commonly used aggregation algorithms, CF-UAMG achieves average speedups of 1.29x, 2.04x, and 1.94x for the three classes of problems, respectively.
Source: arXiv (Cornell University)
Authors: Fan, Zhibo, Hu, Shaoliang, Huang, Silu, Xu, Xiaowen
Language: English
Topic: Matrix Theory and Algorithms
Subfield: Computational Theory and Mathematics
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Science Challenge Project
OpenAlex ID: W7226012822
