Optimal Polynomial Smoothers for Parallel AMG

In this paper, we propose some Chebyshev polynomials of the 1st-kind which produce optimal bound for a polynomial dependent constant involved in the AMG V-cycle error bound and do not require information about the spectrum of matrices. We formulate a variant of a minimax problem already proposed in [J. Lottes, Optimal polynomial smoothers for multigrid V-cycles, Numer. Lin. Alg. with Appl., 30 (2023), p. e2518, https://doi.org/10.1002/nla.2518] and define Chebyshev polynomial of the 1st-kind as acceleration for a weighted-Jacobi smoother; we also describe efficient GPU kernels for the application of the polynomial smoother and compare results with accelerators defined in [J. Lottes, Optimal polynomial smoothers for multigrid V-cycles, Numer. Lin. Alg. with Appl., 30 (2023), p. e2518, https://doi.org/10.1002/nla.2518] on usual benchmarks at very large scales.