A mixed precision LOBPCG algorithm

DOI10.1007/S11075-023-01550-9arXiv2302.12528MaRDI QIDQ6141529FDOQ6141529

Meiyue Shao, Yuxin Ma, Daniel Kressner

Publication date: 19 December 2023

Published in: Numerical Algorithms (Search for Journal in Brave)

Abstract: The locally optimal block preconditioned conjugate gradient (LOBPCG) algorithm is a popular approach for computing a few smallest eigenvalues and the corresponding eigenvectors of a large Hermitian positive definite matrix A. In this work, we propose a mixed precision variant of LOBPCG that uses a (sparse) Cholesky factorization of A computed in reduced precision as the preconditioner. To further enhance performance, a mixed precision orthogonalization strategy is proposed. To analyze the impact of reducing precision in the preconditioner on performance, we carry out a rounding error and convergence analysis of PINVIT, a simplified variant of LOBPCG. Our theoretical results predict and our numerical experiments confirm that the impact on convergence remains marginal. In practice, our mixed precision LOBPCG algorithm typically reduces the computation time by a factor of 1.4--2.0 on both CPUs and GPUs.

Full work available at URL: https://arxiv.org/abs/2302.12528

zbMATH Keywords

symmetric eigenvalue problemLOBPCG algorithmmixed precision algorithm

Mathematics Subject Classification ID

Computational methods for sparse matrices (65F50) Numerical computation of eigenvalues and eigenvectors of matrices (65F15) Preconditioners for iterative methods (65F08) Numerical analysis (65-XX)

Cites Work

• Algorithm 589: SICEDR  : A FORTRAN Subroutine for Improving the Accuracy of Computed Matrix Eigenvalues
• Toward the optimal preconditioned eigensolver: Locally optimal block preconditioned conjugate gradient method
• Title not available (Why is that?)
• Numerical Methods for Large Eigenvalue Problems
• Accuracy and Stability of Numerical Algorithms
• Perturbed preconditioned inverse iteration for operator eigenvalue problems with applications to adaptive wavelet discretization
• Basis selection in LOBPCG
• A geometric theory for preconditioned inverse iteration applied to a subspace
• Iterative refinement for symmetric eigenvalue decomposition
• Mixed precision algorithms in numerical linear algebra
• Convergence theory for preconditioned eigenvalue solvers in a nutshell
• Accelerating the Solution of Linear Systems by Iterative Refinement in Three Precisions
• Mixed-Precision Cholesky QR Factorization and Its Case Studies on Multicore CPU with Multiple GPUs
• Iterative refinement for symmetric eigenvalue decomposition. II. Clustered eigenvalues
• A Robust and Efficient Implementation of LOBPCG
• Iterative refinement for singular value decomposition based on matrix multiplication
• Iterative refinement of Schur decompositions

This page was built for publication: A mixed precision LOBPCG algorithm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6141529)