A GPU / CPU faster Block Arnoldi Method for Solving Large-Scale Lyapunov Equation

Articles in Press

Document Type : Research Article

Author

ENSA OUJDA

Abstract

Krylov methods have proven effective in solving large-scale matrix equations with sparse coefficients, in particular through the use of the extended Arnoldi process, which involves the inverse of the matrix coefficients in the projection subspace. This approach has significantly reduced the time and number of iterations needed to find a suitable solution. In this paper, we are interested on solving the low-rank Lyapunov equation whether in the continuous or discrete case. We propose to enhance the convergence time by modifying the Block Arnoldi process so that the Krylov projection subspace contains additional blocks from the inverse of the square coefficient of this equation. Our intention is to benefit from these additional informations, similar to the extended Arnoldi version, without incorporating them at each iteration, thus preventing any impact on the convergence speed. To confirm the effectiveness of the proposed method, some numerical results obtained using CPU and GPU implementations are provided.

Keywords

Main Subjects

Journal of Mathematical Modeling

Articles in Press, Corrected Proof
Available Online from 18 April 2026

Files

Share

How to cite

Statistics