Speaker
Xiaobo Liu
Description
We investigate the use of mixed precisions in an iterative refinement framework for solving low-rank Lyapunov matrix equations, where the sign function method with Newton iteration is used as the solver. We also discuss the possibility of using mixed precisions within the Newton iteration itself and propose a mixed-precision truncation scheme based on the QR factorization with column pivoting. Numerical experiments are presented to verify the accuracy of our mixed-precision algorithms, and accordingly, we also discuss the potential speedup based on current and future hardware.
