Iterative refinement (iterative improvement) to enhance solutions of direct factorizations

MaRDI QIDQ6825845FDO

After solving Ax=b via a direct factorization, compute the residual r=b−Ax in higher precision, solve Ad=r, and correct x←x+d; repeat until convergence. Modern variants enable mixed-precision workflows that recover high-precision accuracy at the cost

Described by source

• Iterative refinement for linear systems and LAPACK
• LAPACK Users' Guide

Implemented by

• MAGMA — GPU-accelerated dense linear algebra

This page was built for algorithm: Iterative refinement (iterative improvement) to enhance solutions of direct factorizations