Combining Sparse Approximate Factorizations with Mixed-precision Iterative Refinement

Abstract

The standard LU factorization-based solution process for linear systems can be enhanced in speed or accuracy by employing mixed precision iterative refinement. Most recent work has focused on dense systems. We investigate the potential of mixed precision iterative refinement to enhance methods for sparse systems based on approximate sparse factorizations. In doing so we first develop a new error analysis for LU- and GMRES-based iterative refinement under a general model of LU factorization that accounts for the approximation methods typically used by modern sparse solvers, such as low-rank approximations or relaxed pivoting strategies. We then provide a detailed performance analysis of both the execution time and memory consumption of different algorithms, based on a selected set of iterative refinement variants and approximate sparse factorizations. Our performance study uses the multifrontal solver MUMPS, which can exploit block low-rank (BLR) factorization and static pivoting. We evaluate the performance of the algorithms on large, sparse problems coming from a variety of real-life and industrial applications showing that mixed precision iterative refinement combined with approximate sparse factorization can lead to considerable reductions of both the time and memory consumption.

TrustDL: Use of trust-based dictionary learning to facilitate recommendation in social networks

“…Soft thresholding and coordinate descent were utilized in (Permiakova & Burger, 2022) for solving the optimization problem. The sparse solver, an approach based on neural networks and involving a competition within a set of neurons to represent the input signal b leading to sparse solutions, was used in (Amestoy et al, 2022) to solve sparse coding as in Eq. (1).…”

The stability of split-preconditioned FGMRES in four precisions

“…Their numerical experiments show that they can obtain a backward error of the level of double precision. Variants of left-preconditioned GMRES using various numbers of precisions have been analyzed as inner solvers within GMRES-based iterative refinement for solving linear systems of equations; Vieublé [17] analyzed left-preconditioned GMRES in four precisions with a general preconditioner, following the earlier works [6] and [3], which analyzed left-preconditioned GMRES with an LU preconditioner in two and three precisions, respectively. In general, different precisions can be used for computing the preconditioner, matrix-vector products with A, matrix-vector products or solves with the general preconditioner(s) and the remaining computations.…”

Optimal Solutions of Well-Posed Linear Systems via Low-Precision Right-Preconditioned GMRES with Forward and Backward Stabilization

“…Hence, it is desirable to evaluate the preconditioner in lower precision. From a practical point of view, it is also quite invasive to implement higher-precision triangular solves in a well-crafted sparse solver (such as MUMPS [3,2]). These complications make GMRES-IR beyond the reach of average users.…”