•  
  •  
 

Turkish Journal of Electrical Engineering and Computer Sciences

Author ORCID Identifier

HAYDAR SELBEŞ: 0009-0007-0334-350X

MURAT MANGUOĞLU: 0000-0002-9639-3907

Abstract

Solution of sparse linear systems of equations is often the most time-consuming operation in many large-scale science and engineering applications. For such problems, usually  classical single-level preconditioned or multilevel (inner-outer)  iterative methods are preferred. Fine-tuning inner-outer Krylov subspace methods, however, still remains as a challenge due to the problem-dependent nature of optimal solver and preconditioner configurations. Parameters such as solver stopping tolerances and preconditioner settings are often chosen by trial and error. In this work, we introduce a machine learning framework coupled with an optimization process to predict a good combination of Krylov subspace solver(s), preconditioner, and associated tolerance settings. Our approach predicts the algebraic surrogate cost of various solver-preconditioner pairs across a range of tolerance thresholds, enabling the selection of the most cost-effective configuration for a given problem instance.

DOI

10.55730/1300-0632.4292

Keywords

Krylov subspace iterative solvers, sparse matrices, machine learning, preconditioner, optimization, tolerance values

First Page

1109

Last Page

1125

Publisher

The Scientific and Technological Research Council of Türkiye (TÜBİTAK)

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Share

COinS