Fast randomized least-squares solvers can be just as accurate and stable as classical direct solvers

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Epperly, Ethan N., Meier, Maike, Nakatsukasa, Yuji
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866913998327250944
author Epperly, Ethan N.
Meier, Maike
Nakatsukasa, Yuji
author_facet Epperly, Ethan N.
Meier, Maike
Nakatsukasa, Yuji
contents One of the greatest success stories of randomized algorithms for linear algebra has been the development of fast, randomized algorithms for highly overdetermined linear least-squares problems. However, none of the existing algorithms is backward stable, preventing them from being deployed as drop-in replacements for existing QR-based solvers. This paper introduces sketch-and-precondition with iterative refinement (SPIR) and FOSSILS, two provably backward stable randomized least-squares solvers. SPIR and FOSSILS combine iterative refinement with a preconditioned iterative method applied to the normal equations and converge at the same rate as existing randomized least-squares solvers. This work offers the promise of incorporating randomized least-squares solvers into existing software libraries while maintaining the same level of accuracy and stability as classical solvers.
format Preprint
id arxiv_https___arxiv_org_abs_2406_03468
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Fast randomized least-squares solvers can be just as accurate and stable as classical direct solvers
Epperly, Ethan N.
Meier, Maike
Nakatsukasa, Yuji
Numerical Analysis
65F10, 65F20, 65G50, 65K10, 68W20
One of the greatest success stories of randomized algorithms for linear algebra has been the development of fast, randomized algorithms for highly overdetermined linear least-squares problems. However, none of the existing algorithms is backward stable, preventing them from being deployed as drop-in replacements for existing QR-based solvers. This paper introduces sketch-and-precondition with iterative refinement (SPIR) and FOSSILS, two provably backward stable randomized least-squares solvers. SPIR and FOSSILS combine iterative refinement with a preconditioned iterative method applied to the normal equations and converge at the same rate as existing randomized least-squares solvers. This work offers the promise of incorporating randomized least-squares solvers into existing software libraries while maintaining the same level of accuracy and stability as classical solvers.
title Fast randomized least-squares solvers can be just as accurate and stable as classical direct solvers
topic Numerical Analysis
65F10, 65F20, 65G50, 65K10, 68W20
url https://arxiv.org/abs/2406.03468