EUVIMEDEUVIMED entdecken
Crossref

Mixed precision low-rank approximations and their application to block low-rank LU factorization

Patrick Amestoy, Olivier Boiteau, Alfredo Buttari, Matthieu Gerest, Fabienne Jézéquel, Jean-Yves L’Excellent, Theo Mary

IMA Journal of Numerical Analysis · 03.08.2023 · 43 (4) · 2198-2227

DOI: 10.1093/imanum/drac037 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IMA Journal of Numerical Analysis, online erschienen am 23.08.2022. Verfasst von Patrick Amestoy und 6 weiteren Autor:innen. Die Arbeit wurde bisher 24-mal zitiert (Crossref) und nennt 35 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We introduce a novel approach to exploit mixed precision arithmetic for low-rank approximations. Our approach is based on the observation that singular vectors associated with small singular values can be stored in lower precisions while preserving high accuracy overall. We provide an explicit criterion to determine which level of precision is needed for each singular vector. We apply this approach to block low-rank (BLR) matrices, most of whose off-diagonal blocks have low rank. We propose a new BLR LU factorization algorithm that exploits the mixed precision representation of the blocks. We carry out the rounding error analysis of this algorithm and prove that the use of mixed precision arithmetic does not compromise the numerical stability of the BLR LU factorization. Moreover, our analysis determines which level of precision is needed for each floating-point operation (flop), and therefore guides us toward an implementation that is both robust and efficient. We evaluate the potential of this new algorithm on a range of matrices coming from real-life problems in industrial and academic applications. We show that a large fraction of the entries in the LU factors and flops to perform the BLR LU factorization can be safely switched to lower precisions, leading to significant reductions of the storage and expected time costs, of up to a factor three using fp64, fp32, and bfloat16 arithmetics.

Zitieren

APA

Amestoy, P., Boiteau, O., Buttari, A., Gerest, M., Jézéquel, F., L’Excellent, J., & Mary, T. (2023). Mixed precision low-rank approximations and their application to block low-rank LU factorization. IMA Journal of Numerical Analysis, 43(4), 2198-2227. https://doi.org/10.1093/imanum/drac037

Vancouver

Amestoy P, Boiteau O, Buttari A, Gerest M, Jézéquel F, L’Excellent J, et al. Mixed precision low-rank approximations and their application to block low-rank LU factorization. IMA Journal of Numerical Analysis. 2023;43(4):2198-2227. doi:10.1093/imanum/drac037

BibTeX
@article{Amestoy2023,
  author = {Amestoy, Patrick and Boiteau, Olivier and Buttari, Alfredo and Gerest, Matthieu and Jézéquel, Fabienne and L’Excellent, Jean-Yves and Mary, Theo},
  title = {Mixed precision low-rank approximations and their application to block low-rank LU factorization},
  journal = {IMA Journal of Numerical Analysis},
  year = {2023},
  volume = {43},
  number = {4},
  pages = {2198--2227},
  doi = {10.1093/imanum/drac037}
}
Kostenlos starten →