We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.NA

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Mathematics > Numerical Analysis

Title: Thick-restarted joint Lanczos bidiagonalization for the GSVD

Abstract: The computation of the partial generalized singular value decomposition (GSVD) of large-scale matrix pairs can be approached by means of iterative methods based on expanding subspaces, particularly Krylov subspaces. We consider the joint Lanczos bidiagonalization method, and analyze the feasibility of adapting the thick restart technique that is being used successfully in the context of other linear algebra problems. Numerical experiments illustrate the effectiveness of the proposed method. We also compare the new method with an alternative solution via equivalent eigenvalue problems, considering accuracy as well as computational performance. The analysis is done using a parallel implementation in the SLEPc library.
Subjects: Numerical Analysis (math.NA); Mathematical Software (cs.MS)
Cite as: arXiv:2206.03768 [math.NA]
  (or arXiv:2206.03768v3 [math.NA] for this version)

Submission history

From: Jose E. Roman [view email]
[v1] Wed, 8 Jun 2022 09:31:06 GMT (31kb)
[v2] Sun, 25 Dec 2022 10:22:48 GMT (32kb)
[v3] Fri, 12 May 2023 16:02:14 GMT (40kb)

Link back to: arXiv, form interface, contact.