Lanczos with compression for symmetric eigenvalue problems
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866908850526879744 |
|---|---|
| author | Casulli, Angelo A. Kressner, Daniel Shao, Nian |
| author_facet | Casulli, Angelo A. Kressner, Daniel Shao, Nian |
| contents | The Lanczos method with implicit restarting is one of the most popular methods for finding a few exterior eigenpairs of a large symmetric matrix $A$. Usually based on polynomial filtering, restarting is crucial to limit memory and the cost of orthogonalization. In this work, we propose a novel strategy for the same purpose, called Lanczos with compression. Unlike polynomial filtering, our approach compresses the Krylov subspace using rational approximation and, in doing so, it sacrifices the structure of the associated Krylov decomposition. Nevertheless, it remains compatible with subsequent Lanczos steps and the overall algorithm is still solely based on matrix-vector products with $A$. On the theoretical side, we show that compression introduces only a small error compared to standard (unrestarted) Lanczos and therefore has only a negligible impact on convergence. Comparable guarantees are not available for commonly used implicit restarting strategies, including the Krylov--Schur method. On the practical side, our numerical experiments demonstrate that compression often outperforms the Krylov--Schur method in terms of matrix-vector products. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_20948 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Lanczos with compression for symmetric eigenvalue problems Casulli, Angelo A. Kressner, Daniel Shao, Nian Numerical Analysis 65F15 The Lanczos method with implicit restarting is one of the most popular methods for finding a few exterior eigenpairs of a large symmetric matrix $A$. Usually based on polynomial filtering, restarting is crucial to limit memory and the cost of orthogonalization. In this work, we propose a novel strategy for the same purpose, called Lanczos with compression. Unlike polynomial filtering, our approach compresses the Krylov subspace using rational approximation and, in doing so, it sacrifices the structure of the associated Krylov decomposition. Nevertheless, it remains compatible with subsequent Lanczos steps and the overall algorithm is still solely based on matrix-vector products with $A$. On the theoretical side, we show that compression introduces only a small error compared to standard (unrestarted) Lanczos and therefore has only a negligible impact on convergence. Comparable guarantees are not available for commonly used implicit restarting strategies, including the Krylov--Schur method. On the practical side, our numerical experiments demonstrate that compression often outperforms the Krylov--Schur method in terms of matrix-vector products. |
| title | Lanczos with compression for symmetric eigenvalue problems |
| topic | Numerical Analysis 65F15 |
| url | https://arxiv.org/abs/2602.20948 |