pipette
ENEnglish

Randomized algorithms in numerical linear algebra for scientific computing

Ethan N. W. Epperly, Daniel Kressner, Yuji Nakatsukasa

Preprint

En palabras de los autores

Randomized numerical linear algebra (RNLA) has fundamentally changed the landscape of scientific computing over the last twenty to thirty years. While early breakthroughs focused primarily on low-rank approximation and least-squares problems, randomized techniques are now used to solve a broad range of linear algebra problems, supporting applications spanning data science, optimization, and computational engineering. This paper surveys some of the main ideas of RNLA, focusing on methods that are effective in practice and on simple theoretical bounds that can be used to understand them. We highlight the core benefits of randomization: improved scalability, algorithmic simplicity, and---contrary to what the name may suggest---remarkable reliability and numerical stability. Topics include sketching, low-rank approximation, least squares, orthogonalization and approximate orthogonalization, randomized Krylov subspace methods, preconditioning, trace and spectral density estimation, row and column selection algorithms, posterior error estimation, randomized regularization, and tensor-structured randomness.

Resultado principalEl resumen no menciona limitaciones.

Apareció: viernes, 30 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: 186 pages; to appear in Acta Numerica