English

Randomized biorthogonalization through a two-sided Gram-Schmidt process

Numerical Analysis 2025-09-05 v1 Numerical Analysis

Abstract

We propose and analyze a randomized two-sided Gram-Schmidt process for the biorthogonalization of two given matrices X,YRn×mX, Y \in\mathbb{R}^{n\times m}. The algorithm aims to find two matrices Q,PRn×mQ, P \in\mathbb{R}^{n\times m} such that range(X)=range(Q){\rm range}(X) = {\rm range}(Q), range(Y)=range(P){\rm range}(Y) = {\rm range}(P) and (ΩQ)TΩP=I(\Omega Q)^T \Omega P = I, where ΩRs×n\Omega \in\mathbb{R}^{s \times n} is a sketching matrix satisfying an oblivious subspace ε\varepsilon-embedding property; in other words, the biorthogonality condition on the columns of QQ and PP is replaced by an equivalent condition on their sketches. This randomized approach is computationally less expensive than the classical two-sided Gram-Schmidt process, has better numerical stability, and the condition number of the computed bases Q,PQ, P is often smaller than in the deterministic case. Several different implementations of the randomized algorithm are analyzed and compared numerically. The randomized two-sided Gram-Schmidt process is applied to the nonsymmetric Lancozs algorithm for the approximation of eigenvalues and both left and right eigenvectors.

Cite

@article{arxiv.2509.04386,
  title  = {Randomized biorthogonalization through a two-sided Gram-Schmidt process},
  author = {Laura Grigori and Lorenzo Piccinini and Igor Simunec},
  journal= {arXiv preprint arXiv:2509.04386},
  year   = {2025}
}

Comments

26 pages, 5 figures, 3 tables

R2 v1 2026-07-01T05:21:33.644Z