A stable one-synchronization variant of reorthogonalized block classical Gram--Schmidt
Abstract
The block classical Gram--Schmidt (BCGS) algorithm and its reorthogonalized variant are widely-used methods for computing the economic QR factorization of block columns due to their lower communication cost compared to other approaches such as modified Gram--Schmidt and Householder QR. To further reduce communication, i.e., synchronization, there has been a long ongoing search for a variant of reorthogonalized BCGS variant that achieves loss of orthogonality while requiring only \emph{one} synchronization point per block column, where represents the unit roundoff. Utilizing Pythagorean inner products and delayed normalization techniques, we propose the first provably stable one-synchronization reorthogonalized BCGS variant, demonstrating that it has loss of orthogonality under the condition , where represents the condition number. By incorporating one additional synchronization point, we develop a two-synchronization reorthogonalized BCGS variant which maintains loss of orthogonality under the improved condition . An adaptive strategy is then proposed to combine these two variants, ensuring loss of orthogonality while using as few synchronization points as possible under the less restrictive condition . As an example of where this adaptive approach is beneficial, we show that using the adaptive orthogonalization variant, -step GMRES achieves a backward error comparable to -step GMRES with BCGSI+, also known as BCGS2, both theoretically and numerically, but requires fewer synchronization points.
Cite
@article{arxiv.2411.07077,
title = {A stable one-synchronization variant of reorthogonalized block classical Gram--Schmidt},
author = {Erin Carson and Yuxin Ma},
journal= {arXiv preprint arXiv:2411.07077},
year = {2025}
}
Comments
25 pages, 8. figures