4.5 Article

Mixed precision low-rank approximations and their application to block low-rank LU factorization

期刊

IMA JOURNAL OF NUMERICAL ANALYSIS
卷 43, 期 4, 页码 2198-2227

出版社

OXFORD UNIV PRESS
DOI: 10.1093/imanum/drac037

关键词

numerical linear algebra; rounding error analysis; floating-point arithmetic; mixed precision algorithms; multiprecision algorithms; block low-rank matrices; data sparse matrices; LU factorization; linear systems; low-rank approximations; singular value decomposition

向作者/读者索取更多资源

This research introduces a novel approach that exploits mixed precision arithmetic for low-rank approximations, specifically in the context of block low-rank (BLR) matrices. An LU factorization algorithm is proposed to take advantage of the mixed precision representation of the blocks. Rounding error analysis demonstrates that the use of mixed precision arithmetic does not compromise the numerical stability, leading to significant reductions of storage and time costs.
We introduce a novel approach to exploit mixed precision arithmetic for low-rank approximations. Our approach is based on the observation that singular vectors associated with small singular values can be stored in lower precisions while preserving high accuracy overall. We provide an explicit criterion to determine which level of precision is needed for each singular vector. We apply this approach to block low-rank (BLR) matrices, most of whose off-diagonal blocks have low rank. We propose a new BLR LU factorization algorithm that exploits the mixed precision representation of the blocks. We carry out the rounding error analysis of this algorithm and prove that the use of mixed precision arithmetic does not compromise the numerical stability of the BLR LU factorization. Moreover, our analysis determines which level of precision is needed for each floating-point operation (flop), and therefore guides us toward an implementation that is both robust and efficient. We evaluate the potential of this new algorithm on a range of matrices coming from real-life problems in industrial and academic applications. We show that a large fraction of the entries in the LU factors and flops to perform the BLR LU factorization can be safely switched to lower precisions, leading to significant reductions of the storage and expected time costs, of up to a factor three using fp64, fp32, and bfloat16 arithmetics.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.5
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据