Randomized Rank-Structured Matrix Compression by Tagging

Abstract: In this work, we present novel randomized compression algorithms for flat rank-structured matrices with shared bases, known as uniform Block Low-Rank (BLR) matrices. Our main contribution is a technique called tagging, which improves upon the efficiency of basis matrix computation while preserving accuracy compared to alternative methods. Tagging operates on the matrix using matrix-vector products of the matrix and its adjoint, making it particularly advantageous in scenarios where accessing individual matrix entries is computationally expensive or infeasible. Flat rank-structured formats use subblock sizes that asymptotically scale with the matrix size to ensure competitive complexities for linear algebraic operations, making alternative methods prohibitively expensive in such scenarios. In contrast, tagging reconstructs basis matrices using a constant number of matrix-vector products followed by linear post-processing, with the constants determined by the rank parameter and the problem's underlying geometric properties. We provide a detailed analysis of the asymptotic complexity of tagging, demonstrating its ability to significantly reduce computational costs without sacrificing accuracy. We also establish a theoretical connection between the optimal construction of tagging matrices and projective varieties in algebraic geometry, suggesting a hybrid numeric-symbolic avenue of future work. To validate our approach, we apply tagging to compress uniform BLR matrices arising from the discretization of integral and partial differential equations. Empirical results show that tagging outperforms alternative compression techniques, significantly reducing both the number of required matrix-vector products and overall computational time. These findings highlight the practicality and scalability of tagging as an efficient method for flat rank-structured matrices in scientific computing.