arXiv Science⌕ Search
Home/Archive/arXiv

arXiv · 2610.13905

Fast and Stable Arbitrary Precision Approximate Matrix Multiplication

Ioana Dumitriu·Olga Holtz·Abraham Hsu·Nicholas Liskij·Yoav Moran·Ryan Schneider·Oded Schwartz·Shuncheng Yuan
View original source ↗Read source document ↗Export metadata ↓

Abstract

We present several new and surprising results on the theory and practice of Arbitrary Precision Approximate (APA) matrix multiplication, which refers to algorithms that output merely approximations to matrix multiplication even in exact arithmetic. While APA algorithms have been around for nearly half a century, and despite the fact that they have potentially lower (asymptotic) cost than many fast exact matrix multiplication schemes, they have not been studied as closely, to the extent that relatively little is known about their numerical properties. In this paper, we shine a spotlight on these algorithms by presenting three important results. First, we provide a new search algorithm for APA routines that is easy to use, automatically generates encoding matrices with functional entries, and can match or beat the best known APA algorithms for several sizes. In particular, we improve the upper bound on the border rank of $\langle 2,4,5 \rangle$ matrix multiplication. Second, we combine this insight with alternative bases to minimize the number of additions in an APA algorithm, thereby improving its leading coefficient. Finally, we prove the same kind of stability that characterizes fast exact matrix multiplication for an algorithm of Bini [Calcolo 1980] that computes an exact product of two matrices using a modified APA scheme (with or without alternative bases).

Explore related subjects

math.NAcs.NA

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Ioana Dumitriu, Olga Holtz, Abraham Hsu, Nicholas Liskij, Yoav Moran, Ryan Schneider, Oded Schwartz, Shuncheng Yuan. 2026-10-09. Fast and Stable Arbitrary Precision Approximate Matrix Multiplication. https://arxiv.org/abs/2610.13905

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

High order integration of stochastic dynamics on Riemannian manifolds with frozen flow methods

We present a new class of numerical methods for solving stochastic differential equations with additive noise on general Riemannian manifolds with high weak order of accuracy. In opposition to the popular approach with projection methods, the proposed methods are intrinsic: they only rely on geometric operations and avoid coordinates and embeddings. We provide a robust and general convergence analysis and an algebraic formalism of exotic planar Butcher series for the computation of order conditions at any high order. To illustrate the methodology, an explicit method of second weak order is introduced, and several numerical experiments confirm the theoretical findings and extend the approach for the sampling of the invariant measure of Riemannian Langevin dynamics.

math.NA↗
arXiv2026

Convergence analysis of GMRES applied to Helmholtz problems near resonances

The finite element solution of Helmholtz problems near resonant or quasi-resonant frequencies poses significant challenges, as iterative solvers typically suffer from severely degraded convergence. We analyze the convergence behavior of GMRES applied to linear systems arising from such configurations. Theoretical convergence estimates are derived based on harmonic Ritz values, highlighting their proximity to small eigenvalues as a key determining factor. We further examine deflation strategies and their interplay with preconditioning techniques, using the Complex Shifted Laplacian preconditioner as a case study. Numerical experiments on resonant and quasi-resonant test cases validate the theoretical framework and demonstrate the effectiveness of deflation strategies. This study provides new insights and practical guidance for analyzing and improving iterative solvers for time-harmonic problems near resonances.

math.NA↗
arXiv2026

A general framework for the OSRC-preconditioned EFIE in computational electromagnetics

We investigate on-surface radiation condition (OSRC) preconditioners for the electric field integral equation (EFIE), considering Electric-to-Magnetic (EtM) and Magnetic-to-Electric (MtE) approximations applied from both the left and the right. A pseudodifferential analysis establishes the well-posedness of the four resulting formulations and shows that, locally, the OSRC EtM and MtE approximations present a similar structure to the EFIE itself and its inverse, respectively. At the discrete level, both MtE formulations are compatible with the standard RWG/SNC discretization and can be implemented without barycentric refinement, whereas the EtM formulations require an appropriate discrete duality map and BC/RBC spaces. Numerical experiments show that the OSRC preconditioners substantially improve GMRES convergence compared with the unpreconditioned EFIE, with the refinement-free MtE formulations also providing the lowest computational cost among the formulations considered.

math.NA↗