Hybrid Digital-Analog Approximate Inverse Preconditioning for Krylov Methods

DOI10.48550/arXiv.2606.17227MaRDI QIDQ7269507zbMATH OpenarXivFDO

Authors Shikhar Shah, Ruipeng Li, Tayfun Gokmen, Vassilis Kalantzis, Lior Horesh, Yuanzhe Xi, Rui Peng Liu, Rui Peng Li

Publication date 2026

Published in arXiv

Tldr We propose a novel hybrid digital-analog preconditioning strategy using analog in-memory computing for Krylov methods, where preconditioner applications are performed through analog crossbar matrix-vector multiplications while the precision-sensitive operations are performed digitally. The key insight is that stronger digital preconditioners may be less effective after accounting for analog non-idealities.

Full work available at URL https://doi.org/10.48550/arXiv.2606.17227

arXiv classification

math.NA

This page was built for publication: Hybrid Digital-Analog Approximate Inverse Preconditioning for Krylov Methods

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7269507)