Error Analysis of Matrix Multiplication Emulation Using Ozaki-II Scheme Journal: Nuclear Physics B

Yuki Uchino, Katsuhisa Ozaki, Toshiyuki Imamura · 2026

Read Error Analysis of Matrix Multiplication Emulation Using Ozaki-II Scheme Journal: Nuclear Physics B by Yuki Uchino, Katsuhisa Ozaki, Toshiyuki Imamura (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02549.txt.

Description

Yuki Uchino and 2 co-authors deliver a strict error analysis for a scheme that emulates high-precision matrix multiplication on fast low-precision hardware, in a 2026 arXiv paper. The Ozaki-II scheme uses the Chinese remainder theorem to rebuild exact products from a sequence of low-precision multiplications. More rounds buy a closer match.

Earlier studies had shown single- and double-precision emulation beating standard library routines on AI hardware with fast eight-bit units. Quality degrades, however, when input exponents spread widely, forcing many rounds. The new analysis is exact rather than based on trials. It shows exactly how quality behaves as rounds build up. It also estimates in advance how many low-precision multiplications a target quality level requires. That turns a lab trick into an engineering tool with guarantees one can plan on, for scientific computing on modern hardware.

Catalog record

CREATORYuki Uchino, Katsuhisa Ozaki, Toshiyuki Imamura
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02549

Browse the full library · About the Tower of Babel