Id: arxiv/2611.19687 Url: https://arxiv.org/abs/2611.19687 Tags: math.NA Score: 0 Title: Adaptive Precision Cholesky with Error Guarantees Domain: math.NA Source: arxiv Authors: Sudhanva Kulkarni, James W. Demmel, Laura Grigori Reviews: 0 Subject: math.NA Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-11-15 Page: https://delphi.science/r/arxiv/2611.19687 Modern accelerators increasingly support low-precision formats such as 8-bit floats and integers to speed up machine learning. We show that a Cholesky factorization in which both GEMM and TRSM are implemented using the fixed-point Ozaki 1 scheme satisfies the same backward error bounds as the standard floating-point Cholesky factorization, and that the same holds for a Cholesky factorization computed using a low-precision MicroScaled GEMM. We then use our error analyses to construct an adaptive precision Cholesky factorization that exploits low-precision formats while respecting a prescribed…