Mixed Precision General Alternating-Direction Implicit Method for Solving Large Sparse Linear Systems
Jifeng Ge, Bastien Vieublé, Juan Zhang* · 2025
Read Mixed Precision General Alternating-Direction Implicit Method for Solving Large Sparse Linear Systems by Jifeng Ge, Bastien Vieublé, Juan Zhang* (2025) 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-2512-21164.txt.
Description
Jifeng Ge, Bastien Vieuble, and Juan Zhang split the arithmetic itself to solve huge sparse linear systems faster. Their method builds on the alternating-direction implicit family, which splits a matrix so the full problem becomes a chain of easy structured subsystems. In their three-precision version, the subsystems are solved in cheap low precision. Residuals and solution updates are computed in high precision to protect accuracy. Modern graphics hardware rewards that division of labor.
A rounding-error analysis pins down how far forward and backward errors converge and which splitting matrices guarantee success for a given precision mix. A Gaussian-process model trained on small problems picks the regularization parameter. On an A100 graphics card, problems reaching 130 million unknowns run between two and three times faster than full double precision, across convection-diffusion and reaction-diffusion tests. Simulation pipelines gain speed without surrendering accuracy guarantees.
Catalog record
| CREATOR | Jifeng Ge, Bastien Vieublé, Juan Zhang* |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-21164