Faster Simulation Codes through Data-Driven Algebraic Multigrid
Li, Rui Peng, Petrides, Socratis, Thopalli, Kowshik et al. · 2026
Read Faster Simulation Codes through Data-Driven Algebraic Multigrid by Li, Rui Peng, Petrides, Socratis, Thopalli, Kowshik et al. (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/doe-osti-17019158-faster-simulation-codes-through-data-dri.txt.
Description
Faster simulation is the goal of this abbreviated final report from Lawrence Livermore, dated September 2026. Rui Peng Li, Socratis Petrides, Kowshik Thopalli, and Jin Yao combined machine learning with algebraic multigrid. That method is a leading solver for the huge linear equation systems that eat most run time in physics codes. Tuning it by hand for each new problem is hard.
The team taught software to learn coarse level operators, smoothing steps, and coarse variable choices, and built a shared test environment called AMGLearn. In sample tests, learned coarse operators cut coarse level complexity by at least 44 percent without hurting solver strength. A side study also cut error in a mesh free shock wave method by tuning its kernel and starting layout.
Catalog record
| CREATOR | Li, Rui Peng, Petrides, Socratis, Thopalli, Kowshik et al. |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 7 KB |
| LICENSE | Public Domain |
| RIGHTS | PUBLIC DOMAIN |
| SOURCE | DOE OSTI |