GMT : A Geometric Multigrid Transformer Solver for Microstructure Homogenization

Yu Xing, Yang Liu, Tianyang Xue, Lin Lu · 2026

Read GMT : A Geometric Multigrid Transformer Solver for Microstructure Homogenization by Yu Xing, Yang Liu, Tianyang Xue, Lin Lu (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-2604-26518.txt.

Description

Xing and three colleagues built GMT, a geometric multigrid transformer solver for lattice metamaterial homogenization. Neural stand-ins are fast but often lack engineering-grade accuracy and stability, so GMT mirrors geometric multigrid by running a point transformer across sparse multigrid levels. Physics-aware position codes enforce repeating patterns, and guesses of finest-level answers plus multi-level fixes need only one multigrid V-cycle to converge. Errors reached ten to the minus five with 160 times speedup over leading GPU solvers at equal accuracy.

Catalog record

CREATORYu Xing, Yang Liu, Tianyang Xue, Lin Lu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE103567
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.26518

Browse the full library · About the Tower of Babel