Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries
Hang Zhou, Haixu Wu, Haonan Shangguan, Yuezhou Ma, Huikun Weng, Jianmin Wang, Mingsheng Long · 2026
Read Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries by Hang Zhou, Haixu Wu, Haonan Shangguan, Yuezhou Ma, Huikun Weng, Jianmin Wang, Mingsheng Long (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-2602-04940.txt.
Description
Hang Zhou and 6 co-authors wrote this 2026 paper on neural PDE solvers, programs that learn to imitate physics simulations. Engineers want these solvers for huge designs, such as full aircraft, but meshes with more than a hundred million cells overwhelm graphics card memory. Speed means little if the model cannot hold the geometry at all.
Their system, Transolver-3, restructures the computation so slices of the physical state can be tiled and multiplied more cheaply. Training uses random subsets of the full mesh, and a caching step reuses physical states during prediction. The result handled meshes above 160 million cells and performed well on three hard benchmarks, including aircraft and car design tasks. The authors released the code publicly.
Catalog record
| CREATOR | Hang Zhou, Haixu Wu, Haonan Shangguan, Yuezhou Ma, Huikun Weng, Jianmin Wang, Mingsheng Long |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 64 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04940