Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhou, Hang, Wu, Haixu, Shangguan, Haonan, Ma, Yuezhou, Weng, Huikun, Wang, Jianmin, Long, Mingsheng
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914306621177856
author Zhou, Hang
Wu, Haixu
Shangguan, Haonan
Ma, Yuezhou
Weng, Huikun
Wang, Jianmin
Long, Mingsheng
author_facet Zhou, Hang
Wu, Haixu
Shangguan, Haonan
Ma, Yuezhou
Weng, Huikun
Wang, Jianmin
Long, Mingsheng
contents Deep learning has emerged as a transformative tool for the neural surrogate modeling of partial differential equations (PDEs), known as neural PDE solvers. However, scaling these solvers to industrial-scale geometries with over $10^8$ cells remains a fundamental challenge due to the prohibitive memory complexity of processing high-resolution meshes. We present Transolver-3, a new member of the Transolver family as a highly scalable framework designed for high-fidelity physics simulations. To bridge the gap between limited GPU capacity and the resolution requirements of complex engineering tasks, we introduce two key architectural optimizations: faster slice and deslice by exploiting matrix multiplication associative property and geometry slice tiling to partition the computation of physical states. Combined with an amortized training strategy by learning on random subsets of original high-resolution meshes and a physical state caching technique during inference, Transolver-3 enables high-fidelity field prediction on industrial-scale meshes. Extensive experiments demonstrate that Transolver-3 is capable of handling meshes with over 160 million cells, achieving impressive performance across three challenging simulation benchmarks, including aircraft and automotive design tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2602_04940
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries
Zhou, Hang
Wu, Haixu
Shangguan, Haonan
Ma, Yuezhou
Weng, Huikun
Wang, Jianmin
Long, Mingsheng
Machine Learning
Deep learning has emerged as a transformative tool for the neural surrogate modeling of partial differential equations (PDEs), known as neural PDE solvers. However, scaling these solvers to industrial-scale geometries with over $10^8$ cells remains a fundamental challenge due to the prohibitive memory complexity of processing high-resolution meshes. We present Transolver-3, a new member of the Transolver family as a highly scalable framework designed for high-fidelity physics simulations. To bridge the gap between limited GPU capacity and the resolution requirements of complex engineering tasks, we introduce two key architectural optimizations: faster slice and deslice by exploiting matrix multiplication associative property and geometry slice tiling to partition the computation of physical states. Combined with an amortized training strategy by learning on random subsets of original high-resolution meshes and a physical state caching technique during inference, Transolver-3 enables high-fidelity field prediction on industrial-scale meshes. Extensive experiments demonstrate that Transolver-3 is capable of handling meshes with over 160 million cells, achieving impressive performance across three challenging simulation benchmarks, including aircraft and automotive design tasks.
title Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries
topic Machine Learning
url https://arxiv.org/abs/2602.04940