GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training

TL;DR

GeoPT introduces a novel pre-training approach for physics simulators that incorporates synthetic dynamics into geometric data, significantly reducing data needs and speeding up convergence for various physics tasks.

Contribution

The paper proposes lifted geometric pre-training with synthetic dynamics, enabling scalable, dynamics-aware neural physics simulation without extensive labeled data.

Findings

Reduces labeled data requirements by 20-60%.

Accelerates convergence by 2×.

Improves performance across fluid and solid mechanics benchmarks.

Abstract

Neural simulators promise efficient surrogates for physics simulation, but scaling them is bottlenecked by the prohibitive cost of generating high-fidelity training data. Pre-training on abundant off-the-shelf geometries offers a natural alternative, yet faces a fundamental gap: supervision on static geometry alone ignores dynamics and can lead to negative transfer on physics tasks. We present GeoPT, a unified pre-trained model for general physics simulation based on lifted geometric pre-training. The core idea is to augment geometry with synthetic dynamics, enabling dynamics-aware self-supervision without physics labels. Pre-trained on over one million samples, GeoPT consistently improves industrial-fidelity benchmarks spanning fluid mechanics for cars, aircraft, and ships, and solid mechanics in crash simulation, reducing labeled data requirements by 20-60% and accelerating convergence by 2$\times$. These results show that lifting with synthetic dynamics bridges the geometry-physics gap, unlocking a scalable path for neural simulation and potentially beyond. Code is available at https://github.com/Physics-Scaling/GeoPT.