GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training

2026-02-23

AI Summary

This work addresses the scalability bottleneck of neural simulators, which stems from the high cost of generating high-fidelity physical data and the inability of static-geometry-only pretraining to capture dynamic physical behaviors—often leading to negative transfer. To overcome this, we propose GeoPT, a “lifted” geometric pretraining framework that injects synthetic dynamics into static geometry for the first time, enabling dynamics-aware self-supervised learning without real-world physical labels. GeoPT bridges the gap between geometric representation and physical modeling, establishing a foundational model for general-purpose physics simulation. Evaluated on industrial benchmarks spanning aerodynamics of cars and aircraft, hydrodynamics of ships, and solid mechanics in collision scenarios, GeoPT significantly improves performance, reduces labeled data requirements by 20–60%, and accelerates convergence by up to 2×.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Low Level & Physics-based Vision

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Web data generation and simulation

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.

Problem

• neural simulation
• physics simulation
• geometric pre-training
• data efficiency
• negative transfer

Innovation

• lifted geometric pre-training
• synthetic dynamics
• neural simulation
• self-supervision
• physics-informed learning