![[2409.00730] Generating Physical Dynamics under Priors](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2409.00730] Generating Physical Dynamics under Priors
Summary
This article presents a novel framework for generating physically realistic dynamics in data-driven contexts by incorporating physical priors into diffusion-based generative models.
Why It Matters
The integration of physical priors into generative models addresses a significant gap in existing methodologies, which often violate fundamental physical laws. This advancement has implications for various fields, including AI4Physics, enhancing the accuracy and reliability of simulations and predictions in physical systems.
Key Takeaways
- Introduces a framework that incorporates physical priors into generative models.
- Utilizes distributional and physical feasibility priors to ensure realistic dynamics.
- Demonstrates high-quality results across diverse physical phenomena.
- Addresses limitations of existing methodologies that overlook physical laws.
- Potentially advances data-driven studies in fields like AI4Physics.
Computer Science > Machine Learning arXiv:2409.00730 (cs) [Submitted on 1 Sep 2024 (v1), last revised 13 Feb 2026 (this version, v4)] Title:Generating Physical Dynamics under Priors Authors:Zihan Zhou, Xiaoxue Wang, Tianshu Yu View a PDF of the paper titled Generating Physical Dynamics under Priors, by Zihan Zhou and 2 other authors View PDF Abstract:Generating physically feasible dynamics in a data-driven context is challenging, especially when adhering to physical priors expressed in specific equations or formulas. Existing methodologies often overlook the integration of physical priors, resulting in violation of basic physical laws and suboptimal performance. In this paper, we introduce a novel framework that seamlessly incorporates physical priors into diffusion-based generative models to address this limitation. Our approach leverages two categories of priors: 1) distributional priors, such as roto-translational invariance, and 2) physical feasibility priors, including energy and momentum conservation laws and PDE constraints. By embedding these priors into the generative process, our method can efficiently generate physically realistic dynamics, encompassing trajectories and flows. Empirical evaluations demonstrate that our method produces high-quality dynamics across a diverse array of physical phenomena with remarkable robustness, underscoring its potential to advance data-driven studies in AI4Physics. Our contributions signify a substantial advancement in the fiel...
