AI agents have started discovering the laws of physics on their own, without any help from human scientists.
April 15, 2026
Original Paper
Agentic Exploration of PDE Spaces using Latent Foundation Models for Parameterized Simulations
arXiv · 2604.09584
The Takeaway
For centuries, discovering the math that describes how fluids flow or how heat moves has been the job of brilliant humans. This paper describes a team of AI 'agents' that can autonomously explore complex physics equations and discover new 'scaling laws' for airflow just by running their own simulations. They found divergent patterns in how air moves past cylinders that were previously unknown. Instead of a human spending a decade in a lab, the AI hypothesized, tested, and verified these laws in a fraction of the time. This marks the beginning of an era where scientific discovery is automated, with AI acting as a 'discovery engine' for the physical world.
From the abstract
Flow physics and more broadly physical phenomena governed by partial differential equations (PDEs), are inherently continuous, high-dimensional and often chaotic in nature. Traditionally, researchers have explored these rich spatiotemporal PDE solution spaces using laboratory experiments and/or computationally expensive numerical simulations. This severely limits automated and large-scale exploration, unlike domains such as drug discovery or materials science, where discrete, tokenizable represe