A simple grayscale video of ink spreading in water can now be automatically turned into a perfect mathematical law of physics.
Extracting the underlying rules of a moving fluid usually requires high-end sensors and millions of data points. This new AI pipeline takes raw, uncalibrated video footage and identifies the specific partial differential equation that governs the movement. It ignores the messy details of the video and finds the core truth of the physics involved. This means we can now discover new laws of nature just by pointing a camera at complex phenomena. It turns every smartphone into a high-powered laboratory for discovering the mathematics of the world. This could revolutionize how we track forest fires, oil spills, and weather patterns.
From Video-to-PDE: Data-Driven Discovery of Nonlinear Dye Plume Dynamics
arXiv · 2605.04535
Inferring continuum models directly from video is hampered by two facts: the recorded field is uncalibrated image intensity rather than a physical state, and direct numerical differentiation of noisy frames is unstable. We develop a video-to-PDE pipeline that converts grayscale recordings of an ink plume into a normalised scalar field $u(x,y,t)$, isolates a bulk drift $\mathbf{v}(t)$ from intrinsic spreading via the intensity-weighted centroid, and identifies an effective transport law by weak-f