Speaker
Description
Back in the 1950s, the famous Fermi-Pasta-Ulam-Tsingou (FPUT) numerical experiment began with a simple question: what happens if you take a line of coupled masses and springs, give them a push, and wait for the energy to spread out evenly? Instead of mixing smoothly into thermal chaos, the system surprised the researchers by neatly returning almost back to where it started.
As physics students, we spend hours solving differential equations and watching these recurrences play out on screen. In this project, we wondered: can everyday machine learning models learn the quirks of this classic chain without crunching through every single numerical time-step?
To find out, we simulated nonlinear mass-spring chains and used the trajectory data to train simple machine learning models. Our goal was modest: see if data-driven tools could pick up on energy sharing between normal modes, predict recurrence tendencies, and spot when the chain starts drifting from neat patterns into real chaos. The initial results are exciting—even straightforward models manage to follow the collective rhythm of the masses quite well. This work is just a hands-on attempt to see how friendly AI tools can help students explore classic nonlinear problems from a fresh perspective.