Title: Learning force laws in many-body systems

Abstract: Scientific laws that describe natural systems may be more complex than our intuition can handle, and thus how we discover laws must change. Machine learning (ML) models can analyze large quantities of data, but their structure should match the underlying physical constraints to provide useful insight. Here we demonstrate a ML approach that incorporates the physics intuition about the underlying system to infer forces and learn new laws from experimental data in dusty plasmas. In dusty plasmas, micron-sized particles levitate in a low temperature plasma and interact through plasma-mediated electrostatic forces. We track the motion of these particles using 3D tomographic imaging. The model is then trained on the 3D experimental particle trajectories, and accounts for the inherent symmetries and varying number of non-identical particles, extracts the mass and charge of each particle, and learns the effective non-reciprocal force law governing their motion. These results guide new routes of discovery using physics-tailored ML in many-body systems.

Host: Mary Elting