Physicists at Emory University applied a machine-learning method to find unexpected details about non-reciprocal forces in a many-body system. Their findings appeared in PNAS and came from a neural network model, combined with lab experiments on dusty plasma. Dusty plasma is an ionized gas mixed with suspended dust particles.
This work stands out because it used artificial intelligence (AI) to reveal new physical laws, rather than just handling data or making predictions. The physicists noted that the AI method is clear and understandable, not mysterious, and could apply to various many-body systems. The study gave the most detailed view yet of dusty plasma physics, with precise calculations for non-reciprocal forces accurate to over 99 percent. It also fixed some incorrect common ideas about these forces.
Experiments in the lab tracked the three-dimensional movement of individual particles in dusty plasma. Researchers hope this AI approach can help uncover rules in other many-body systems, such as colloids, which are mixtures like paint where tiny particles are dispersed in a liquid, or groups of cells in living things.
Insights from plasma and AI
Plasma consists of ionized gases with free-moving charged particles, making up most of the visible universe, like solar winds or lightning. Dusty plasma happens in space, such as Saturn's rings or the moon's surface where dust levitates due to weak gravity, and on Earth during wildfires when charged soot mixes with smoke.
In the lab, researchers used a vacuum chamber with plastic particles in plasma to model these systems. They created a tomographic-imaging technique, which builds three-dimensional images from slices, to follow particle paths over time. The neural network was designed to work with limited data, separating effects like drag force, a slowing due to resistance, environmental forces like gravity, and particle interactions.
It revealed non-reciprocal forces, like how a leading particle attracts a trailing one but the trailing repels the leading, using a boat-wake analogy on a lake. The work corrected theories, showing particle charge does not grow exactly with size and force decrease depends on particle size. Experiments confirmed these results. This framework, run on a regular computer, could extend to studying collective motion in biology, such as cell movement in cancer.