AI Finds Laws in Plasma
- Emory University physicists reported a Proceedings of the National Academy of Sciences study showing a physics-tailored neural network learned hidden force laws in laboratory dusty plasma from three-dimensional particle trajectories. - The model inferred one-way, non-reciprocal particle forces with R2 above 0.99, cross-checked particle masses in two independent ways, and found charge and screening values that diverged from common assumptions. - The paper frames the method as a route to infer laws directly from dynamics in many-body systems, from colloids to living cells. (pnas.org)
Plasma is an ionized gas, and a dusty plasma adds tiny charged grains that push and pull on each other in ways physicists struggle to write down. (pnas.org) A team at Emory University said it used a physics-tailored neural network to learn those interaction rules directly from experiments instead of starting with a standard force law. (news.emory.edu) (pnas.org) The experiments tracked the three-dimensional motion of individual dust particles in a laboratory plasma. Those trajectories became the training data for a model built to respect physical symmetries and handle particles that are not identical. (news.emory.edu) (pnas.org) The target was a class of forces called non-reciprocal forces, where particle A can affect particle B differently than B affects A. In dusty plasma, those one-way interactions can draw energy from the surrounding nonequilibrium environment. (pnas.org) The paper said the model learned the effective non-reciprocal forces with “exquisite accuracy,” reporting R2 greater than 0.99. The researchers also inferred particle masses in two independent ways and said the answers agreed. (pnas.org) That accuracy let the team estimate particle charge and screening length, a measure of how the plasma weakens electric forces over distance. The paper said those estimates showed large deviations from common theoretical assumptions. (pnas.org) The study was published in Proceedings of the National Academy of Sciences after being received on March 24, 2025, accepted on July 1, 2025, and later highlighted by Emory on July 30, 2025. (pnas.org) (news.emory.edu) The authors were Wentao Yu, Eslam Abdelaleem, Ilya Nemenman, and Justin C. Burton. Emory said Yu is now at the California Institute of Technology and Abdelaleem is now at Georgia Tech. (pnas.org) (news.emory.edu) The paper casts the method as a template for other many-body systems, where large numbers of particles interact at once and simple equations break down. The authors named colloids and living organisms as possible next targets. (pnas.org) (news.emory.edu) The claim here is not that AI replaced physics. It is that a model built with physical constraints pulled a cleaner force law out of messy motion than standard approximations had managed. (pnas.org)
