AI uncovers non‑reciprocal plasma forces
- Emory University physicists and collaborators reported that a machine-learning model, validated against lab measurements, mapped nonreciprocal forces in dusty plasma and found standard approximations miss key parts of how particles push and pull. - The team said its physics-tailored neural network fit three-dimensional particle trajectories with R² above 0.99, then used those inferred forces to recover particle masses in two independent consistency checks. - The paper appeared in PNAS in August 2025 after a July 1 acceptance, adding a peer-reviewed result to earlier preprint claims. (pnas.org)
Plasma is a gas so energized that electrons break free from atoms, and dusty plasma adds tiny solid particles that move inside that charged soup. In a new PNAS paper, Emory University researchers said machine learning exposed force rules in that system that standard formulas had missed. (pnas.org) (news.emory.edu) In ordinary mechanics, if one object pushes another, the second pushes back equally. In dusty plasma, the surrounding charged medium can distort that balance, producing nonreciprocal forces in which particle A’s effect on particle B differs from particle B’s effect on particle A. (pnas.org) (nature.com) That idea is not brand new. A 2020 Scientific Reports paper said direct experimental determination of nonreciprocal interparticle forces in complex plasmas had remained an unsolved problem even after decades of theory and simulation. (nature.com) The new study used three-dimensional trajectories of individual dust particles from laboratory experiments, then trained what the authors called a physics-tailored machine-learning model on those motions. The model was built to respect symmetries and differences between nonidentical particles instead of treating the system as a generic black box. (pnas.org) (arxiv.org) The authors reported an R² above 0.99 for the inferred nonreciprocal forces. They then checked the result by recovering particle masses in two separate ways and said the answers agreed. (pnas.org) (news.emory.edu) Those fits let the team estimate particle charge and screening length, which describes how the plasma medium weakens electrical interactions over distance, like fog dimming a headlight beam. The paper said those measurements showed large deviations from common theoretical assumptions. (pnas.org) The work was led by Wentao Yu, Eslam Abdelaleem, Ilya Nemenman, and Justin C. Burton. PNAS said the manuscript was received on March 24, 2025, accepted on July 1, 2025, and published in Volume 122, Issue 31 dated August 5, 2025. (pnas.org 1) (pnas.org 2) Emory’s write-up said the Burton lab developed methods to track particles in 3D and used experiments to validate the model’s inferences. Nemenman said the resulting description of nonreciprocal forces reached “more than 99%” accuracy. (news.emory.edu) Dusty plasma matters outside one lab chamber because it appears in places such as Saturn’s rings and interstellar space, and because related many-body systems show up in colloids, industrial materials, and living tissues. The authors said the same inference framework could be applied beyond plasma physics. (pnas.org) (news.emory.edu) So the news here is narrower, and stronger, than a viral social post suggested: the paper does not claim AI found a brand-new category of force from nowhere. It says a physics-guided model, checked against experiment, measured known-but-hard-to-pin-down nonreciprocal interactions in dusty plasma with unusual precision and exposed where common approximations break down. (pnas.org) (nature.com)
