Physicists at Emory University have used artificial intelligence to uncover new physics in dusty plasma, a system representing the fourth state of matter.
The AI model, a custom neural network trained on laboratory data, described non-reciprocal forces between charged particles with over 99% accuracy, revealing inaccuracies in long-standing theoretical assumptions.
Non-reciprocal forces occur when one particle affects another differently than the reverse interaction, a phenomenon difficult to measure in complex many-body systems like dusty plasma.
The researchers combined experimental measurements with theoretical modeling, allowing the AI to identify patterns that corrected prior misunderstandings about how particles in ionized gas with dust grains influence each other.
Justin Burton, professor of experimental physics at Emory and senior co-author, said the method is not a black box: the team understands how and why the AI reaches its conclusions, making the framework transparent and reusable.
Ilya Nemenman, professor of theoretical physics and co-senior author, noted that the AI’s high precision enabled the team to see particle interactions in exquisite detail, exposing flaws in common simplifications used in plasma physics models.
The study, published in PNAS, was led by Wentao Yu, an Emory PhD student now at Caltech, with contributions from Eslam Abdelaleem, now a postdoc at Georgia Tech.
Funding came primarily from the National Science Foundation, with additional support from the Simons Foundation.
Slava Lukin, NSF Plasma Physics program director, said the approach could extend beyond plasma to other many-body systems, such as industrial materials like paint and ink, or even biological systems like groups of living cells.
The research demonstrates that AI can go beyond data analysis or prediction to actively discover new physical laws by identifying subtle, correctable errors in established theory.
What is dusty plasma and why is it called the fourth state of matter?
Dusty plasma is an ionized gas containing charged particles and tiny dust grains; plasma is termed the fourth state of matter because it differs from solids, liquids, and gases in that its atoms are ionized, allowing collective electromagnetic behaviors.
How did the AI correct existing theoretical assumptions?
By modeling particle interactions with over 99% accuracy, the AI revealed that some widely used approximations for non-reciprocal forces in dusty plasma do not fully capture the actual dynamics, enabling physicists to refine their theories based on empirical data.
