Physicists at Emory University have used a machine-learning method to identify surprising new twists on the non-reciprocal forces governing a many-body system. The researchers employed a neural network model and data from laboratory experiments on dusty plasma, a type of ionized gas containing suspended dust particles.
The findings, published in the Proceedings of the National Academy of Sciences, provide detailed descriptions for the physics of dusty plasmas, yielding precise approximations for non-reciprocal forces with an accuracy of over 99%.
According to researchers, common theoretical assumptions about these forces are not entirely accurate. The AI approach corrected these inaccuracies due to its ability to see details in exquisite detail.
The researchers hope that their method will serve as a starting point for inferring laws from the dynamics of many-body systems. These systems can be found in various environments, including living organisms.
A dusty plasma is an ionized gas containing suspended dust particles, which can be observed in space and planetary environments. Charged particles levitating above the moon’s surface, for instance, are an example of a dusty plasma.
Researchers Justin Burton and Ilya Nemenman led the project, with Wentao Yu and Eslam Abdelaleem contributing to its development. The team used AI to investigate the dynamics of collective motion in a dusty plasma system.
The researchers employed a tomographic-imaging technique to track the three-dimensional motion of particles in the plasma. The AI model accounted for inherent symmetries, non-identical particles, and learned the effective non-reciprocal forces with accuracy.
Their findings also corrected some wrong assumptions about dusty plasmas, such as the relationship between dust particle size and charge.
The researchers verified their results through experiments conducted in a vacuum chamber filled with plasma.
Source: https://phys.org/news/2025-08-ai-reveals-unexpected-physics-dusty.html