AI Uncovers Hidden Physics in the Fourth State of Matter

A team at Emory University has used artificial intelligence to surface previously unknown rules of plasma physics, in a study covered by ScienceDaily on April 22 and published in PNAS. Plasma is the fourth state of matter, the soup of charged particles and free electrons that fills stars, lightning, and the inside of fluorescent tubes. The Emory experiment focused on dusty plasma, where micrometer-scale grains drift inside an ionized gas.

The researchers, led by experimental physicist Justin Burton and theoretical physicist Yu Yu, built a tomographic imaging rig that scans a laser sheet through the chamber while a high-speed camera records the result. Stacking those images reconstructed the 3D motion of dozens of particles across a few centimeters and several minutes. That dataset went into a neural network trained to look for the underlying force laws governing how the dust grains push and pull on one another.

What the model found

The most striking result is that a textbook claim about dust particle charge does not hold. Larger grains do carry more charge, but not in proportion to their size. The relationship shifts with plasma density and temperature, which earlier models had treated as fixed. The AI also produced what the authors describe as the most detailed account so far of non-reciprocal forces between dust particles, the kind of asymmetric pull where particle A acts on particle B more than B acts on A.

"We showed that we can use AI to discover new physics," Burton said in comments reported by SciTechDaily.

A different use of AI in science

Most physics applications of machine learning are about speed: faster simulations, faster analysis of telescope data, faster sorting through accelerator output. This work is different. The neural network was a tool for hypothesis discovery. It surfaced a rule the researchers then verified against the raw measurements, with the model guiding theory rather than crunching its predictions.

The findings have practical consequences beyond plasma research. Dusty plasma turns up in semiconductor fabrication, where stray dust contaminates wafers, and in astrophysics, where it shapes ring systems and interstellar clouds. A more accurate model of how those particles charge and interact could feed into both fields.

The result also fits a broader pattern. Earlier this year, an Emory-style approach was used in chemistry and protein folding work, where neural networks have started to act less as oracles and more as collaborators in formal scientific reasoning. Plasma physics is now on that list.

Sources: ScienceDaily, SciTechDaily, Mirage News.