Physicists at Emory University have achieved a groundbreaking milestone by leveraging artificial intelligence to uncover previously unknown physical laws governing particle interactions in plasma, the fourth state of matter. This discovery, detailed in a recent study published in the Proceedings of the National Academy of Sciences, demonstrates how AI can transcend traditional data analysis to reveal fundamental principles directly from experimental observations. By analyzing three-dimensional trajectories of microscopic particles suspended in a laboratory-created dusty plasma, researchers developed a custom neural network that identified non-reciprocal forces with over 99 percent accuracy. This approach not only challenges long-held theoretical assumptions but also opens new avenues for understanding complex many-body systems prevalent in astrophysics, fusion energy research, and even biological processes.

The fourth state of matter, plasma, consists of ionized gas where electrons are stripped from atoms, creating a soup of charged particles. Comprising more than 99 percent of the visible universe—from the sun's core to nebulae—plasma behaves in ways that defy simple solid, liquid, or gas dynamics due to electromagnetic interactions. Dusty plasma takes this further by incorporating micron-sized dust grains that become charged within the plasma environment, mimicking conditions in Saturn's rings, interstellar dust clouds, or the ionosphere during wildfires.

🔬 The Experimental Setup at Emory University

In the Burton Lab at Emory University, scientists recreated dusty plasma conditions inside a vacuum chamber. Tiny plastic microparticles, levitated in a radio-frequency driven plasma sheath above a flat electrode, served as model dust grains. A innovative tomographic imaging technique brought the system to life: a laser sheet scanned up and down the chamber while high-speed cameras captured particle positions. These two-dimensional slices were reconstructed into precise 3D trajectories over centimeter scales for several minutes, providing rich data on collective particle motion.

This setup allowed control over variables like gas pressure, which influences plasma density and temperature, enabling systematic studies. The experiments revealed chaotic yet patterned dances among particles, driven by Coulomb forces screened by the plasma and complicated by ion wakes—focused streams of ions trailing each particle like a boat's wake on a lake.

Physics-Informed Machine Learning: The AI Breakthrough

The core innovation was a physics-tailored machine learning model comprising three neural networks. One network modeled inter-particle forces, another environmental influences like gravity, and the third drag from particle velocity. Trained on sparse 3D trajectory data, the AI incorporated physical symmetries—such as action-reaction principles where applicable—and handled non-identical particles of varying sizes.

Unlike black-box AI, this framework was interpretable: it decomposed motion into drag, environmental, and interaction components, enforcing Newton's second law at every step. Running on a standard desktop computer, the model learned effective force laws without prior assumptions about their form, distilling complex dynamics into simple, predictive equations.

Non-Reciprocal Forces: A Surprising Revelation

The AI unveiled non-reciprocal interactions, where the force one particle exerts on another differs from the reverse. For particles at different heights, a leading particle attracts a trailing one via its ion wake, but the trailing particle repels the leader. This asymmetry peaks at close horizontal separations under 0.6 millimeters and vanishes for particles at the same height, where forces follow a screened Coulomb potential.

Quantitatively, the model fitted these forces with R-squared values exceeding 0.99, predicting accelerations frame-by-frame across test datasets. Visualizations showed force fields warping around particles, highlighting how plasma-mediated screening lengths—distances over which charges influence each other—increase with particle size, contrary to expectations.

Challenging Decades-Old Theories

Traditional orbital-motion-limited theory predicted dust charge scaling as the particle radius to the power of 1/3. The AI found exponents between 0.30 and 0.80, rising with gas pressure, indicating flows and collisions in the plasma sheath alter charging. Similarly, screening lengths grew linearly with size, not fixed by plasma properties alone.

Particle masses inferred two ways—from drag coefficients and interaction strengths—matched closely, validating the model. These deviations underscore gaps in theory, urging refined models for dusty plasmas in real-world scenarios.

Experimental Validation and Precision Measurements

Beyond prediction, the AI turned particles into plasma probes. Drag coefficients revealed local gas densities, while force fits yielded charges and screening parameters with unprecedented detail. Additional experiments at varied pressures confirmed trends, bridging lab data to natural phenomena.

• Charge-mass relation: q ∝ m^p, p=0.30-0.80
• Screening length λ increases ~20-30% with larger particles
• Non-reciprocity strongest for vertical offsets <0.1 mm
• Mass agreement within 5% across methods

Implications for Fusion Energy Research

Dusty plasmas plague fusion reactors, where wall erosion produces dust that absorbs heat, quenches plasmas, and risks explosions. Understanding these interactions could optimize dust removal, enhancing reactor stability. Universities like Emory contribute foundational science, training experts for national labs like PPPL or ITER projects. For more on fusion-related opportunities, see the original PNAS study.

Biological and Materials Science Crossovers

Non-reciprocal forces mirror active matter in biology: cells propel asymmetrically during metastasis or wound healing. The framework could decode flocking in birds, bacterial swarms, or cancer invasion, linking physics to medicine. In industry, it applies to colloids in paints, inks, or semiconductors, predicting rheology for better formulations. Emory's interdisciplinary approach exemplifies higher education's role in such convergences. Explore Emory's lab details.

Spotlight on Emory Researchers and Academic Careers

Lead experimentalist Justin Burton probes amorphous materials and plasmas, fostering a lab blending experiment with computation. Theoretical physicist Ilya Nemenman, bridging biophysics, eyes applications in collective behavior. Alumni Wentao Yu (now Caltech postdoc) and Eslam Abdelaleem (Georgia Tech postdoc) highlight career pipelines from PhD to prestigious fellowships.

This work underscores booming demand for physics PhDs skilled in AI/ML. Salaries for plasma physicists average $120,000+, with 15% premiums for AI expertise. Programs at Emory prepare students for academia, national labs, or tech, emphasizing interdisciplinary tools.

Future Outlook: AI as a Discovery Engine

As AI evolves, expect broader adoption in physics, from quantum many-body problems to cosmology. Emory's method, scalable and interpretable, sets a template for sparse-data regimes. Funded by NSF and Simons Foundation, it signals growing support for AI-physics hybrids. Challenges remain: ensuring generalizability and human oversight for validation.

In higher education, this discovery inspires curricula integrating ML with core physics, attracting top talent. It positions universities as hubs for transformative research, driving innovations in energy, health, and beyond.

Photo by Luke Jones on Unsplash