Physics-tailored machine learning reveals unexpected physics in dusty plasmas.

Area of Science:

• Plasma physics
• Complex systems
• Machine learning applications

Background:

• Dusty plasma, a mixture of charged particles, exhibits complex, nonconservative, and nonreciprocal forces.
• Understanding these forces is crucial for space and planetary environments.
• Existing models often struggle to capture the intricacies of dusty plasma interactions.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for inferring force laws in laboratory dusty plasma.
• To incorporate physical intuition and constraints into ML models for accurate force law discovery.
• To demonstrate the utility of ML in uncovering novel physics from experimental data.

Main Methods:

• Training a 3D particle trajectory ML model on experimental dusty plasma data.
• Incorporating physical symmetries and nonidentical particle properties into the ML model.
• Validating the ML model by inferring particle masses and comparing results.

Main Results:

• The ML model accurately learned effective nonreciprocal forces with R≈0.99.
• Particle masses were inferred consistently through two independent methods.
• Precise measurements of particle charge and screening length revealed significant deviations from theoretical assumptions.

Conclusions:

• ML models, when designed with physical constraints, can accurately infer complex force laws in dusty plasmas.
• This approach enables precise measurements and the discovery of unknown physics.
• The methodology offers a new pathway for scientific discovery in diverse many-body systems.