Skip to Main Content

Space Physics (293): Zackary Pine – Reconstructing Plasma Dynamics from Sparse Observations with Physics-Informed Machine Learning: From Laboratory Experiments to Space

Speaker: Zackary Pine

Affiliation: Physics and Astronomy, UCLA

Date: Friday, May 8, 2026

Time: 3:30 PM


Abstract

Accurately diagnosing and characterizing plasma dynamics in laboratory experiments and space plasmas is essential for advancing basic plasma science. Modern high-repetition-rate

experiments and multi-spacecraft missions provide increasingly rich spatiotemporal measurements, but computational tools that can fully exploit sparse, noisy, and incomplete data

remain lacking. Physics-informed neural networks (PINNs) offer a promising route by combining partial measurements with fundamental plasma equations to reconstruct physically

consistent quantities that were not directly measured.

In this talk, I will illustrate this approach using shear Alfvén waves, a fundamental mode of magnetized plasma that transports electromagnetic energy along magnetic fields and is relevant

to auroral energy flow [1]. Using synthetic magnetic-field measurements from fully kinetic particle-in-cell simulations, we show that PINNs can reconstruct the full two-fluid plasma

state—including electric fields, ion and electron velocities, and density perturbations—from sparse magnetic-field data alone. The method achieves approximately 10% relative accuracy

under adequate sampling and remains robust to substantial measurement noise.

I will then discuss first applications to experimental measurements of shear Alfvén waves on the Large Plasma Device at UCLA [2], including challenges posed by real data. Finally, I will briefly

outline how this framework could be extended to multi-spacecraft observations, where detailed measurements of the magnetic-field and the plasma distribution function are available but very

sparse.