Physics-informed machine learning reconstruction of shear alfvén wave dynamics on the LAPD

The Large Plasma Device (LAPD) at UCLA is a unique experimental platform for basic plasma science, enabling studies relevant to space physics, astrophysics and controlled nuclear fusion. The LAPD's high reproducibility, high repetition rate, and comprehensive diagnostic suite enable detailed spatiotemporal measurements of complex plasma dynamics. However, the diagnostics used in the LAPD inevitably capture only a partial picture of the underlying physics. Techniques from the field of physics-informed machine learning, such as physics-informed neural networks (PINNs), are enabling new ways to overcome this observation gap. These techniques can be used to combine partial measurements from multiple diagnostics with theoretical plasma models expressed as partial differential equations, enabling the physically-consistent reconstruction of unmeasured or difficult-to-measure quantities in dynamically evolving plasmas. In this work, we explore the potential of PINNs to reconstruct the plasma dynamicsof a shear Alfvén wave (SAW) propagating in cold plasma, based solely on partial measurements of magnetic field fluctuations. We conduct numerical experiments using fully kinetic particle-in-cell simulations of SAWs in conditions relevant to the LAPD, and determine the conditions under which PINNs can be used to accurately reconstruct plasma dynamics associated with SAWs. We discuss next steps towards the application of this methodology to experimental B-dot measurements of SAWs on the LAPD.