Title: Plasma turbulence in stellarators: physics insights from machine learning
Abstract: The stellarator is a device for confining charged particles and plasma using geometrically optimized magnetic fields. It has applications both for fusion energy and for fundamental plasma physics. Turbulence in stellarator plasmas limits their temperature, and this turbulence depends strongly on the magnetic field geometry. To understand the nature of this dependence, we look for patterns in a new dataset of over 200,000 numerical turbulence simulations. We apply machine learning methods that respect physical invariances and are interpretable, uncovering analytic expressions of the geometry that correlate strongly with the turbulent heat flux. This example demonstrates one way that machine learning can go beyond black-box interpolation, to work in concert with traditional physics analysis.