Abstract Details
Abstracts
Author: Chris McDevitt
Requested Type: Poster
Submitted: 2026-03-20 06:56:45
Co-authors:
Contact Info:
University of Florida
PO Box 116400
Gainesville, FL 32611
United States
Abstract Text:
Deep learning methods offer the promise of drastically reducing the computational cost of describing several plasma physics processes. This poster will describe progress on the development of physics-informed machine learning descriptions of energetic electrons and ions in magnetized plasma devices. Each of these models is trained only on the underlying PDE, boundary conditions and initial conditions without any experimental or simulation data. In both cases, an adjoint formulation is employed through which quantities of interest of the energetic particle population are predicted rather than directly computing the ion or electron distribution. We demonstrate how such adjoint approaches can be used to predict quantities such as the runaway electron (RE) density or current, needed to close the MHD equations when a large runaway population is present, or directly predict confinement metrics of the fast ion population. This is achieved by defining distinct adjoint problems, whereby the quantity of interest is directly predicted. In the context of REs, an adjoint problem is derived to predict the rate that REs are generated during a tokamak disruption. These generation rates, which are based on fully kinetic solutions to the relativistic Fokker-Planck equation, allow for a fluid model of REs to be closed, thus enabling an efficient and accurate treatment of RE generation. In the context of fast ions, an adjoint problem is defined to evaluate the fast ion confinement time as a function of the ion's initial phase space location. Such a metric provides an efficient means of identifying the quality of confinement of the fast ion population in a given magnetic field geometry. Ongoing work is focused on extending this description to non-axisymmetric magnetic field geometries, while incorporating a broader range of fast ion metrics such as the slowing down probability and the probability of an ion being captured by a fast ion diagnostic.
Characterization: 6.0
Comments:
I would prefer to present my poster Tuesday afternoon if possible. If a late Tuesday poster session is available (last year there was a 4-6pm poster session) that would work best.