CUOS Seminar | Optics Seminar
CUOS Noon Seminar: Do Graph Neural Networks dream of Landau Damping?
This event is free and open to the publicAdd to Google Calendar
In recent years, in order to obtain computational speed ups, there have been significant efforts to combine plasma simulation codes with machine learning surrogate models.
In this presentation, I will focus on the possibility of fully replacing a plasma physics kinetic simulator with a graph neural network-based simulator (GNS). This class of surrogate models can be of particularly interest for kinetic simulations due to the similarity between their message-passing update mechanism and the traditional physics solver update, as well as the potential for enforcing known physical priors into the graph construction and update process.
I will demonstrate how the GNS learns the kinetic plasma dynamics of the one-dimensional plasma model, which serves as a predecessor to contemporary kinetic plasma simulation codes, and that it is capable of recovering a wide range of well-known kinetic plasma processes without being specifically trained to reproduce them. Examples shown will include plasma thermalization, electrostatic fluctuations about thermal equilibrium, the drag on a fast sheet, and Landau damping.
Additionally, I will compare the performance of the GNS against the original plasma model in terms of run-time, conservation laws, and the temporal evolution of key physical quantities. To conclude, I will address the limitations of the model and discuss potential directions for developing higher-dimensional surrogate models for kinetic plasmas.
Pizza will be served!