Kinetic Simulations of Space Plasmas using Machine Learning
Title: |
Kinetic Simulations of Space Plasmas using Machine Learning |
DNr: |
Berzelius-2022-177 |
Project Type: |
LiU Berzelius |
Principal Investigator: |
Shahab Fatemi <shahab.fatemi@umu.se> |
Affiliation: |
Umeå universitet |
Duration: |
2022-10-01 – 2023-04-01 |
Classification: |
10303 |
Homepage: |
http://www.amitiscode.com |
Keywords: |
|
Abstract
Kinetic simulations in space plasma physics require tracking 10s of billions of charged particles often for 100s of thousands time steps. In these simulations the equation of motion is integrated for a system of many particles using numerical time-integrators. Similar to classical simulations in molecular dynamics, the time-integrators in kinetic plasma models require taking small time-steps to maintain accuracy and stability of the system that directly impacts the computational cost of a simulation run. In addition, electromagnetic waves and instabilities and their interaction with plasma often mandates choosing smaller time steps than those required for particle integrators. We have introduced operators using recurrent neural networks (RNN) to solve the equation of motion for plasma. We have developed this method as part of the particle solver package in a massively parallel kinetic model of plasma, Amitis, that runs on multiple GPU nodes. This advanced method allows us taking ~10x larger simulation time steps compared to the classical time steps required in particle integrators, and the development was (partly) achieved through our previous computation project on Berzelius (project Berzelius2022-52). The goal of this computation project is to continue using Berzelius GPU nodes and together with our advanced kinetic model of plasma combined with Machine Learning techniques we are going to provide a detailed understanding on the interaction between plasma and different planetary bodies inside and outside our solar system. Our primary focus is on the global and local structures of plasma interactions with outer planetary moons (e.g., Ganymede, Europa, and Callisto) and the stellar wind interaction with terrestrial bodies (e.g., Mercury, Mars, Earth, and Exoplanets). We have currently three research projects funded by the National Research Council (VR), Swedish National Space Agency (Rymdstyrelsen), and Kempestiftelserna to pursue our science objectives and goals using plasma simulations and through comparison with observations.