Abstract Details
Abstracts
Author: Emmanouil G. Drimalas
Requested Type: Poster
Submitted: 2026-03-14 19:57:00
Co-authors: F. Fraschetti, C.-K. Huang, Q. Tang
Contact Info:
University of Arizona
2867 N TYNDALL AVE
TUCSON, AZ 85719
United States
Abstract Text:
Plasmas are fundamentally multiscale systems. Even though kinetic effects can drive fluid scale behavior, the scale separation by several orders of magnitude remains a major theoretical challenge. This in turn makes holistic simulations of plasmas prohibitively expensive and impractical.
A primary computational bottleneck in plasma kinetic simulations is the particle pusher, the routine responsible for the time evolution of particle position and velocity according to the Lorentz equation. The sheer volume of particles that need to be evolved and resolution requirements for stability make this process slow even for highly efficient, lower order methods, like the Boris algorithm.
In this work, we propose the utilization of neural networks as surrogates of such pushers. In particular, by expressing the dynamics of charged particles through a Hamiltonian and using a symplectic neural network [1,2] to learn the evolution map of the system, one can obtain a map that effectively behaves like the analytical solution, even for problems that cannot be solved analytically and are computationally hard to tackle. Moreover, the symplectic nature of these surrogates impose hard constraints on the learned map, guaranteeing certain desired properties, like long term energy stability through bounded errors and phase space volume preservation.
In this poster, we discuss the feasibility of such an approach and its performance in terms of computational cost and accuracy for test particle acceleration at astrophysical shock waves. We outline a recently published application [3] to 1D magnetic discontinuity, along with new preliminary results for more complicated field backgrounds.
[1] J.W. Burby, Q. Tang, and R. Maulik, Plasma Phys. Control. Fusion 63, 024001 (2021).
[2] C.-K. Huang, Q. Tang, Y.K. Batygin, et al., J. Phys. Conf. Ser. 2687, 062026 (2024).
[3] E.G. Drimalas, F. Fraschetti, C.-K. Huang, Q. Tang, Phys. Plasmas 32, 103901 (2025).
Characterization: 4.0
Comments: