Surrogate Modelling of Ion Acceleration and Overdense Laser-Plasma Interactions
Surrogate Modelling of Ion Acceleration and Overdense Laser-Plasma Interactions
Miethlinger, T.; Garten, M.; Göthel, I.; Hoffmann, N.; Schramm, U.; Kluge, T.
Abstract
Interaction of an overdense plasma with ultra-intense laser pulses represents a promising route to enable the development of compact ion sources. Prospective applications of high-energetic protons and ions include, but are not limited to, medical applications, materials science and nuclear fusion. However, current records for maximum proton energies (94 MeV, Higginson Nat Commun 9, 724 2018) are still well below the required values for many applications (typically 150-250 MeV) and many challenges remain unsolved to this day. In particular, a high-dimensional parameter space, as well as considerable effort per observation, make it impossible to uniformly sample the parameter space by means of simulations, let alone experimentally, while simultaneously strong nonlinearities limit the coarseness of the grid. Consequently, a combination of modern sampling approaches, optimized simulation codes and powerful data-based methods are essential for building realistic surrogate models. More specifically, we want to employ invertible neural networks (Ardizzone arXiv:1808.04730, 2018) for bidirectional learning of input and output, and convolutional autoencoder (Vincent J. Mach. Learn. Res. 11, 12 2010) to reduce intermediate field data to a lower-dimensional latent representation.
Keywords: Laser-Plasma; Ion Acceleration; Particle-in-cell; Machine Learning; Surrogate Modeling
Beteiligte Forschungsanlagen
- Strahlungsquelle ELBE DOI: 10.17815/jlsrf-2-58
Verknüpfte Publikationen
- DOI: 10.17815/jlsrf-2-58 is cited by this (Id 33798) publication
-
Poster
(Online Präsentation)
Helmholtz AI Virtual Conference 2021, 14.-15.04.2021, Online, Deutschland
Permalink: https://www.hzdr.de/publications/Publ-33798