Deep learning surrogate models to solve time-dependent NLTE absorption and emission spectra

Non-local thermodynamic equilibrium (NLTE) absorption and emission spectra are crucial in indirect drive inertial confinement fusion (ICF) simulations. Meanwhile, they are one of the most computationally expensive parts in ICF simulations. In some special physics scenarios, we need to calculate non-stationary ion states instead of stationary ones in NLTE problems. Although previous works have developed some effective methods to calculate stationary states of plasmas in NLTE conditions, they cannot be directly used for calculating non-stationary states. In this paper we propose a deep learning surrogate model method to solve time-dependent NLTE spectra. This new method fits data generated by time-dependent radiation-hydrodynamics simulations quite well and achieves about tens to hundreds of times acceleration on Average Atom Model (AAM) and about twenty to fifty thousand times acceleration on Multi-Average Ion Collisional-Radiative Model (MAICRM) respectively.