Developing a generalized NLTE spectral ML model for HED applications

Non-local thermodynamic equilibrium (NLTE) physics are important in high energy density (HED) plasmas. These effects must be included in simulations and analyses of HED systems to obtain accurate results. However, calculating the needed NLTE data is computationally expensive. Machine learning (ML) based NLTE models have been developed to address these inefficiencies, but published methods are generally trained on small plasma parameter spaces and low-resolution spectra. In this talk, we present an approach to develop a generalized, high-resolution NLTE ML spectral model that covers 0.01–10 keV in temperature and 10¹⁹–10²² cm^-3 in ion density. The ML model is therefore applicable to nearly every HED experiment. Initial tests show that the ML model is faster, more memory efficient, and more accurate than tabular spectral interpolators. The ML training process reveals the need for multiple models to cover the entire plasma parameter space. Dividing along average plasma ion charge bounds has shown promising results. We also discuss incorporating non-Planckian radiation fields into the training set. Once fully developed, our ML model can be applied to simulations and analyses of HED experiments and significantly enhance scientific understanding of these systems.