Rate coefficient function estimation using Gaussian process regression

Collisional-radiative (CR) models operate with a library of transition rate coefficient information whose memory requirements can increase dramatically depending on the number of atomic states and types of reaction pathways that are considered. Multiple methodologies have been explored in the literature seeking to minimize the memory footprint of the CR transition library. This work introduces Gaussian process (GP) regression as an alternative method to calculate and store transition rate coefficient values for collisional-radiative simulations. Unlike previous methods in the literature that primarily character-ize each transition's cross section function, the GP regression algorithm devised in this work creates fits directly over the integrated rate coefficient profiles for Maxwellian electron energy distributions using sparse and discrete observations. Results are shown in this work for several neutral and singly-ionized xenon transitions of electron-impact excitation, electron-impact ionization, and radiative recombination. Relative errors measuring approximately less than 1% were observed in comparisons between rate coeffi-cient predictions and exact solutions. (c) 2022 Elsevier Ltd. All rights reserved.