Collisional-radiative rate coefficient function estimation using Gaussian process regression

Calculating rate coefficients for atomic kinetics simulations can require prolonged computational times and additional memory reserves whenever a large number of atomic levels and transitions must be resolved. These requirements are amplified when strong multifluid phenomena are observed such that multifluid rate coefficients must be employed, thereby expanding the precomputed, 1D rate coefficient profile into a 2D map due to the additional relative kinetic energy axis. This work explores using Gaussian process regression (GPR) to model any general atomic transition's rate coefficient profile or map. Through a sparse set of training data points, this nonparametric, Bayesian approach can provide an estimation of the entire transition profile and map over prescribed test data points. Preliminary results will show the set of parameters and features used to generalize GPR to a variety of transitions, along with the performance improvements attained when compared to an exhaustive, point-by-point calculation of the rate coefficient field.