Towards noise-aware surrogate models of carrier population dynamics in optically-excited GaN

Monte Carlo (MC) methods have been shown to capture the ultrafast dynamics of carriers and resulting luminescence in photo-excited GaN. These calculations have assumed a homogeneous material under uniform illumination. We investigate the development of surrogate models to represent the carrier dynamics with fewer degrees of freedom, with the ultimate goal of computing the response of structured materials under inhomogeneous illumination profiles. These models are presently based on the extraction of mean and variance information from single data sets, through the use of fits to s piece-wise linear basis. Two models are then developed, one based on the mean and variance directly, and one on a Gaussian process regression model. New carrier populations with similar statistical profiles are then generated from these models to determine the amount of data reduction that can be obtained.