Machine Learning Technique for Fast Calculation of Radiative Processes

Radiative processes such as synchrotron radiation and Compton scattering play an important role in astrophysics. Radiative processes are fundamentally stochastic in nature, and the best tools currently used for resolving these processes computationally are Monte Carlo (MC) methods. These methods are used to sample from complicated probability distributions such as the differential cross section for electron-photon scattering, and a large number of samples are collected to compute the radiation properties such as angular distribution, spectrum, and polarization. In this work we propose a machine learning (ML) technique for fast, efficient sampling from arbitrary known probability distributions that can be used to accelerate the calculation of radiative processes in astrophysical simulations. In particular, we apply our technique to inverse Compton radiation and find that our ML method can achieve a speed approximately 20x faster than traditional MC methods currently in use.