Machine Learning Acceleration of Neutron Star Pulse Profile Modeling

Abstract

Ray tracing algorithms that compute pulse profiles from rotating neutron stars are essential tools for constraining neutron-star properties with data from missions such as NICER. However, the high computational cost of these simulations presents a significant bottleneck for inference algorithms that require millions of evaluations, such as Markov Chain Monte Carlo methods. In this work, we develop a residual neural network model that accelerates this calculation by predicting the observed flux from the surface of a spinning neutron star as a function of its physical parameters and rotational phase. Leveraging GPU-parallelized evaluation, we demonstrate that our model achieves many orders-of-magnitude speedup compared to traditional ray tracing while maintaining high accuracy. We also show that the trained network can efficiently accommodate complex emission geometries, including non-circular and multiple hot spots, by integrating over localized flux predictions.