2024 workshop program
Accelerating neutron star light curve simulation and parameter inference through neural networks
The work
Abstract
Understanding neutron stars and their equation of state is a fundamental goal in astrophysics, with implications for our understanding of dense matter and fundamental physics. NASA missions such as NICER and Fermi play a pivotal role in advancing this field. However, the large computational overhead of simulating neutron star emission has prohibited research groups from efficiently inferring information from the existing observations.
In this study, we leverage neural networks to address this computational bottleneck. We employ ResNet-like neural networks to speed up light curve simulations by a factor of 300 and efficiently explore a vast, non-convex parameter space. This allows the inclusion of more degrees of freedom and more physically realistic effects. Additionally, it now becomes computationally feasible to model the Fermi gamma-ray light curves in addition to the X-ray light curves, which constrains the larger parameter space by offering a unique perspective on a neutron star’s magnetic field.