Speaker
Description
Constraining neutron-star structure requires models that are physically faithful yet fast enough for the evaluations required by Bayesian sampling. We begin with static vacuum field (SVF) magnetospheres, where direct light-curve synthesis is too slow for practical MCMC; our neural-network (NN) surrogate matches SVF profiles with high fidelity and provides ~400x speedups, enabling multipolar fits to NICER X-ray pulse profiles. We then extend to force-free (FF) magnetospheres and promote mass M and radius R to free parameters that enter GR ray tracing and the multipole-dipole balance at the light cylinder. Because FF simulations are ~2000x costlier than vacuum, we pretrain on SVF and fine-tune the full network on a compact FF dataset to obtain surrogates with ~10^6x speedups. We also train an NN to emulate FF gamma-ray light curves, making joint NICER-Fermi-LAT likelihoods feasible at MCMC scale. In FF, the current-sheet geometry specifies where gamma rays originate and their light-curve morphology, so the combined fit yields stricter, physically grounded constraints on M, R, and magnetic topology; we illustrate the workflow on PSR J0030+0451. Tools are in place for energy-dependent X-ray modeling, and the framework extends to future radio observables, enabling comprehensive multiwavelength studies and a clear path to equation-of-state constraints.
