Speaker
Description
The Galactic Center GeV excess is one of the most promising indirect detections of dark matter, but robust interpretation of the signal is bottlenecked by the cost of simulating the cosmic-ray-induced $\gamma$-ray background. It is intractable to run the gold-standard GALPROP code over the full space of plausible propagation models, whose cosmic-ray distributions are highly uncertain and sensitive to physics across many energy scales. As a proof of concept, we train lightweight Physics-Informed Neural Networks (PINNs) that directly learn the solution to the cosmic-ray transport equation for primary $e^-$ over energies from 100 MeV to 1 PeV. We demonstrate that PINNs can emulate the GALPROP solution to high accuracy at substantially lower computational cost during both training and inference, and in some cases offer more reliable convergence to the equilibrium cosmic-ray flux.