Kilonova Light-Curve Inference Using a Neural Network

Kilonovae are astrophysical transients which are powered by the decay of radioactive elements following a neutron-star-merger. The modeling of kilonova observables, such as light curves, is done using radiative transfer simulations. As these models encapsulate more realistic physics, their computational cost becomes prohibitive to broad parameter space sampling. Emulators, such as neural networks, are frequently employed to minimize computational cost while retaining high simulation fidelity. The use of emulators enables the generation of millions of light curves in a matter of minutes; however, it also introduces additional systematic uncertainty to an already complex problem with compounded and unknown uncertainties. In this talk, we present AT2017gfo parameter inference results using the neural network model presented in an associated poster. We also highlight important considerations with regard to our treatment of the systematic uncertainty and implications for similar future analyses.