Rapid Supernova Model Inference using Amortized Posterior Estimation
Abstract
Current physical models of supernovae use high dimensional semi-analytical models or hydrodynamical simulations combined with time consuming Approximate Bayesian Computation (ABC) methods, such as Markov chain Monte Carlo (MCMC), to infer their physical properties. While very accurate, this method of inference takes many CPU hours on average, and must be repeated whenever new observations become available, thereby making it practically infeasible to infer the explosion properties of the hundreds of thousands of supernovae that future modern all sky surveys will discover. In this work, we use Amortized Neural Posterior Estimation (ANPE), which is a simulation-based inference method that uses neural networks to estimate posterior probability distributions, to accurately estimate the posteriors of the supernovae models implemented in the Modular Open-Source Fitter for Transients (MOSFIT) python package. Once trained, it can rapidly infer model parameters of any event orders of magnitudes faster (<1 second) than traditional methods. This approach of adapting physical models to neural networks will be essential to estimate the posterior of model parameters and conduct population scale studies of supernovae for the terabytes of data per night that future all-sky surveys, such as the Vera Rubin Observatory, will produce.
- Publication:
-
American Astronomical Society Meeting Abstracts #243
- Pub Date:
- February 2024
- Bibcode:
- 2024AAS...24326038G