Unveiling core-collapse supernova progenitors: characterization and physical insights through explainable artificial...

Core-collapse supernovae (CC-SNe) represent the final evolutionary explosive phase of sufficiently massive stars. Their characterization consists in understanding the progenitor star's physical properties such as stellar mass, radius, and explosion energy, and is crucial in many fields like astrophysics, cosmology, and multi-messenger astronomy. However, current methods require significant human expertise and are computationally prohibitive, taking weeks to months per event, and cannot keep up with the increasing number of SNe observations, especially with the advent of large-scale transient surveys. Here, we present a machine learning framework that can infer the physical parameters of CC-SN events with sub-second computation on standard hardware, enabling the rapid and accurate characterization of thousands of CC-SNe. Our deep learning model, trained on synthetic light curves from astrophysical simulations, achieves errors below 5% for most physical parameters when tested on real observations. Using explainable artificial intelligence techniques, we identify which phases of SN evolution are most informative for determining progenitor properties, providing insights for optimizing observational strategies.