Exploring Supernova Gravitational Waves with Machine Learning

Core-collapse supernovae (CCSNe) emit powerful gravitational waves (GWs). Since GWs emitted by a source contain information about the source, observing GWs from CCSNe may allow us to learn more about CCSNs. We study if it is possible to infer the iron core mass from the bounce and early ring-down GW signal. We generate GW signals for a range of stellar models using numerical simulations and apply machine learning to train and classify the signals. We consider an idealized favourable scenario. First, we use rapidly rotating models, which produce stronger GWs than slowly rotating models. Second, we limit ourselves to models with four different masses, which simplifies the selection process. We show that the classification accuracy does not exceed ~70%, signifying that even in this optimistic scenario, the information contained in the bounce and early ring-down GW signal is not sufficient to precisely probe the iron core mass. This suggests that it may be necessary to incorporate additional information such as the GWs from later post-bounce evolution and neutrino observations to accurately measure the iron core mass.