Speakers
Description
Core-collapse supernovae have been the subject of theoretical studies for over half a century and of observational investigations for even longer. Furthermore, decades of simulations suggest that these explosions occur only if the neutrinos emitted by the proto-neutron star (PNS) deposit sufficient energy into the shock wave. Thus, neutrinos play a central role in all stages of the collapse and explosion, serving as a vital tool for directly probing the cores of stars at the end of their lives. If a core-collapse supernova from a massive star occurs in the Milky Way or a nearby galaxy, thousands to millions of neutrinos would be detected on Earth over the course of a few seconds. In this work, we trained a Long Short-Term Memory (LSTM) neural network to relate the time evolution of a PNS's mass and radius to the neutrinos it emits during the first few seconds following its formation. The neutrino emission data used for training the network were obtained from spherically symmetric (1D) numerical simulations of the core collapse of massive stars, using the FLASH code for progenitor stars with various equations of state and heating factors designed to emulate multidimensional effects within 1D simulations. Evaluation was performed on a test set comprising core-collapse supernova simulations covering a progenitor star mass range of 12 to 60 $M_\odot$ in 1 $M_\odot$ increments; the network had no access to any of these cases during or prior to training. In the analyzed sample, we demonstrated that the neural network was able to reproduce the evolution of the PNSs' gravitational mass and radius with small deviations. We are currently in the final stages of network optimization, and further analyses are underway.
| Keywords | supernovae, neutrinos, neural networks, LSTM |
|---|---|
| Proceedings Publication Intent | Yes |
| Student Award Eligibility | no |
| Financial Support Request | yes |