ΗΜΕΡΑ/ΩΡΑ: Τετάρτη 18/11/2026, 18:00 ΑΙΘΟΥΣΑ: Μικρή Αίθουσα Αστεροσκοπείου, ΑΠΘ Η παρουσίαση θα μεταδοθεί και στον ακόλουθο σύνδεσμο: https://authgr.zoom.us/j/93194715211?pwd=aUAEvUO26L5YPh1vIFo3RwmkFyi2nq.1 ΟΜΙΛΗΤΗΣ: Θεοφάνης Τουλιόπουλος ΕΠΙΒΛΕΠΩΝ: Νικόλαος Στεργιούλας ΣΥΝΕΠΙΒΛΕΠΟΝΤΕΣ : Oliver Just, Andreas Bauswein, Zewei Xiong (GSI Darmstadt) ΤΙΤΛΟΣ: Predicting the electron fraction (Ye) in neutron star merger simulations using machine learning models ΠΕΡΙΛΗΨΗ: A crucial aspect of modelling rapid neutron capture process (r-process) nucleosynthesis that is enabled during binary neutron star (BNS) mergers is accounting for the interactions that neutrinos have with the astrophysical plasma, specifically the effect on the electron fraction Ye. Including these interactions in a hydrodynamical simulation using some weak interaction implementation is, however, not trivial. In addition, it is computationally very expensive, while also not being available for all hydrodynamical schemes. In order to circumvent these issues, we opted to train a machine learning model (ML) PYTHEA (Predicting Ye Through Emulation Algorithm) with the purpose of predicting Ye using physical parameters of the system which constitutes a novel approach in the field. For the purposes of training the ML model we used data from hydrodynamic BNS merger simulations that include a neutrino treatment. In order to test the inference capability of the model, PYTHEA was applied to simulations without neutrinos, while benchmarking was conducted by applying the model to unseen simulations with neutrinos. We discuss the accuracy and efficiency of the obtained ML models. Nucleosynthesis calculations were also conducted to test whether the Ye values given by the model result in nucleosynthesis patterns consistent with our simulations that include neutrinos, in both benchmark and inference predictions. Another important application of PYTHEA is that it will allow the modelling of kilonova signals in simulations lacking a neutrino treatment in a computationally efficient manner, thereby enabling the exploration of larger sets of kilonova models.