Extrapolating Mixture Density Network predictions: application to the astrophysical r-process

Understanding the origin of elements in our Universe is one of the outstanding questions in science. The astrophysical rapid neutron capture process (r-process) is believed to be responsible for creating half of the heavy isotopes up to bismuth and all of thorium and uranium. In modeling the r-process, masses are used as a basis for nuclear input for the network calculations. In previous studies, the relation between the r-process abundance and the mass model has been investigated, and it was found that different mass models have strong influences on the calculated abundance pattern. However, the uncertainty of the calculated r-process abundance either comes from different mass models or arises from varying individual masses within the same range. Here, we use a sophisticated Machine Learning based mass model utilizing the probabilistic Mixture Density Network (MDN) to predict masses. This method accurately predicts the masses with root mean square around 300 keV and also provides quantified uncertainties of each mass. We simulate the r-process abundance pattern that arise from the application of our MDN mass model. The MDN model encodes complex correlations among nuclei, so we use this information to estimate abundance uncertainties.