NucleiML: A machine learning framework of ground-state properties of finite nuclei for accelerated Bayesian exploration

Abstract

The global behavior of the nuclear equation of state (EoS) is commonly studied using data from finite nuclei (FN), heavy-ion collisions, and astrophysical observations of neutron stars (NS). The constraints derived from FN such as binding energies and charge radii play the most crucial role in shaping the EoS up to saturation density. The computational cost associated with explicitly incorporating these constraints presents a significant challenge especially when the aim is to explore the model uncertainties rather than optimizing a single model. We address this by introducing NucleiML (NML), a machine learning framework trained on ground-state properties of a few finite nuclei generated by a relativistic mean-field model. NML allows us to integrate FN and NS properties within a Bayesian inference framework in an efficient manner. The results demonstrate reasonable accuracy and a speedup of $\sim 10^4$ times for calculation of FN properties for a single parameter set, yielding roughly $\sim 10^3 \times$ speed up in the Bayesian framework. The present study makes the case for extending the work to a larger set of nuclei, potentially enabling future studies of NS properties to incorporate the whole nuclear chart.