Insights on metal-poor stars from machine learning and nucleosynthesis calculations

Metal-poor stars are a key probe of rare processes such as the rapid neutron capture process (r-process). They are a standard benchmark of comparison with nucleosynthesis calculations, derived from a combination of astrophysical conditions from simulations and nuclear physics inputs. Often such comparisons seek to probe whether metal-poor star patterns point uniquely to a particular nucleosynthesis processes or astrophysical site. In this talk I will discuss applications of machine learning to decipher stellar patterns and their classifications [1], and show developments to extract which elements are most influential in their identification. I will also highlight work that considers network calculations with a grid of neutron densities to represent a full range of possible neutron capture processes from i-process to r-process. I will discuss the different mechanisms for termination and the relative importance of given nuclear inputs for such calculations, as well as how each case compares to a bank of stellar patterns. Agnostic nucleosynthesis calculations and ML analyses such as these represent an important path forward toward better identifying and understanding the source of heavy elements in our Universe.

[1] Vassh, Wang, Woloshyn, Kuchera, et al. ApJ 992, 36 (2025)