Application of Neutron Capture Reaction Rate from Kernel Ridge Regression Prediction in s-process Simulation

Abstract

The origin of elements in the universe is a basic scientific problem. The slow neutron capture process (s-process) is believed to be responsible for the nucleosynthesis of about half of the elements heavier than iron. The neutron capture reaction rates are crucial nuclear physics inputs for the s-process, as they can affect the reaction flow and the s-process branchings along the s-process path. Over the past decades, increasingly powerful computers and advances in machine-learning (ML) methods have driven explosive applications of ML in many fields of physics, including nuclear physics. Recently, the ML kernel ridge regression (KRR) approach have been successfully employed to improve the theoretical predictions of neutron capture reaction rates. The corresponding results indicate that the corrections might lead to positive influences on the s-process simulations. This work aims to investigate the detailed effects of the KRR corrections of the neutron capture reaction rates on the s-process simulations. The s-process simulations were performed with the nuclear reaction network calculations based on the NucNet tools. The neutron capture reaction rates were taken from the Talys calculations and the KRR predictions, respectively, when the experimental data in the KADoNiS database were not available. Other nuclear physics inputs, including β decay rates and α decay rates, were taken from the JINA Reaclib database. The astrophysical conditions of the s-process simulations were taken as the typical values, i.e., temperature kT=30 keV and neutron density nn=1.6×107 cm-3. The final s-process abundances were obtained by the superposition of the abundances from simulations with different irradiation time with the weights following an exponential distribution. It is found that the final s-process abundances from both simulations based on the Talys calculations and the KRR predictions can reasonably reproduce the Solar s-abundances. The abundances produced by both simulations are actually very similar, which means that the KRR corrections on the neutron capture reaction rates do not impose significant effects on the whole pattern of the s-process abundances. However, it can be also found that the abundances for several special nuclei that are important for the s-process, namely 72Ge, 73Ge, 79Se, 131Xe, 154Eu, 193Pt and 205Pb, are different by about 30% between the simulation based on the KRR predictions and the one based on the Talys calculations. The reasons for these differences are analyzed in detail. The results indicate that the ML corrections on the neutron capture reaction rates can influence the abundances of some important s-process nuclei, which lie on the sprocess path.

Keywords

• s-process
• neutron capture reaction rate
• machine-learning