$\beta$-decay half-lives studied using neural network method
Abstract
Nuclear $\beta$-decay half-lives determine the time scale of the rapid neutron capture process (r-process) in astrophysics; thus, their accurate description is crucial for the study of the r-process. In this work, we use the machine learning method to predict the nuclear $\beta$-decay half-lives and their error bars for the whole nuclear chart by constructing three different neural networks. We study the influences on the results of different neural network inputs, number of neurons, and activation functions. Compared with the quasiparticle random-phase approximation based on the finite range droplet model (FRDM+QRPA), the accuracy of the description of half-lives is improved by about 2.6 times, and the root mean square (rms) deviation from experimental data reaches $10^{0.43}$. For the nuclei with half-lives of $<$ 1 s, the rms value reaches $10^{0.22}$, which can have an important effect on the r-process simulation study.
