Exploring the nuclear landscape using neural network techniques

This study presents a comprehensive Deep Neural Network (DNN) model that utilizes the 2016 Atomic Mass Evaluation (AME2016) data set to train and predict the nuclear observables across the entire nuclear landscape. The focus of our model encompasses fundamental nuclear properties such as binding energy per nucleon (BE/A), beta decay energy ([Formula: see text]-DE), and separation energies. By incorporating input parameters including neutron number (N), proton number (Z), the pairing term ([Formula: see text]), and a term contingent upon magic numbers (m), the DNN model achieves a multifaceted understanding of nuclear behaviors. Our investigation involves comparing the predicted nuclear observables of a diverse set of randomly selected nuclei across low-, medium- and high-mass regions to assess the efficacy of the AME2016 data set. The analysis revealed the impact of magic numbers and odd-even staggering on nuclear properties. This study highlights the model’s potential utility in predicting nuclear observables of selected isotopes, whose experimental data are not available in the recently published 2020 Atomic Mass Evaluation (AME2020) data set. Our model’s remarkable achievement lies in its ability to accurately portray the binding energy of several nuclei with a root-mean-square error of approximately [Formula: see text] [Formula: see text]MeV. This performance exceeds that of both microscopic and macroscopic models, as well as recently published artificial neural network models, thus affirming its superiority in forecasting nuclear observables. In essence, this research showcases a potent approach to comprehending and forecasting nuclear properties by amalgamating advanced machine-learning techniques and intricate nuclear data sets.