Deep Neural Network for Detecting Nucleon-Nucleon Bound States

Abstract

There is a long-standing ambiguity in the interpretation of near-threshold enhancement in hadron-hadron scatterings. The origin of enhancement can reveal the nature of the interaction between the two hadrons. However, there is no straightforward approach to probe the nature of near-threshold enhancement in a model-independent manner. The present study aimed to formulate a deep learning approach to detect a two-hadron bound state given only the partial scattering cross-section. To ensure that analyticity and unitarity were satisfied, an S-matrix model with a parameterized background was used for the training dataset. A deep neural network (DNN) model was designed and developed using the Adam and AMSGrad optimizers. To demonstrate that the trained DNN model can generalize beyond the training dataset, two variants of exact amplitudes of separable potential were used for validation. Finally, without using the deuteron’s known properties, such as the binding energy and its magnetic moment, the model identified the correct origin of threshold enhancement in the nucleon-nucleon scattering data. The proposed method can be applied to analyze the recently discovered near-threshold enhancements observed in scattering experiments.