Generative modeling for the nucleon-nucleon interaction

Developing precision models for nuclear interactions remains one of the primary challenges in nuclear physics and is key to a better understanding of theoretical uncertainties for many nuclear many-body calculations. In recent years, Chiral Effective Theory has become the most widely used framework for constructing nuclear potentials due to its ability to systematically generate contributions order-by-order in a well-defined expansion parameter. However, the uncertainties related to the truncation of order in the chiral expansion and the choice of regulator cutoff have not been fully estimated. To address this issue, an effective method to generate a range of chiral potentials with different truncations and cutoffs is required. We propose the application of the Glow model, a generative machine learning model which can learn the properties of a set of data and generate new samples for the nuclear interaction model. Our results show that a well-trained Glow model can accurately reproduce the original interactions used for training and in addition generate realistic nuclear potentials at different momentum-space cutoffs that reproduce very well nucleon-nucleon scattering phase shifts.