Neural Networks Crack the Code of Atomic Nuclei and Neutron Stars

Scientists have made progress in modelling atomic nuclei and neutron stars using artificial neural networks. The new approach improves how researchers calculate the behaviour of complex quantum systems. This method has now been successfully applied to nuclei as large as oxygen-16.

The work comes from a collaboration between Argonne National Laboratory, INFN-TIFPA Trento Institute, and the University of Oslo. Their findings could also benefit other fields studying many-body quantum systems.

The study focuses on approximating the many-body wave function—a mathematical description of how multiple interacting particles behave in a quantum system. Traditional methods struggle with larger nuclei, but neural networks provide a more flexible and efficient way to represent these wave functions. This allows for more precise solutions to the Schrödinger equation.

The team tested different wave function models, including Pfaffian-Jastrow and Backflow correlations, to improve accuracy. They also used Gaussian cutoff functions with controlled radii to smooth transitions between regions in nuclear models. This reduced boundary artifacts and helped match experimental data from neutron and proton scattering more closely.

A major achievement was extending calculations to oxygen-16, one of the most complex nuclei simulated so far. The new techniques not only handle larger systems but also offer a framework for studying nuclear clustering and superfluidity in greater detail.

The research demonstrates how artificial neural networks can advance nuclear physics by modelling intricate quantum states. The methods developed here have already improved the accuracy of nuclear calculations. Their potential applications stretch beyond this field, offering tools for other areas dealing with many-body quantum problems.