Machine learning predictions of nuclear stability

Machine learning (ML) methods have become a useful tool in many areas of physics, including Nuclear Physics. ML methods' ability to take in aggregate information about the behaviour of the system and predict trends is able to make relevant and verifiable predictions. The existence of super heavy stable nuclei in currently experimentally inaccessible regions has been predicted by the nuclear shell effect, however, its location and extension are still in dispute by different models (e.g. Z=120 N=172 from relativistic models) [1]. We aim to apply ML tools to develop accurate statistical models to predict isotopic lifetimes in regions of heavy nuclei, such as the fabled stability island. We explore various ML methods, and study the predictive power of different nuclear representations. We use ML models both with and without theoretical bias from nuclear models, such as isotopic magic numbers, over different radioactive decay channels to possibly offer a glimpse of the next stability region.

[1] Sorlin, O., and M-G. Porquet. "Nuclear magic numbers: New features far from stability." Progress in Particle and Nuclear Physics 61.2 (2008): 602-673.