Machine Learning for global predictions of nuclear excited states

CGS18

Invited presentation on 06/2026

The comprehensive characterization of nuclear excited states is fundamental to our understanding of nuclear structure and is a cornerstone for modern applications. In this work, we introduce a robust machine learning (ML) framework designed for the global prediction of nuclear energy levels with high precision. Utilizing the Evaluated Nuclear Structure Data File (ENSDF) as a foundational dataset, we employ a data-driven approach to map spectral properties across the nuclear landscape. Despite using a sparse training set of only 40%, the model reproduces the remaining 60% of known levels with a mean deviation within 100 keV. The model demonstrates a capacity for extrapolation into presently unreachable regions of the nuclear chart, providing a predictive roadmap for future rare-isotope beam facilities.