Abstract
We present a hybrid framework that integrates nuclear density functional theory with machine learning to investigate spin-parity effects in nuclear binding energies. Our approach employs a Skyrme energy density functional augmented with isospin-symmetry breaking terms calibrated through analysis of mirror displacement energies (MDEs). The machine learning component utilizes a multi-layer neural network architecture that learns the mapping between mean-field quantities and their symmetry-restored counterparts after angular-momentum and isospin projection. Our analysis reveals sensitivity to nucleon spin-parity alignments, with the symmetric K02 (aligned) configuration demonstrating improved predictive performance (RMSE = 0.26 MeV) compared to the anti-aligned K01 arrangement (RMSE = 0.31 MeV) when benchmarked against AME2020 experimental data. The framework exhibits reasonable extrapolation capabilities for nuclear mass predictions across the nuclear chart, showing performance comparable to established theoretical models including DZ28, FRDM2012, HFB25, WS, WS3, and WS4. The model reproduces MDEs with typical deviations below 0.3 MeV for most cases, although some discrepancies persist, indicating that isospin-symmetry breaking effects are not yet fully captured in all regions. This work demonstrates the potential of combining nuclear theory with physics-informed machine learning for nuclear mass predictions, while also highlighting areas where further theoretical development is needed.

This publication has 44 references indexed in Scilit: