How are computational advances changing our understanding of atomic nuclei?

Bhoomika Maheshwari (Variable Energy Cyclotron Centre (VECC), Kolkata India)

Nucleons are bound in a nucleus by the underlying strong nuclear force but understanding their behavior poses a dual challenge: the incompleteness of unified theoretical models and the complexity of modern experimental data. Although extensive nuclear databases and a wide range of successful nuclear structure models, from phenomenological to fully microscopic approaches, are available, achieving a consistent description of diverse nuclear spectroscopic observables remains difficult. Motivated by these challenges, this talk explores how modern computational techniques can extract new physical insights beyond those accessible through traditional nuclear models, with a particular focus on a physics-driven machine learning framework combining numerical and symbolic regression.

The talk presents a machine learning methodology specifically designed for the small and highly skewed datasets typical of nuclear physics data. By combining numerical regression with symbolic regression, the framework enables robust predictions while mitigating the risk of overfitting. To move beyond the “black-box” nature of machine learning, feature importance and correlation analyses are employed to assess the physical significance of the learned relationships. Symbolic regression is then used to reverse-engineer interpretable analytical expressions from the numerically learned models. Using our recent work on nuclear charge radii as an example, we demonstrate how this approach provides both predictive power and physical interpretability. The methodology is readily applicable to a broad range of nuclear observables and can be naturally integrated into both theoretical and experimental research.

The talk further examines quantum computing as a promising approach to addressing the complexity of nuclear many-body Hamiltonians. We present our recent progress in mapping shell-model Hamiltonians onto qubit representations, enabling single-step quantum simulation of the complete low-lying shell-model spectrum. The discussion concludes with an outlook on quantum machine learning, where the complementary strengths of quantum algorithms and machine learning offer new opportunities for predicting nuclear properties.