Microscopic derivation of the interacting boson model parameters with machine learning

TL;DR

This paper introduces a physics-guided neural network that derives interacting boson model parameters from microscopic nuclear energy landscapes, enhancing nuclear structure predictions without manual tuning.

Contribution

A novel machine learning approach that integrates nuclear structure information to microscopically determine interacting boson model parameters.

Findings

Successfully reproduces microscopic energy landscapes for rare-earth nuclei.

Provides nuclear model parameters and spectra reflecting structural evolution.

Offers a robust alternative to traditional microscopic descriptions.

Abstract

Machine learning is applied to derive microscopically parameters of the interacting boson model for nuclear spectroscopy. A physics-guided neural network is proposed, which is trained to map the potential energy landscapes that are calculated within the nuclear density functional theory onto the bosonic parameter space. To incorporate the underlying nuclear structure information and mitigate parameter degeneracy, the network integrates a global quadrupole collectivity indicator and valence nucleon numbers as key input features. In its applications to rare-earth nuclei, by reproducing the microscopic energy landscapes without any manual parameter tuning, the trained network is shown to provide a set of the model parameters and energy spectra that reflect the nuclear structural evolution, offering a robust alternative microscopic description of nuclear collectivity.