Abstract:A deep neural network (DNN) has been developed to generate the distributions of nuclear charge density, utilizing the training data from the relativistic density functional theory and incorporating available experimental charge radii of 1014 nuclei into the loss function. The DNN achieved a root-mean-square (rms) deviation of 0.0193 fm for charge radii on its validation set. Furthermore, the DNN can improve the description in both the tail and central regions of the charge density, enhancing agreement with experimental findings. The model's predictive capability has been further validated by its agreement with recent experimental data on charge radii. Finally, this refined model is employed to predict the charge density distributions in a wider range of nuclide chart, and the parameterized charge densities, charge radii, and higher-order moments of charge density distributions are given, providing a robust reference for future experimental investigations.

What problem does this paper attempt to address?

This paper aims to develop a deep neural network (DNN) model for predicting the charge density distribution of atomic nuclei worldwide. Researchers used the training data of the relativistic density functional theory (RDFT) and existing experimental charge radii to construct this model. The DNN achieved a root mean square deviation of 0.0193 fm for the charge radii on the validation set and improved the description of the tails and central region of the charge density distribution to better match the experimental results. The paper first introduces the importance of charge density distribution, which reflects nuclear structural information and electromagnetic properties. Currently, the research on charge density based on RDFT mainly focuses on charge radii, while DNN can improve the accuracy of the description by incorporating experimental data. In this study, the authors used a DNN with four hidden layers to learn the Fourier-Bessel coefficients of the charge density distribution calculated by RCHB theory. Then, the experimental charge radius information was included in the loss function to constrain the charge density and obtain more accurate results. The paper demonstrates the performance of the DNN model on different nuclide graphs. Compared to RCHB theory, DNN shows improvement in predicting charge radii in the light to medium mass region. However, for the charge radius evolution trend of certain specific nuclei, such as calcium isotopes, DNN fails to accurately capture it, which may be attributed to complex physical effects. Nevertheless, DNN still shows advantages in global prediction and detailed description, and it also has some extrapolation capability for nuclides without experimental data. Overall, the proposed method in the paper provides reliable predictions of charge density distribution for future experimental studies, but there is still room for improvement, especially in interpreting neural network results and dealing with complex nuclear physics phenomena.

Global prediction of nuclear charge density distributions using deep neural network

Prediction of Nuclear Charge Density Distribution with Feedback Neural Network

Predictions of Nuclear Charge Radii and Physical Interpretations Based on the Naive Bayesian Probability Classifier

Taming nucleon density distributions with deep neural network method

Predictions of Nuclear Charge Radii Based on the Convolutional Neural Network

A Novel Bayesian Neural Network Approach for Nuclear Root-Mean-square Charge Radii

Improved naive Bayesian probability classifier in nuclear charge radius prediction

Shell quenching in nuclear charge radii based on Monte Carlo dropout Bayesian neural network

Calibration of nuclear charge density distribution by back-propagation neural networks

Nuclear mass prediction using a neural network approach

Data-driven Density Functional Model for Atomic Nuclei

Nuclear charge radii in Bayesian neural networks revisited

A Data-Driven Density Functional Model for Nuclear Systems

Deep Learning for Nuclear Physics

Deep learning on nuclear mass and $α$ decay half-lives

A Neural Network Approach Based on More Input Neurons to Predict Nuclear Mass *

Deep learning approach to nuclear masses and -decay half-lives

Uncertainties of Nuclear Level Density estimated by Bayesian Neural Networks

Nuclear mass predictions based on deep neural network and finite-range droplet model (2012)

Novel Bayesian neural network based approach for nuclear charge radii

Deep Learning: A Tool for Computational Nuclear Physics