Abstract:The nuclear charge radius can provide direct information about the nuclear structure. In recent years, many statistical methods such as neural networks and Gaussian processes have been used to study nuclear charge radii. Based on previous research, this paper proposes an improved naive Bayesian probability (iNBP) classifier to study the nuclear charge radius. The key strategy of the iNBP method is to treat the prediction of nuclear charge radii as a classification problem, interpreting the predicted charge radii as the most reasonable expectations. First, the residuals are divided into groups by the k-means method to generate prior and conditional probabilities. During the calculation, a weight function is introduced to describe the local relationship between the nuclear charge radii. Next, the expected posterior probability is further calculated by the Bayesian formula. We choose the expectation with the highest probability as the final prediction. The reliability of the iNBP method is evaluated by global optimization and extrapolation ability based on three different theoretical models. Using the proposed method, the accuracy of charge radius prediction for the HFB model, RMF model, and the semiempirical formula was improved by 50%, 56%, and 44%, respectively. The calculations also show that the iNBP method has a strong extrapolation ability to predict charge radii in unknown regions in the nuclear chart. By analyzing the charge radii of the isotopes, the iNBP method can reproduce the parity and shell effects in the isotope chains. Combining the global description with the local relationship of nuclear charge radii, the iNBP method allows considerable fine-tuning of the theoretical results of complex nuclear models. The method proposed in this paper can also be used in other areas of nuclear physics.

Improved naive Bayesian probability classifier in predictions of nuclear mass

Improved naive Bayesian probability classifier in nuclear charge radius prediction

Novel Bayesian probability method in predictions of nuclear masses

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

Nuclear mass predictions based on Bayesian neural network approach with pairing and shell effects

Nuclear mass predictions with machine learning reaching the accuracy required by $r$-process studies

Nuclear masses learned from a probabilistic neural network

Nuclear Mass Predictions for the Crustal Composition of Neutron Stars: A Bayesian Neural Network Approach

Nuclear mass predictions with the naive Bayesian model averaging method

Refining mass formulas for astrophysical applications: a Bayesian neural network approach

Nuclear binding energy predictions based on BP neural network

A New Method to Improve the Generalization Ability of Neural Networks: A Case Study of Nuclear Mass Training

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

Uncertainties of Nuclear Level Density estimated by Bayesian Neural Networks

From neutron star observations to nuclear matter properties: A machine learning approach

Improved predictions of phenomenological nuclear charge radius formulae with Bayesian optimization approach

Neural Network Approach to Improve the Quality of Atomic Nuclei

Validating neural-network refinements of nuclear mass models

Bayesian averaging for ground state masses of atomic nuclei in a Machine Learning approach

Bayesian Neural Network improvements to nuclear mass formulae and predictions in the SuperHeavy Elements region

Performance of the Levenberg–Marquardt Neural Network Approach in Nuclear Mass Prediction