Predictions of nuclear charge radii based on the convolutional neural network

Cao, Ying-Yu; Guo, Jian-You; Zhou, Bo

doi:10.1007/s41365-023-01308-x

Cited by 9 publications

(1 citation statement)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In nuclear physics, machine learning methods are also widely used to study various nuclear properties [39], such as nuclear mass [40][41][42][43], αdecay [44,45], β-decay half-life [46,47], low-lying excitation spectra [48][49][50], and fission yield [51,52], etc. Various machine learning methods including artificial neural networks [53,54], Bayesian neural networks [55][56][57], naive Bayesian probability classifiers [58], kernel ridge regression [59], and convolutional neural networks [60] have also been used to predict nuclear charge radii [61,62]. And machine learning methods can generally achieve higher prediction accuracies compared to the traditional nuclear theoretical models.…”

Section: Introductionmentioning

confidence: 99%

Nuclear charge radius predictions based on eXtreme Gradient Boosting

Li,

Zhang,

Fang

2024

Phys. Scr.

View full text Add to dashboard Cite

Nuclear charge radii with 8 ≤ Z ≤ 100 are studied based on the eXtreme Gradient Boosting (XGBoost) method. Besides the proton, neutron, and mass numbers, the physical quantities related to the isospin, shell, and pairing eﬀects are important to improve the performance of the XGBoost method by including them as the input variables. The XGBoost method describes the nuclear charge radii better than the Skyrme-Hartree-Fock-Bogoliubov (HFB)-21 model, especially for odd-Z nuclei. The root-mean-square deviation with respect to the experimental data is reduced from 0.025 fm of the HFB-21 model to 0.012 fm of the XGBoost method in the learning set. It is found that the XGBoost method has reliable extrapolation ability at least for the nuclei not far from the learning region, which is veriﬁed by comparison with the data in the newly measured experimental data. When extrapolated to the unknown region, the XGBoost predictions of charge radii are close to the HFB-21 results, and the deviations between them are generally less than 0.1 fm even within about 20 steps from the known region.20 steps from the known region.

show abstract