XBXiao Jun Bao

Xiao Jun Bao

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Xiao Jun Bao has a total of 50 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Research on Actinide Fission Yields: A Physics-informed Data-driven Approach

    by Qing Qing Yao, Hao Wen Sheng, Tian Liang Zhao, Xiao Jun Bao

    To address the need for accurate fission-yield data in applications such as r -process nucleosynthesis, and the limitations of existing models, this work proposes a physics-informed multimode fission-yield (PI-MFY) prediction model. The model is based on multimode fission theory and, adopting Bohr’s compound-nucleus independence hypothesis as a guiding assumption, constructs a four-dimensional physical feature space. Within this framework, an inverse-distance-weighted K -nearest-neighbors (IDW-KNN) algorithm is adopted to infer the model parameters nonparametrically from neighboring systems, enabling accurate predictions of the eight crucial parameters in multimode fission theory and subsequently yielding the corresponding pre-neutron-emission fission–fragment-mass distributions. Comparison of the calculated results of multiple fission systems with experimental data in the EXFOR database shows that the model achieves an RMSE of 0.00163 on the test set not involved in training, effectively avoiding overfitting and demonstrating good generalization capability. The model accurately reproduces the peak positions of the asymmetric bimodal structure arising from shell effects. Moreover, it exhibits greater flexibility than the traditional empirical formula and the general description of the fission observables (GEF) model in describing the energy dependence of the fission yields associated with the gradual damping of microscopic shell effects as the excitation energy increases. The present model is able to provide fission–fragment-mass distributions in good agreement with experimental data for fissioning systems that lie within or reasonably close to the experimentally covered nuclear region, thereby serving as a complementary approach to traditional methods. At this stage, however, the approach is more appropriately regarded as an auxiliary tool that provides a comparative reference for analyzing trends in fission–fragment-mass distributions in neutron-rich regions relevant to r -process nucleosynthesis.

  • Machine-Learning-Based Prediction of Heavy-Ion Capture Cross Sections with System-Level Cross-Validation

    by Jie Liu, Tianliang Zhao, Xiao-Jun Jun Bao

    This work investigates machine-learning approaches for predicting capture cross sections of heavy-ion capture reactions. Three models are implemented: Tabular Prior-data Fitted Network (TabPFN), Gaussian Process Regression (GPR), and Backpropagation Neural Network (BP). All models are trained and tested on capture-cross-section data comprising 592 projectile-target combinations and 14\,523 experimental data points. Two data-splitting strategies are adopted: point-wise random split (subscript P, for interpolation on known systems) and system-wise split (subscript S, for system-level test-set prediction), both at an 8:2 training-to-test ratio. Under point-wise-split interpolation, \(TabPFN_{P}\) achieves the lowest test-set MAE of 0.0773, outperforming \(GPR_{P}\) and \(BP_{P}\). Under the more challenging system-wise-split protocol, \(TabPFN_{S}\) still yields the smallest test MAE of 0.1878, whereas \(GPR_{S}\) suffers severe performance degradation. We further perform leave-one-system-out cross-validation (LOSO-CV) for \(TabPFN_{S}\), including ablation experiments comparing the 5-feature baseline and the full 13-feature model to examine the utility of additional entrance-channel-related physical features. The statistics show that 95\% of unseen projectile-target pairs satisfy $\overline{\mathrm{MAE}} \le 0.5$. Binned-residual statistics are also carried out to analyse residual distributions across different relative-energy bins. For three fully-external local in-distribution test cases of unseen projectile-target pairs, \(TabPFN_{S}\) delivers reasonable mean-point predictions together with physically plausible 95\% prediction bands. These results indicate that \(TabPFN_{S}\) provides a promising computational tool for unseen reaction pairs within the training-data-covered parameter domain, especially for systems that are experimentally inaccessible or difficult to measure.

  • Systematic study of proton radioactivity using the transfer matrix method

    by Haitao Yang, Zhongxia Zhao, Zhangyan Li, Xiaopan Li, Gongming Yu, Xiaojun Bao

  • Open Access

    Precise Inference of Nuclear Level Density Parameters Far from the Stable Line in Nuclear Astrophysics: The Advantages and Verification of the Physical-constraint Neural Network

    by Tian Liang Zhao, Yi Xiang Diao, Xiao Jun Bao

    This study tackles the accurate inference of level density parameters for nuclides far from stability in nuclear astrophysics. It systematically demonstrates the superiority of physics-constrained neural networks over traditional Bayesian inference. Using the extended Thomas–Fermi model, four progressive training strategies were designed: pure data-driven, physical formula parameterization, embedded prior knowledge, and imposed physical constraints. Based on resonance spacing data for 217 nuclides, an end-to-end optimization of the parameters a , b , c , and shell damping energy E d was performed via a Transformer architecture with differentiable physical formulas. Results show that the strategy with no prior information plus explicit physical constraints performs best, achieving a root mean square error of 0.471—significantly better than the Bayesian benchmark (0.643). Test set accuracy exceeds training set accuracy, indicating strong generalization. By embedding basic physical laws into the loss function, this approach avoids nonphysical oscillations from pure data fitting and prior-bias overfitting, while preserving the ability to capture diverse parameter trends across nuclide groups. Residual analysis and parameter distribution verification confirm that the model’s multimodal outputs align systematically with mass number and isospin, successfully reproducing the expected parameter peak near magic numbers due to shell effects. This work establishes a reliable machine learning framework for uncertainty propagation in next-generation nucleosynthesis calculations and offers methodological guidance for nuclear model parameterization.

  • Systematic study of proton radioactivity based on the double-folding potential model

    by Haitao Yang, Zhong-Xia Zhao, Xiao-Pan Li, Yu-Hui Luo, Xun Xue, Xiao-Jun Jun Bao

    We study proton radioactivity in proton-rich nuclei with $53 \le Z \le 83$ within a semi-microscopic framework in which the emitted proton is described by a finite-size density distribution. Two proton-density prescriptions are considered, namely a Gaussian profile and a Fermi profile. We find that, when microscopic spectroscopic factors are combined with a double-folding potential based on finite-size proton and daughter densities, the Fermi-density prescription yields the best overall agreement with experiment, reducing the global root-mean-square (RMS) deviation of $\log_{10} T_{1/2}$ to $\sigma = 0.37$. In addition, a modified universal decay law that embeds the same spectroscopic factors yields an even smaller deviation, $\sigma = 0.28$, thereby providing a high-precision global description of proton-radioactivity half-lives. Specifically, we construct the proton–daughter interaction by folding a Skyrme-type effective nucleon–nucleon interaction with finite-size proton densities and spherical daughter densities, and we evaluate proton-emission half-lives within the Wentzel–Kramers–Brillouin (WKB) approximation for 39 known proton emitters in this region. State-dependent spectroscopic factors $S_p^{\mathrm{QYB}}$ and $S_p^{\mathrm{ZHF}}$, obtained from relativistic mean-field plus Bardeen–Cooper–Schrieffer (RMF+BCS) calculations, are employed to account for proton preformation. We then confront the calculated half-lives with experimental data and analyze the residuals by grouping them according to the minimum orbital angular momentum and whether the decay proceeds from the ground state or an isomeric state. Finally, we extend the universal decay law for proton emission (UDLP) by including an explicit spectroscopic-factor term, thereby obtaining a modified formula (MUDLP) that provides a compact global parametrization of experimental proton-radioactivity half-lives. \textcolor{red}{Using the best-performing semi-microscopic prescription together with the MUDLP parametrization, we provide conditional half-life estimates for several candidate proton emitters near the proton drip line, where the input $Q_{p}$ values are taken from AME2020. These results provide quantitative reference values for future experimental searches for new proton-radioactive nuclei.


  • Investigation of the quasifission process in the synthesis of superheavy nuclei

    by Qing Qing Yao, Shu Qing Guo, Tian Liang Zhao, Xiao Jun Bao

  • α -decay half-lives of superheavy nuclei with support-vector regression

    by Haitao Yang, Xiaopan Li, Xiefei Song, Dianxu Ma, Gongming Yu, Xiaojun Bao


  • Testing the reliability of the method for calculating the survival probability

    by Yi Xiang Diao, Qing Qing Yao, Tian Liang Zhao, Xiao Jun Bao