Abstract:To ensure agreement between theoretical calculations and experimental data, parameters to selected nuclear physics models, are perturbed, and fine-tuned in nuclear data evaluations. This approach assumes that the chosen set of models accurately represents the `true' distribution. Furthermore, the models are chosen globally, indicating their applicability across the entire energy range of interest. However, this approach overlooks uncertainties inherent in the models themselves. As a result, achieving satisfactory fits to experimental data within certain energy regions for specific channels becomes challenging, as the evaluation is constrained by the deficiencies of the selected models. In this work, we propose that instead of selecting globally a winning model set and proceeding with it as if it was the `true' model set, we instead, take a weighted average over multiple models within a BMA framework, each weighted by its posterior probability. The method involves executing a set of TALYS calculations by randomly varying multiple nuclear physics models and their parameters to yield a vector of calculated observables. Next, the likelihood function was computed at each considered incident energy point for selected cross sections by comparing the vector of calculated observables with that of the selected differential experimental data. As the cross sections and elastic angular distributions were updated locally on a per-energy-point basis, the approach typically results in discontinuities or "kinks" in the curves, and these were addressed using spline interpolation. The proposed BMA method was applied to the evaluation of proton induced reactions on $^{58}$Ni within 1 - 100 MeV. The results demonstrate favorable comparisons with experimental data, as well as with the TENDL-2021 evaluation.

Nuclear mass predictions with the naive Bayesian model averaging method

Novel Bayesian probability method in predictions of nuclear masses

Uncertainty quantification of mass models using ensemble Bayesian model averaging

Nuclear masses learned from a probabilistic neural network

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

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

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

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

Bayesian Model Averaging (BMA) for nuclear data evaluation

New Extrapolation Method for Predicting Nuclear Masses

Nuclear mass predictions with radial basis function approach

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

Nuclear binding energy predictions based on BP neural network

Radial basis function approach in nuclear mass predictions

Validating neural-network refinements of nuclear mass models

Further improvements on a global nuclear mass model

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

Improved naive Bayesian probability classifier in predictions of nuclear mass

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

High Precision Nuclear Mass Predictions Towards a Hundred Kilo-Electron-volt Accuracy

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