Machine learning the nuclear mass

Machine learning in nuclear physics at low and intermediate energies

“…the magic numbers around Z = 120 and N = 184 in the superheavy nuclei region. A recent development of the gradient boosting decision tree, the LightGBM was also used to refine nuclear mass models, which achieves very high accuracies for both training and testing data [45]. Moreover, the importance of input fea-tures in refining mass models was analyzed by investigating the correlation between the input characteristic quantities and the output, which may provide new insights for further developing nuclear mass models [45].…”

“…A recent development of the gradient boosting decision tree, the LightGBM was also used to refine nuclear mass models, which achieves very high accuracies for both training and testing data [45]. Moreover, the importance of input fea-tures in refining mass models was analyzed by investigating the correlation between the input characteristic quantities and the output, which may provide new insights for further developing nuclear mass models [45]. In addition, the RBF approach [46] and its improved version of RBF with odd-even effects (RBFoe) [47] have also been widely used to improve the predictions of nuclear masses [48][49][50][51][52][53][54].…”

“…Thus the probability of model or parameters can be assigned, which is thought more reliable and flexible than the frequentist approach in quantifying uncertainties on the reaction observables. [29] and [45] in panel (a) are taken from Ref. [116].…”

Recent progress in two-proton radioactivity

“…Exploring the candidate 2p emitters using density functional theory. -The ground-state 2p decay usually occurs beyond the 2p dripline; here, single-proton emission is energetically forbidden or strongly suppressed owing to the odd-even binding energy staggering attributable to the pairing effect [17][18][19][133][134][135]. Meanwhile, the presence of the Coulomb barrier has a confining effect upon the nucleonic density; hence, relatively long-lived 2p emitters should be expected.…”

Multi-task learning on nuclear masses and separation energies with the kernel ridge regression

“…For nuclear physics, ML applications can be traced back to early 1990s [28,29], and recently, it has been widely adopted to many aspects of nuclear physics, e.g., β-decays [30,31], fusion reaction crosssections [32], charge radii [33,34,35], excited states [36,37,38], nuclear landscape [39,40,41], fission yields [42,43], variational calculations [44,45], extrapolations for manybody physics [46,47,48,49], nuclear energy density functional [50], etc. In particular for nuclear masses, many ML approaches have been employed to improve its description, such as the kernel ridge regression (KRR) [51,52,53], the radial basis function (RBF) [54,55,56,57,58], the Bayesian neural network (BNN) [59,60,61,62], the Levenberg-Marquardt neural network [63], the gaussian process regression [64], the light gradient boosting machine [65], the Bayesian probability classifier [66], etc. By training the ML network with the mass model residuals, i.e., deviations between experimental and calculated masses, ML approaches can significantly reduce the corresponding rms deviation to about 200 keV [51,55,…”