A physics-informed neural network predicts nuclear binding energies to about 0.1 MeV and reproduces pairing and shell effects, with extrapolation tested against new AME2020 data.
Microscopic mass formulae
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
By assuming the existence of a pseudopotential smooth enough to do Hartree-Fock variations and good enough to describe nuclear structure, we construct mass formulae that rely on general scaling arguments and on a schematic reading of shell model calculations. Fits to 1751 known binding energies for N,Z$\geq 8$ lead to rms errors of 375 keV with 28 parameters. Tests of the extrapolation properties are passed successfully. The Bethe-Weizs\"acker formula is shown to be the asymptotic limit of the present one(s). The surface energy of nuclear matter turns out to be probably smaller than currently accepted.
citation-role summary
background 1
citation-polarity summary
fields
nucl-th 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Validation and extrapolation of atomic mass with physics-informed fully connected neural network
A physics-informed neural network predicts nuclear binding energies to about 0.1 MeV and reproduces pairing and shell effects, with extrapolation tested against new AME2020 data.