The authors introduce the alchemical harmonic approximation (AHA) as a method to efficiently model the absolute electronic energy of charge-neutral iso-electronic diatomic molecules. The AHA relies on a parabolic approximation of the energy as a function of the difference in nuclear charges between the two atoms, with only a single calibration point required.
To account for changes in interatomic distance, the authors combine the AHA with a novel potential energy function that captures the correct behavior at short and long distances. This joint model covers the entire two-dimensional potential energy surface spanned by distance and nuclear charge differences.
The authors assess the accuracy of the AHA model by comparing it to legacy interatomic potentials like the harmonic oscillator, Lennard-Jones, and Morse potentials. They find that the AHA provides comparable accuracy for individual diatomics, but significantly better predictive power when extrapolating to the entire iso-electronic series.
The authors also investigate using the AHA as a baseline for ∆-machine learning models of diatomic energies. They show that this baseline leads to a systematic improvement, effectively reducing the training data needed to reach chemical accuracy by up to an order of magnitude compared to direct learning or using the Morse potential as a baseline.
toiselle kielelle
lähdeaineistosta
arxiv.org
Syvällisempiä Kysymyksiä