Fig. 4: Benchmark of the latent Ewald summation (LES) model on the dipeptide test set.
From: Machine learning of charges and long-range interactions from energies and forces
a Top: A snapshot of a dipeptide conformer from the test set. Bottom: The predicted charge (q) distribution. The color bar is in linear scale. b The predicted charges from LES compared to minimal basis iterative stockholder (MBIS) charges in SPICE43. c The predicted dipole components computed from the LES charges (\({{\mathbf{\mu }}}=\mathop{\sum }_{i=1}^{N}{q}_{i}{{{\bf{r}}}}_{i}\)) compared to the density functional theory (DFT) dipole components in SPICE. d The predicted traceless quadrupole components computed from the LES charges (\(Q\,=\,\mathop{\sum }_{i=1}^{N}{q}_{i}{{{\bf{r}}}}_{i}\otimes {{{\bf{r}}}}_{i}\)) compared to the DFT quadrupole components in SPICE. e The predicted diagonal born effective charge (BEC, \({Z}_{\alpha \alpha }^{*}=\partial {{{\boldsymbol{\mu }}}}_{\alpha }/\partial {{{\bf{r}}}}_{\alpha }\)) components compared to BECs calculated with the ωB97M-D3BJ DFT functional in the Def2SVP basis. f The comparison for the off-diagonal BEC components (\({Z}_{\alpha \beta }^{*}=\partial {{{\boldsymbol{\mu }}}}_{\alpha }/\partial {{{\bf{r}}}}_{\beta }\)). Squared Pearson correlation coefficients R2 are shown in each plot where quantitative agreement is expected.
