Improving energies and densities in machine-learned quantum functionals
Stationarity Constraints Improve Both Energies and Densities of Machine-Learned Exchange-Correlation Functionals
Artificial Intelligence
Summary
Predicting how electrons behave in molecules is important but tricky, especially when using machine learning to improve calculations. The authors found that measuring a certain error in their trained functionals helps fix both energy predictions and the electron distributions at the same time. By including this check during training, they avoid complex computations and achieve better results than some existing methods on new molecules. This advances more accurate computer models of molecules.
What this means in practice
- •For computational chemists: Enhance quantum chemistry simulations with functionals that improve both energy accuracy and electron density predictions for molecules.
- •For materials design engineers: Use more reliable density functional approximations to predict molecular properties critical for designing new materials and drugs.
Authors
Yuhao Chen, Donald G. Truhlar, Xiao He
Abstract
Empirical exchange-correlation functionals are fitted on fixed orbitals but used self-consistently, so better energies often come with worse densities. We show that the Brillouin residual of a trained functional on the Kohn-Sham orbitals of an accurate density controls, to first order, both the self-consistent density shift and the gap between fitted and self-consistent energies. Penalizing it requires no self-consistent-field solution during training. Trained on 6 energies and 11 CCSD densities, a neural meta-GGA surpasses SCAN in both atomization energies and densities of unseen molecules.