O-kRystal targets electronic-band predictions beyond standard DFT
Its creators say it predicts HSE06 bands for nearly 6,000 chemically diverse three-dimensional materials using lower-cost PBE calculations.
TLDR
The team behind O-kRystal says its graph neural network uses orbital information from lower-cost PBE calculations to predict corrections to electronic states. They report HSE06 band predictions across nearly 6,000 three-dimensional materials and say the framework, trained separately, predicts G0W0 energies in two-dimensional materials. On a separate benchmark of strongly correlated materials, they report a 68.7% reduction in PBE band-energy mean absolute error.
Combined views
546
1 Source, first seen ago
