How accurate are DFT forces? Unexpectedly large uncertainties in molecular datasets.

Training of general-purpose machine learning interatomic potentials (MLIPs) relies on large datasets with properties usually computed with density functional theory (DFT). A prerequisite for accurate MLIPs is that the DFT data are well converged to m…

Preview only. Read the full abstract at the source