2018/12/05 by Andrew T. Fowler, Chris J. Pickard, Fowler, Andrew T. +3
Chemistry · Computer Science · Materials Science · #Computational Drug Discovery Methods #Computational Physics (physics.comp-ph) #FOS: Physical sciences #Machine Learning in Materials Science #Mass Spectrometry Techniques and Applications #Materials Science (cond-mat.mtrl-sci)
paper · pdf · doi:10.48550/arxiv.1812.01966
openalex publication_date 2018/12/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Faithful representations of atomic environments and general models for\nregression can be harnessed to learn electron densities that are close to the\nground state. One of the applications of data-derived electron densities is to\norbital-free density functional theory. However, extrapolations of densities\nlearned from a training set to dissimilar structures could result in inaccurate\nresults, which would limit the applicability of the method. Here, we show that\na non-Bayesian approach can produce estimates of uncertainty which can\nsuccessfully distinguish accurate from inaccurate predictions of electron\ndensity. We apply our approach to density functional theory where we initialise\ncalculations with data-derived densities only when we are confident about their\nquality. This results in a guaranteed acceleration to self-consistency for\nconfigurations that are similar to those seen during training and could be\nuseful for sampling based methods, where previous ground state densities cannot\nbe used to initialise subsequent calculations.\n