Deep neural networks for accurate predictions of crystal stability

Predicting the stability of crystals is one of the central problems in materials science. Today, density functional theory (DFT) calculations remain comparatively expensive and scale poorly with system size. Here we show that deep neural networks utilizing just two descriptors—the Pauling electronegativity and ionic radii—can predict the DFT formation energies of C 3 A 2 D 3 O 12 garnets and ABO 3 perovskites with low mean absolute errors (MAEs) of 7–10 meV atom −1 and 20–34 meV atom −1 , respectively, well within the limits of DFT accuracy. Further extension to mixed garnets and perovskites with little loss in accuracy can be achieved using a binary encoding scheme, addressing a critical gap in the extension of machine-learning models from fixed stoichiometry crystals to infinite universe of mixed-species crystals. Finally, we demonstrate the potential of these models to rapidly transverse vast chemical spaces to accurately identify stable compositions, accelerating the discovery of novel materials with potentially superior properties. Crystal stability prediction is of paramount importance for novel material discovery, with theoretical approaches alternative to expensive standard schemes highly desired. Here the authors develop a deep learning approach which, just using two descriptors, provides crystalline formation energies with very high accuracy.