Space-group determination by combining deep learning-based prediction and Le Bail analysis: Perovskite structure as an example

Abstract

When dealing with crystalline inorganic compounds, a great deal of skill and expertise are entailed in constructing an initial input structure for the Rietveld analysis from their powder X-ray diffraction (XRD) profiles. For newly synthesized materials in particular, determining the space group to which the compound belongs, the first step to begin the Rietveld analysis, often relies on experimentalists’ intuition and experience. Occasionally, we need to try several candidate space groups through trial and error, which takes a significant amount of time. To streamline this initial step, this study demonstrates a scheme for determining the valid space group and lattice parameters of an unknown compound using the Le Bail analysis. This scheme uses a deep-learning model, SpgVolNet, developed in our previous study to predict candidate space groups from an experimental XRD profile. Although we had known the structure of the model compound in this study, perovskite-type La_0.6Sr_0.4Mn_0.9Co_0.1O₃, we assumed it to be “unknown.” The SpgVolNet predicted that this compound belongs to R3c with the highest probability. Le Bail analysis under R3c and several other plausible space groups led us to the “correct” conclusion that the compound indeed belongs to R3c with reliable lattice parameters.