We report the development of a combined machine learning and high-throughput density functional theory (DFT) framework to accelerate the search for new ferroelectric materials. The framework can predict potential ferroelectric compounds using only elemental composition as input. A series of machine-learning algorithms initially predict the possible stable and insulating stoichiometries with polar crystal structures, necessary for ferroelectricity, within a given chemical composition space. A classification model then predicts the point groups of these stoichiometries. A subsequent series of high-throughput DFT calculations finds the lowest-energy crystal structure within the point group. As a final step, nonpolar parent structures are identified using group theory considerations, and the values of the spontaneous polarization are calculated using DFT. By predicting the crystal structures and the polarization values, this method provides a powerful tool to explore new ferroelectric materials beyond those in existing databases.

• Received 27 September 2022
• Revised 24 January 2023
• Accepted 3 March 2023

DOI:https://doi.org/10.1103/PhysRevResearch.5.023122

Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to the author(s) and the published article's title, journal citation, and DOI.

Published by the American Physical Society