Accelerated search for new ferroelectric materials

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