Characterizing ferroelectricity in KTaO₃ with first-principles calculations and machine learning

Perovskite oxides exhibit a variety of structural instabilities which lead to emergent phenomena such as ferroelectricity. Understanding how such phenomena can be controlled via external fields (electric field, strain, etc.) is therefore of great importance for technological applications. In this study, we use decision trees and first-principles calculations to model the soft polar phonon in KTaO₃ based on the magnitudes of different strain-related degrees of freedom. Our decision tree models generate predictions that overlap with both DFT and experimental findings. Furthermore, these models provide an interpretable framework that yields physically meaningful conclusions about the driving force behind ferroelectricity under strain. This method can also be extended to multiferroics which we demonstrate by identifying relevant order parameters related to magnetic ordering in EuTiO₃.