Phase diagrams classification based on machine learning and phenomenological investigation of physical properties in K1 − xNaxNbO3 thin films

Abstract

In this work, we have predicted and classified the temperature-misfit strain phase diagrams of (001)-oriented K1 − xNaxNbO3 (KNN, 0 ≤ x ≤ 0.5) thin films using three classical machine learning algorithms: k-nearest neighbors, support vector machine, and deep neural networks, which have a very excellent prediction accuracy rate of about 99%. Furthermore, various physical properties including ferroelectric, dielectric, piezoelectric, and electrocaloric properties have been calculated and studied based on the phenomenological Landau–Devonshire theory. The calculated results show that the dielectric constant ɛ33, piezoelectric coefficient d33, and isothermal entropy change ΔS of the KNN thin films can be enhanced at the orthorhombic–rhombohedral phase boundary. This work will provide theoretical guidance for experimental studies of KNN thin films.