Machine Learning Lattice Parameters of M2AX Phases

Abstract

MAX phases, with a general composition of M_n+1AX_n, are layered materials with hexagonal symmetry that have increasingly captivated a lot of attention because of their unique way of combining ceramic and metallic properties into a homogeneous bulk material. We developed a machine learning approach to predict the lattice parameters a and c of M₂AX phases. This approach consists of training an ensemble model on a data set collecting all experimentally synthesized M₂AX phases’ lattice parameters. Our approach combines a data augmentation scheme with state-of-the-art regression models and hyperparameter optimization tools. We tested our model on newly synthesized compositionally complex high-entropy M₂AX phases with positive results. Finally, we also show that our machine learning predictions of lattice parameters are useful as initial values for variable-cell relaxations of M₂AX structures with the density functional theory.