Thermal conductivity of the layered titanate K0.8Li0.27Ti1.73O4 explored by a deep learning interatomic potential

Abstract

The theoretical prediction of thermal conductivity in many layered oxides remains challenging, primarily due to their structural complexity and low symmetry. The traditional Boltzmann transport equation method is highly accurate but limited by the low-order phonon scattering model, which makes it difficult to resolve the high-order scattering effects of low symmetry layered materials. The classical molecular dynamics calculation is efficient but lacks accuracy due to the missing multi-component potential function. In this study, we develop a strategy to predict the thermal conductivity of K0.8Li0.27Ti1.73O4 (KLTO), a model of layered oxides by machine-learning using a deep neural network model to acquire the interatomic potential of KLTO. The deep learning potential (DLP) is in excellent agreement with density functional theory in predicting atomic force, energy, and elastic properties. In addition, the calculated out-of-plane thermal conductivity values based on the DLP (0.37 W m−1 K−1) are close to experimental results (0.28 W m−1 K−1). This machine-learning framework for constructing interatomic potentials can be extended to other layered materials, offering a promising approach for advancing the theoretical study of such systems.