Rattling induced bonding hierarchy in Li–Cu–Ti chalcotitanates for enhanced thermoelectric efficiency: a machine learning potential approach

Abstract

The nature of chemical bonding in crystalline solids significantly influences heat conduction, impacting lattice thermal conductivity and, consequently, thermoelectric (TE) performance. In this study, we report the development of the first principles-based machine learning interatomic potentials to predict TE efficiency in chalcogenide-based materials. We present unique lattice dynamics characterized by a particle-like phonon transport model, defined by anharmonic scattering rates and flat bands in phonon spectra, further influencing the thermal conductivity. The presence of Li in such chalcogenide-based systems exhibits a strong rattling effect, leading to a hierarchy in bonding and resulting in ultralow thermal conductivity. The presence of a bonding hierarchy, arising from both strong and weak lattice interactions in Li–Cu–Ti chalcotitanates, allows for the modulation of both electronic and thermal transport properties. Additionally, bonding analysis and atomic displacement parameters reveal the presence of Li and Ti/Cu rattling cations in LiCu₃TiTe₄, which strongly scatter the heat-carrying phonons and lead to overdamped atomic vibrations, significantly reducing thermal conductivity. Thus, our machine learning potential-based work provides a comprehensive understanding of the bonding hierarchy in these systems due to the presence of rattling cations and we show that interactions between such rattling cations can improve the TE efficiency significantly.