Beyond Boltzmann transport: Green–Kubo prediction of lattice thermal conductivity with machine-learned potentials

Abstract

We investigate the microscopic origin of the ultralow lattice thermal conductivity in Rb2ZnTe by combining anharmonic lattice dynamics based on first-principles density functional theory with the Boltzmann transport equation (BTE). Full cubic and quartic anharmonicity is included, accounting for phonon renormalization and both three-phonon and four-phonon scattering processes. In addition, due to the unusually strong coherence effects in Rb2ZnTe, we consider off-diagonal contributions beyond the conventional BTE framework. The calculated phonon transport approaches the Ioffe–Regel limit, indicating the breakdown of the quasiparticle picture. To obtain an accurate thermal conductivity, we employ a machine-learning potential and perform Green–Kubo molecular dynamics simulations using GPUMD. At 300 K, the computed thermal conductivity increases by 38% compared to previous estimates, yet remains close to the amorphous limit, highlighting Rb2ZnTe as a promising thermoelectric material. The methodology presented here provides a robust framework for understanding lattice thermal transport in strongly anharmonic crystals and guiding the rational design of low-thermal-conductivity materials.