Molecular dynamics simulations for thermoelectric materials

Abstract

High temperature copper chalcogenides such as Cu2S are promising thermoelectric materials due to the combination of a solid S lattice and liquid Cu atoms in the context of the so-called phonon-liquid electron-crystal (PLEC) mechanism. Hexagonal β-Cu2S existing at lower temperatures also exhibits such hybrid nature with anisotropic diffusion channels for Cu atoms in the presence of hexagonal S layers. We performed ab initio molecular dynamics simulations over 50 picoseconds for β-Cu2S at 450 K. The Cu atoms are diffusive when they travel between S layers and yield a calculated diffusion coefficient close to 5×10−6 cm2s−1 in the in-plane direction and a 50% smaller diffusion coefficient in the vertical direction. We further investigate the surface melting effect in hexagonal β-Cu2S at 450 K. The diffusion coefficient of interlayer Cu atoms increases to 1.14×10−5 cm2s−1 near the surface, about twice of that in the bulk and is within the range of a typical liquid. Besides liquid-like atoms, strong anharmonicity can also result in a low thermal conductivity for solids such as SnSe. We explore the possibility of applying machine-learning techniques to systematically sort out the significant higher-order interactions in solids and estimate the thermal conductivity via molecular dynamics simulations. The fitting accuracy is above 99% for Si at 500 K, and above 98% for the low-temperature NaCl and SnSe.