Cubic halide perovskites as potential low thermal conductivity materials: A combined approach of machine learning and first-principles calculations

Thermal conductivity is the key factor affecting thermoelectric properties of materials. Here, machine-learning techniques combined with first-principles calculations are used to identify the cubic halide perovskites ${\text{Cs}B\text{Br}}_{3}$ (B = Ca, Cd, and Sn) with ultralow thermal conductivity. Based on the Boltzmann transport equation within the relaxation time approximation, we demonstrate this type of perovskites have remarkably low lattice thermal conductivities ${\ensuremath{\kappa}}_{L}\ensuremath{\sim}\phantom{\rule{4pt}{0ex}}0.4\ensuremath{-}1$ W/mK at 300 K. We employ the self-consistent phonon theory incorporating both cubic and quartic anharmonicity, which is considered from the bubble and loop self-energy diagrams rather than many-body perturbation theory. We show that the approach yields a cubic-tetragonal phase transition of ${\mathrm{CsCaBr}}_{3}$ at temperature ${T}_{c}=226\ensuremath{-}265\phantom{\rule{0.16em}{0ex}}\mathrm{K}$, in good agreement with the experimental value of 239 K. An anomalously temperature dependence of ${\ensuremath{\kappa}}_{L}$ is observed in ${\mathrm{CsCdBr}}_{3}$, where the coherent term account for 26% of the total lattice thermal conductivity. We also demonstrate that the hardening of vibrations in low-lying phonon modes offset the phonon population effect as temperature increases by reducing the available phase space.