Probing Lattice Anharmonicity and Thermal Transport in Ultralow‐κ Materials Using Machine Learning Interatomic Potentials

Crystalline solids with ultralow lattice thermal conductivity (κ) are highly sought after for thermoelectric energy conversion and thermal barrier coating applications. However, a comprehensive theoretical understanding of heat transport in strongly anharmonic materials remains limited, as conventional perturbative frameworks such as the Boltzmann transport equation (BTE) break down when anharmonicity is too strong to be treated as a "small" perturbation. Herein, machine learning interatomic potentials (MLIP) are developed to investigate thermal transport in TlAgSe, a metal chalcogenide, and Cs₂PbI₂Cl₂, an all-inorganic layered Ruddlesden-Popper perovskite. The anharmonic lattice dynamics, structural properties, and finite-temperature distortions are examined using MLIP-driven molecular dynamics (MD) simulations, revealing local symmetry breaking typical of ultralow-κ (<1 Wm^-1K^-1) materials. The linear response theory-based Green-Kubo (GK) framework, implemented via equilibrium MD simulations, is employed to calculate the κ. The non-perturbative GK framework captures all anharmonic effects of underlying interatomic potentials and yields κ closely matching experimental values. Phonon scattering rates exceeding the Ioffe-Regel limit and the degree of anharmonicity σ^A > 0.5 confirm the strongly anharmonic nature of both materials. This MLIP-integrated theoretical and numerical framework enhances the physical understanding of heat transport and guides the design of ultralow-κ materials.