Exploring Combined Stereochemically Active Lone-Pair and Rattling Effects in Thermoelectrics with Machine Learning Potentials

Stereochemically active lone pairs (SCALPs) are recognized for their ability to break local symmetry, induce lattice anharmonicity, and influence thermoelectric properties. Similarly, rattling atoms influence the thermal conductivity by introducing additional vibrational modes that disrupt phonon transport. SCALPs containing pnictogen chalcogenides alongside rattling atoms pose challenges for calculating thermal transport properties using ab initio methods because of their noncentrosymmetric structures. Here, we employ machine learning interatomic potentials (MLIPs) to explore the combined effects of SCALPs and rattling atoms in AAsSe₂ (A = Li or Na). The strong anharmonicity in the γ-NaAsSe₂ system arises from rattling modes and active As 4s² SCALP-induced electrostatic interactions, leading to reduced lone-pair angles. The unique chemical bonding behind the high anharmonicity is attributed to antibonding states near the valence band and the flat vibrational behavior of rattler atoms, induced phonon softening, structural distortions, and enhanced phonon scattering. This work highlights the combined effects of SCALPs and rattling atoms, leveraging MLIPs to accelerate thermoelectric material design.