From DFT to MLFF: Accurate and Efficient Modeling of Strain-Tuned Lattice Thermal Conductivity in MoS₂ Monolayer

On-the-fly machine learning force fields (MLFFs), trained using ab initio molecular dynamics, offer a fast and accurate alternative to density functional theory (DFT) for predicting lattice thermal conductivity (κ _L ) in two-dimensional materials. This study focuses on MoS₂ monolayer, where the MLFF demonstrates excellent agreement with DFT, yielding low root-mean-square errors (RMSEs) of 0.303 meV·atom^-1 for energies and 0.013 eV·Å^-1 for atomic forces. Leveraging this accuracy and the computational efficiency of MLFFs, we investigated the effects of biaxial tensile strain on κ _L . Our results show an 80% reduction in κ _L under 16% strain, which is attributed to suppressed phonon group velocities (v _λ), shortened phonon relaxation times (τ_λ), and enhanced anharmonicity. These mechanisms were understood through detailed analysis of strain-induced changes in phonon dispersions, τ_λ, and v _λ. This study on MoS₂ monolayer reveals that MLFFs provide a powerful and efficient framework for calculating κ _L in low-dimensional systems under strain, significantly advancing our understanding of strain-modulated thermal transport.