📝 SummarXiv

Abstract

We employ Molecular Dynamics (MD) simulations to study atom vibrational amplitudes in carbon-supported Molybdenum Disulphide (MoS2) nanoparticles. Enhanced and correlated atom vibrational amplitudes are observed as the nanoparticle edges are approached from the bulk, consistent with recent experimental High-Resolution Transmission Electron Microscopy (HR-TEM) observations by Chen et al (Nature Communications 12, 5007 (2021). Analysis of phonon modes in finite systems explains the experimental observation by low-energy phonon modes confined at the nanoparticle edge, underscoring the need of full MD modeling for accurate HR-TEM image interpretation. Noticeably, we introduce a workflow for training Equivariant Neural Network-based machine learning potentials using limited Density Functional Theory (DFT) calculations. This approach effectively captures both covalent and van der Waals interactions, enabling accurate extrapolations of DFT calculations to larger systems with built-in error estimation.