Benchmarking CHGNet Universal Machine Learning Interatomic Potential Against DFT and EXAFS: Case of Layered WS2 and MoS2 (2509.08498v1)

Abstract: Universal machine learning interatomic potentials (uMLIPs) deliver near ab initio accuracy in energy and force calculations at low computational cost, making them invaluable for materials modeling. Although uMLIPs are pre-trained on vast ab initio datasets, rigorous validation remains essential for their ongoing adoption. In this study, we use the CHGNet uMLIP to model thermal disorder in isostructural layered 2Hc-WS2 and 2Hc-MoS2, benchmarking it against ab initio data and extended X-ray absorption fine structure (EXAFS) spectra, which capture thermal variations in bond lengths and angles. Fine-tuning CHGNet with compound-specific ab initio (DFT) data mitigates the systematic softening (i.e., force underestimation) typical of uMLIPs and simultaneously improves alignment between molecular dynamics-derived and experimental EXAFS spectra. While fine-tuning with a single DFT structure is viable, using ~100 structures is recommended to accurately reproduce EXAFS spectra and achieve DFT-level accuracy. Benchmarking the CHGNet uMLIP against both DFT and experimental EXAFS data reinforces confidence in its performance and provides guidance for determining optimal fine-tuning dataset sizes.