Neural Network Potentials for Finite-Temperature Studies of Transition Metal Dichalcogenide Moiré Superlattices

In the past decade, 2D transition metal dichalcogenides (TMDs) have become an ideal playground to explore novel low-energy phenomena in condensed matter physics. While density functional theory can accurately predict the structural properties of such systems, it scales poorly with system size, making it inefficient for modeling large-scale TMD moiré superlattices. In this study, we develop a neural network potential (NNP) for multilayered TMDs based on the equivariant message passing MACE model [1], accounting for long-range van der Waals interactions to accurately describe weak interlayer interactions. We will demonstrate the accuracy of our NNP in computing the thermodynamic properties of multilayered TMDs at finite temperatures, and compare the results with the MACE universal potential and a recently developed dispersion-aware NNP based on invariant representations [2]. We discuss the extent to which this new ML-based interatomic potential will increase the accuracy and efficiency of predicted thermal properties of layered TMDs relevant to thermoelectric and optoelectronic applications.

[1] I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Csanyi, MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields, Advances in Neural Information Processing Systems 35, 11423 (2022).

[2] Shaidu Y., Naik M., Louie S. and Neaton B., Accurate Dispersion-Aware Neural Network Potentials for Twisted Bilayer Transition Metal Dichalcogenides (in preparation)