| Abstract Scope |
Two-dimensional (2D) materials offer exceptional potential for electronics, photonics, energy harvesting, and electrochemical applications, owing to their atomically thin structures and unique quantum-confined properties. However, their intrinsic brittleness poses critical challenges to mechanical reliability under operational stresses. Accurate prediction of their non-equilibrium mechanical response is thus essential for understanding deformation, thermal effects, and failure mechanisms. In this study, we assess two machine learning (ML) interatomic potentials—SNAP and Allegro—for modeling the mechanical behavior and fracture of monolayer MoSe₂, using a density functional theory (DFT)-derived training set. We benchmark their performance by evaluating accuracy, transferability, and computational efficiency. Both ML models exhibit near-DFT accuracy, with Allegro outperforming SNAP due to its more advanced neural network architecture. These results highlight ML-based potentials as robust, scalable tools for simulating complex mechanical processes in 2D materials, offering new pathways for predictive modeling and design of next-generation nanoscale materials and systems |