Generalizability of Graph Neural Network Force Fields for Predicting Solid-State Properties

Machine‐learned force fields (MLFFs) promise to offer a computationally efficient alternative to ab initio simulations for complex molecular systems. However, ensuring their generalizability beyond training data is crucial for their wide application in studying solid materials. This work investigates the ability of a graph neural network (GNN)‐based MLFF, trained on Lennard–Jones Argon, to describe solid‐state phenomena not explicitly included during training. The MLFF's performance is assessed in predicting phonon density of states (PDOS) for a perfect face‐centered cubic (FCC) crystal structure at both zero and finite temperatures. Additionally, vacancy migration rates and energy barriers are evaluated in an imperfect crystal using direct molecular dynamics (MD) simulations and the string method. Notably, vacancy configurations are absent from the training data. These results demonstrate the MLFF's capability to capture essential solid‐state properties with good agreement to reference data, even for unseen configurations. Data engineering strategies are further discussed to enhance the generalizability of MLFFs. The proposed set of benchmark tests and workflow for evaluating MLFF performance in describing perfect and imperfect crystals pave the way for reliable application of MLFFs in studying complex solid‐state materials.