Enabling Structure-Only Initialization and Out-of-Distribution Generalization in GNN-based Molecular Dynamics Simulators

ArXi:2605.09495v1 Announce Type: cross Machine learning-based simulators offer the potential to model the dynamics of complex systems efficiently than classical approaches, while retaining differentiability, a key property for materials design. Graph neural network (GNN)-based simulators have shown strong performance across a range of physical domains, including molecular dynamics. However, their reliance on temporal context for accurate prediction limits their use in inverse design settings, where simulations must be initialized from a single static configuration.