Predicting Dynamics in Sodium Silicate Glasses using Graph Neural Networks

Understanding the dynamics of atoms in glasses is crucial for unraveling their transport and dynamical properties, but it is challenging to identify the underlying structural features controlling atom dynamics. Recent studies have used machine learning models such as graph neural networks (GNNs) to predict long-term dynamics, but the focus has so far been on model systems like Kob-Andersen-type Lennard-Jones mixtures. This study extends this approach by using GNNs and molecular dynamics simulations to investigate the dynamics across varying timescales in a realistic system that forms the basis for most industrial glasses, namely sodium silicate glasses. By harnessing the capabilities of graph neural networks, our method provides an effective means for predicting the long-term dynamics of ions in glassy systems based solely on initial atom positions, without relying on handcrafted features. We compare our predictions with those of previously proposed methods. Our findings pave the way for designing glass formulations with tailored dynamical properties.