Enhanced sampling of Crystal Nucleation with Graph Representation Learnt Variables
This provides a method for improved sampling in computational chemistry, though it appears incremental as it applies existing techniques to new systems.
The study tackled the problem of sampling crystal nucleation by using graph neural network-derived variables to bias enhanced sampling, achieving accurate free energy calculations that agree with experiments for iron and glycine systems.
In this study, we present a graph neural network-based learning approach using an autoencoder setup to derive low-dimensional variables from features observed in experimental crystal structures. These variables are then biased in enhanced sampling to observe state-to-state transitions and reliable thermodynamic weights. Our approach uses simple convolution and pooling methods. To verify the effectiveness of our protocol, we examined the nucleation of various allotropes and polymorphs of iron and glycine from their molten states. Our graph latent variables when biased in well-tempered metadynamics consistently show transitions between states and achieve accurate free energy calculations in agreement with experiments, both of which are indicators of dependable sampling. This underscores the strength and promise of our graph neural net variables for improved sampling. The protocol shown here should be applicable for other systems and with other sampling methods.