Predicting Atomistic Transitions with Transformers
This work addresses the computational bottleneck of simulating atomistic transitions in materials science, but the results are preliminary and limited to nano-clusters.
The authors demonstrate that transformers can predict atomistic transitions in nano-clusters, offering a fast surrogate model to replace computationally intensive simulations. They also show how to evaluate prediction validity and generate diverse microstates by varying input data.
Accurate knowledge of the atomistic transition pathways in materials and material surfaces is crucial for many material science problems. However, conventional simulation techniques used to find these transitions are extremely computationally intensive. Even with large-scale, accelerated material simulations, the computational cost constrains the applicable domain in practice. Machine learning models, with the potential to learn the complex emergent behaviors governing atomistic transitions as a fast surrogate model, have great promise to predict transitions with a vastly reduced computational cost. Here, we demonstrate how transformers can be trained to predict atomistic transitions in nano-clusters. We show how we evaluate physical validity of the predictions and how a multitude of additional, different microstates can be generated by slightly varying the data provided to the model.