Two-tier machine learning acceleration and dimensionality reduction of molecular dynamics for predicting catalytic kinetics

Direct ab initio molecular dynamics of slow catalytic reactions can be prohibitive due to the poor scaling and long simulation time needed to accumulate sufficient statistics. On the other hand, enhanced sampling techniques can accelerate the simulation but require collective variables which can be hard to design for complex reactions. A two-tier machine learning approach is introduced to accelerate MD to address these two problems.

In this two-tier approach, an accurate NequIP deep equivariant neural network force field is first trained to replace the costly ab-initio calculations. Second, a machine-learned reaction coordinate is learned with a multitask encoder framework. In this framework, one upstream neural network is trained to map atomic configurations to a lower-dimensional reaction coordinate latent space, while two additional downstream neural networks are trained to map the latent space to potential energies and metastable state labels. The trained latent space is then used as the reaction coordinate for enhanced sampling to obtain free energy barriers of reactions. The approach is demonstrated for the catalytic process of formate dehydrogenation on Cu(110) surfaces.