Electron Structure Theory Trained Machine Learning Force Field for Molecular Dynamics of Material Growth and Catalytic Reactions

We will use electron structure theory (EST) calculations as a foundation to train a machine learning force fields (MLFFs) used for long timescale molecular dynamics (MD) studies of materials growth and catalytic synthesis on metal surfaces to the microsecond scale.

MD simulations with EST can only be performed for (sub)nanosecond scales due to the computationally demanding quantum mechanical methods, but if trained properly an MLFF can replace EST and the time and computational resource requirements are radically reduced while maintaining accuracy.

This will allow us to study materials growth, and catalytic reactions in solution, in order to help steer these processes with regard to choice of metal, temperature, pH, etc. with the use of MLFF in MD simulations for the relevant timescales and sizes, which is currently not possible. Active learning with minimization of model deviation will be employed through use of an ensemble of MLFFs.

In the four years of the project, we at Applied Physics at LTU will extend our current MLFF from two elements to a dozen and establish modelling protocols for its further extension.

To be able to model material growth and molecular synthesis with dynamic simulations will open a new dimension for the development of future nanomaterials and making catalytic synthesis cheaper.