Density functional theory provides a good description of materials on a quantum mechanical level. However, the high computational cost associated with it prevents studies of models at larger length and time scales. The latter are a prerequisite to explore important properties of real materials such as, for example, the impact of extended defects, diffusion, and thermal conductivity. Due to their unique combination of efficiency and high accuracy machine-learned potentials (MLPs) offer a route to access such properties.

This session introduces MedeA's Machine-Learned Potential Generator (MLPG), which allows to generate MLPs in a highly efficient manner for a given set of training structures. You will learn how to use MLPG to generate kernel-regression based spectral neighbour analysis potentials (SNAP) as well as neural network potentials (NNP) with Behler-Parrinello atom-centered symmetry functions. Once generated these MLP's can be used like any other forcefield within the MedeA environment leveraging the capability of various MedeA modules to study the physical property of interest.