Machine-learned interatomic potentials (MLPs) have become an indispensable and central part of multiscale modeling by bridging the gap between ab-initio and phase-field approaches. While inheriting the accuracy of DFT from calculations for comprehensive sets of training structures these potentials offer unprecedented capabilities to investigate large and complex atomic structures at long time scales. Thereby they open the door to materials properties and phenomena, which reach beyond the limitations of DFT methods with respect to system sizes and time scales, and at the same time provide a basis for continuum approaches to materials.

Here we demonstrate the full integration of MLPs in the MedeA software environment combining efficient ways for full-scale training-set calculations with the MLP Generator to provide potentials for direct use within MedeA. This opens a plethora of capabilities for materials property calculations extending beyond the calculation of energies and forces. The presentation will showcase new results obtained from the latest GRACE potentials for highly accurate electronic properties.