Agenda

Résumé : Molecular crystals are a common and important class of crystalline materials. However, modeling molecular crystals based on first principles (eg. with density functional theory) is often difficult due to the size of a typical unit cell. Therefore, high-throughput calculations for the discovery of useful properties are rare. Mechanical and thermal properties are even harder to model correctly as a standard harmonic approximation is often not accurate enough. In this presentation, we will show how machine-learned interatomic potentials can enable accurate and fast calculations of mechanical and thermal properties of molecular crystals enabling a high-throughput search for materials with the desired properties. In principle, to train machine learning potential one would need to create a sufficiently large database of molecular crystals calculated with the desired accuracy. This is also a very challenging task and we will show how to avoid this step using transfer learning and existing databases of small systems.