Dresden 2017 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 20: Computational Materials Modelling - Defect structure and formation

MM 20.1: Talk

Monday, March 20, 2017, 17:00–17:15, IFW B

Prediction of Morphologies at Inorganic/Organic Interfaces with Machine Learning on the Example of TCNE/Cu(111) — •Michael Scherbela, Lukas Hörmann, Veronika Obersteiner, and Oliver T. Hofmann — Institute of Solid State Physics, NAWI Graz, Graz University of Technology, Austria

Many properties of thin films, such as their solubility or conductivity, depend strongly on the crystal structure of the adsorbed molecules. A major step towards designing new materials is therefore understanding and predicting the polymorphs that form at interfaces. This could in principle be done by finding low energy structures using ab-initio methods. However, the many degrees of freedom lead to a rich polymorphism that prohibits an exhaustive search for the global minimum. We show on the example of Tetracyanoethylene/Cu(111) that this challenge can be tackled with a combination of coarse-graining and machine learning:

First we determine adsorption geometries that isolated molecules adopt on the substrate. We then build supercells by combining these isolated adsorption geometries to generate a set of possible "guess polymorphs". This discretizes the configurational space to a huge, but finite size. Using optimal design methods we select a small, representative subset, calculate its energies using DFT and extract effective interactions between adsorbates. This provides an efficient energy prediction for all remaining guess polymorphs which is exploited by sampling the energetically most promising structures and iterated relearning. For TCNE/Cu(111), we predict stable polymorphs for different coverages and explain an experimentally observed phase transition.