DescriptionOrganic adlayers on inorganic substrates often display unique properties. These properties depend, apart from the chemical composition, on the structure of the adlayer. Yet, due to the enormous number of possible polymorphs, predicting the global minimum structure using conventional stochastic algorithms is not feasible.
To overcome this challenge, we coarse grain the search space by using a building block approach to generate polymorph candidates. The building blocks are the adsorption geometries of isolated molecules on the surface. This allows investigating commensurate, dense-packed structures, with good agreement with experiment. To determine the adsorption energies of single molecules as well as polymorphs, we use density functional theory augmented with Bayesian machine leaning.
We extend this approach to incommensurate structures by generalizing our machine learning model. This allows considering higher-order commensurate structures containing up to 100 molecules per unit cell and enables predicting properties such as work functions and coherent fractions (obtained via X-ray standing wave measurements) for these polymorphs. As an outlook we use the generalized model to predict the shear forces that occur when laterally moving commensurate and incommensurate adlayers across a substrate.
|Period||7 Jun 2022|
|Held at||University of Warwick, United Kingdom|