Surface Structure Search using Coarse Grained Modeling and Bayesian Linear Regression

Hörmann, L. (Speaker)

Activity: Talk or presentationTalk at conference or symposiumScience to science

Description

The key information about a monolayer of molecules on a substrate, aside from chemical composition, is arguably the polymorph it forms. First-principles prediction of such polymorphs is a major challenge, due to the large number of possible arrangements of molecules. To meet this challenge, we develop SAMPLE[1,2], which uses physically motivated coarse graining and statistical learning to explore the potential energy surface of commensurate organic monolayers on inorganic substrates.

We first determine adsorption geometries of isolated molecules on the substrate. By generating commensurate arrangements of these geometries, we compile a large number of possible polymorphs. Using experimental design theory, we select subsets of these polymorphs and calculate their adsorption energies using dispersion-corrected density functional theory. These subsets serve as training data for an energy model, based on molecular interactions. Using Bayesian linear regression, we determine the model parameters, yielding meaningful physical insight and allowing the prediction of adsorption energies for millions of possible polymorphs with high accuracy.

We demonstrate this on three complimentary systems: naphthalene on Cu(111), TCNE on Cu(111), and benzoquinone on Ag(111).

[1] Hörmann et al., arXiv:1811.11702 (2018)
Period4 Apr 2019
Event titleDPG-Frühjahrstagung: null
Event typeConference
LocationRegensburg, Germany, Bavaria
Degree of RecognitionInternational

Keywords

  • Surface Structure Search
  • Machine Learning