SurfaceScience21 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 42: Poster Session III: Poster to Mini-Symposium: Machine learning applications in surface science I
O 42.7: Poster
Dienstag, 2. März 2021, 10:30–12:30, P
Ab initio structure search of flexible molecules at interfaces — •Dmitrii Maksimov1,2 and Mariana Rossi1,2 — 1Fritz Haber Institute of the Max Planck Society, Berlin, Germany — 2Max Planck Institute for the Structure and Dynamics of Matter, Hamburg, Germany
We investigate how the accessible conformational space of two flexible amino acids, Arg and Arg-H+, changes upon adsorption, by building and analyzing a database of thousands of structures optimized at Cu(111), Ag(111) and Au(111) surfaces with the PBE functional including screened pairwise (vdW) interactions. We employ an unsupervised dimensionality reduction procedure that enables us to understand the alteration of the high-dimensional conformational space [1]. The creation of this database, which is paramount to train further diverse machine-learning models, suffers from well-known bottleneck related to the efficiency of the geometry optimizer. We introduce a flexible way of preconditioning approximate Hessian matrices in the BFGS algorithm that is tailored to accelerate the relaxation of vdW bonded structures that can handle large structural changes. An automated sampling of these systems is implemented within a random structure search package [2] that can take explicitly into account the flexibility of molecules, their position and orientation with respect to fixed surroundings and interfaces.
[1] Maksimov et. al., Int. J. Quantum Chem., e26369 (2020)
[2] https://github.com/sabia-group/gensec