Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 93: Scanning Probe Techniques: Method Development
O 93.12: Vortrag
Donnerstag, 21. März 2024, 17:45–18:00, MA 043
Automated prediction of three-dimensional molecular structures from Atomic Force Microscopy images — •Joakim S. Jestilä1, Shuning Cai1, Niko Oinonen1, Peter Liljeroth1, and Adam S. Foster1,2 — 1Department of Applied Physics, Aalto University, 00076 Aalto, Espoo, Finland — 2Nano Life Science Institute (WPI-NanoLSI), Kanazawa University, Kanazawa 920-1192, Japan
Identification of three-dimensional features in surface-adsorbed molecules imaged by Atomic Force Microscopy (AFM) represents a great challenge. While the structures of planar molecules can often be recognised by human users, deviation from planarity contributes to images that are non-intuitive and difficult to interpret, even for experts. Fortunately, neural networks are well-suited for extracting the embedded information in such images. Still, the latter cannot directly determine the placement of atoms that do not contribute to the image contrast, such as atoms eclipsed by those closest to the AFM-tip. In an attempt to access the hidden atoms in the lower layers, we supplement the prediction of the upper atoms with an algorithm that provides candidate structures based on their physical feasibility, evaluated hierarchically in terms of chemical connectivity and the corresponding density functional theory energy. We demonstrate the applicability of the method in a case study of a model system for surface-adsorbed lignocellulosic molecules: 4-nitrophenyl-α/β-D-galacturonide on Au(111).
Keywords: Convolutional Neural Networks; Atomic Force Microscopy; Scanning Tunneling Microscopy; Structure determination; Surface adsorption