Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 20: Data Driven Material Science: Big Data and Workflows III

MM 20.6: Vortrag

Dienstag, 19. März 2024, 11:45–12:00, C 243

Experiment-driven atomistic materials modeling: Combining XPS and MLPs to infer the structure of a-CO_x — •Tigany Zarrouk and Miguel Caro — Aalto University, Espoo, Finland

One facet of materials modelling is to gain insights from experimental results, which necessitates an effective strategy for identifying atomic structures that align with experimental data, e.g. spectra. Conventional approaches for amorphous materials involve generating numerous configurations through Molecular Dynamics and selecting one with the closest predicted spectrum to experiment. However, this process is inefficient and lacks assurance of spectrum conformity. We introduce a Grand-Canonical Monte Carlo methodology to generate configurations that concur with both experimental data and ab-initio calculations. Utilising a SOAP-based [1] X-Ray Photoelectron Spectroscopy (XPS) model trained on GW and Density Functional Theory (DFT) data, in conjunction with CO Gaussian Approximation Potential (GAP), we identify oxygenated amorphous carbon structures compliant with experimental XPS predictions that are also energetically favourable within DFT. Clustering and embedding SOAP descriptors provides a data-driven deconvolution of the XPS spectrum into motif contributions, revealing the significant inaccuracies present in experimental XPS interpretation. This method generalises to multiple sets of experimental data and allows for the elucidation of specific experimental results, enhancing the applicability of materials modelling.

[1] Albert P. Bartók et al.: On representing chemical environments, Phys. Rev. B 87, 184115

Keywords: Machine-Learned Potentials; Amorphous Materials; Carbon; X-Ray Photoelectron Spectroscopy; Monte-Carlo