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

O: Fachverband Oberflächenphysik

O 34: Catalysis and surface reactions I

O 34.2: Vortrag

Dienstag, 10. März 2026, 10:45–11:00, HSZ/0204

Machine-learning Driven Approach of Surface Adsorption Energy Prediction — •Karlo Sovic and Johannes Margraf — University Bayreuth, Bayreuth, Germany

Heterogeneous catalysis is a cornerstone technology across industrial chemistry, materials science, and environmental engineering, effectively promoting and redirecting key chemical reactions in various applications. In these processes, control of surface adsorption energies is paramount for designing efficient catalysts. However, in the case of large adsorbates, the high computational cost of accurate DFT calculations limits comprehensive exploration of the corresponding potential energy surface. Machine-learning interatomic potentials offer a promising solution to this challenge. Leveraging recently reported pre-trained models, the complex interactions crucial for accurate surface chemistry can be captured with effectively via fine-tuning.

Here, a data-efficient workflow for describing the glycerol hydrodeoxygenation mechanisms leading to propanediols on M(111) and M(211) surfaces (M=Pt, Cu, Ni) is reported. With our approach, the fine-tuned model yields high-fidelity adsorption energies with near-DFT accuracy at low computational cost. Furthermore, analysis of thermodynamic properties was carried out to accurately determine the overall Gibbs free energetics of investigated reaction pathways. Based on these comprehensive insights, we propose a versatile workflow for the accelerated screening of catalytic systems, enabling the rapid construction of adsorption energy databases and exploring relevant reaction mechanisms.

Keywords: catalysis; adsorption; machine-learning; MACE