Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 85: Heterogeneous Catalysis I
O 85.3: Vortrag
Donnerstag, 21. März 2024, 11:00–11:15, TC 006
Infra-red spectra prediction of functionalized copper nanoparticles using machine learning force fields — •Nitik Bhatia, Ondrej Krejci, and Patrick Rinke — Department of Applied Physics, Aalto University, FI-00076 AALTO, Finland
Vibrational spectroscopy serves as a powerful tool for in vivo chemical reaction analysis, allowing the identification of distinctive peaks as markers for specific chemical groups. Despite yielding valuable insight, the qualitative and quantitative analysis of IR spectra is aggravated by the presence of other, neighboring species. In this work we investigate the vibrational spectra of several small molecules adsorbed on copper nanoparticles, commonly used in heterogeneous catalysis. We compute the infrared spectra using ab initio molecular dynamics (AIMD) [1]. We achieve excellent agreement with experimental results, but at a high computational cost. We therefore train neural network force fields like FieldSchNet [2] on our AIMD data to capture pertinent spectral properties more efficiently (∼2500 times faster). FieldSchNet achieves remarkable accuracy for simple molecules (<20 cm−1 shift in peak positions). However, functionalized copper nanoparticles introduce a more complex chemical environment and the predicted peak positions are only accurate to ∼100 cm−1. To improve the predictive accuracy, we are enhancing our dataset quality and exploring alternative machine learning strategies for fine-tuning the models.
[1] M.-P. Gaigeot, M. Sprik, J. Phys. Chem. B 107, 10344-10358 (2003). [2] M. Gastegger, K. T. Schütt, K.-R. Müller, Chem. Sci. 12, 11473-11483 (2021).
Keywords: Infra-red spectroscopy; Machine learning force fields; AIMD; Functionalized copper nanoparticles; Heterogeneous catalysis