Regensburg 2025 – wissenschaftliches Programm

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

HL: Fachverband Halbleiterphysik

HL 20: Poster I

HL 20.1: Poster

Dienstag, 18. März 2025, 10:00–12:30, P3

Accuracy Requirements for Polarizabilities in MD-based Raman Spectra — •Markus Amaseder¹, Manuel Grumet¹, Tomáš Bučko^2,3, and David A. Egger¹ — ¹TUM School of Natural Sciences, Technical University of Munich — ²Faculty of Natural Sciences, Comenius University Bratislava — ³Institute of Inorganic Chemistry, Slovak Academy of Sciences

Raman spectroscopy provides a versatile and accessible method for characterizing atom dynamics in materials. While frozen phonon approaches have proven well for the prediction of Raman spectra in many cases, they do not inherently include anharmonic and temperature-dependent effects. Raman spectra calculated from molecular dynamics (MD) offer an alternative [1] and have received considerable interest. However, they remain computationally challenging as they require many single-point polarizabilities. We have shown previously that machine learning (ML) can aid in the speed-up using a delta learning approach [2]. However, it is still to be fully understood how accurate single-point polarizabilities need to be in order to provide sufficiently correct spectra. We present an evaluation of polarizabilities from density functional theory and ML for MD Raman spectra, investigating the effects of the functional and further parameters. This is relevant both in terms of training data and ML predictions. Since many single-point calculations are needed, the trade-off between accuracy and computational cost is crucial for the practical application of MD-based Raman spectra. [1] Thomas, et al. Phys. Chem. Chem. Phys. 15, 6608-6622 (2013) [2] Grumet, et al. J. Phys. Chem. C 128, 6464-6470 (2024)

Keywords: Machine Learning; Raman Spectroscopy; Molecular Dynamics; Polarizabilities