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

MM: Fachverband Metall- und Materialphysik

MM 31: Data-driven Materials Science: Big Data and Worksflows

MM 31.3: Vortrag

Donnerstag, 20. März 2025, 15:30–15:45, H10

Advanced Machine Learning of ¹⁷O NMR in Non-Magnetic Oxides: High-Throughput Calculation, Prototype Compound Analysis, and Transfer Learning — •Zhiyuan Li, Bo Zhao, Hongbin Zhang, and Yixuan Zhang — Institute of Materials Science, TU Darmstadt, 64287 Darmstadt Germany

The study of ¹⁷O NMR spectroscopy is crucial for understanding the local structure of oxides, where the naturally occurring NMR-active oxygen isotope, ¹⁷O, provides unique insights into local environments due to its large chemical shift range and quadrupolar nature. In this work, we present a high-throughput workflow integrating AiiDa and CASTEP to calculate the NMR parameters of over 7100 compounds from the Materials Pro ject database, followed by utilizing machine learning models to predict ¹⁷O NMR parameters. Furthermore, taking BaTiO₃ as an example, we identify prototypical ABO₃ crystal structures, construct BaTiO₃ analogs via substitution, perform ab initio molecular dynamics simulations to generate 3000 perturbated structures, and evaluate the NMR parameters. The results of our machine learning modeling with such additional dataset reveal that incorporating perturbated structures enhances the accuracy of the machine learning model. Moreover, by leveraging transfer learning, using previously trained model from our high-throughput dataset, the predictivity for the newly generated BaTiO₃ analogs can be further improved.

Keywords: NMR parameters; High-Throughput Calculation; Transfer Learning