Regensburg 2022 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 33: Computational Materials Modelling: Process Schemes / Oxides

MM 33.8: Vortrag

Donnerstag, 8. September 2022, 17:45–18:00, H44

Uncertainty in Predicting Thermodynamic Properties of TiO2 Polymorphs — •Olga Vinogradova, Pin-Wen Guan, Siying Li, and Venkatasubramanian Viswanathan — Carnegie Mellon University, Pittsburgh, USA

Polymorphism of crystals directly leads to materials with vastly different chemical and physical properties. However the lowest energy polymorphs often differ by only small amounts of energy. This makes it challenging to predict relative properties using first-principles density functional theory (DFT), which is significant in designing a material for the desired application. In this work we apply computational uncertainty within DFT to quantify the accuracy of stability and phase transition predictions under finite temperature and pressure. We study six polymorphs of TiO2 using a set of six exchange-correlation functionals to present a detailed sensitivity analysis using uncertainty capabilities within the Bayesian Error Estimation Functional. We show that a prediction confidence metric is particularly important for comparing the stability of numerically close predictions. We show how the choice of functional significantly affects predictions of phase transitions and identify which structures and properties that have inherently large uncertainties. From the trends observed in stability, finite-temperature, and phase transition pressure predictions we propose that uncertainty quantification provides a valuable insight in problems where drawn conclusions are highly sensitive to the choice of the functional.