Regensburg 2025 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 17: Development of Calculation Methods

MM 17.7: Vortrag

Mittwoch, 19. März 2025, 12:00–12:15, H22

Performance and limits of finite-temperature DFT for SiO2 — •Axel Forslund, Jong Hyun Jung, Blazej Grabowski, and Yuji Ikeda — Institute for Materials Science, University of Stuttgart, Germany

Silicon dioxide (SiO2) is a widely studied compound, yet far from fully understood. It exists in a variety of different phases, several of which are dynamically stabilized. These structures require dynamic vibrations of the atoms to not transform, and pose a challenge from an atomistic modeling point of view. Even in recent publications where state-of-the-art machine-learning interatomic potentials (MLIPs) have been used, the predictions differ significantly from experiments. For example, the transition between the two dynamically stabilized phases beta-quartz and cristobalite is very sensitive, and a single meV/atom shift can change the transition by 100 K. Not only the accuracy of the MLIP and the method for free-energy calculations matters, but also the underlying ab initio data play a crucial role. We demonstrate this sensitivity, and provide a simplified, yet precise method of estimating the quartz-cristobalite transition temperature. This approach is accurate enough to closely estimate the transition temperature using new density-functional-theory (DFT) functionals, and we demonstrate this for several functionals and on-top corrections. Our method is also efficient enough for using the random-phase approximation (RPA), which provides a transition temperature in very good agreement with the average CALPHAD value, and thus serves as a benchmark for the development of improved DFT functionals.

Keywords: Random phase approximation; Thermodynamic integration; Finite temperature; Machine learning interatomic potentials; Silicon dioxide