Regensburg 2025 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
O: Fachverband Oberflächenphysik
O 85: New Methods: Theory
O 85.7: Vortrag
Donnerstag, 20. März 2025, 12:00–12:15, H25
Uncertainty quantification for DFT calculations — •Teitur Hansen1, Thomas Bligaard2, and Karsten W. Jacobsen1 — 1DTU Physics, Kongens Lyngby, Denmark — 2DTU Energy, Kongens Lyngby, Denmark
Density functional theory (DFT) is widely used for understanding material properties and chemical reactions. DFT simulations use approximations which result in errors when comparing to experiments. There are errors in different types of energies, lattice constants and other properties. Quantitative uncertainty estimates would be helpful for DFT simulations to highlight which results are trustworthy. Quantifying uncertainties not only improves trust in calculations but also identifies high-uncertainty predictions that can subsequently be revisited and reanalyzed.
We propose a method to model uncertainties using a probability distribution over exchange-correlation functionals. The probability distribution is trained on experimental values of molecular atomization energies, cohesive energies of solids, reactions energies, heat of formations, and lattice constants. The optimized probability distribution is then used to define an ensemble method which can be used to estimate the bias and variance on simulated materials properties. I will demonstrate that the method gives systematic improvements in error estimates over the Bayesian error estimation class of functionals (BEEF).
The proposed method is general enough to be applied to any simulation methodology where accurate reference benchmarks exist.
Keywords: Uncertainty; functional; exchange-correlation; error; DFT