Berlin 2024 – wissenschaftliches Programm

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

MM: Fachverband Metall- und Materialphysik

MM 43: Data Driven Material Science: Big Data and Workflows V

MM 43.2: Vortrag

Mittwoch, 20. März 2024, 16:00–16:15, C 243

A robust, simple and efficient algorithm to converge GW calculations — •Max Großmann, Malte Grunert, and Erich Runge — Theoretische Physik I, TU Ilmenau, Germany

In the era of high-throughput computation for materials science and machine learning, the demand for heuristics and efficient convergence schemes in ab initio calculations has become essential to speed up computational workflows. A prime example of computationally demanding ab initio material-science calculations is the GW method, which is crucial for accurate band gap predictions and the subsequent computation of dielectric functions using TDDFT/BSE. The main hurdle within GW calculations lies in the convergence of two interdependent parameters in the dynamic screening W which must be optimized simultaneously. To overcome this obstacle we introduce a straightforward, "cheap first, expensive later" coordinate-search algorithm, which is tested against a commonly used state-of-the-art method on a diverse dataset of 50 semiconducting and insulating solids. Additionally, the practical independence between the k-point grid and parameters in W is checked using both algorithms on five different k-meshes of increasing density for all investigated materials, starting at Γ-only calculations. The results show how to converge GW calculation in a robust, simple and efficient way.

Keywords: High-Throughput; Workflow; Many-Body Perturbation Theory; GW; Convergence