Regensburg 2019 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
HL: Fachverband Halbleiterphysik
HL 40: Perovskite and Hybrid Photovolatics I (joint session HL/CPP)
HL 40.13: Vortrag
Donnerstag, 4. April 2019, 12:45–13:00, H36
Exploring the stability of halide perovskite alloys by combining density-functional theory and machine learning — •Guo-Xu Zhang1,2, Lauri Himanen1, Jingrui Li1, and Patrick Rinke1 — 1Department of Applied Physics, Aalto University, Finland — 2School of Chemistry and Chemical Engineering, Harbin Institute of Technology, China
Halide perovskites (ABX3) have attracted considerable attention in recent years due to their breakthrough performance as photovoltaic materials in perovskite solar cells (PSCs). We here consider the materials space of perovskites spanned by A = Cs and Rb, B = Sn and Pb, and X = Cl, Br, and I. Since this space is too large to peruse with density-functional theory (DFT) alone, we combine DFT with machine learning. We use the recently proposed many-body tensor representation (MBTR) [1] as structural descriptor. We then train a kernel ridge regression (KRR) model for fast energy prediction with DFT energies for 2×2×2 and 4×4×4 perovskite supercell models of varying composition. We analyse the effect of MBTR parameters on the KRR learning quality and then use KRR to explore the vast alloy space. We compute the convex-hull of several binary alloy series, for example CsxRb1−xPbI3, CsPbxSn1−xI3, and CsxRb1−xPbySn1−yClz1Brz2I3−z1−z2. Our prediction accuracy for the cohesive energy of different alloys is as low as few meV/atom. This suggests that KRR in combination with the MBTR can be used to speed up the discovery of stable halide perovskite alloys.
0.cm [1] Huo and Rupp, arXiv 1704.06439.