Regensburg 2025 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 3: Data-driven Materials Science: Big Data and Worksflows
MM 3.6: Vortrag
Montag, 17. März 2025, 11:45–12:00, H10
Active learning workflow for mixed-halide perovskite stability and electronic band-structure — •Tim Bechtel1,2, Santiago Rigamonti1, and Claudia Draxl1,2 — 1Humboldt-Universität zu Berlin, Germany — 2Max Planck Institute for Solid-State Research, Stuttgart, Germany
Mixed-halide perovskites are most promising materials for stable and efficient light harvesting and emission applications. Their composition can be tailored to match relevant regions of the light spectrum [1]. First-principles calculations can provide insight into stability, ground-state proper- ties, and electronic structure [2, 3]. Comparison with experimental results for “real” materials is, however, challenging. For example, the consideration of chemical (dis)order requires huge super- cells, which is computationally out of reach with state-of-the-art methodology. For the family of CsPb(ClxBryI1−x−y)3 compounds, we bridge this gap with an active learning workflow. It is based on a fine-tuned machine learning interatomic potential [4] that interpolates between already seen compositions, and actively explores new composition ranges. This approach allows for data-efficient predictions of the stability of these materials through finite- temperature phase diagrams and their optical properties for a wide range of compositions.
[1] H. Näsström, PhD Thesis, https://doi.org/10.18452/24939
[2] F. Pan, et al.; https://doi.org/10.1021/acs.chemmater.4c00571
[3] J. Laakso, et al.; https://doi.org/10.1103/PhysRevMaterials.6.113801
[4] I. Batatia, et al.; https://doi.org/10.48550/arXiv.2401.00096
Keywords: Mixed-halide Perovskites; Active learning; Workflow; Machine learning