SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 31: Data Driven Materials Science: Big Data and Work Flows – Machine Learning
MM 31.7: Vortrag
Mittwoch, 29. März 2023, 17:30–17:45, SCH A 251
Machine learning discovery of new materials — •Jonathan Schmidt1,2, Hai-Chen Wang2, Noah Hoffman2, Tiago Cerqueira3, Pedro Borlido3, Pedro Carrico3, Love Pettersson4, Claudio Verdozzi4, Silvana Botti1, and Miguel Marques2 — 1Friedrich-Schiller-University Jena, Germany — 2Martin-Luther-University Halle-Wittenberg, Germany — 3University of Coimbra, Portugal — 4Lund University, Sweden
Graph neural networks for crystal structures typically use the atomic species and atomic positions as input. We construct crystal-graph attention networks replacing these precise bond distances with embeddings of graph distances. This allow us to perform high-throughput studies based on both compositions and crystal structure prototypes. Combining a newly curated dataset of 3M materials and the networks we have already scanned more than two thousand prototypes spanning a space of more than 5 billion materials and identified tens of thousands of theoretically stable compounds. We also demonstrate the effectiveness of transfer learning to adapt the networks to new domains such as of two dimensional structures.
Schmidt et al. Crystal graph attention networks for the prediction of stable materials, Sci. Adv. 7.49 (2021)
Schmidt et al., Large-scale machine-learning-assisted exploration of the whole materials space, arXiv:2210.00579 (2022)
Wang et al., Symmetry-based computational search for novel binary and ternary 2D materials, submitted (2022)