Regensburg 2016 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 15: Poster session I
MM 15.4: Poster
Montag, 7. März 2016, 18:00–20:00, Poster B3
Machine Learning of Structural and Electronic Properties of Semiconductors — Benedikt Hoock1,2, Ute Werner1, Karsten Hannewald1,2, Luca Ghiringhelli2, Matthias Scheffler2, and •Claudia Draxl1,2 — 1Humboldt-Universität zu Berlin, Berlin, DE — 2Fritz-Haber-Institut der MPG, Berlin, DE
High-level solid-state computational methods enable very precise calculations of material properties such as lattice parameters and band structures. However, they usually also require a considerable computational effort. In order to circumvent such time-consuming calculations, recently machine learning techniques have emerged as an alternative predictive tool with potentially high accuracy. For example, Ghiringhelli et al. [*] could predict the crystal structure of binary octet semiconductors with the LASSO regression technique applied on an extended feature space. Using a similar methodology, we demonstrate that the lattice parameter can be learned from purely atomic and dimer data. Further, we explore the viability of learning ab initio band gaps from atomic and dimer data and/or low cost tight-binding calculations.
[*]: L.M. Ghiringhelli et al., Phys. Rev. Lett. 114, 105503 (2015)