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

MM: Fachverband Metall- und Materialphysik

MM 55: Topical session: Data driven materials design - structure maps

MM 55.4: Vortrag

Donnerstag, 23. März 2017, 11:00–11:15, BAR 205

Predicting lattice parameters of ternary compounds by compressed sensing — •Benedikt Hoock^1,2, Santiago Rigamonti¹, Luca Ghiringhelli², Matthias Scheffler^1,2, and Claudia Draxl^1,2 — ¹Humboldt-Universität zu Berlin — ²Fritz-Haber-Institut der Max-Planck-Gesellschaft, Berlin

Data analytics is emerging as a new branch of materials science enabling the interpolation and even (moderate) extrapolation of high-level computational results. In our present work, we demonstrate the successful prediction of lattice parameters for a set of 438 group-IV zincblende ternary compounds from density-functional theory (DFT) results. We use a compressed-sensing based method that combines the least absolute shrinkage and selection operator (LASSO) and ℓ₀ -regularized optimization on a feature space consisting of basic features and a large range of simple mathematical combinations of them. The basic features comprise atomic and dimer DFT data as well as properties of relaxed tetrahedral clusters. We achieve a root mean square error of ∼ 0.04 Å( ∼ 0.4 %) for the fit and a similar prediction accuracy, as estimated by a leave-10%-out cross-validation.