hide
Free keywords:
Condensed Matter, Materials Science, cond-mat.mtrl-sci
Abstract:
Discovery of topological insulators remains a challenge because it is usually laborious, high cost, and time consuming. High-throughput computational prescreening is an effective way to reduce the set of candidate systems. Herein, based on compressed sensing technique, we derive an optimized two-dimensional descriptor which can quickly predict potential topological
insulators in the tetradymite family. With only two kinds of fundamental constants of the constituent elements (the atomic number and the electronegativity) as input features, the proposed descriptor effectively classifies topological insulators and normal insulators from a training data containing 230 tetradymite compounds. The predicative accuracy as high as 97% demonstrates that the descriptor really capture the essential nature of topological insulators, and can be further used to fast screen other potential
topological insulators beyond the input dataset.
