Mining for Crystals

Abstract

MATERIALS SCIENCE Predicting the crystal structure of an alloy is challenging, because even small changes in composition can lead to large changes in the way the atoms prefer to coordinate. Fischer et al. have developed a technique that mines the existing crystal database to determine top candidate structures, which are then evaluated using quantum mechanical calculations. The model determines correlations for structural motifs that jointly appear in a single alloy system at different compositions, and thereby assigns probabilities to candidate structures, given those already known in the system. In one test, the authors considered the Ag-Mg alloy with 75% Mg content, for which the exact crystal structure is undetermined. The top candidate highlighted by their model was the Cu2.82P structure, an uncommon motif that nonetheless was computed to have the lowest ground-state energy. They also tested the model by selectively removing specific compositions from the database to see if the remaining data could be successfully used to predict the correct structures; this approach succeeded 90% of the time in placing the true missing structure among the top five candidates. — MSL Nat. Mater. 5 , 641 (2006).