A handle on the scandal: Data driven approaches to structure prediction

Abstract

Structure-property relationships play a central role in condensed matter physics, chemistry, and materials science. However, the problem of predicting the structure of a material, given its chemical composition, remains immensely challenging. Here, we review some of the progress that has been made in this area for both crystalline materials and atomic clusters. Early work consisted of heuristic rules-of-thumb or structure maps using descriptors that were obtained largely by inspection. Increasingly, these approaches are being expanded to use descriptors that have been obtained by applying machine learning techniques to big data containing information from the experiment and/or first principles calculations. Improved techniques for global optimization in the multi-dimensional coordinate space have also led to major advances in the field.