(272i) Autonomous Crystal Structure Characterization with Neighborhood Graph Analysis

We have previously developed a template-free method for characterizing local crystalline structure called Neighborhood Graph Analysis. Unlike conventional methods, our technique uses nonlinear manifold learning to infer structural relationships directly from particle tracking data. While it provides exceptional flexibility and discriminating power, in the past our graph-based classifier was prohibitively slow to evaluate, making it impractical for everyday use. We present a modification to the original method which results in nearly identical results while speeding up the calculation by three orders of magnitude. As with the original technique, we are able to characterize particles near defects, grain boundaries, and interfaces of close-packed crystal structures. However, the increased speed of our new algorithm allows us to extend the analysis to more exotic structures such as ices, clathrates, perovskites, and pyrochlores. We can even handle superlattices composed of two or more species. We demonstrate how the resulting analysis can be used to enhance structural analysis whether the relevant structures are known or not.