Physics-informed machine learning applied to the identification of high-pressure elusive phases from spatially resolved X-ray diffraction large datasets
Multi-technique high resolution X-ray mapping enhanced by the recent advent of 4th generation synchrotron facilities can produce colossal datasets, challenging traditional analysis methods. Such difficulty is clearly materialized when probing crystal structure of inhomogeneous samples, where the number of diffraction patterns quickly increases with map resolution, making the identification of crystal phases within a vast collection of reflections unfeasibly challenging by direct human inspection. Here we develop a novel analysis approach based on unsupervised clustering algorithms for identifying independent phases within a diffraction spatial map, which allowed us to identify the material distribution across a high-pressure cerium hydride. By investigating the specific compound, we also contribute to the understanding of synthesis inhomogeneities among the superhydrides, a prominent superconductor class in condensed matter physics whose characterization is highly challenging even for state-of-the-art materials techniques. The analysis framework we present may be readily extended to any correlated set of curves whose features are tied to specific entities, such as structural phases.