Machine‐Learning Spectral Indicators of Topology

Abstract

Topological materials discovery has emerged as an important frontier in condensed matter physics. While theoretical classification frameworks have been used to identify thousands of candidate topological materials, experimental determination of materials’ topology often poses significant technical challenges. X‐ray absorption spectroscopy (XAS) is a widely used materials characterization technique sensitive to atoms’ local symmetry and chemical bonding, which are intimately linked to band topology by the theory of topological quantum chemistry (TQC). Moreover, as a local structural probe, XAS is known to have high quantitative agreement between experiment and calculation, suggesting that insights from computational spectra can effectively inform experiments. In this work, computed X‐ray absorption near‐edge structure (XANES) spectra of more than 10 000 inorganic materials to train a neural network (NN) classifier that predicts topological class directly from XANES signatures, achieving F1 scores of 89% and 93% for topological and trivial classes, respectively is leveraged. Given the simplicity of the XAS setup and its compatibility with multimodal sample environments, the proposed machine‐learning‐augmented XAS topological indicator has the potential to discover broader categories of topological materials, such as non‐cleavable compounds and amorphous materials, and may further inform field‐driven phenomena in situ, such as magnetic field‐driven topological phase transitions.

X-ray Absorption Spectroscopy Studies of a Molecular CO2-Reduction Catalyst Deposited on Graphitic Carbon Nitride

“…The lack of knowledge of the geometry of the catalyst-support sites hampers such studies in which atomically dispersed catalysts of size-selective clusters are analyzed by “inverting” their XANES spectra. This method opens a new opportunity for structural analysis of such photocatalysts using a machine learning approach. , Namely, multiple spectra calculated theoretically may be used for training a neural network classifier, similar to an analogous approach developed recently for predicting new topological insulators . Such a classifier will be used for interpreting the spectra of different hybrid photocatalysts in terms of different classes of local structure and local geometry of the support, for subsequent structural refinement.…”

“…69,70 Namely, multiple spectra calculated theoretically may be used for training a neural network classifier, similar to an analogous approach developed recently for predicting new topological insulators. 71 Such a classifier will be used for interpreting the spectra of different hybrid photocatalysts in terms of different classes of local structure and local geometry of the support, for subsequent structural refinement. In comparison, current efforts are mostly limited to chemometrics-based approaches and require a large number of experimental spectra for such a classification.…”

Topological superconductors from a materials perspective

“…Last but not least, integrating ML with experiments offers an alternative pathway to augment the capability of experimental techniques for throughput and accuracy [189,190]. For instance, it has been shown that the experimental spatial resolution to identify proximity effect -one key approach to realize TSC -can be enhanced by a factor of two through ML-based analysis [191].…”

Toward the Golden Age of Materials Informatics: Perspective and Opportunities

“…An alternative approach is therefore necessary. In particular, feature extraction from spectra data is perhaps the key to machine learning training. − , For instance, peak heights, width, numbers of peaks, and slope can be extracted as descriptors and used during training. Furthermore, it is also demonstrated that semi-supervised machine learning can be implemented to classify the crystal structure .…”