Researchers Develop Techniques Using Machine Learning to Predict the Electronic Band Structure of Materials from Band Mapping Data

The band structure of a material describes the energy levels that electrons in the material can occupy and is important for understanding its physical and chemical properties. Photoemission spectroscopy can be used to measure the band structure of a material, but interpreting the resulting data can be challenging, especially for materials with complex band structures. In this article, the authors propose a computational framework for predicting the electronic band structure of materials from photoemission spectroscopy data using machine learning techniques.

The authors propose a method that combines the advantages of two existing approaches for interpreting photoemission spectra. The first approach is physics-based and involves fitting one-dimensional lineshapes (energy or momentum distribution curves) to the data using least-squares methods. This approach is accurate and interpretable but can be computationally inefficient when applied to large datasets. The second approach is based on image processing and involves data transformations to improve the visibility of dispersive features in the data. This approach is more efficient but does not allow for the reconstruction of the band structure and is not suitable for quantitative analysis.

The authors’ proposed method uses a probabilistic machine learning model to fit a model to the data, with the energy values of the electronic band structure as the variables to be extracted. The model uses a nearest-neighbour Gaussian distribution, describing the proximity of energy values at nearby momenta. The maximum a posteriori estimation in probabilistic inference is used to find the optimal fit to the data. This formulation allows for the incorporation of imperfect physical knowledge, such as impurities or defects in the material, and can also handle noise in the data.

The authors demonstrate their method’s effectiveness on various materials, including graphene, the transition metal dichalcogenides MoS2 and WS2, and the topological insulator Bi2Se3. They show that their way can accurately reconstruct the band structures of these materials and is scalable to multidimensional datasets. The authors also demonstrate that their method can reproduce the band structures obtained from other ways, including density functional theory calculations and experimental data from other sources.

Overall, the authors’ proposed method provides a promising approach for accurately predicting the electronic band structure of materials from photoemission spectroscopy data and could be useful for understanding and interpreting complex photoemission data.

Check out the Paper. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our Reddit page and discord channel, where we share the latest AI research news, cool AI projects, and more.