A machine learning route between band mapping and band structure

Electronic band structure (BS) and crystal are the two complementary identifiers of solid state materials. While convenient instruments reconstruction algorithms have made large, empirical, databases possible, extracting quasiparticle dispersion (closely related to BS) from photoemission mapping data is currently limited by available computational methods. To cope with growing size scale data, we develop a pipeline including probabilistic machine learning associated processing, optimization evaluation methods for reconstruction, leveraging theoretical calculations. The reconstructs all 14 valence bands semiconductor shows excellent performance on benchmarks other materials datasets. uncovers previously inaccessible momentum-space structural information both global local scales, while realizing path towards integration science databases. Our approach illustrates potential combining domain knowledge scalable feature extraction in multidimensional data.