Demystifying Quantum Materials with Deep Learning and Angle-Resolved Photoemission Spectroscopy

Modern materials often exhibit behavior that cannot be explained by current condensed matter physics. These elusive properties, like cuprate superconductivity, emerge from quantum phenomena and are yet to be fully understood. Angle-resolved photoemission spectroscopy (ARPES) can probe these properties by determining the allowable energy-momentum states. Specifically, inhomogeneities within these states often accompany emergent behavior. However, spatially-resolved ARPES generates noisy high-dimensional data difficult for human analysis. In this paper, we develop a deep autoencoder that extracts and characterizes electron spectral features, before comparing our methodology against currently used techniques. We then dissect data from the high-superconductor Bi
2
Sr
2
CaCu
2
O
8
(BSCCO) and correlated insulator TaSe
2
, utilizing our approach to automatically extract Fermi levels, band structures, surface inhomogeneities, and potentially electron mass renormalization kinks hidden under noise.