Achieving Smaller Effective Spot Sizes in nano-ARPES with Machine Learning

Nano-ARPES is emerging as a critical and interpretable technique for the study of van der Waals heterostructures and devices, but extending the applicability of the nano-ARPES to traditional quantum materials and stretching the reach of ARPES experiments with larger beams remains a practical challenge for the time being. We demonstrate that convex optimization and machine learning enhanced nano-ARPES allows for resolving the individual contributions of sub-beam domains to the ARPES spectra of discrete domain patterned materials. By using this method instead of naïve averaging—the typical approach for photoemission experiments—arbitrarily larger beams can be used. We explore the conditions for applicability of our technique, improvements that can be made by leveraging the physical invariants of the photoemission measurement, and the applicability of the technique to the study of phase transitions in quantum materials.