Pairing glue in cuprate superconductors from the self-energy revealed via machine learning

Abstract

Recently, machine learning was applied to extract both the normal and the anomalous components of the self-energy from photoemission data at the antinodal points in Bi-based cuprate high-temperature superconductors [Y. Yamaji {\it et al.}, arXiv:1903.08060]. It was argued that both components do show prominent peaks near 50 meV, which hold information about the pairing glue, but the peaks are hidden in the actual data, which measure only the total self-energy. We analyze the self-energy within an effective fermion-boson theory. We show that soft thermal fluctuations give rise to peaks in both components of the self-energy at a frequency comparable to superconducting gap, while they cancel in the total self-energy; all irrespective of the nature of the pairing boson. However, in the quantum limit $T \to 0$ prominent peaks survive only for a very restricted subclass of pairing interactions. We argue that the way to potentially nail down the pairing boson is to determine the thermal evolution of the peaks.