The GoogLeNet-assisted phase transition detectors

TL;DR

This paper introduces a GoogLeNet-based AI model to accurately distinguish between superconducting transition temperature and BKT transition in nanowire systems with overlapping heat capacity signals, advancing phase transition detection.

Contribution

The novel application of a GoogLeNet-assisted AI model to differentiate closely overlapping phase transition anomalies in superconducting nanowires.

Findings

The model accurately separates Tc and TBKT in complex systems.

It demonstrates high interpretability of phase transition features.

The approach enhances detection precision in superconductivity research.

Abstract

In the presence of the phase fluctuations in superconducting nanowires array, the electrical resistance of the superconducting nanowires is always non-zero unless the system undergoes Berezinskii-Kosterlitz-Thouless (BKT) transition where the superconducting vortices and anti-vortices form pairs. The two-dimensional XY model can mimic the superconducting transition temperature Tc and the BKT transition at a lower critical temperature TBKT by observing the heat capacity anomalies upon cooling. If the Josephson coupling across the nanowires is strong, the heat capacity anomalies almost overlap with each other so that it is difficult to distinguish between the Tc and the TBKT. To solve this issue, we apply an artificial-intelligence technique to split the nearly overlapped heat capacity anomalies. After the GoogLeNet-assisted phase transition detector is built, the GoogLeNet model can learn from the features of the phase transitions and then interpret the Tc and TBKT in the unseen system precisely. Our work opens a path for the GoogLeNet model to enter the world of magnetism and superconductivity.