Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Machine learning protocol to identify pairing symmetries via quasiparticle interference imaging in Ising superconductors

Adam Hložný, Jozef Haniš, Martin Gmitra, Marko Milivojević

TL;DR

A machine-learning-guided strategy to determine pairing symmetry from quasiparticle interference data, which integrates first-principles calculations, tight-binding modeling, and symmetry-based classification of the superconducting pairing function, demonstrates that machine-learning-assisted QPI analysis provides a promising pathway for precise learning of superconducting pairing functions in quantum materials.

Abstract

Identifying the pairing symmetry in unconventional superconductors is essential for reliably characterizing their superconducting states and for enabling their integration into realistic quantum devices. Here, we introduce a machine-learning-guided strategy to determine pairing symmetry from quasiparticle interference (QPI) data, which integrates first-principles calculations, tight-binding modeling, and symmetry-based classification of the superconducting pairing function. We demonstrate the approach on monolayer NbSe2 as an experimentally accessible probe of superconductivity in real materials, within a single scalar-impurity Bogoliubov-de Gennes framework. Our analysis shows that the QPI-to-parameter inverse problem can be solved with high accuracy for most superconducting pairing channels in this setting, indicating that QPI carries rich, learnable information about the superconducting gap structure. Taken together, these results demonstrate that machine-learning-assisted QPI analysis provides a promising pathway for precise learning of superconducting pairing functions in quantum materials.

Machine learning protocol to identify pairing symmetries via quasiparticle interference imaging in Ising superconductors

TL;DR

A machine-learning-guided strategy to determine pairing symmetry from quasiparticle interference data, which integrates first-principles calculations, tight-binding modeling, and symmetry-based classification of the superconducting pairing function, demonstrates that machine-learning-assisted QPI analysis provides a promising pathway for precise learning of superconducting pairing functions in quantum materials.

Abstract

Identifying the pairing symmetry in unconventional superconductors is essential for reliably characterizing their superconducting states and for enabling their integration into realistic quantum devices. Here, we introduce a machine-learning-guided strategy to determine pairing symmetry from quasiparticle interference (QPI) data, which integrates first-principles calculations, tight-binding modeling, and symmetry-based classification of the superconducting pairing function. We demonstrate the approach on monolayer NbSe2 as an experimentally accessible probe of superconductivity in real materials, within a single scalar-impurity Bogoliubov-de Gennes framework. Our analysis shows that the QPI-to-parameter inverse problem can be solved with high accuracy for most superconducting pairing channels in this setting, indicating that QPI carries rich, learnable information about the superconducting gap structure. Taken together, these results demonstrate that machine-learning-assisted QPI analysis provides a promising pathway for precise learning of superconducting pairing functions in quantum materials.
Paper Structure (9 sections, 19 equations, 2 figures, 1 table)

This paper contains 9 sections, 19 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Theoretical framework: superconductivity, QPI, and machine learning
  3. Bogoliubov-de Gennes description of superconductivity in monolayer NbSe$_2$
  4. Quasiparticle interference details
  5. Machine-learning details
  6. Results
  7. Conclusions
  8. Superconducting gap function
  9. Architecture of the neural network

Figures (2)

  • Figure 1: Schematic overview of the proposed workflow. Ab-initio calculations and tight-binding approach are used to model the normal phase of NbSe$_2$, whereas the superconducting pairing functions of NbSe$_2$ are determined using the group-theory approach. Secondly, a large dataset of simulated QPI spectra is generated across different superconducting parameters. This dataset is then used to train and optimize a neural network model.
  • Figure 2: Architecture of the neural network. The model is based on VGG16 SZ15DZM+21 architecture, input channels reduced from 3 to 1, prediction head replaced with two custom prediction heads, one for classification (IRs), the other one for regression (parameters).