Cuprate superconducting materials above liquid nitrogen temperature from machine learning.

Area of Science:

• Condensed matter physics
• Materials science
• Artificial intelligence in scientific discovery

Background:

• Superconductivity in cuprates is a complex challenge with significant implications for future technologies.
• High-temperature superconductors are sought for applications exceeding liquid nitrogen temperatures.

Purpose of the Study:

• To investigate machine learning (ML) models for predicting superconducting properties in materials.
• To identify key physical factors governing the superconducting critical temperature (Tc).

Main Methods:

• Employed deep neural networks (DNNs) with two distinct atomic feature sets: element symbolic (AFS-1) and physics-informed (AFS-2).
• Utilized SHapley Additive exPlanations (SHAP) to determine feature importance.
• Developed a virtual high-throughput search workflow.

Main Results:

• Cuprates were identified as having high potential for superconductivity.
• Covalent bond length and hole doping concentration were found to be critical factors influencing Tc.
• ML model analysis revealed insights into the underlying physics of superconductivity.

Conclusions:

• Machine learning, particularly DNNs with combined descriptors, can effectively guide the search for novel superconducting materials.
• The findings reinforce the importance of covalent bond length and doping in cuprate superconductivity.
• The proposed workflow enhances the robustness and practicality of ML-driven material discovery.