3DSC - a dataset of superconductors including crystal structures

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Data-driven methods, especially machine learning, can accelerate materials discovery by identifying patterns in existing data.
• The discovery of new superconductors is hindered by a scarcity of accessible, comprehensive datasets.

Purpose of the Study:

• To introduce the 3DSC, a novel, publicly available superconductivity dataset.
• To augment existing databases by including approximate 3D crystal structures alongside critical temperature (Tc) data.
• To demonstrate the utility of structural information in predicting superconducting properties.

Main Methods:

• Compilation of a new superconductivity dataset (3DSC) including Tc values and non-superconductor data.
• Augmentation of the dataset with approximate 3D crystal structures.
• Statistical analysis and machine learning experiments to evaluate the impact of structural data on Tc prediction.

Main Results:

• The 3DSC dataset provides critical temperature (Tc) and 3D structural information for superconductors and non-superconductors.
• Incorporating 3D structural data significantly improves the prediction accuracy of the critical temperature (Tc) compared to composition-only data.
• The study validates the enhanced predictive power of machine learning models when utilizing comprehensive structural information.

Conclusions:

• The 3DSC dataset represents a valuable resource for advancing superconductivity research.
• Access to 3D structural data is crucial for improving machine learning-based predictions of critical temperature.
• This work paves the way for more efficient discovery of novel superconducting materials through data science.