Machine learning modeling of superconducting critical temperature (1709.02727v2)

Abstract: Superconductivity has been the focus of enormous research effort since its discovery more than a century ago. Yet, some features of this unique phenomenon remain poorly understood; prime among these is the connection between superconductivity and chemical/structural properties of materials. To bridge the gap, several machine learning schemes are developed herein to model the critical temperatures ($T_{\mathrm{c}}$) of the 12,000+ known superconductors available via the SuperCon database. Materials are first divided into two classes based on their $T_{\mathrm{c}}$ values, above and below 10 K, and a classification model predicting this label is trained. The model uses coarse-grained features based only on the chemical compositions. It shows strong predictive power, with out-of-sample accuracy of about 92%. Separate regression models are developed to predict the values of $T_{\mathrm{c}}$ for cuprate, iron-based, and "low-$T_{\mathrm{c}}$" compounds. These models also demonstrate good performance, with learned predictors offering potential insights into the mechanisms behind superconductivity in different families of materials. To improve the accuracy and interpretability of these models, new features are incorporated using materials data from the AFLOW Online Repositories. Finally, the classification and regression models are combined into a single integrated pipeline and employed to search the entire Inorganic Crystallographic Structure Database (ICSD) for potential new superconductors. We identify more than 30 non-cuprate and non-iron-based oxides as candidate materials.

• The paper presents an ML classification model that predicts T₍c₎ categories with 92% accuracy using features from chemical compositions.
• Regression models are developed for cuprate, iron-based, and low-temperature superconductors, revealing the interplay between composition and superconductivity.
• An integrated ML pipeline screens the ICSD database and forecasts 30 novel candidate compounds, propelling high-T₍c₎ material discovery.

Machine Learning Modeling of Superconducting Critical Temperature

The pursuit of understanding and predicting superconductivity in materials has taken a modern turn with the application of ML strategies. This paper presents a significant endeavor to model and predict superconducting critical temperatures ($T_{\mathrm{c}}$) through various ML approaches using extensive databases of known superconductors. The research leverages the SuperCon database, which archives information about over 12,000 superconductors. This paper not only outlines the development of ML models but also demonstrates their utility in predicting superconducting behavior and potentially discovering new superconductors.

Central to the paper is the division of known superconducting materials into two categories based on their critical temperatures—those above and below 10 K. A classification model is trained to predict the $T_{\mathrm{c}}$ category using coarse-grained features extracted solely from chemical compositions. Remarkably, the classification model achieved an out-of-sample accuracy of approximately 92%, underscoring the strength of machine learning in handling complex material properties.

In addition to classification, separate regression models tailor predictions of $T_{\mathrm{c}}$ values for specific material families such as cuprates, iron-based superconductors, and a general class of low-temperature superconductors. These models highlight the nuanced interplay between composition, structure, and superconductivity across different material classes, revealing potential insights into the mechanisms driving high-temperature superconductivity.

A particular strength of this work lies in its methodology to enhance model accuracy and interpretability by incorporating new features sourced from the Inorganic Crystallographic Structure Database (ICSD), through the use of online repositories. The data-driven approach fundamentally shifts from traditional theories that heavily rely on known microscopic interactions, offering predictive results that are validated by electron band structures.

The culmination of these efforts is the synthesis of classification and regression models into an integrated pipeline, designed for high-throughput screening of untapped compounds in the ICSD for potential new superconductors. Notably, the pipeline forecasts around 30 candidate materials, primarily non-cuprate and non-iron-based oxides, as promising superconductors. The often-adapted ML frameworks in material discovery demonstrate the potential of ML to extrapolate beyond existing datasets to generate novel insights and predictions.

The practical implications of these advancements are substantial. By providing a robust predictive framework, the paper enables targeted experimental validation, potentially accelerating the discovery of high-$T_{\mathrm{c}}$ materials. The theoretical contributions extend to understanding the relationship between elemental properties and superconductivity, which could reshape foundational models in the field.

Looking ahead, this research opens avenues for the continuous refinement of knowledge on superconducting materials. Machine learning's ability to assimilate vast spans of compositional and structural data underscores its role as a pivotal tool in both understanding longstanding phenomena and predicting future material innovations. As more high-quality data becomes available, it is plausible to expect ML models to deliver even more accurate predictions, hopefully ushering in new paradigms in superconductor discovery and application.