Superconductors, defying the electrical resistance, exhibit zeo resistance when cooled below a critical temperature. This fantastic property of superconductors opens the door to numerous real-world applications in energy, transportation, and cutting-edge electronics. Over the last decade, significant progress has been made in the search for high critical temperature superconductors. In this paper, researchers from Georgia Tech and Hanoi University of Science and Technology (Vietnam) have presented the first step in incorporating atomic-level information into machine learning pathways to discover new conventional (or BCS) superconductors, especially at ambient pressure.
Prediction of high-temperature superconductivity at zero temperature was a challenging task for the research scholar due to lack of atomic level information. The researchers have carefully curated a dataset of 584 atomic structures with more than 1100 values of ? and ?log computed at different pressures. Ml models were developed for ? and ?log and used to screen over 80,000 entries of the Materials Project database, unveiling (by first-principles computations) two thermodynamically and dynamically stable materials whose superconductivity may exist at Tc approximately equal to 10-15K and P = 0. They employed a matminer package to convert atomic structures into numerical vectors and used Gaussian process regression as the ML algorithm to achieve this.
The researchers used the ML models for predicting superconducting properties for 35 candidates. Among them, six had the highest predicted Tc values. Some were unstable and needed further stabilization calculations. After verifying the stability of the remaining two candidates, namely cubic structures of CrH and CrH2, their superconducting properties were computed using first-principles calculations. The researchers validated their predictions and performed additional calculations using local-density approximation (LDA) XC functional, confirming the predicted results’ accuracy within 2-3% of the reported values. Also, the researchers investigated the synthesizability of superconductors by tracing their origin in the Inorganic Crystalline Structure Database (ICSD). They found that these were experimentally synthesized in the past and hope future tests will confirm their predicted superconductivity.
In future research, researchers plan to enhance their ML approach by enlarging and diversifying the dataset, using deep learning techniques, and integrating an inverse design strategy to explore the practically infinite materials efficiently. The researchers envision further improvements to their approach, which could facilitate the discovery of high Tc superconductors and collaborate with experimental experts for real-world testing and synthesis.