In the quest for high-temperature superconductors, a team of researchers from the University of Tokyo, led by Kazuaki Tokuyama, has developed a machine learning approach to identify promising candidates among ternary hydrides. Superconductors are materials that can conduct electricity without resistance, and high-temperature superconductors could revolutionize the energy industry by enabling more efficient power transmission and storage.
The researchers focused on ternary hydrides, which are compounds consisting of three elements, one of which is hydrogen. These materials have shown potential for high superconducting transition temperatures, but the vast compositional space makes experimental exploration time-consuming and costly. To tackle this challenge, the team constructed an ensemble of 30 machine learning models trained on a curated dataset of approximately 2000 binary and ternary hydride entries.
The models were used to screen a broad set of A-B-H compositions at pressures of 100, 200, and 300 GPa. The screening outcomes were evaluated based on prediction consistency across the ensemble members. The analysis highlighted several high-scoring compositional systems, including Ca-Ti-H, Li-K-H, and Na-Mg-H, which were not explicitly included in the training dataset. This suggests that the models can generalize well to new compositions.
Feature-importance analysis revealed that elemental properties such as ionization energy and atomic radius contribute significantly to the learned composition-level trends in superconducting transition temperature. This insight could guide future experimental work in the field.
The practical applications for the energy sector are significant. High-temperature superconductors could enable more efficient power transmission and storage, reducing energy losses and improving the overall efficiency of the energy grid. Additionally, the machine learning approach developed by the researchers could be applied to other materials science problems, accelerating the discovery of new materials with desirable properties.
The research was published in the journal npj Computational Materials. The study demonstrates the utility of ensemble-based machine learning as a primary screening tool for identifying promising regions of chemical space in superconducting hydrides, paving the way for more efficient and targeted experimental work in the future.
This article is based on research available at arXiv.