Superconductivity allows electric conductance with no energy losses when the ambient temperature drops below a critical value (T c ). Currently, the machine learning (ML)-based prediction of potential superconductors has been limited to chemical formulas without explicit treatment of material structures. Herein, we implement an efficient structural descriptor, the smooth overlap of atomic position (SOAP), into the ML models to predict the T c values with explicit atomic structural information. Using a data set containing 5713 compounds, our ML models with the SOAP descriptor achieved a 92.9% prediction accuracy of coefficient of determination (R 2 ) score via rigorous multialgorithm cross-verification procedures, exceeding the 86.3% accuracy record without atomic structure information. Several new high-temperature superconductors with T c values over 90 K were predicted using the SOAP-assisted ML model. This study provides insights into the structure−property relationship of high-temperature superconductors.

The figure of merit (zT) is a key parameter to measure the performance of thermoelectric materials. At present, the prediction of zT values via machine leaning has emerged as a promising method for exploring high-performance materials. However, the machine learning-based predictions still suffer from unsatisfactory accuracy, and this is related to the size of the data set, the hyperparameters of models, and the quality of the data. In this work, 5038 pieces of data of thermoelectric materials were selected, and several regression models were generated to predict zT values. This large data set-driven light gradient boosting (LGB) model with 57 features performed with an excellent accuracy, achieving a coefficient of determination (R 2) value of 0.959, a root mean squared error (RMSE) of 0.094, a mean absolute error (MAE) of 0.057, and a correlation coefficient (R) of 0.979. Owing to the large size of the data set, the prediction accuracy exceeds that of most reported zT predictions via machine learning. The “ME Lattice Parameter” was verified as the most important feature in the zT prediction. Furthermore, nine potential candidates were screened out from among one million pieces of data. This study solves the problem of the data set size, adjusts the hyperparameters of the models, uses feature engineering to improve data quality, and provides an efficient strategy to perform wide-ranging screening for promising materials.

Identifying new superconductors with high transition temperatures (T c > 77 K) is a major goal in modern condensed matter physics. The inverse design of high T c superconductors relies heavily on an effective representation of the superconductor hyperspace due to the underlying complexity involving many-body physics, doping chemistry and materials, and defect structures. In this study, we propose a deep generative model that combines two widely used machine learning algorithms, namely, the variational auto-encoder (VAE) and the generative adversarial network (GAN), to systematically generate unknown superconductors under the given high T c condition. After training, we successfully identified the distribution of the representative hyperspace of superconductors with different T c , in which many superconductor constituent elements were found adjacent to each other with their neighbors in the periodic table. Equipped with the conditional distribution of T c , our deep generative model predicted hundreds of superconductors with T c > 77 K, as predicted by the published T c prediction models in the literature. For the copper-based superconductors, our results reproduced the variation in Tc as a function of the Cu concentration and predicted an optimal T c = 129.4 K, when the Cu concentration reached 2.41 in Hg 0.37 Ba 1.73 Ca 1.18 Cu 2.41 O 6.93 Tl 0.69 . We expect that such an inverse design model and comprehensive list of potential high Tc superconductors would greatly facilitate future research activities in superconductors.