Formation ability descriptors for high-entropy carbides established through high-throughput methods and machine learning

Establishing formation ability descriptors is vital for accelerating the discovery and design of novel high-entropy carbides (HECs). However, very few studies have been reported so far. Here, we propose a combination of formation ability descriptors of HECs through high-throughput synthesis and calculation combined with the machine learning approach. Specifically, 91 HEC samples have been synthesized by the self-developed high-throughput ultrafast high-temperature synthesis apparatus, and 22 formation ability descriptors have been calculated by the high-throughput calculations simultaneously. Based on these results, eight formation ability descriptors are selected as the optimal combination to predict the phase formation of HECs through the machine learning method and the genetic algorithm. The validation accuracy of our established formation ability descriptors reaches 93.4%, at least 25.3% higher than that of previously reported ones. This study may provide theoretical guidance for discovering and designing novel HECs.