AI-based descriptor for predicting alloy formation energy

Because of their rich geometric structure and electronic properties, metal alloys have been widely used in catalysis and materials science. Among them, alloys formation energy has an important influence on the formation and catalytic activity of metal alloys. With the development of artificial intelligence and databases in recent years, machine learning has been used to rationally design new materials. Based on the multi-task compressed sensing algotithm in artificial intelligence, the alloy formation energy descriptor of the AB2 alloy formation energy database was investigated. A universal descriptor of the corresponding alloy formation energy was established, and the sensitivity analysis of features revealed the importance of electronic and geometrical properties of metal alloys. The results show that this descriptor has a prediction error lower than 8.10kJ·mol-1 and a better physical interpretation. Finally, the formation energy of a large number of unknown metal alloys was predicted.