Enhanced discovery of transition-metal carbides and nitrides via machine learning
Purposes—To predict the composability and thermodynamic stability of 748 candidate metal carbon/nitride(MAX)materials which are designed in advance.Methods—With the dataset ob-tained from the open quantum materials database(OQMD),a deep neural network(DNN)model based on a machine learning method is used to predict the relative formation energies of candidate MAX materials and explore the correlation between such materials and their chemical properties.Re-sults—The quantitative relationship between relative formation energy and stability can be elucidated by several compositional and structural descriptors.The synthesis of 339 out of the total 748 MAX candidates is highly probable.In comparison to nitride MAX candidates,carbide MAX materials ex-hibit a greater probability of successful synthesis.Conclusions—This work not only discovers several promisingly stable MAX compounds but also develops an accurate and efficient ML on small data sets to reveal the relations between physical and chemical descriptors and thermodynamic stability of MAX phases.
MAXstabilitymachine learningrelative formation energy