Prediction of MAX-to-Mxene Convertibility Using Constrained Clustering

Abstract Since 2011, MXenes have emerged as promising materials with diverse applications. However, their synthesis through selective etching of recurring layers in MAX phases remains challenging to predict due to a lack of comprehensive understanding of the factors influencing the synthesis protocol. This work seeks to accelerate the discovery of promising MAX phases for conversion to MXenes by employing machine learning. Inspired by previous work using positive–unlabeled learning for this problem, we employed a combination of a constrained clustering algorithm for identifying convertible materials with a classifier for scoring their convertibility. Our proposed algorithm learns to weigh multiple kernels projecting elemental and calculated properties of MAX phases, thus leveraging clusters that respect constraints holding convertible MAX phases together. We score the convertibility of the identified clusters using a support vector classifier with the previously learned kernel, and retain samples with high convertibility scores. To validate the model, we developed a specific timeline validation that incorporates the time of discovery of each MAX-to-MXene conversion, and show that the proposed pipeline is able to correctly suggest convertible MAX phases before their discovery at a given time. Our results suggest Ta3AlC2, and Ti4GaC3 as notable materials for synthesizing novel MXenes. The model also validates the successful conversions of Nb2GaC to NbC and Zr2AlC to Zr2C despite these not being labeled as positive. Moreover, we also identified 4 MAX phases as alternative candidates to synthesize already discovered MXenes: Ta2GaC, Ti2GaN, Ti3GaC2, and Nb2InC.