Machine learning screening the feasibility for self-propagating reactions of the MAX and MAB phases with ab initio dataset

Abstract

As a rapid, scalable, and eco-friendly synthesis method, the self-propagating high-temperature synthesis (SHS) is widely applied for ceramics, however, whose development is always limited by high experimental costs. To address this challenge, a workflow for its feasibility is established by combining first-principles calculations and machine learning to quickly predict SHS reactions of MAX and MAB phases. Based on feasibility criteria of SHS (adiabatic combustion temperature T ad  > 1800 K), 60 MAX and 19 MAB phases are predicted to be feasible for direct-ignition SHS under ideal adiabatic assumptions, with 17 experimentally validated ones. Furthermore, some high- T ad phases are successfully synthesized by SHS, confirming the practical utility of the calculated thermodynamic properties and T ad . It follows that all the data of T ad as well as elemental properties are fed to train a Random Forest Regression model and a SISSO-derived analytical model. Moreover, the synergistic effect of low-VEC transition metals and high-VEC main-group elements significantly improves the heat release performance. Of much interest, a new MAB phase V 5 PB 2 is experimentally discovered by SHS with the aid of machine-learning models. This screening workflow is expected to be a valuable tool for future large-scale synthesis and optimization of reaction conditions for materials.