Machine learning-accelerated evolutionary Monte Carlo for rapid phase exploration of compositionally complex materials in reactive environments

Compositionally complex materials, e.g. high-entropy alloys/oxides, have long been of research interest for their unique mechanical and functional properties thanks to the synergistic effects between the large number of components. More recently, they have emerged as a promising platform for the development of novel heterogeneous catalysts for thermo- and electro-chemical applications. However, computational investigation of the phase stability of such catalyst materials remains challenging due to the large design space and the complex active site dynamics under reaction conditions. In this paper, we present a novel evolutionary Monte Carlo algorithm which allows fast exploration of stable compositionally-complex catalyst phases in reactive environments and can be generalized for any catalyst materials and any reaction conditions. In detail, we will showcase that this method can be used in combination with finetuned machine learning potentials for predicting bulk and surface mixing/segregating behaviors of compositionally-complex catalysts under various reaction conditions with reasonable computational cost.