Data-driven methods to predict the stability metrics of catalytic nanoparticles

Abstract

A prevailing challenge in computational catalyst design is to discover nanostructures which are thermodynamically stable and synthesizable in practice. Important metrics for the stability of nanostructures include the chemical potential of supported nanoparticles, cohesive energies of nanoparticles, surface and adhesion energies of crystal planes that bound the nanoparticle, and segregation energies in bimetallic nanoparticles. Ab initio methods can calculate these metrics but are computationally intensive due to the large configurational space that these nanostructures span. Moreover, sub-nanometer nanoparticles are structurally flexibile under reaction conditions. Hence, physics-based and machine-learning-derived data-driven approaches are becoming prevalent to determine the stability of nanostructures. In this review we discuss the recent advances in data-driven methods to predict stability metrics of nanoparticles.