A generalizable machine learning potential of Ag–Au nanoalloys and its application to surface reconstruction, segregation and diffusion

Abstract Owing to the excellent catalytic properties of Ag–Au binary nanoalloys, nanostructured Ag–Au, such as nanoparticles and nanopillars, has been under intense investigation. To achieve high accuracy in molecular simulations surface must be modeled with first-principles precision. In this work, we constructed a generalizable machine learning interatomic potential for nanoalloys based on deep neural networks trained from database calculations. This is highlighted by accurate prediction Au (111) reconstruction segregation toward nanoalloy surface, where empirical force field (EFF) failed both cases. Moreover, regarding adsorption diffusion adatoms surfaces, overall performance our better than EFFs. We stress that reported are blind modeling sense none configurations explicitly included training database; therefore, expected have strong generalization ability wide range play key role investigating evolution, descriptions free surfaces necessary.