Robust, Multi-Length-Scale, Machine Learning Potential for Ag–Au Bimetallic Alloys from Clusters to Bulk Materials

Materials composed of Ag, Au, and Ag–Au alloys remain of great interest despite decades of intense research scrutiny. We interpret these efforts as an impetus for developing robust, accurate, and relatively fast computational methods for modeling these materials. Herein, we describe the training, development, and validation of a machine learning deep neural-network potential (DNP) for improved modeling of Ag–Au systems. This DNP was iteratively trained using density functional theory (DFT) to produce a robust multi-length-scale potential, which yields results comparable to DFT on a wide range of properties such as equilibrium and nonequilibrium lattices, mechanical properties, and defect energies. Further, this DNP can well describe adatom (Ag or Au) energy barriers for diffusion on 100-, 110-, and 111-terminated surfaces (Ag or Au), in agreement with previously reported works. We utilized the DNP to study the nucleation and growth of simulated seeded core–shell Ag and Au nanoparticles (NP). We show that both nanoalloys grow such that 111 facets significantly increase at the expense of the 100 ones. In contrast, the Ag core NP is found to have a more disordered inner structure than the Au one and that Ag adatoms in Au@Ag NP have a more pronounced penetration power than Au in Ag@Au NP. These findings are rationalized in terms of adatom adsorption and diffusion energies.