Predicting the shapes of Au55 and Au147: Force fields vs density-functional theory

Abstract

Gold nanocrystals have been widely used in sensing and medicine, where nanocrystal shape can profoundly influence properties. To describe and predict the structure and properties of Au nanomaterials, first-principles studies are the most accurate. Force fields can provide effective surrogates for first-principles calculations, and in the case of Au, many such force fields exist. To clarify the current state of Au structure prediction using force fields, we used parallel tempering molecular dynamics simulations to explore the temperature-dependent shape distributions of Au147 and Au55 from three different Embedded-Atom Method (EAM) force fields. We used machine learning to classify nanoparticle shapes and observe a wide variation in the temperature-dependent shape distributions among the various force fields, accompanied by a wide range of melting temperatures. We also compared the EAM structures to those from literature studies employing an Artificial Neural Network (ANN), a Gaussian process regression machine-learning force field, density functional tight binding, the Gupta potential, and Density Functional Theory (DFT). We re-optimized the lowest-energy structures from each force field/study using DFT and found that the lowest-energy structures for both Au147 and Au55 are amorphous. These structures were predicted by the ANN and DFT. Most of the 30 structures with minimum energy for each size are either hollow or partially to totally disordered and not icosahedral, as the sizes of these nanocrystals might imply.