Quantifying temperature- and composition-dependent structures of AgCu nanocrystals using machine learning

We explore the structure classification of silver–copper (AgCu) nanocrystals using machine learning. Using parallel-tempering molecular dynamics simulations, we generated equilibrium shapes of Ag \(_{147-N}\) $_{147-N}^{}$ Cu \(_N\) $_N^{}$ nanoparticles. To derive structural descriptors of the nanoparticles, we employed Common Neighbor Analysis, followed by dimensionality reduction using Principal Component Analysis and clustering using K-Means and the Gaussian Mixture Model (GMM). We evaluated the clustering models using the silhouette score and the gap statistic. These measures showed that GMM provides more physically meaningful clustering than K-Means. Through GMM clustering, we obtained five distinct shape and composition classes and 14 different sub-classes. Our analysis shows these different classes and sub-classes reflect the shape, Cu(Ag) content, temperature, and fraction of Ag(Cu) on the nanoparticle surface. This approach of combined atomic-scale simulations and machine learning methods enhances our understanding of the structures in bimetallic nanoparticles and provides valuable insights for designing nanomaterials with tailored properties for applications.