2024 AIChE Annual Meeting

(279d) Molecular Dynamics Simulations and Machine Learning Assisted Size and Temperature Dependent Shape Analysis of Metal Nanoparticles

Authors

Fichthorn, K., Pennsylvania State University
Metal nanocrystals are useful in a host of applications, ranging from catalysis to plasmonics to electronics to various personal and health applications, and many studies show there are optimal nanocrystal sizes and morphologies for these applications. In the interest of reliably synthesizing optimal nanocrystals, it is important to understand the shape trajectories nanocrystals follow as they grow. In a self-seeding synthesis, the small (likely single-nm) nanocrystal seeds that form after nucleation are fluxional and we expect them to assume an equilibrium shape distribution as they grow. As the nanocrystals grow larger, they may become locked into kinetic shapes. It is important to understand the thermodynamic-to-kinetic shape transition to design processing routes by which nanocrystals with specific sizes and shapes could be synthesized. Such an understanding hinges on resolving the thermodynamics of nanocrystal shapes.

In this study, we conducted Parallel-Tempering Molecular Dynamics (PTMD) simulations to investigate the temperature- and size-dependent equilibrium shapes for both mono-metallic (Cu and Ag) and bi-metallic (AgCu) nanocrystals in the 100- to 200-atom size range. Additionally, we employed Common Neighbor Analysis (CNA) and unsupervised machine learning methods to classify the shapes for each nanoparticle into different structural motifs.

Our PTMD studies of Cu nanocrystals reveal a preference for shapes intermediate between decahedra and icosahedra, which we designate as Dh-Ih, due to vibrational entropy effects. Analyzing temperature-dependent shape distributions across various sizes highlights the significant role of entropy in shaping Cu nanocrystals, indicating that studies solely focused on determining minimum-energy shapes may not accurately predict experimental shapes at varied temperatures. Additionally, significant shape changes are observed with changes in size, including alterations occurring with just a single atom. Our simulations of Ag nanocrystals also exhibit the Dh-Ih. Entropy-related phenomena are also prominent in Ag nanocrystals. For instance, pure FCC nanocrystals are less common than those containing stacking faults, and minimum potential energy shapes are not always favored across all temperatures.

In the study of AgCu bi-metallic nanocrystals, we frequently encountered nanocrystals with distinctive combinations of CNA signatures that deviate from the commonly applied empirical classification guidelines. Thus, we opted to employ machine learning methods to classify these shapes in a more robust and automatic manner. By combining the descriptors of nanoparticle structures based on their atomic environment and composition with unsupervised learning methods such as K-Means and Gaussian mixture method, we can distinguish between different and more detailed structural motifs. Overall, our study underscores the intricate interplay between temperature, size, and chemical identity in shaping the morphology of metallic nanocrystals. The insights gained from this study hold potential for informing the development of processing strategies aimed at achieving selective nanoparticle shapes.