Quantitative Image Simulation and Analysis of Nanoparticles

Materials science increasingly relies on powerful microscopes to study the relationship between a material property and its underlying structure. Understanding this relationship is critical for catalysis, due to the importance of structure at the nanoscale. High-Resolution Transmission Electron Microscopy (HRTEM) has become a routine analysis tool for structural characterization at atomic resolution, and with the recent development of in-situ TEMs, it is now possible to study catalytic nanoparticles under reaction conditions. However, the connection between an experimental image, and the underlying physical phenomena or structure is not always straightforward. The aim of this thesis is to use image simulation to better understand observations from HRTEM images. Surface strain is known to be important for the performance of nanoparticles. Using simulation, we estimate of the precision and accuracy of strain measurements from TEM images, and investigate the stability of these measurements to microscope parameters. This is followed by our efforts toward simulating metal nanoparticles on a metal-oxide support using the Charge Optimized Many Body (COMB) interatomic potential. The simulated interface structures are used as input for image simulations, to understand how support-induced strain influences a HRTEM image. This thesis also introduces two novel analysis tools for atomic-resolution images. The first tool is an automatic method for calculating strain from HRTEM images with several advantages over previous methods. The second tool, is a neural network based algorithm for recognition of the local structure in images. The neural network was trained entirely from image simulations, but is capable of making correct predictions on experimental images.