Composition- and Ordering-Dependent Evolution of Simulated Kikuchi Patterns of Au-Ni Alloys

Abstract

Understanding how Kikuchi patterns behave with subtle material variations is essential for developing machine learning (ML) based indexing methods for structurally and chemically complex cases, such as phases with potential sub-lattice order or (meta)stable defects with segregation (defect phases). Simulated Kikuchi patterns of the binary Au-Ni system were systematically analysed to investigate the effects of lattice parameter, chemical composition, partial site occupancy and ordering. The full compositional range from pure nickel to pure gold was considered, including a hypothetical ordered L$1_2$ Au$_3$Ni structure. Simulation parameters were optimised by comparison with experimental patterns. Normalized cross correlation showed limited sensibility to subtle differences between patterns. EMsoft simulation results revealed a systematic increase in mean intensity and more reflections contribute significantly as gold content increases. A surprising four-fold increase in mean intensity from 99 % gold to the pure gold sample highlighted limitations of partial site occupancy simulation by EMsoft. Difference maps showed enhanced normalised intensity along the {111} and {200} bands for the sample with 20 % gold compared to higher gold compositions. By isolating lattice parameter and chemical composition effects, chemical composition contributes predominantly to the mean intensity and strong reflections count, although the normalised intensity distribution was affected by both. Introducing L$1_2$ ordering increased the number of strong reflections and mean intensity, and redistributed normalised intensity along {111} and {200} bands and selected zone axis. Normalised intensity distribution and band width, are key descriptions to distinguish the Kikuchi patterns, which can be incorporated in future representation learning based indexing methods.