Modelling phase separation in Fe-Cr system using different atomistic kinetic Monte Carlo techniques

Atomistic kinetic Monte Carlo (AKMC) simulations were performed to study α-α' phase separation in Fe-Cr alloys. Two different energy models and two approaches to estimate the local vacancy migration barriers were used. The energy models considered are a two-band model Fe-Cr potential and a cluster expansion, both fitted to ab initio data. The classical Kang-Weinberg decomposition, based on the total energy change of the system, and an Artificial Neural Network (ANN), employed as a regression tool were used to predict the local vacancy migration barriers 'on the fly'. The results are compared with experimental thermal annealing data and differences between the applied AKMC approaches are discussed. The ability of the ANN regression method to accurately predict migration barriers not present in the training list is also addressed by performing cross-check calculations using the nudged elastic band method.