Machine learning based on-the-fly kinetic Monte Carlo simulations of sluggish diffusion in Ni-Fe concentrated alloys

Abstract

Defect diffusion in concentrated alloys plays a key role on governing their unique mechanical and physical properties. In such alloys, defect diffusion depends on its complex local atomic environment and varies from site to site due to the chemical disorder. On-the-fly determination of the defect migration barrier at every site using the standard nudged elastic band (NEB) method is computationally expensive and often impractical. In this work, we couple machine learning and kinetic Monte Carlo (KMC) to study vacancy-mediated sluggish diffusion in concentrated Ni-Fe model alloys. Based on about 32,000 pre-calculated NEB barriers, an artificial neural network (ANN) based machine learning model is developed to accurately predict the vacancy migration barriers for arbitrary local atomic environments, including both random solution configurations and alloys with short-range orders. The ANN model is then coupled with KMC (ANN-KMC) to determine the vacancy migration barriers on-the-fly, enabling an efficient study of the vacancy diffusion in the full composition range at a wide range of temperatures. In addition, a composition and temperature dependent jump attempt frequency model is developed. Upon calibration, the ANN-KMC modeling can predict nearly identical vacancy diffusivities as those obtained from independent molecular dynamics (MD) and temperature accelerated dynamics (TAD) simulations at their accessible temperatures. The sluggish diffusion mechanisms in this specific alloy system at both high and low temperatures are discussed based on the ANN-KMC results.