Uncertainty Quantification in the Atomistic Modelling of Grain Boundaries

Atomistic simulations have been successful in the prediction of materials properties in many applications. Commonly used atomistic simulation methods such as density functional theory (DFT) and molecular dynamics (MD) possess inherent uncertainties that may greatly impact simulation results. However, the uncertainties are rarely quantified, mostly due to the limitations in computational resources and the complexity of the methods for uncertainty quantification.<br/><br/>Grain boundaries are interfaces between grains in polycrystalline materials, being in an intermediate state in between crystalline and amorphous materials, has many unique properties, such as the ability to provide fast transport paths for atoms and ions. Grain boundary motion is a key process that governs the microstructural evolution in materials. In energy materials such as batteries, grain boundaries are crucial to ionic transport. Due to the limitations of current experimental methods in understanding the mechanisms of grain boundary motion at the atomic level, atomistic simulations have become an essential tool for mechanistic studies.<br/><br/>In our study, we adopt elemental metallic materials as model materials, and investigate the migration and diffusion of grain boundaries using MD simulations. We demonstrate that some sources of uncertainty can significantly impact simulation results, but were usually ignored in prior works. In the simulations, atomic velocities are randomly assigned according to the Maxwell-Boltzmann distribution. The different assignments of initial velocities can significantly impact the simulation results of grain boundary migration rate and diffusion coefficient, but the effects are rarely quantified. This work highlights the impact of different assignments of initial velocities and the quantification of the uncertainty in the measurement of grain boundary migration rate and diffusion. We also show that the simulation box sizes at which similar types of simulations are usually performed may cause unphysical artifacts that can be resolved by using larger simulation box sizes.