Atomistic design of nanocrystalline samples: A Bayesian approach

Grain boundary energy influences multiple properties like grain-growth, complexion-transition, intergranular corrosion, etc. In atomistic simulations, a straightforward method of designing polycrystalline samples with the desired average grain boundary energies is difficult to obtain on account of multiple orientational degrees of freedom. Here we present a strategy based on Bayesian optimization, which aims at producing simulated nanocrystalline samples by minimizing the relative error between the computed and targeted grain boundary energy. Analysis of the atomistic structures of the optimized samples reveals that an increase in the average grain boundary energy primarily results from an increase in the fractions of grain boundary atoms with very low and very high free volumes.