Atomistic Insights into The High Temperature Dislocation Dynamics of Cu using a Bayesian Force Field from First Principles

High temperature dislocation dynamics present a difficult simulation task for existing classical and<br/>ab initio methods due to the required accuracies and length-scales. These limitations ultimately<br/>prohibit advanced understanding of plastic deformation of materials under relevant stimuli. Here, we<br/>develop a Bayesian machine-learned force field (MLFF) from ab initio (first principles) training data<br/>that extends quantum-mechanical accuracy to large length-scale molecular dynamics simulations<br/>which permit reliable description of high-temperature dislocation dynamics in Cu and direct<br/>comparison to experimental observations. In concert, a general and intuitive training protocol<br/>is defined for construction of MLFFs for the description of dislocations, which can be employed for<br/>other systems of interest. The resulting MLFF provides excellent predictions of both static bulk<br/>properties (e.g. bulk modulus and elastic tensor), stacking fault energies, and the dynamic evolution<br/>of edge and screw dislocations, as well as cross-slip mechanisems and activation energy across a<br/>broad range of temperatures and applied shears. Such simulations allow for the unbiased, atomistic<br/>insight into plastic deformation under various stimuli, helping to ultimately explain experimental<br/>observations through an atomic-lense.