Automated Data-Driven Upscaling of Transport Properties in Materials

Transport properties of complex defects are crucial factors that control the performance of many material systems, e.g., the radiation tolerance of materials for nuclear fusion or fission applications. Characterizing the transport of complex defects is however notoriously tedious and<br/>time-consuming, especially as the defects grow, leading to a combinatorial explosion in the number of possible conformations and local transition pathways. We will present a large-scale data-driven approach to automatically obtain reduced-order models of defect evolution, transport coefficients, as well as effective continuum transport equations, from large number of short molecular dynamics (MD) simulations. The optimal MD simulations to carry out are identified on-the-fly using a Bayesian uncertainty quantification framework and automatically executed on a massively-parallel task-execution infrastructure. We show how this microscopic information can be systematically and efficiently upscaled into meso and macro-scale representations that can inform microstructure evolution models.