Application-agnostic software for model adaptivity of two-scale coupled problems

Surrogate modeling promises tremendous speed-ups for coupled multi-scale problems. While significant advances in the actual surrogate models have been achieved in recent years, software engineering (SE) and high-performance computing (HPC) aspects are oftentimes still ignored. In this project, we target SE aspects by studying how surrogate modeling in coupled two-scale problems can be used irrespective of the concrete application, the used simulation software, and the surrogate models themselves, with the goal to modularize monolithic software concepts. Moreover, concerning HPC, we want to study how adaptive on-the-fly switching within a hierarchy of surrogate models can be efficiently executed on heterogeneous compute hardware.

Today, such model hierarchies are (at best) available in specialized and hard-wired research codes to the best of our knowledge. More precisely, their integration remains highly application and implementation specific. We will, thus, enhance the existing preCICE Micro Manager, an application-agnostic two-scale coupling software, with new functionality for offline snapshot generation and online adaptive model switching, focusing on efficiency and reusability. We will test the new software approach on nonlinear two-scale solid material problems. The new approach should help domain experts to focus on the surrogate modeling itself, while offloading SE and HPC challenges to research software engineering experts.