In the EXASTEEL project, experts on scalable iterative solvers, computational modeling in material
science, performance engineering, and parallel direct solvers joined forces to develop new computational
algorithms and implement software for a grand challenge problem from computational material science,
i.e., the predictive simulation of the Nakajima test to obtain forming limit curves (FLCs) for dual-phase steel
materials. EXASTEEL is a project in the German exascale initiative "Software for Exascale
Computing" which is funded as a priority program by the German Science Foundation
(Deutsche Forschungsgemeinschaft - DFG).
There is an increasing need for predictive simulations of the macroscopic behavior of complex new materials. In the EXASTEEL project, this problem is considered for modern micro-heterogeneous (dual- phase) steels, attempting to predict the macroscopic properties of new materials from those on the microscopic level. It is the goal to develop algorithms and software towards a virtual laboratory for predictive material testing in silico. A bottleneck is the computational complexity of the multiscale models needed to describe the new materials. Therefore, new ultra-scalable nonlinear implicit solvers will be developed and combined with a highly parallel computational scale bridging approach (FE2), intertwined with a consequent and permanent performance engineering, to bring the challenging engineering application of a virtual laboratory for material testing and design to extreme scale computing.
In the talk, we present results on the current status of the project. This includes scalability results of our computational homogenization software FE2TI with more than a million parallel tasks for plasticity problems from material science. Parallel scalability of our solvers, based on a combination of FETI-DP domain decomposition methods and algebraic multigrid methods, on up to more than a million MPI ranks and up to several hundred thousands of cores will be shown as well. New nonlinear domain decomposition methods as a robust alternative to the standard Newton-Krylov-DD approach are discussed as well as their potential to save energy. If time allows, we will also discuss the parallel simulation of a contact problem discretized by the FE2 method. Such a problem is necessary to be able to simulate the Nakajima test for computing the FLCs. The computational results are obtained on different architecture including the BG/Q Juqueen (Forschungszentrum Juelich, Germany) and Mira (Argonne National Laboratory, USA) as well as the KNL machines Theta (Argonne National Laboratory, USA) and Oakforest-PACS (Joint Center for Advanced High Performance Computing, Japan).