Finite element method with hierarchical domain decomposition : enabling experimentally relevant mesoscale models

Abstract

Mesoscale materials such as metallic glass present a difficult modeling challenge because their time and length scales place them in a gap where neither continuum mechanics nor quantum mechanics-based models are computationally tractable. The STZ dynamics model is a mesoscale approach to modeling this class of materials. However, modeling the response of such amorphous metals to deformation is still very computationally expensive. As meshes get larger, the runtimes of the mesoscale models get much longer, particularly in three dimensions; in fact, the computation is currently not efficient enough to run on experimentally relevant length scales. This thesis focuses on a hierarchical domain decomposition method that will be combined with other strategies to speed up the current models. A hierarchical mesh was generated, and then used to make the finite-element portion more efficient. The runtime and error of this accelerated model were then studied in order to assess the usefulness of the technique. The results show a mediocre runtime speedup that could become more impressive after optimization. More importantly, error drops off superlinearly with distance from the strained element, so accuracy is not sacrificed when using the accelerated method. Therefore, hierarchical domain decomposition can be used with the other speedup strategies to enable larger mesoscale simulations.