Parallelization of an efficient 2D-Lagrangian model for massive multi-domain simulations.

Abstract

The parallelization of algorithms is an essential step towards the optimization of large-scale computations. The modeling of evolving multi-domain problems is not an exception to this rule, specifically when it is applied to the context of microstructural evolutions. A new method for the simulation of evolving microstructures has been introduced in a previous work, consisting on a modified front-tracking approach where the main originality is that not only interfaces between domains are discretized but also their bulks. This new model has obtained promising results in terms of accuracy and numerical performance, however, it has been implemented in a sequential environment and is not readily usable in modern HPC units for parallel computations. This article proposes a parallel implementation for the new model using a distributed-memory approach developed with the standard protocol ‘message passing interface’. The new parallel methodology has been tested in an HCP station with up to 140 cores for problems involving motion by curvature flow in polycrystals, i.e. by considering pure grain growth. Good results were obtained in terms of CPU-time and speed-up for large polycrystals (with up to 560 000 initial grains), showing that this model can lead to fast and/or large computations of microstructural evolutions in a full-field context.

A 2D Front-Tracking Lagrangian Model for the Modeling of Anisotropic Grain Growth

“…As an alternative to model microstructural evolutions with anisotropic GB properties, we propose the TRM model presented in [17][18][19]; this article will present the needed implementations in order to model fully anisotropic grain properties with the TRM model. Special attention will be given to the development of a robust high order Multiple Junction (MJ) decomposition algorithm and to the reformulation of the velocity equation at triple junctions extending the methodology presented in [15] to an anisotropic context, using the notions used in [12] for its discrete formulation.…”