ESR8 - Optimization and highly parallel implementation of domain decomposition based solvers for water – turbines simulations

Recruiting Institution/Company: MAVEL AS (Czech Republic)

PhD awarded by: Vysoka Skola Banska – Technicka Univerzita Ostrava (Czech Republic)

Objectives: The project focuses on the development, implementation, and optimization of the highly parallel solvers of contact problems for large-scale water – turbines simulations, so that these solvers are able to fully utilize modern HPC architectures. The idea is based on the combination of the FETI type domain decomposition methods developed by the IT4I team and efficient algorithms employing active set strategy (MPRG, SMALSE, semismooth Newton method) and using all three parallelization techniques: MPI + OpenMP + vectorization to reduce the communication and to preserve the efficiency with increasing problem size. These solvers will be used for such problems as the simulation of the water – turbine bearing/shaft system, topological and parameter optimization of chosen water – turbine components or lifetime and limit states computation for stressed parts of water – turbines.

Expected Results:
1) SW – highly parallel and efficient solver library for large scale water – turbines simulations,
2) benchamarking – numerical simulations of selected large scale water – turbine problems with billions of unknowns,
3) best practices – performance evaluation and comparison with existing approaches.

Planned secondment(s):
TUM: Learning new approaches in solution of friction contact problems,
IT4I: Learning domain decomposition solver techniques and their efficient implementation.