The Helmholtz-Zentrum Dresden-Rossendorf is a member of the Helmholtz Association of German Research Centres pursuing new insights that will allow us to maintain and improve all of our lives. That’s why the HZDR conducts research in the sectors health, energy, and matter in Dresden and at three other locations. Three of our five large-scale facilities are also available to external guests from around the world to help answer the decisive questions of our society.
Institute of Fluid Dynamics – Department Magnetohydrodynamics
The Institute of Fluid Dynamics was founded in 2012 within the Helmholtz-Zentrum Dresden-Rossendorf. The institute’s work aims for the exploration of dynamic systems in the fields of thermo-fluid dynamic as well as in magnetohydrodynamics. The department magnetohydrodynamics is engaged in the basics of interaction of magnetic fields with electrical conductive liquids. Key activities are experimental and numerical research on flow control by means of electromagnetic forces in conducting fluids, investigation of the local properties of liquid metal two-phase flows, development and application of measuring techniques to determine liquid metal flow quantities, and numerical simulations and optimization of magnetic field self-excitation in liquid metal flows.
Scientific Projects with GPU background
The computaional work in the MHD group at HZDR pursues two main objectives. Particle Image Velocimetry (PIV) and Contactless Inductive Flow Tomography (CIFT) are utilized to reconstruct the velocity fields in transparent fluids (PIV) as well as in liquid metals (CIFT). In particular the latter method iteratively executes an inversion and regularization of a fully occupied linear system which is extremely demanding in order to achieve a real time computation of the velocity field.
A further application that considerably benefits from CUDA is the stringent computation of (realistic)
insulator boundary conditions which significantly influence magnetic field geometry and growth rates
in numerical simulations of liquid metal laboratory experiments. The consideration of an insulating
medium (e.g. air) surrounding the computational domain imposes a non-local problem which can be
solved using a modified integral equation approach, the so called Boundary Element Method (BEM).
Advanced methods, like Fast Multipole Methods (FMM) need less memory at the expense of computational costs although their complexity only scales / O(N). The realization of FMMs on the GPU architecture is not straightforward because the hierarchical structure of FMMs requires implementation strategies that are different from those on CPUs.
Future implementation of the referred algorithms will have to make use of multiple GPUs in order to
allow real time velocity field reconstruction (CIFT) or to achieve highest resolutions in simulations
of magnetohydrodynamic instabilities and/or dynamos that they will be conducted in the framework
of the project DRESDYN (DREsden Sodium facility for DYNamo and thermohydraulic studies) at
Contact: Dr. Andre Giesecke, firstname.lastname@example.org