Hybrid parallelism für den Jülicher DFT code FLEUR
The full-potential all-electron code FLEUR enables modelling of quantum materials by performing first-principles calculations based on the density-functional theory (DFT).
A major difficulty in the applicability of FLEUR is the insufficient scaling behavior for large structures, which limits the possibility to simulate large unit-cells frequently encountered in disordered or multicomponent systems or in situations when structural or chemical defects become of significant importance.
In particular, the increasing demand to simulate complex heterostructures, defects and diffusion processes will lead to a further increase of system sizes and required computational power.
The present project aims at implementing an efficient shared memory OpenMP parallelization extending the current MPI realization such that current and future supercomputers can be used to perform large DFT calculations.
The final version of the program should not only enable the use of several hundreds of nodes but also show a high single-node performance.
June 2014 - June 2016
Prof. Dr. S. Blügel, Peter Grünberg Institut (PGI-1) und Institute for Advanced Simulation (IAS-1), Forschungszentrum Jülich
Prof. Dr. M. Müller, IT Center, RWTH Aachen University
Dr. D. Wortmann, Peter Grünberg Institut (PGI-1) und Institute for Advanced Simulation (IAS-1), Forschungszentrum Jülich
Dr. U. Alekseeva, HPC Group, IT Center, RWTH Aachen