This title appears in the Scientific Report :
2012
Please use the identifier:
http://hdl.handle.net/2128/4912 in citations.
RealSpace FiniteDifference PAW Method for LargeScale Applications on Massively Parallel Computers
RealSpace FiniteDifference PAW Method for LargeScale Applications on Massively Parallel Computers
Simulations of materials from first principles have improved drastically over the last few decades, benefitting from newly developed methods and access to increasingly large computing resources. Nevertheless, a quantum mechanical description of a solid without approximations is not feasible. In the...
Saved in:
Personal Name(s):  Baumeister, Paul Ferdinand (Corresponding author) 

Contributing Institute: 
QuantenTheorie der Materialien; IAS1 QuantenTheorie der Materialien; PGI1 
Imprint: 
Jülich
Foschungszentrum Jülich GmbH Zentralbibliothek, Verlag
2012

Physical Description: 
VI, 212 S. 
Dissertation Note: 
RWTH Aachen, Diss., 2012 
ISBN: 
9783893368365 
Document Type: 
Book 
Research Program: 
Spinbased and quantum information 
Series Title: 
Schriften des Forschungszentrums Jülich. Schlüsseltechnologien / Key Technologies
53 
Link: 
OpenAccess 
Publikationsportal JuSER 
Simulations of materials from first principles have improved drastically over the last few decades, benefitting from newly developed methods and access to increasingly large computing resources. Nevertheless, a quantum mechanical description of a solid without approximations is not feasible. In the wide field of methods for $\textit{ab initio}$ calculations of electronic structure, it has become apparent that density functional theory and, in particular, the local density approximation can also make simulations of large systems accessible. Density functional calculations provide insight into the processes taking place in a vast range of materials by their access to an understandable electronic structure in the framework of the KohnSham single particle wave functions. A number of functionalities in the fields of electronic devices, catalytic surfaces, molecular synthesis and magnetic materials can be explained by analyzing the resulting total energies, ground state structures and KohnSham spectra. However, challenging physical problems are often accompanied by calculations including a huge number of atoms in the simulation volume, mostly due to very low symmetry. The total workload of wavefunctionbased DFT scales at best quadraticallywith the number of atoms. This means that supercomputersmust be used. In the present work, an implementation of DFT on realspace grids has been developed, suitable for making use of the massively parallel computing resources of modern supercomputers. Massively parallel machines are based on distributed memory and huge numbers of compute nodes, easily exceeding 100,000 parallel processes. An efficient parallelization of density functional calculations is only possible when the data can be stored processlocal and the amount of internode communication is kept low. Our realspace grid approach with threedimensional domain decomposition provides an intrinsic data locality and solves both the Poisson equation for the electrostatic problemand the KohnSham eigenvalue problem on a uniform realspace grid. The derivative operators are approximated by finite differences leading to localized operators which only require communication with the nearest neighbor processes. This leads to excellent parallel performance at large system sizes. Treating only valence electrons, we apply the projectoraugmented wave method for accurate modeling of energy contributions and scattering properties of the atomic cores. In addition to realspace grid parallelization, we apply a distribution of the workload of different KohnSham states onto parallel processes. This second parallelization level avoids the memory bottleneck for large system sizes and introduces even more parallel speedup. Calculations of systems with up to 3584 atoms of Ge, Sb and Te were performed on (up to) all 294,912 cores of JUGENE, the massively parallel supercomputer installed at Forschungszentrum Jülich. 