2009 guest student programme
The 2009 guest student programme ran from 3 August to 9 October 2009 with 9 students.
Speck, Robert (Ed.) (2009):
Proceedings 2009, JSC Guest Student Programme on Scientific Computing (PDF, 26 MB),
Technical Report FZJ-JSC-IB-2009-04
Parallel Monte Carlo for Condensed Polymers
Jonathan Groß, Universität Leipzig
Advisers: Thomas Neuhaus, Marcus Richter, Binh Trieu, JSC
A simple model of a condensed polymer was simulated using the parallel tempering algorithm. The implementation makes use of multiple cores in modern workstations utilizing POSIX threads. Also there is an outlook to porting the parallel tempering algorithm to GPGPUs using NVIDIA's CUDA.
Threshold Determination for the Fault-Tolerant Implementation
Ricardo Kennedy, Universität zu Köln
Adviser: Stefan Dürr, JSC
The controlled-NOT (CNOT) gate is implemented fault-tolerantly using the Juelich Massively Parallel Ideal Quantum Computer Simulator (JUMPIQCS) and its performance is tested under the influence of two different error sources: For decoherence errors, we find that the threshold for effective error correction (10-6 < pthr< 10-5) is comparable to the single qubit threshold found in . In the case of operational errors, we find that the threshold is of the same order of magnitude as in the single qubit case, but dependent on the initial state. It is determined to be σthr1=(5.5±0.3)*10-2 for a separable initial state |11> and σthr2=(2.0±0.3)*10-2 for the maximally entangled state 1/√2*(|01>-|10>)
Algorithmic optimization issues in quenched lattice gauge theory
Malik Kirchner, Humboldt University Berlin
Adviser: Stefan Dürr, JSC
Calculations in lattice are very time consuming and need an enormous amount of memory and computing power. Inverting the Dirac-operator is very costly. Many architectures compute faster in single than double precision. It is possible to speed up the inversion of the Dirac-operator by iterative "mixed precision" solvers by a factor of almost two. The Monte-Carlo integration is done by a multi-hit Metropolis algorithm based on random uniformly distributed (N) elements.
I/O Problems in Massively Parallel Applications
Lukasz Kucharski, Adam Mickiewicz University Poznan
Advisers: Jan Meinke, Sandipan Mohanty, Wolfgang Frings, JSC
Recently ProFASi application has encountered the problem of excessive file usage on massively parallel systems. This became a real concern in terms of performance degradation and the file number quota limits. This work focuses on integrating SIONlib into ProFASi by introducing a new middle layer library called SFFS. SFFS aims to extend SIONlib capabilities to make it a general parallel I/O replacement for standard C/C++ paradigms. This paper introduces SFFS concepts and discusses ProFASi performance gains and the integration process.
Simulation of Drift-Kinetic Dynamics of Charged Particles in Magnetized Plasmas
Stilianos Louca, Friedrich Schiller Universität Jena
Advisers: Paul Gibbon, Benjamin Berberich, JSC
In particle codes for fusion plasma simulations, an upper bound for the integration time-step is set by the high frequency gyro-motions (∼ THz) of electrons. A method for replacing the exact electron positions by their guiding centers is discussed and tested on single particle simulations. It is shown, that for typical electromagnetic fields found in fusion experiments as the TOKAMAK, the time-step can be increased by a factor of up to 100, while still maintaining deviations of the guiding centers in the order of 10-2 gyro-radii.
Moreover, a divergence-free interpolation method is introduced for cylinder-symmetric magnetic fields defined on triangular meshes, which results in an almost everywhere smooth field. The interpolation is reduced to 1-dimensional interpolations and can include an arbitrary number of base-points.
Electronic Structure of Transition Metal Complexes with Phthalocyanines
Stefan Maintz, RWTH Aachen University
Adviser: Thomas Müller, JSC
Density functional theory (DFT) calculations have become an elemental tool for various natural sciences over the past decades. In this work calculations on 3d-transition metal complexes with Phthalocyanine ligands have been carried out using the B3-LYP hybrid functional, in order to evaluate the performance of DFT applied to open-shell complexes that contain many electronic excited states in the vincinity of the ground state. The results have been compared with predictions and expectations arising from ligand field theory (LFT). It was found that DFT apparently has inherent errors as not even expected trends within the 3d-period were optained.
Analysis of a parallel preconditioner
Hannah Rittich, Bergische Universität Wuppertal
Adviser: Bernhard Steffen, JSC
In this report we will analyse the quality and performance of the library Hypre, a library wich provides highly parallel preconditioners and solvers for large sparse systems of linear equations. First we will give a short introduction into a certain class of iterative methods and preconditioners. Afterwards we will present results from different experiments analysing different parts of Hypre.
Simulation of NMR signal formation
Martin Rückl, Julius-Maximilians-Universität Würzburg
Adviser: Walter Nadler, JSC
Constrictions of heart arteries called stenosis can be detected by analysing the configuration of capillaries in myocardium using nuclear magnetic resonance (NMR). In this work a simulation model for the NMR signal formation was developed to check influence of capillary configuration on transversal relaxation processes. It has been found that faster diffusion strongly reinforces geometric influence.
Theodros Zelleke, Ruhr-Universität Bochum
Adviser: Godehard Sutmann, JSC
A new parallel algorithm for classical Molecular Dynamics is presented. The algorithm is an extension to a common domain decomposition algorithm and is suitable for simulation systems with short range forces and an inhomogenous distribution of the particles and for low density systems. The implementation is presented in the context of the combined Molecular Dynamics/ Multiparticle Collision Dynamics program MP2C.