Personal tools

Project B6

From MACwiki

Revision as of 14:59, 8 March 2017 by Bader (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search


Efficient Parallel Strategies in Computational Modelling of Materials

Current implementations of first principles electronic structure methods for molecules, clusters, and local models of surfaces and solids are limited in their parallel scalability due to the intrinsic diversity of structures and algorithms involved in various subtasks. The inherent structure of the electronic self-consistent field (SCF) problem does not admit a homogeneous parallelisation strategy, which severely limits current opportunities in modelling of complex chemical systems like catalysts, nano structured materials as well as large complexes in solution.

Gold Atom


The project will develop a new paradigm for the parallelisation of density functional theory (DFT) methods for electronic structure calculations and implement this new strategy. Advanced embedding techniques will account for environment effects (e.g. solvent, support) on a system. We propose a strong modularisation of the DFT approach, facilitating task specific parallelisation, memory management, and low-level optimisation. Efficiency will be further increased by dynamical adaptation of varying resource usage at module level and pooling of applications.

For software layout and low-level numerical methods the project relies on the well established software ParaGauss developed by the coordinator. Fundamental restructuring of this software facilitates also extension of features, e.g. development of a direct Gaussian four-centre analytic “integral” module for calculating “exact exchange” energy contributions on the fly during the SCF procedure and evaluation of forces. These computationally very demanding modules are key for developing and verifying the novel strategy. Pertinent demonstration examples derive from application projects of the coordinator, e.g. transition metal nano particles, their reactivity and sorption on support, heterogeneous and homogeneous catalysis, and large metal complexes in aqueous medium. Finally the new quality of simulation performance will be showcased with “real-life” applications from these areas.

Research Focus

The project relies on extensive experience in developing quantum chemistry methods (Rösch). This expertise will be combined with mathematical and computer science expertise of the partners. First, the SCF procedure of the Kohn-Sham-DFT approach will be adapted to the new strategy. Next follow modules for pre-calculating “integrals” and direct four-centre “integral” evaluation, the latter exploiting localisation and screening of interactions. Finally, additional features will be adapted, e.g. solvation treatment, embedding of surface cluster models, and strategies for parallel geometry optimisation. The partners will closely collaborate at the design stage, for algorithm development (Rösch, Bungartz), and concerning hardware aspects and parallelisation (Gerndt, Bode). LRZ brings in optimisation and performance analysis on HPC platforms such as HLRB 2 (Hegering).

Research Highlights

Scheduler for blocked parallel eigenvalue problems (Roderus). We employed a scheduling algorithm for efficiently solving of the eigenvalue problem for block-diagonal Hamilton matrices characteristic for problems exhibiting symmetry.


Dynamic Load Balancing (DLB) (Nikodem). We designed a generic task scheduling framework for use in several parallel algorithms that naturally require dynamic load balancing (DLB). We reformulated the real space grid integration in SCF using this newly designed DLB framework.

Electron repulsion integral (ERI) package and direct SCF (Soini). A challenging aspect of any QC calculation is the evaluation of the so-called four-center electron repulsion integrals, ERIs, which is of the formal complexity of O(N4).

ParaTools: Python framework for parallel exploration of potential energy surfaces (Chaffey-Millar). To explore high-dimensional potential energy surfaces of molecular systems, we implemented a Python framework for path searching algorithms and parallel scheduling of computations of energy/forces at distinct molecular geometries along the path.

Energy Surface


Pricipal Investigators:

Prof. Dr. Dr. h.c. Notker Rösch (coordination) Theoretical Chemistry
Prof. Dr. Arndt Bode Computer Organisation; Parallel Computer Architecture
Prof. Dr. Hans-Joachim Bungartz Scientific Computing in Computer Science
Prof. Dr. Michael Gerndt Computer Organisation; Parallel Computer Architecture
Prof. Dr. Heinz-Gerd Hegering Supercomputing

Project Members:

Dr. Martin Roderus
Dr. Anca Berariu
Thomas Soini
David Tittle


Dr. Astrid Nikodem
Dr. Hugh Chaffey-Millar

Project Team Leader

Dr. Alexei Matveev


Dr. Sven Krüger



Poster (PDF) - Anca Berariu, Hugh Chaffey-Millar, Alexei Matveev, Astrid Nikodem, Martin Roderus,Thomas Soini, David Tittle, Arndt Bode, Hans-Joachim Bungartz, Michael Gerndt, Sven Krüger, Notker Rösch: MAC Workshop Project poster, 2010


Poster (PDF) - Anca Berariu, Hugh Chaffey-Millar, Alexei Matveev, Astrid Nikodem, Martin Roderus,Thomas Soini, David Tittle, Arndt Bode, Hans-Joachim Bungartz, Michael Gerndt, Sven Krüger, Notker Rösch, *Efficient Parallel Strategies in Computational Modeling of Materials*, SimTech 2011