An Implementation Framework for Solving High-Dimensional PDEs on Massively Parallel Computers
Accurate solution of time-dependent, high-dimensional PDEs requires massive-scale parallel computing. In this paper, we describe an implementation framework for block-decomposed structured grids and discuss techniques for optimization on clusters where the nodes have one or more multicore processors. We use a two-level parallelization scheme with message passing between nodes and multithreading within each node, and argue that this is the best compromise for memory efficiency. We present some examples where the time-dependent Schrödinger equation is solved.
KeywordsHamiltonian Matrix Ghost Cell Multicore Processor Implementation Framework Lanczos Algorithm
Unable to display preview. Download preview PDF.
- 4.Gustafsson, M.: A PDE Solver Framework Optimized for Clusters of Multi-core Processors, Master’s thesis UPTEC F09 004, School of Engineering, Uppsala University (2009)Google Scholar
- 6.Lubich, C.: Integrators for Quantum Dynamics: A Numerical Analyst’s Brief Review, In: Grotendorst, J., Marx, D., Muramatsu, A. (eds.) Quantum Simulations of Complex Many-Body Systems: From Theory to Algorithms, vol 10, pp. 459–466. John von Neumann Institute for Computing, Jülich (2002)Google Scholar