Quantum Dynamics Calculations on Massively Parallel Supercomputers: Implementation and Molecular Applications

Wenwu Chen (NCNR)

The eigenvalue/eigenvector and linear solver problems arising in chemical physics and many other fields (e.g. structured grid simulation of partial differential equations in multiple dimensions) often involve large sparse matrices that exhibit a certain block structure. In such cases, specialized iterative methods that employ preconditioners derived from a block Jacobi diagonalization procedure have been found to be very efficient, vis-a-vis reducing the required CPU effort on serial computing platforms. A parallel implementation was presented, based on a non-standard domain decomposition scheme. Excellent parallel scalability was observed for both the specialized block Jacobi and the fundamental matrix--vector product operations up to hundreds of nodes with generalization for arbitrary number of nodes and data sizes, while the isoefficiency analysis suggests fruitful extrapolation to clusters with tens of thousands of nodes. This parallel implementation renders the resultant parallel codes suitable for robust application to a wide range of real problems, such as in the quantum dynamics calculations of scattering cross sections, Green's functions, cumulative reaction probabilities, and thermal rate constants, running on massively parallel computing architectures.

Back to Seminar Home Page

Last modified 12-July-2007 by website owner: NCNR (attn: )