Solving large scale eigenvalue problems by HPC: domain decomposition in Jacobi-Davidson
Staff - Faculty of Informatics
Start date: 1 April 2014
End date: 2 April 2014
The Faculty of Informatics is pleased to announce a seminar given by Menno Genseberger
DATE: Tuesday, April 1 2014
PLACE: USI Lugano Campus, room SI-015, Informatics building (Via G. Buffi 13)
TIME: 10.00
ABSTRACT:
Most computational work in Jacobi-Davidson, an iterative method for large scale eigenvalue problems, is due to a so-called correction equation. For this, a strategy was developed based on a domain decomposition preconditioning technique to reduce wall clock time and local memory requirements.
Furthermore, Jacobi-Davidson consists of two nested iterative solvers: one for the correction equation (the "innerloop"), the other for the eigenvalue problem (the "outerloop"). One may take advantage of the nesting by applying the same domain decomposition technique to the outerloop.
In the talk, the impact of the strategy on the parallel performance will be shown by results of scaling experiments on supercomputers. This is of interest for large scale eigenvalue problems that need high performance computing.
HOST: Prof. Rolf Krause and Dr. Daniel Ruprecht
