Return to Colloquia & Seminar listing
An algebraic multi-level Krylov substructuring method for large-scale eigenvalue computation.
Student-Run Research| Speaker: | Ben-Shan Liao, UC Davis |
| Location: | 2112 MSB |
| Start time: | Wed, Jun 7 2006, 12:10PM |
Description
Algebraic Multi-level substructuring (AMLS) techniques have
demonstrated great success in computing eigen-analysis of extremely
large scale matrices arising from structural dynamics applications,
such as the vibration of car bodies. In these applications,
it is typical that a large number of eigenvalues with relatively
low accuracy are required. However, such a lack of accuracy
is a cause for concern in some applications.
In this talk, we present an algebraic multi-level Krylov
substructuring (AMLKS) method, which preserves the mechanism
of AMLS techniques, such as efficiency and parallelism,
and meanwhile significantly improves the accuracy of computed eigenvalues.
The gist of AMLKS method is to replace the eigenmodes of
interior substructures by proper Krylov modes of the substructures,
which take the coupling among substructures into
the account. The accuracy improvement of the AMLKS method are
demonstrated by numerical results from electromagnetic and MEMS simulations.
This is a joint work with Zhaojun Bai.