[petsc-users] Slepc - SVD routines
Luke Bloy
luke.bloy at gmail.com
Wed Jul 7 10:17:42 CDT 2010
Hi,
First off I apologize for the slepc question, they don't have a users
lists so I'm hoping someone on here might be able to offer some guidance.
Problem:
I'm working on a graph partitioning problem. Basically I have the
laplacian of a large graph and am interested in extracting its connected
sub components. An approach is to compute the smallest eigenvalues and
eigenvectors of the laplacian matrix. it is known that the lowest
eigenvalue of the laplacian is 0, and has a multiplicity of K if the
graph has K subcomponents. These subcomponents can be extracted from the
smallest eigenvectors.
Question:
What is the best approach within slepc to attack this problem?
I've been using SVD solvers using the LANCOS method to attack this
problem, but sometimes have convergence problems. I know one of the
smallest eigen vectors is the vector of all ones, and have tried to
initialize the svd solver with this vector but this has led to very
strange problems.
Would I be better off using one of the EPS solvers? If so which can
correctly extract multiplicities of eigenvalues?
Thanks,
Luke
More information about the petsc-users
mailing list