Hi John,

Thanks very much for your comments and suggestions!

In the Eigen documentation they have a Sparse LU decomposition with Umfpack and 
Super LU support, but I have not checked the implementation.

Unfortunately my matrix has a dimension of 1M \times 1M…

Thanks,
Kathrin

Am 08.12.2016 um 12:01 schrieb John Peterson 
<[email protected]<mailto:[email protected]>>:



On Thu, Dec 8, 2016 at 10:44 AM, Kathrin Smetana 
<[email protected]<mailto:[email protected]>> wrote:
Dear libmesh users/developers,

I have to solve a linear system of equations (system size approximately 10^6) 
very often (about 5000 times). The linear system of equations is the result of 
a FEM discretization of 3D linear elasticity.

I thought about using a sparse Cholesky decomposition as the matrix is 
symmetric or a sparse LU decomposition, depending on availability. I have had a 
look at the Eigen package and their direct LU factorization for instance with 
Umfpack or SuperLU support looks very promising to me.

Do you think that is a good option or do you have any other recommendations for 
me?

I don't think the Eigen direct LU implementation is sparse?  Is your matrix 1M 
x 1M or 1000 x 1000? The former is probably prohibitively large for *any* 
direct solver...

I second what Roy said, but will also add that superlu_dist is a good option.  
You can access the latter by building PETSc with --download-superlu_dist=1 and 
running with:

-pc_type lu -pc_factor_mat_solver_package superlu_dist

--
John

------------------------------------------------------------------------------
Developer Access Program for Intel Xeon Phi Processors
Access to Intel Xeon Phi processor-based developer platforms.
With one year of Intel Parallel Studio XE.
Training and support from Colfax.
Order your platform today.http://sdm.link/xeonphi
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to