I'm trying use Eigen with SuiteSparse, and I am getting incorrect solutions for the Ax = b problem using Eigen's SPQR (SuiteSparseQRSupport) routines. I'm using Eigen 3.2.3 and Suitesparse 4.2.1. I compiled SuiteSparse in Visual Studio 2010 using the suitesparse-metis-for-windows project.

I can get the correct solution from Eigen's built-in SparseQR, but SparseQR is too slow for the large sparse matrices (30000 * 30000, 100,000 non zeros) that I have to deal with. I can also get the correct solution by using SuiteSparseQR directly (by using the CHOLMOD data structures and SuiteSparseQR's solve routine), but I would like to use Eigen for its simpler programming interface.

Any help would be appreciated. Thanks

At this point, the best is probably that you look at Eigen's SPQR wrapper code to see if the calls to SPQR are the same as in your hand coded version. If you don't see anything obvious, then export your matrix (see http://eigen.tuxfamily.org/dox-devel/gr ... tml#title3) and send us a zip.

I can export the matrices to matrix market format. Where do I send them to?

Fixed. The default threshold value was too aggressive.

Devel:
https://bitbucket.org/eigen/eigen/commits/83294dd8b458/

3.2:
https://bitbucket.org/eigen/eigen/commits/5502387f4926/

Thank you for your help!