Hello,

First of all let me tell you this: support for MKL/Pardiso in eigen is for me one of the top feature of Eigen. This has significantly improved the computation time of my problem.

I facing a minor problem though:

As for now I just give the A matrix to the solver, then call .solve() on the solver multiple time, and even though my A matrix was singular, the whole process just goes forward, and sure enough just using the result of .solve() when A is singular is wrong.


In my situation, what would the fastest way (performance wise) to check if the A matrix is singular (I would then just throw an exception)?

S

As far as I know, given a matrix, there is no cheap way to detect the singularity of the matrix without computing the decomposition.
Besides, the Pardiso solver will detect it during the factorization.
You can call first .compute() and before calling .solve(), you check the value of .info(). It should be equal to Numericalissue if the matrix is singular.
Hope it helps!!

Dee33 thanks a lot for the answer. I am having trouble finding which compute() method you are talking about.

As far as the PardisoLU class, there is no compute method for it:

Code: Select all

162:42: error: no matching function for call to ‘Eigen::PardisoLU<Eigen::SparseMatrix<double, 0, int> >::compute()’

Where should I look for it?

S

Yes this is curious. It should have one. All decomposition classes should have a .compute(matrix)
Try analyzePattern() and factorize() http://eigen.tuxfamily.org/dox-devel/cl ... isoLU.html