Hi folks,

I'am doing a Cholesky decomposition on an symmetric Jacobi-matrix using Eigen::SimplicialCholesky. This works well, but in some cases the decomposition fails. Therfore I have put the SuperLU solver behind that to solve this error-cases. If SuperLU also fails, I want to use the old values instead. Problem is, that SuperLU.info() returns sucsess, even if the solution consists of 1.#QNB s . So I can't catch this error easily and my programm crashes.

possible solutions:
1) Not use SuperLU and use only the old values if the solving fails. Possible in this case, but I will certainly need a LU-decomposition on an other problem, were it will also be problematic if SuperLU.info() does not work.

2) Use an other sparse LU solver. How stable and major is the build in version in Eigens devel branch?

3) Test for 1.#QNB in the solution afterwards. But like I mentioned, I will need a reliable LU-solver for a related problem.

greetings, medic

I'm not sure we can do much on our side to fix this issue. I suggest you to give a chance at our built-in sparse LU that should offer you many advantages: simpler to use, faster, as robust, and hopefully better error report. And if you get troubles we'll have full control to address them. Please get back to us even to say it works, that's always good to know.