I am using ArnoldiSolver to compute an eigenvalue problem, it gave eigenvalues around 1e+290. The following error showed up:

“Solve Hessenberg evp with Lapack … error in eigensolver, info = 21”.

I’ve tested several other eigenvalue problems using ArnoldiSolver and they worked well in those case.

I believe the input matrices are not the causes, since “scipy.sparse.linalg.eigs” gave correct eigenvalues (it is just very very slow…).


It’s probably because I’m using mixed elements, where “CouplingType” has to be treated carefully?

Hi Shawn,
Yes, this may be the case. Lapack is somehow trying to invert a singular matrix. Could you provide a small test case? I’ve just tried solving a mixed eigenvalue problem and it worked…


Thank you very much Christopher. I think the ArnoldiSolver is working well. I realized the problem is possibly that I didn’t update ngsolve to the latest version…

I currently have trouble updating ngsolve since I accidentally deleted mac-based-python2.7!