Dear FEniCS Users,
My problem is somehow related to the previous question: Preconditioner for indefinite linear system
I would like to solve some Eddy Current problems using the Curl-Curl Formulation with edge elements. Starting from the undocumented curl-curl demo CG solver combinded with the Hypre AMS preconditioner works very well for time-domain problems.
For frequency domain problems I splitted the complex equations into real and imaginary part (since I am not using complex-number support yet). Setup a MixedFunctionSpace and solving with MUMPS also seems to work. But how can I get the AMS Preconditioner working?
I tried to setup a PETSc field-split preconditioner, using defining the AMS preconditioner for each diagonal block, but the performance is very bad.
Is there a “default” procedure to get a working preconditioner for a complex problem, if one has a proper method for the real-valued case?
Would the new complex number support solve my problem? Could i the simply use the complex Hypre AMS method for one complex Functionspace?
thanks for any advice
best wishes
Florian