Iterative solver and LU generate different solutions


I recently switched from LU solvers in FEniCS to iterative solvers (CG). The solutions looks similar to the eyeball, however, there is the difference between the solutions has a significant norm! I tried to reduce the tolerances (relative and absolute, also the maximum iterations) in the CG method but the difference does not vanish.

Anyone has an idea where I can look for the possible explanation? This problem is a hyperbolic PDE with a very fine and unstructured mesh.

Thank you

Have you checked that the CG solver actually converges?

Without a reproducible example, it is hard to give any guidance.

1 Like