Open IvanYashchuk opened 1 year ago
Not sure what the right answer is here, the problem doesn't see c0
The problem doesn't see c0 and so the gradient wrt c0 should be zero, right? If I'd use finite differences to compute the derivative I would get zero.
I'd argue that the input is illegal here and so you deserve whatever Firedrake-adjoint gives you ;).
It's not clear to me how we would detect this situation and do something different.
I noticed that the gradient wrt DirichletBC might be incorrectly non-zero even when it doesn't have an influence on the result:
Taylor test convergence rate wrt
c0
is 1.0 indicating incorrect implementation.