Element-based topology optimization

[mohamed82008]

gridap/Gridap.jl#633

[stevengj]

See this tutorial: https://gridap.github.io/Tutorials/dev/pages/t018_TopOptEMFocus/#Tutorial-18:-Topology-optimization-1

[mohamed82008]

Thanks Steven. I will take a look at how you worked around the non-differentiability of Gridap. I suppose you have to define one adjoint for each physics. I am hoping that one day Enzyme will be able to diff through the entire code base without needing custom adjoints EnzymeAD/Enzyme.jl#447.

[stevengj]

In my experience any sufficiently complicated problem will require custom adjoints, and the fact that there is currently no good way to do this with Enzyme (EnzymeAD/Enzyme.jl#172) makes it unlikely for me to use Enzyme in any real problem.

(That being said, in this example we probably did more manual differentiation than we needed to. @fverdugo found that he could use ForwardDiff to differentiate the weak form, for example.)

[mohamed82008]

Fair enough. I need to study your example.