Using Dolfin Adjoint for Custom method to estimate the roots
Dear all,
My apologies for bothering you regarding Dolfin Adjoint issues.
I would like to know if you have some examples related to the "Lower-level interfaces."
I need to solve my problem implementing a particular type o full Newton method to estimate the roots of the problem and during the optimization process, the Dolfin Adjoint replaces my solving method when I call the reduced functional. How can I do a more low-level implementation that Dolfin Adjoint can keep my particular solution method?
Is there any example?
Best Regards,
Comments (5)
-
-
reporter Hello Simon,
I implemented a Generalized Displacement Control Method (GDCM) to solve instability structural problems inside o fenics. For that is not possible to use the Newton method.
I would like to use dolfin-adjoint to perform topology optimization, for that, I have to solver the adjoint problem manually for each GDCM step.
I would like to know some example that I can use dolfin-adjoint libraries to perform it.
Ricardo
-
reporter Here the TAO example that I am performing using GDCM.
https://www.dropbox.com/s/pu5ox7fus4rbrjv/deformacao_viga2.avi?dl=0
The TAO solver is not working well in my examples, I don't know why.
Ricardo
-
- changed status to resolved
Hi Ricardo,
The most useful interface for you is probably to construct a
dolfin_adjoint.ReducedFunctional
ordolfin_adjoint.ReducedFunctionalNumpy
. With that you can evaluate the functional and compute its derivatives at different points in the control space.Check out the API documentation to see if this is what you need: http://www.dolfin-adjoint.org/en/latest/documentation/api.html#dolfin_adjoint.ReducedFunctional
-
reporter Hello Simon,
I started to programming a ReducedFunctional for a GDCM, using libadjoint libraries.
I think I will need more information about the variables, but I appreciate a lot your time to have a look in my issue.
If you have more examples or information, it will be very welcome.
Regards,
- Log in to comment
Hi Ricardo,
I think what you are asking is if you can add a custom solve into the dolfin-adjoint pipeline with your own Newton implementation? Is that correct?
Simon