Dear Poulios and Yves,
thank you very much for your kind and prompt replies.
Little by little I'm getting familiar with the logics of this FEM library that I found very interesting and I'd like to go in depth more in the near future, learning also the C++ features, but it will take a good amount of time that in this moment I don't have, do for now I will continue mastering the Python interface.
I believe that you should emphasize in the manuals and push the usage of the GWFL in place of all the other model bricks, since for the beginners it is not obvious and indeed it makes life easier using "add_linear_term" and "add_nonlinear_term" for everything. Thank you for this enlightening clarification.
I've some additional questions that rose up today:
A) If I initialize a theta-method with theta=0 (explicit forward euler) I get the following error
RuntimeError: (Getfem::InterfaceError) -- Error in getfem_models.cc, line 1281 :
Invalid value of theta parameter for the theta-method
but the manual doesn't exclude this option (even though the manual is not complete at the bottom of page The model tools for the integration of transient problems — GetFEM)
B) Let's imagine that our model is perfectly linear and I assemble it with "model.assembly()". What happens if I run "model.solve()" ? Are the equations assembled again or the software somehow understand that the assembly has already been done and skips to the proper solver?
C) Linked to the previous question. If I need to change only the RHS by "model.assembly(option='build_rhs')" and after that I run "model.solve()", does this method assembly again the whole model including the tangent matrix or on the contrary it is smart and skips to the solver?
I ask these things because I'm trying to generate a 3D thermal-electric fully coupled model that I found quite slow in solving the equations since the model is strongly non-linear everywhere. To date, it is still not fully featured.
In the near future I'll probably make some sensitivity analysis so as to try to linearize the model where possible, and for this reason I'm trying to understand possible performance improvements using the built-in standard solver.
Thank you in advance for your kind feedback.
...and compliments for the nice FEM library you're developing.
HAve a nice week-end