Francesc Verdugo

Nice to see this! Is this approach composable? In some situations, we need to wait for a precondition (Request 1) in order to start a non-bloking communication that will generate...

I would follow a quite different approach, which I think opens the door to more optimizations. From the user perspective: ```julia a(u,v) = a(u,v) = ∫( ∇(v)⋅∇(u) )*dΩ # Some...

BTW, you can always implement these geters if you want to consider forward and adjoint problems separately (at the cost of loosing the optimizations that take place when they are...

Transposing lazily or assembling the transpose from the bilinear form / jacobian will depend on the linear solver used for the adjoint. If the linear solver allows to use a...

Hi @aerappa! The low-level functionality is clear. Now, we need to think a nice high-level API to express the solve/estimate/mark/refine loop in such a way that we can consider several...

Hi @aerappa as a first step, we can implement a simpler version of the problem and add more complexity later. I think that it can also be a good exercise...

And for the moment, I would refine ALL cells so that you don't have hanging nodes. FESpaces on models with hanging nodes is still not implemented in Gridap, but we...

> Edit: I have created an interface to sort each entry of cell_node_ids anti-clockwise. However, do I need to resort this array based on GIDs before creating a new UnstructuredGrid?...

@amartinhuertas, is the orientation based on global nodes ids needed any more after your refactoring of the Raviart-Thomas reference element? If not, we can use a more standard orientation for...

Perhaps we can discuss this face 2 face since I don't understand the problem completely. In any case, it sounds like we need to solve it at the `SkeletonPair` level...