INS AD/nonAD version

[mcacace]

Dear all,
I'm trying to "upgrade" a NS INS example to its AD version (see whether this could speed up the solver). I (think I) put almost all things together, though I am stuck with a BC. The problem is that I need to impose a MatchValueBC along a specific boundary (whose values I take from an auxVariable - lagrange), and I'm not sure how to make it working while having the velocity as a lagrange vector variable. I tried to have a lagrange auxvariable (with 3 components) and make use of the BuildArrayVariable, but it is not compatible with lagrange_vec type. This said, before looking at a solution on my own, I though it wise to ask the forum...
I am attaching my two input files (sorry if this violates the posting rules), for anyone who could help.
Thanks a lot
fracture_flow.zip

[GiudGiud]

Hello

So you need a VectorMatchedValueBC?
Or do you need a way to create a vector variable, to then use it in a VectorMatchedValueBC?

Guillaume

[mcacace]

In principle a way to create a VectorVariable (same as per the velocity - Lagrange_vec) to use in a VectorMatchedValueBC

[GiudGiud]

Does the ParsedVectorAux let you do that ?

Or is it restricted to non nodal variables as well

[mcacace]

I can give it a try (guess I can move the auxvariables to nodes). Will report back tomorrow once in the office.
Thanks again

[GiudGiud]

sounds good. Thanks for engaging on this

BuildArrayVariable is for array variables but maybe we need to add a BuildVectorVariable to make it clear what we're doing with vector variables

[mcacace]

@GiudGiud your hints did work indeed! Thanks a lot.

[mcacace]

Of course, the vector variable should be populated from existing lagrangian variables.
Thanks

[lindsayad]

I'm trying to "upgrade" a NS INS example to its AD version (see whether this could speed up the solver)

If you're referring to the INS vs INSAD prefixed objects in the navier_stokes module you're unlikely to see better solver performance with INSAD unless you're running on displacing meshes. We worked pretty hard to get the Jacobians correct for the INS objects

[mcacace]

I see, indeed we are running on a displaced mesh. And, just tested it, did not provide any speed up wrt the INS formulation, so in the end I will stay with the non AD version. Thanks anyway for the support.

[lindsayad]

Are you using PJFNK or NEWTON?