Guidance on MultiApps and Transfers

[robfairh]

Hi folks, I am trying to solve a problem that uses two input files and two different meshes. The first input file solves the temperature in a cylinder. The variable changes in 'z' but not in 'r'. I record the temperature along 'z' in a .csv file using a vector-postprocessor. The second file uses the .csv file data to define the values of a BC on a line. For that, I am using 'FunctionDirichletBC' and defining the function values reading the .csv file.

I am trying to define this with the MultiApps and the Transfer actions to run only one simulation instead of two. I haven't used these actions before. Is there any easy way to implement this?

[aeslaughter]

I am not sure we have a transfer that exists for this case. I think what you need is a VectorPostprocessor transfer that transfers the entire vector from one system to the other. A vector version of this: MultiAppPostprocessorTransfer.

The best approach might be to add this capability to the existing MultiAppVectorPostprocessorTransfer, I will take a look and see how that might work. It might be easy to add.

[aeslaughter]

I am going to try to add this capability quickly, see #16055

[robfairh]

Great! Thank you

[aeslaughter]

As you noticed in the issue we are going to build something a bit more generic that should work for this, but it will take some time before it is available. Will the answer below work for you?

[robfairh]

I'll try it and see if it works!

[vincentlaboure]

I'd think the most direct way to accomplish this is using a LayeredAverage which you can transfer with a MultiAppUserObjectTransfer. Then you might have to create a VariableDirichletBC but that should be pretty straightforward.

[robfairh]

Maybe I am doing something else wrong.
In my sub file, I define the UserObjects action like this:

[UserObjects]
  [./sub_app_temp]
    type = LayeredAverage
    direction = y
    variable = temp
    num_layers = 100
    execute_on = linear
    block = 'cool'
  [../]
[]

And in my master file, I define the MultiApps and the Transfers actions like this:

[MultiApps]
  [./sub]
    type = FullSolveMultiApp
    input_files = simple.i
    no_backup_and_restore = true
  [../]
[]

[Transfers]
  [layered_transfer_from_sub_app]
    type = MultiAppUserObjectTransfer
    direction = from_multiapp
    user_object = sub_app_temp
    variable = sub_app_var
    multi_app = sub
    # displaced_source_mesh = true
    # skip_bounding_box_check = true
  []
[]

which defines an AuxVariable like this:

[AuxVariables]
  [./sub_app_var]
    order = CONSTANT
    family = MONOMIAL
    block = 'cool'
    initial_condition = 1
  [../]
[]

The transfer of the LayeredAverage object is working, and it modifies the AuxVariable. But defining the BC with the AuxVariable I haven't figured that out. Am I missing something obvious here?

[vincentlaboure]

Can you try to model your VariableDirichletBC on PenaltyDirichletBC to impose your boundary condition weakly instead of strongly? The problem with a NodalBC (such as DirichletBC) is that it does not know what to do when the variable you impose is discontinuous, as in your case. On the other hand, an IntegratedBC should work fine.

[robfairh]

I'll try that. Thank you!

[robfairh]

It worked! Thank you so much @vincentlaboure and @aeslaughter for your time. @aeslaughter I will still be willing to try that new capability under development.

I had to play a little bit around with the value of penalty to get it to work. Here is a figure with my result:

The transfer sets the values of the AuxVariable sub_app_var and then, AuxDirichletBC imposes those values to the variable temp.

Here are the .h and the .C just in case someone in the future wants to do something similar.

[aeslaughter]

The new transfer capability was just finished off a few days ago: https://mooseframework.inl.gov/source/transfers/MultiAppReporterTransfer.html