Ignore structure presence for AMR

[Shashank Acharya]

Hello all

I'm trying to run a few cases where I control the which cells are tagged for refinement.

Essentially, there is a tube that has a contraction wave that reduces its diameter. I want to tag cells around the contraction so that the flow in the reduced diamter is resolved properly.

Controlling which cells to tag has already been done here: https://github.com/IBAMR/IBAMR/blob/master/examples/multiphase_flow/ex5/TagLSRefinementCells.cpp#L53

The issue is that there is an IB tagging strategy which tags all the cells inside/near the lagrangian mesh for refinement. How do I disable this?
The only fluid cells I want refined are the ones near the contraction wave. And I know its location in the Eulerian mesh for all time so I can set the tag flag for these cells myself.

Thanks!

[Boyce Griffith]

The structure needs contribute to tagging, because the AMR grid needs to be constructed in a way that ensures that the structure is covered by Cartesian cells from the appropriate grid level.

Do you want to supplement the tagging? If the region where you want to add tagging is static, then I think that you can specify both GRADIENT_DETECTOR and also REFINE_BOXES as the tagging strategy in the input file, and then specify in the input file where you want to tag.

Do you want to do something different from the above?

-- Boyce

Thanks!

--
You received this message because you are subscribed to the Google Groups "IBAMR Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ibamr-users...@googlegroups.com.
To post to this group, send email to ibamr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Shashank Acharya]

Hi I'm so sorry I wasn't clear.

I don't want the entire structure to be on the finest level because that takes too long to solve (especially because the tube is quite long)

If I have 3 levels, I wish for the structure to be on the coarsest level and the finest level must be around the contraction which is just a fraction of the tube's length.

I have a movie attached maybe it helps me explain better: https://youtu.be/aCjEIMgW-yQ

So in this case, only the fluid cells around the elements highlighted in red must be tagged for refinement (I know their location)

If the structure contributes to tagging, too much of the domain will be refined and the problem becomes computationally heavy.

Thanks again!

Just an FYI in the video I attached, the narrowest part of the tube during contraction is spanned only by 5 cells. I wish that number to be at least around 10 or 15.

[Boyce Griffith]

If you are using IBFEMethod, then currently the structure has to go on the finest level --- currently there is no mechanism for assigning meshes to different levels. I guess that if every part has its own FEDataManager (which I think it does), then we could set it up so that structures could be assigned to different levels. If you want to set up a PR that does this, I am happy to take a look, but there could be some difficulties related to Lagrangian-Eulerian interaction on coarse regions that are covered by finer levels.

-- Boyce

[Shashank Acharya]

Oh I see. Thanks for clearing that up. I'll try to dive in to see how I might be able to solve this issue with the different levels.

[Amneet Bhalla]

Having structures on levels other than finest level is not a good strategy. The LE interaction is going to be poor at best and unstable for most of the runs.   

--Amneet 

[Amneet Bhalla]

Looking at your simulation, it seems like you get away with having just a static hierarchy. You can disable regridding all together to speed-up the run time. Current IBhierarchyIntegrator is designed to have regridding on either via regard_interval and regrid_cfl_estimate. Although you can let the former to zero in the input file to disable regridding, you cannot avoid regrid_cfl_estimate. But you can always override IBHierarchyIntegrator::atRegridPointSpecialized() to return false at all times, perhaps except at  initial time. 

--

--Amneet 

[Shashank Acharya]

Thanks I did not realize it would be such an issue.

I'll try your regrid disabling suggestion to speed up run time with a very fine mesh.