Use of Passive Scalars with UserDefinedBoundaryConditions

[hwhitehead]

The wiki states that the current passive scalar implementation is incompatable with the use of UserDefinedBoundaryCondition, with "plans to expand this feature soon" (see here).

I just wanted to check to see if this entry is up to date, and if there were any plans to add this feature?

If not, is this something that could be solved in a somewhat hacky way, for instance:

• While the UserDefinedBoundaryCondition function does not have direct access to the r field, it does have access to its relevant meshblock (as pmb) which in turn has access to the PassiveScalars class as pscalars
• The PassiveScalars class DOES have access to r for the primitive scalar properties

Are there any obvious issues with exploiting this pointer chain to access and set the scalar properties with a UserDefinedBoundaryCondition?

[tomidakn]

I am sorry but it is still the case. I wanted to do this is a general way (see #626) but I have been too busy to do it for myself.

Physical boundary functions are a little bit tricky, because they are applied not only on MeshBlock data but also on what we call "coarse buffer". For AMR prolongation/restriction operations, the coarse buffer stores the MeshBlock data but with half resolution. And before the prolongation, the boundary function must be applied on this coarse buffer. The user-defined boundary function is designed so that it can be applied automatically and consistently while users do not have to be aware of it. This is also true for the passive scalar.

So, if you want to access the passive scalars from the user-defined function, you need to identify if it is being applied on MeshBlock data or on a coarse buffer, if you use AMR. This can be done by checking the range of is,ie/js,je/ks,ke given as the parameters of the boundary function. So it is technically possible, but it is tedious and not elegant.

I guess it won't interfere with other parts (e.g. time integrator) but I have not tried it by myself. If you are not using AMR, it is probably not too difficult and I hope it works for you.

[hwhitehead]

Thanks for the speedy reply. I forgot that these BCs are also applied to the buffer data, that definitely complicates things.

I will have a look to see how complicated this is to implement, could you elaborate on how checking the index range would help identify whether the code is acting on the coarse buffer or the MeshBlock data? If my standard MeshBlock size is 16^3, would the coarse buffer have a size 8^3? Unfortunately AMR (or at the very least SMR) is unavoidable for my system.

[tomidakn]

If my standard MeshBlock size is 16^3, would the coarse buffer have a size 8^3?
Correct. Please try it, and feel free to ask me if it does not work.

[hwhitehead]

Is the coarse buffer only used when MeshBlocks are changing AMR levels? When I run with a simple print to check ku-kl, ju-jl and iu-il, I get values ranging from 15-19 (which makes sense for a 16^3 MeshBlock with 2 ghost cells). I don't appear to get any prints that would correspond to the BCs being called on a coarse buffer.

[tomidakn]

I guess there is no refined MeshBlock directly touching the physical boundary.

[hwhitehead]

At the moment, I can ensure my system never refines MeshBlocks adjacent to the boundary, so there is no case where the BCs are applied to the coarse buffer. If in the future this changes, where can I find the primitive values for the coarse buffer?