Clarification on fill priority in FillPatchTwoLevels with internal boundary conditions

[Afraid31]

I have a question regarding the exact logic used by amrex::FillPatchTwoLevels when all boundary conditions are set to BCType::int_dir.

Here is the usage context:

amrex::FillPatchTwoLevels(mf, time,
                          cmf, ctime, fmf, ftime,
                          0, icomp, ncomp,
                          geom[lev-1], geom[lev],
                          cphysbc, 0, fphysbc, 0,
                          refRatio(lev-1),
                          mapper, bcs, 0);

My current understanding of the fill priority is as follows:

1.For the valid region of mf:

• First, it is fully interpolated from the valid region of cmf using the provided mapper.

• Then, for regions where mf and fmf overlap, the valid data from fmf is used to overwrite those corresponding regions in mf.

2. For the ghost region of mf:

• Ghost cells are filled using interpolation from valid region of cmf.

• Finally, mf.FillBoundary(geom[lev].periodicity()) is called to fill periodic ghost regions.

Is this understanding correct?

Especially, I want to confirm:

1. Are ghost cells from cmf or fmf ever used during the interpolation to fill either the valid or ghost regions of mf?

2. Whether FillBoundary is automatically called inside FillPatchTwoLevels, or needs to be done manually after the call.

Thank you very much!

[WeiqunZhang]

This is what the doxygen documentation says

 FillPatch with data from the current level and the level below.

 First, we fill the destination MultiFab/FabArray with the current
 level data as much as possible. This may include interpolation in
 time. For the rest of the destination MF, we fill them with the 
 coarse level data using interpolation in space (and in time if
 needed).

Hope that answers part of the questions.

Are ghost cells from cmf or fmf ever used during the interpolation to fill either the valid or ghost regions of mf?

No,

Whether FillBoundary is automatically called inside FillPatchTwoLevels, or needs to be done manually after the call.

No and no. (FillBoundary is not the only way to fill boundaries.)

[WeiqunZhang]

I noticed that you are using amrex::FillPatchTwoLevels(mf, time,..., which does not fill the ghost cells mf. If you need to fill the ghost cells of mf, you need to call another version FillPatchTwoLevels (MF& mf, IntVect const& nghost, Real time, ....

Here are the links to the doxygen documents.

https://amrex-codes.github.io/amrex/doxygen/namespaceamrex.html#a3f8f5f13146586dfaa0fb8d082132f12

https://amrex-codes.github.io/amrex/doxygen/namespaceamrex.html#a5b5440802e0895fcba37e6b295116585

[WeiqunZhang]

Note that calling amrex::FillPatchTwoLevels(mf, time, ...) followed by mf.FillBoundary(...) is NOT equivalent to amrex::FillPatchTwoLevels(mf, nghost, time, ...) even when the boundaries all periodic, because fine ghost cells outside the coarse/fine boundary are not always at the periodic domain boundary.

[Afraid31]

Thank you very much for the detailed explanation! Your clarification is very helpful.

[Afraid31]

Thanks again for the previous explanation! I’ve read through the documentation provided on the official AMReX website. However, I’ve encountered something that still puzzles me.
I originally thought that the following call:

amrex::FillPatchTwoLevels(mf, time, 
                          cmf, ctime, fmf, ftime,
                          0, icomp, ncomp,
                          geom[lev-1], geom[lev],
                          cphysbc, 0, fphysbc, 0,
                          refRatio(lev-1),
                          mapper, bcs, 0);

would only fill the valid cells in mf. However, I noticed that this call actually does modify the ghost cells of mf as well.
So I dug into the source code and found that this version of FillPatchTwoLevels internally calls:
detail::FillPatchTwoLevels_doit(mf, mf.nGrowVect(), ...);
This clearly indicates that the function uses the ghost cell width from mf.nGrowVect() even though it wasn't specified explicitly in the argument list.
I’m confused — what’s going on here?
Why does the FillPatchTwoLevels version without an explicit nghost argument still operate on ghost cells?

Any clarification on this behavior would be greatly appreciated. Thank you again!

[WeiqunZhang]

Yes, you are right. The documentation is incorrect. We need to fix it.