Visualization of non-uniform geometry

[YqWangcfd]

Hi. I have a specific visualization requirement for AMReX that involves non-uniform cell size distribution. Within each AMReX box (where the number of cells is guaranteed to be a multiple of 5), I want to group cells into sets of 5 and apply a fixed non-uniform size pattern to each group. For instance, if the standard cell size is dx, each group would follow a pattern like [dx/4, dx/2, 3*dx/2, dx/2, dx/4]. This pattern would repeat consistently across all groups within the box.

It's important to note that this non-uniform grid arrangement is purely for visualization purposes. The actual numerical computations would continue to use logically uniform indices, maintaining the standard AMReX computational framework.

However, I've observed that AMReX typically supports only uniform cell sizes within a box. I'm wondering if there's a built-in solution or workaround in AMReX to accommodate such non-uniform cell size distributions while preserving the logical uniformity for computations?

Thanks.

[WeiqunZhang]

I wonder if you can interpolate the data to uniform mesh for visualization purposes.

[YqWangcfd]

Thanks. This sounds like a valuable approach. Do we have other approaches like modifying the geometryData? If so, is it trivial?

[YqWangcfd]

Sorry, I did not express my problem clearly before. I mean can we modifying the IO functions to add features for non-uniform cell sizes within a box?
I noticed that in the WriteGenericPlotfileHeader function, the current implementation assumes uniform cell sizes within each box. It would be possible to modify the IO functions to support non-uniform cell sizes within a box?

[WeiqunZhang]

Maybe what you are looking for is something like https://github.com/erf-model/ERF/blob/406bf830013e065e7ffe58275462a2d5ebd4c63d/Source/IO/ERF_Plotfile.cpp#L1638. In that ERF function, I think, mf_nd stores the physical locations of nodes. You also need to store some information in the plotfile header. https://github.com/erf-model/ERF/blob/406bf830013e065e7ffe58275462a2d5ebd4c63d/Source/IO/ERF_Plotfile.cpp#L1725. Then you can use Paraview for visualization. https://erf.readthedocs.io/en/latest/Visualization.html#paraview

@asalmgren can tell you more. If you need more help, you could also ask for help on ERF's github.

[YqWangcfd]

Thanks a lot for your help! I think it could temporarily resolve my problem. I'll come back if I need any further clarification!

[YqWangcfd]

Hi, Weiqun

I apologize for bothering you again. I've been trying to implement nodal data visualization using the ERF function by replacing the original WriteMultiLevelPlotfile() in AMReX with the following function:

void WriteMultiLevelPlotfile_SD(const std::string& plotfilename, int nlevels,
const Vector<const MultiFab*>& mf,
const Vectorstd::string& varnames,
const Vector& geom, Real time,
const Vector& level_steps,
const Vector& ref_ratio,
const std::string &versionName,
const std::string &levelPrefix,
const std::string &mfPrefix,
const Vectorstd::string& extra_dirs)
{
BL_PROFILE("WriteMultiLevelPlotfile_SD()");

int finest_level = nlevels - 1;

// Vector of MultiFabs for nodal data
Vector<MultiFab> mf_nd(finest_level+1);
for (int lev = 0; lev <= finest_level; ++lev) {
    BoxArray nodal_grids = mf[lev]->boxArray(); 
    nodal_grids.surroundingNodes();
    mf_nd[lev].define(nodal_grids, mf[lev]->DistributionMap(), 3, 0);
    mf_nd[lev].setVal(0.);
}

WriteMultiLevelPlotfileWithTerrain(plotfilename, finest_level+1,
                                    GetVecOfConstPtrs(mf),
                                    GetVecOfConstPtrs(mf_nd),
                                    varnames,
                                    geom, time, level_steps, ref_ratio);

}

I expected to see the nodal data in ParaView (even with zeros everywhere), but it's not showing up, though the cell data displays correctly. Could you help me understand if I'm missing something obvious here?

Thanks!

[YqWangcfd]

Sorry, It's just the Paraview version problem. I've updated it and I can see the terrain data now.