Regarding the use of “pointInterpolation_t” after modifying the mesh by usrdat2()

[DiscoverXV]

Hi, I have a question about using pointInterpolation_t。

After modifying a smooth wall to a rough wall by usrdat2(), I attempted to import a custom point file ('.his') and use the pointInterpolation_t function for point interpolation and output.

However, the results were unsatisfactory; it seems pointInterpolation_t still interpolated onto the original smooth wall. I tried testing pointInterpolation_t with several boundary points (y-values ​​higher than 0) of the rough surface, but pointInterpolation_t still treated y = 0 as "inside", shows in below figure.

I have a question about the mesh used by pointInterpolation_t, specifically where it comes from. Because I've also output nrs->mesh and [x, y, z] = mesh->xyzHost(); and all the points do indeed show rough surface characteristics.

I hope to get your help, thank you.

Cheers,
DiscoverXV

[yslan]

What version of nekrs are you using?
Can you also show your usrdat2?
Where is the test called?

[DiscoverXV]

Hi yslan,

Thank you for your help.

In usrdat2, I called outpost(vx,vy,vz,pr,t,'...') and used paraview to check, and found that the roughness elements were imported into nek, as shown in the image below.

However, in UDF_Setup(), I used auto mesh = nrs->mesh;
auto [x, y, z] = mesh->xyzHost(); and exported x, y, and z. When I tried to read them using Matlab reads this, the result shows smooth wall, which means roughness element do not import into nrs->mesh;

I'm currently using the .f file exported via outpost(vx,vy,vz,pr,t,'...') as the mesh +X, and the .f file of the smooth wall as +U+P, i.e., startFrom = "tbl_8000_10m0.f00101+X,tbl_8000_10m0.f00100+U+P. it can read the boundary points of the rough wall.

I would like to ask if this approach is reasonable?

Thank you again for your help.

Cheers,
DiscoverXV

[yslan]

Restart from multiple files are supported in v24-dev

Not sure what's going on. Changing coordinates in usrdat2 should work. The coordinates should load into the device arrays. We have examples using this:
https://github.com/Nek5000/nekRS/blob/v24-development/examples/periodicHill/periodicHill.usr
mesh->xyzHost() simply copies from device back to the host.

Is your visualization tests using the same numbers of MPI ranks?

[DiscoverXV]

Hi yslan,

I discovered the problem comes from startFrom; if the relevant .f file was loaded i.e., startFrom = "Smooth.f00001", it's mesh would overwrite the mesh modifications in usrdat2.

I'm now using the outpost(vx,vy,vz,pr,t,'...') function you provided to generate Rough.f00001 with a rough surface. Then, combining this with the smooth wall Smooth.f00001, i.e., startFrom = "Rough.f00101+X,Smooth.f00001+U+P, I can import the smooth wall data into the rough wall.

The only question now is whether the generated the new .f file after simulation with startFrom = "Rough.f00101+X,Smooth.f00001+U+P will use a rough mesh or a smooth mesh.

My visualization is relatively simple: I wrote all the GLL points (x,y,z) of all ranks into a file, and then read it by using Matlab. Therefore, the visualization did not use MPI.

Thanks for your help.

Cheers,
DiscoverXV

[yslan]

if the relevant .f file was loaded i.e., startFrom = "Smooth.f00001", it's mesh would overwrite the mesh modifications in usrdat2.

Yes. That's expected. This is typically for simulations that has moving mesh. Therefore, simulation can also reads mesh from the checkpoint files.
It makes sense now.

startFrom = "Rough.f00101+X,Smooth.f00001+U+P
This reads mesh from Rough.f00101, velocity and pressure from Smooth.f00001.

Here are the some examples of multiple files restart:

nekRS/examples/ethier/ci.inc

Line 894 in 36f8977

run("4", "NEK", "ethier0.f00000+X,ethier0.f00001+U+P+S", true);

I had a concern that your reader and writer must use the same #mpi ranks. That's why I suggest using *0.f0* files.

[DiscoverXV]

Hi yslan,

Yes, the number of MPIs used for my output and input GLL points are the same #mpi ranks.

Furthermore, I verified that outputting using startFrom = "Rough.f00001+X,Smooth.f00001+U+P" also make the next output .f file also shows rough surface results.

Thank you so much for your help!

Cheers,
Song