Parallel run shows partial lid velocity in 3D cavity flow

[春风复苏]

Hello Nek5000 community,

I'm experiencing an issue with parallel execution of a 3D lid-driven cavity case and would appreciate your insights.

**Problem:**
- Serial run (`nekbmpi ldc 1`): Correct results - full lid moves as expected
- Parallel run (`nekbmpi ldc 4`): Only partial lid velocity observed, rest of domain shows zero velocity or just end caculating just 5 step with zero velocity.
- No error messages in either case

**Setup:**
- Mesh: 984 elements, 216 GLL points/element
- Generated via: CGNS → cgns2nek → genmap → makenek
- SIZE parameters:

parameter (lelg=1100)
parameter (lpmin=1)
parameter (lelt=lelg/lpmin + 3)

**What I've checked:**
- Boundary conditions set correctly in usrdat2 using bc(5,...)
- Single processor works perfectly
- No compilation or runtime errors

The issue seems related to parallel decomposition - boundary conditions aren't being fully applied across all partitions.

Has anyone encountered similar parallel boundary condition issues? Any suggestions for debugging or SIZE parameter adjustments?

Thank you for your help!

[春风复苏]

I initially used id_face = nint(bc(5,iside,ieg,1)) to read boundary IDs (a method I came across at https://github.com/Nek5000/Nek5000/tree/master/tools/cgns2nek but unfortunately didn't fully understand its applicable scope). This approach fails to account for boundary mapping after mesh partitioning. The correct approach should be using id_face = BoundaryID(iside,e). I'd like to particularly emphasize this critical detail for parallel computing to everyone's attention.

[YuHsiang Lan]

Hi,

I'm glad you figured out and have a working version. However, I'm still curious what causes the issue.

Getting boundary ID from bc should also work because that's how Nek5000 initialize the variable "boundaryID".

  https://github.com/Nek5000/Nek5000/blob/ccc0c521bc70337a80dee283eff9a8bfe37ffb99/core/connect2.f#L115-L119

> id_face = nint(bc(5,iside,ieg,1))

The face that you are using ieg is a bit alarming. If you really go through ieg=1,nelgt, it then indeed only works for single processor.

If you still have your code, we can figure it out. Here is a short table to distinguish those two: 

  https://nek5000.github.io/NekDoc/problem_setup/usr_file.html?highlight=nelgt#id5

Thanks,

Yu-Hsiang

--