Passive scalar BC and some elements vanish if NEL > 150000

[Nils Benedikt]

Hi Neks, 

I'm trying to run a simulation with ~211.000 elements. The case runs fine on every mesh that I try as long as the number of elements is lower than 150.000. I'm using Nek5000 v19 with KTH Framework.

I am aware of the following two problems which occur if I use a mesh with more than 150.000 elements:

1. The BC for the passive scalar all seem to be zero now and therefore the PS is zero everywhere, even though it shouldn't be. 

2. The more elements I have, the more of them don't seem to be part of the simulation anymore. The initial conditions for the velocity show weird gaps exactly along element lines, see attached screenshot.

I'm diagnosing all of this using ParaView, it might also be a ParaView problem, but Visit also has its problems with the files so I assume that it is more than a visualisation problem. Specifically, Visit says that it can't open the file because there is no mesh contained in the first snapshot. What makes me suspicious however, is that Nek doesn't throw a single error.

Here is what I assume could be the problem: 150.000 elements is coincidentally also the default maximum for the Nek tools. I changed this to 250.000 in the makefile and makefile.inc (in Nek5000/tools) but the problem is still there. Could it be that the maximum of 150.000 is also set somewhere else in the code where I have to change it? Also the mesh was generated with gmsh, so it might be some problem with gmsh2nek and genmap, but I executed both tools (compiled for 250.000 elements) on two different machines and the results were always the same (faulty). 

What I tried to fix the problem:

1. Multiple attempts of cleaning and recompiling Nek and all Nek tools (especially gmsh2nek and genmap).

2. Compiling without HYPRE 

I'm thankful for any help or suggestions and please let me know if you need any case files or outputs to make sense of the problem. 

Best regards

Nils

 

[X. Zhou]

Hi Nils,

I am a newcomer to Nek5000.

I think we can not figure out what the operating conditions you try to simulation, and what is your numerical settings in Nek5000.

I calculated the three-dimensional channel flow (very simple cubic geometry) with about 450,000 elements at 7-th order GLL element, and Nek5000 works fine. The HYPRE is also utilized in my setting.

Besides, we could find some "big" cases successfully simulated by Nek5000 in Nek5000 Gallery, in which the number of GLL elements is expected to beyond 150,000.

Hence 211,000 GLL elements may also could be handled by Nek5000.

I think could you please upload your logfile of your simulation. Maybe it would be helpful for other experts to check it out.

Yours,

Zhou.

[Nils Benedikt]

Hi Zhou, 

Thank you very much for looking at my problem! I also think that there shouldn't be any problems with this number of elements, that's why I'm so surprised that I can't figure out the problem. 

How did you generate your Mesh for the case with 450.000 elements? Did you also use gmsh+gmsh2nek?

I attached the logfile, please don't hesitate to ask if you need any other case files. 

Best
Nils

[X. Zhou]

Hi Nils,

I directly adopt the "genbox" tool in Nek5000 because the computational domain is a simple cubic box.

Yours,

Zhou.

[Nils Benedikt]

Hi all, 

To everyone who might find this thread in the future and wonder what the problem was: It was just a display problem in ParaView. I started ParaView directly on the server now and connected remotely and then everything was displayed as it should. 

Best

Nils