Vanishing Jacobian?

[matthew....@gmail.com]

Hello,

I created a mesh of a pipe with objects inside for running a fluid simulation in Nek5000 using Cubit. The mesh has HEX20 elements that all have a positive Jacobian value and I resized the mesh for a non-dimensionalized case. However, whenever I run it in Nek5000, I'm getting "vanishing Jacobian" errors and the simulation stops. This error usually means there are low-quality elements in the mesh. However, I don't see any elements with very low shape quality or negative Jacobians that would cause such an error.

Is there any obvious reason why the Nek log might be crashing with "Vanishing Jacobian" errors? There is a screenshot of my geometry from Cubit using a slice down the center so you can see the internal flow geometry. The fluid enters from the right and moves toward the left. The solid objects in the flow are actually negative space, so the flow moves around the various cylinders within that pipe.

Could it be that the negative space within the flow may be the cause of the runtime errors?

Thanks,

Matt

[Emmanuel Gillyns]

Hi,

You seem to have well checked your mesh, so I guess the problem is elsewhere.

How did you generate the mesh ? was it .msh that you converted to Nek5000 ? In this case you could try to scale it up in your mesh generator tool, then scale it back down in Nek5000 in the usrdat2 subroutine (I know it seems stupid, but I experienced a problem from there before using the genmap tool on a too small mesh) :

n = lx1 * ly1 * lz1 * nelt

scale = 0.1

call cmult ( xm1 , scale , n )

call cmult ( ym1 , scale , n )

call cmult ( zm1 , scale , n )

I also experienced a vanishing Jacobian when using a defective version of OpenMPI, do you have any vanishing Jacobian when running example 3D cases ?

Yours,

Emmanuel

[Gabriel Moldovan]

I also have vanishing jacobian when running simple cases in 3D. I am trying to run the turbPipe case with just 4 procs on a cluster and I get a vanishing jacobian error. I am using gnu 7.3.0 / openmpi 2.0.4. Has anyone got this problem before?

[Stefan K.]

Are you using the latest repo version?

On 27 Apr 2021, at 12:40, Gabriel Moldovan <gabrie...@gmail.com> wrote:

I also have vanishing jacobian when running simple cases in 3D. I am trying to run the turbPipe case with just 4 procs on a cluster and I get a vanishing jacobian error. I am using gnu 7.3.0 / openmpi 2.0.4. Has anyone got this problem before?

--
You received this message because you are subscribed to the Google Groups "Nek5000" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nek5000+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nek5000/a368ea9a-acb0-4436-a906-9e19c788e52cn%40googlegroups.com.

[Gabriel Moldovan]

I'm using the 19.0 version downloaded from the website

[Stefan K.]

Please try the latest repo version. I fixed a related bug recently. 

To view this discussion on the web visit https://groups.google.com/d/msgid/nek5000/6840e56a-e051-4dec-9b70-d68b676f3c13n%40googlegroups.com.

[Gabriel Moldovan]

Thanks Stefan for your suggestion. I run the case on the cluster with the latest repo version and I no longer have issues with vanishing jacobians. I compiled with gnu 8.3.0, openmpi 2.0.4 and arms-forge 20.2.1.

Best regards