Page 1 of 1

Navier-Stokes solver convergence issues

Posted: 11 Jan 2016, 08:52
by LizR
Hello,

I am having some problems with the Navier-Stokes solver. I would like to perform a simple diagnostic test to apply pressure to the top of a cube of ice and get the stresses as an output. I have modified the ForceToStress test case to create a simple example. However, if I replace the ForceToStress solver with the Navier-Stokes solver it won't run. I keep getting this error - ERROR:: IterSolve: Failed convergence tolerances.

I have tried increasing the convergence tolerance but if I make it too large the answers won't mean anything and I don't think tight tolerances are the cause of the problem.

I have been trying to do essentially the same thing in a few different ways with different .sif files and different meshes but I keep hitting this problem. What am I missing?

Re: Navier-Stokes solver convergence issues

Posted: 11 Jan 2016, 10:28
by raback
Hi

Your problem has a large null space. Assume a solution and add constant velocity to that and it will also be a solution. Hence you cannot have a solution for a linear system. You should add some BCs which determine the velocity field uniquely.

Apart from that I don't really understand the setting. You seem to solve the N-S in a 2D plate where also the external pressure is given. It would seem more natural to give the BC for a lower dimensional entity.

-Peter

Re: Navier-Stokes solver convergence issues

Posted: 15 Feb 2016, 09:21
by LizR
Since the last post I have made the boundary conditions and the mesh geometry more realistic and I understand a lot more about what the boundary conditions do now.
However, I still can't seem to get the Navier-Stokes solver to do a diagnostic run. Using the ElasticSolver and equivalent BCs works just fine.
I don't understand what the difference is, and why the Navier-Stokes solution will not converge.
Any help would be much appreciated.
-Liz

Re: Navier-Stokes solver convergence issues

Posted: 18 Feb 2016, 02:23
by raback
Hi Liz

Elasticity solver has a Lagrangian formulation and the displacements are directly linked to the shape of the domain. One solver is therefore sufficient even with a moving surface.

The N-S solver is naturally written in Eulerien frame of reference. Velocity is not directly associated with the shape of the domain. Instead you need a free surface solver, a mesh adaptation solver, and an ALE formulation in the N-S solver. There are many examples among tests, and the Elmer/Ice community can probably provide you with real cases.

-Peter

Re: Navier-Stokes solver convergence issues

Posted: 18 Feb 2016, 02:30
by LizR
Hi Peter
Thank you very much for telling me this. It is good to know that the issue was things that were missing, not things that were incorrect per se. I will continue working on it.
-Liz