Page 1 of 1

Zero velocities in a finer mesh resolution

Posted: 25 Mar 2020, 12:45
by martim
Hello,

I am trying to simulate ice flow around an obstacle and its influence on the ice surface elevation. For that I am using an "ice slab" domain with the obstacle in the center and using a tilted gravity vector instead of surface slope to drive ice flow. My problem is that for a mesh with a resolution high enough that allows the FreeSurface Solver to properly solve the region around the obstacle, I start getting zero velocities everywhere.

I have tried applying only a local mesh refinement and keeping the elements away from the obstacle relatively coarse, and it did not work either. I also tried using the refined mesh on a flat slab with no obstacle at all, and I run into the same problem, making me believe that the problem lies on the mesh resolution and not on the presence of the obstacle. Does anybody know what could be the problem?

I am attaching my sif file (that does not need input data for the flat slab), a mesh for which I get reasonable velocities, and a refined version of the mesh for which the same sif results in zero velocities everywhere.

Many thanks beforehand!

Martim

Re: Zero velocities in a finer mesh resolution

Posted: 27 Mar 2020, 17:05
by joeatodd
Hi Martin,

I had a quick look at your case. I think what's actually happening is that MUMPS is throwing an allocation error and giving up, resulting in no velocity computation. Basically, your mesh is too fine for whatever system you are running on (and indeed too large for my own desktop computer).

Code: Select all

DirectSolver: Using direct method: mumps
 ** ERROR RETURN ** FROM DMUMPS INFO(1)=  -13
 ** INFO(2)=           -3539
 ** ERROR RETURN ** FROM DMUMPS INFO(1)=   -3
 ** INFO(2)=               3
SolveLinearSystem: Calculating nodal loads

Hope it helps,

Joe

Re: Zero velocities in a finer mesh resolution

Posted: 03 Apr 2020, 15:31
by martim
Hey Joe,

Thanks for the reply and for spotting the issue! However, I have been experimenting for the entire week with different linear solvers and if I could find a compromise for the mesh resolution, but to no success. In the end I run into convergence issues which make my free surface go crazy. After discussing this for some time with Eef, a potential solution could be applying a progressive refinement of the mesh towards the center of the obstacle, but I still could not get anything stable. Do you have any suggestions or advice to what or how I should do it? A refinement of up to 100 m at the center could be a good compromise. Once again, thanks a lot for the help!

Best,

Martim

Re: Zero velocities in a finer mesh resolution

Posted: 04 Jun 2020, 09:25
by tzwinger
Martim,
I know this answer comes vastly delayed, but could you try to use the new vectorized Navier Stokes solver with the block preconditioner to avoid running into memory issues?

You can find examples on how to deploy it under the latest course pages http://elmerfem.org/elmerice/wiki/doku. ... es:courses (check for "Arctic Valley Glacier")

Regards,
Thomas