Divergence Stokes solver with mesh refinement

[robelino]

Hi there,

I am running a transient simulation to make a diagnostic of a Scandinavian glacier.
I use a solver which compute the mass balance in function of elevation in the glacier and then actualise the free surface elevation with the free surface solver.

I made some tests with a resolution of 100 m x 100 m and everything works, I can do my simulation with 59 time-steps as I want. Nevertheless, I want to refine my mesh with a resolution of 40 m x 40 m and did not get a convergence with the Stokes solver. I tried different combination with direct and iterative solvers.
Here is the parameters used with the resolution of 100 m x 100 m:

Code: Select all

Solver 5
  Equation = "Stokes-Vec"
  Procedure = "IncompressibleNSVec" "IncompressibleNSSolver"

  !linear settings:
  !------------------------------
  Linear System Solver = Direct
  Linear System Direct Method = mumps

  !Non-linear iteration settings:
  !------------------------------ 
  Nonlinear System Max Iterations = 50
  Nonlinear System Convergence Tolerance = 1.0e-08
  Nonlinear System Newton After Tolerance = 1.0e-03

  !------------------------------ 
  ! steady-statte settings
  !------------------------------ 
  Steady State Convergence Tolerance = Real 1.0e-4

  ! 1st iteration viscosity is constant
  Constant-Viscosity Start = Logical True
End

For the refine mesh, I removed the newton iterations, which allow to have a convergence for few more iterations but not enough. I tried to used :

Code: Select all

Stokes Flow = logical true

I use a parallel computation, so I also try the simulation with a non-split mesh.
Unfortunately, all these try did not change the solution anymore.

Until know, when I got convergence problem, it was for the bigger mesh as well and once the problem was solved, I could refine the mesh then.
So I am wondering if anyone have a solution to this problem ? For information, the area of the glacier is only 2.77 km².

Thanks in advance for your help.
Best,
Olivier

[raback]

Hi Oliver,

How about setting "Nonlinear system relaxation factor = 0.7", or 0.5.

-Peter