[CIG-LONG] Gale again

[Walter Landry]

Rainer Nerlich <rainer at simula.no> wrote:
> *Hi Walter,*
> 
> *thanks again! I did not scale my viscosities because from the user guide I
> understood that I have to multiply my velocities with the lowest viscosity
> that I want to assume in my model so that the velocities become high and
> the viscosities low.*
> 
> *My velocities come from a different numerical model and were originally
> very small. As I wanted three layers (crust, upper and lower mantle), I
> used the viscosity (e.g. 10^21 Pa*s) of the upper mantle to scale my
> velocities. By setting eta0 to 1 for the upper mantle and 100 and 40 for
> the crust and lower mantle, respectively, I achieved a 100/40 times higher
> viscous crust/lower mantle. **This worked well for viscosities of 10^22
> Pa*s or higher. The velocity field looked like it should. 

You should set your parameters so that everything is self-consistent.
If you want to do a run with viscosity=1e22, then make sure that the
velocities and diffusion coefficients are consistent with that.  So it
is good that you used 1, 40, and 100 for your viscosities, because
your velocities are consistent with that.

> For viscosities smaller than 10^22 Pa*s, the velocity field looked
> wrong 

This is because of the drunken seaman effect.  Essentially, there is a
numerical instability that occurs with free surfaces and very low
viscosities.  The only fix in Gale for that is to use specify smaller
time steps with "dt" or "dtFactor".

> and the time steps were varying.*

This is odd.  I can not reproduce that in serial or parallel.  I get
slightly different time steps for serial and parallel, but that is not
unexpected.  It is always the same from run to run.  Do you see that
effect if you run with only one processor?

I have pushed the changes needed to read from up to 30 files into the
repository.  Can you try that version and see if you still see the
time steps varying from run to run?

> *However, I scaled now my viscosities by including 1e22, 40e22 and 100e22
> for eta0 in the input file, and the velocity field looked correct for lower
> viscosities as well (see attached figures). **For higher viscosities the
> velocity field looks identical but I noticed that the magnitude of the
> velocities differs now (compare figures: 10_22_scaled and 10_22_unscaled),
> at least for 10^22 Pa*s. I guess, the scaled one is correct (?).*

Yes, the scaled one is correct.  By replacing 1 with 1e22, you end up
with an effective viscosity of 1e44.

Cheers,
Walter Landry