Test variable grid spacing for C grid - part2

[JFLemieux73]

See also issue #786.

[eclare108213]

I checked the discretization in the overleaf document. It looks okay to me, just one question about a sign in the replacement pressure, which should not affect symmetry.

There might be a typo in the translation from the document to the code. There, we should be able to replace i with j and j with i (which will change off-diagonal array elements but not the diagonal i,j and i+1,j+1 elements), and come up with the same discretization for each term but rotated 90 degrees and probably flipped in sign for some of them.

[JFLemieux73]

Thanks for checking the discretization. Is the error in the code or the document?

[eclare108213]

The comment is in the document. Not sure whether it's an error.

[JFLemieux73]

Ok I will have a look. Thanks!

[JFLemieux73]

Let's start with the uniform grid (dx=constant). For some reason my VP run finished successfully but the last output files are not there. So I will look at 2005-01-26 (after 25 days). I used 100 nonlinear iterations. Here is the hi field:

[JFLemieux73]

Ok the issue with the missing outputs was just the walltime allowed not set properly. No worries. Here is the B grid EVP solution:

and for the C grid EVP:

[JFLemieux73]

It is weird for the C grid. Is it due to the fact that P*=2.75e4? I have seen problems before with stiff problems (e.g. issue #697). Let's try with P*=1e4.

[JFLemieux73]

Ok there is clearly something going on in the corners. Here is aice after 2 weeks (range is between 0.9 and 1.0). I don't think it is related to the EVP; the pattern is symmetrical.

The problem is seen using a uniform grid. I will open a new issue with a better title.

[JFLemieux73]

As shown in issue #792, the checkerboard pattern in the corner is related to the remapping. I am going to use upwind for the rest of the tests here to see if there is still thickening in corners with the C grid.

dt=1h
ndte=2400
elasticDamp=0.1
advection='upwind'
ice_data_conc = 'p8'
Pstar = 1e4

Let's look at the solutions after 1 month.

[JFLemieux73]

Here is the hi B grid solution:

And the C grid:

Note that the ice is thicker than in previous tests because Pstar = 1e4.

[JFLemieux73]

Here is C-B:

The differences are small but we do see thicker ice with the C grid but not exactly in the corner.

[JFLemieux73]

I have verified and the diff between EVP (B grid) and VP (B grid) are very small (~1 mm, not shown).

[JFLemieux73]

The C grid noise has been eliminated with l_fixed_area=T (see issues #809 and #811). The code does not crash anymore for gx simulations (see PR #833).