[MITgcm-devel] Lowering viscosity in high-res cube

[Chris Hill]

Hi All,

 Dimitris and I spoke briefly today. Here is a quick summary of where we are
and whats happening:

 1. the 1/4 latlon model can still run with significantly lower viscosity.
This is a concern because there is a qualitative impression (not quantified
yet) that the cube run is more viscous.

 2. when the cube run blows up with low viscosity it has so far blown up
near a cube face, but not that near a cube corner.

 3. the exch2_ fix that I have that gives the same numbers as exch1_ only
changes what is done right at the cube corner.

 4. I get the general impression that things are less stable with non-linear
free surface on particularly with wider overlaps. 

 5. when a cube run blows up due to too high viscosity it blows up at a
corner. This makes sense since the grid spacing is smallest and weirdest
there.

 Here is what I think needs to be done at this stage

 1. I need to redo my exch2_ tests with various different revisions of last
weeks checked in code. I will do this late today and check in the revised
exch2_ if all goes well. 

 2. We need to evaluate whether the lowest stable cube viscosity is still
too viscous relative to the latlon case. Its unclear to me how low we need
to go. Some maps of the spatially varying viscosity would help here as well
as some better measures of how syrupy i.e. standard deviation's at the same
points for latlon and cube.

 3. We need to think of a good way to make sure we are confident that
biharmonic friction and the associated wider halo exchanges are correct.
Getting exch1_ and exch2_ identical is useful but they could both have the
same error. Once (1.) is done I will try a halo of 4 and 5 and make sure the
results stay they same.

Chris