[MITgcm-devel] problem in seaice_lsr with cubed sphere

[Martin Losch]

OK, so I’ll make an exact rl-copy of fill_cs_corner_rs, just with _RS replaced by _RL in the declarations of uFld/vFld and I’ll call it like this in seaice_lsr:

Do bj=…
Do bi=…
call fill_cs_corner_uv_rs(.true.,uice(:,:,bi,bj),vice(:,:,bi,bj),bi,bj,myThid)
ENDDO
ENDDO

Martin


On Dec 3, 2014, at 5:53 PM, Jean-Michel Campin <jmc at ocean.mit.edu> wrote:

> Hi Martin,
> 
> Should be OK to keep a local (no bi,bj) argument within S/R 
> fill_cs_corner_uv_rl since it only involves local copy from the
> same tile.
> And this way, it could also be re-used to fill the cormner of some
> local (no bi,bj) array.
> 
> Cheers,
> Jean-Michel
> 
> On Wed, Dec 03, 2014 at 05:29:41PM +0100, Martin Losch wrote:
>> Hi Jean-Michel,
>> 
>> thanks for this trick: I inserted this routine after the exch_uv_xy_rl at the end of the linear iteration and this fixes the problem.
>> 
>> Now, do we need to make a _RL version that can also handle 4D arrays (with bi,bj indices) or can we just call it like this:
>> 
>> call fill_cs_corner_uv_rs(.true.,uice(:,:,bi,bj),vice(:,:,bi,bj),bi,bj,myThid)
>> 
>> I have the feeling that this only works for me because nSy = 1, so I would actually make a routine that takes 4D arrays (but that wouldn’t be consistent with the _RS version, what’s best in your opinion?
>> 
>> Martin
>> 
>> On Dec 3, 2014, at 2:26 PM, Jean-Michel Campin <jmc at ocean.mit.edu> wrote:
>> 
>>> Hi Martin,
>>> 
>>> Did you try something like one or more calls to S/R FILL_CS_CORNER_UV
>>> (we currently only have the _RS version but would not be difficult
>>> to make an _RL ; and as long as _RS = real*8, you can certainly give 
>>> it a try just calling the _RS version).
>>> 
>>> Cheers,
>>> Jean-Michel
>>> 
>>> On Wed, Dec 03, 2014 at 01:47:33PM +0100, Martin Losch wrote:
>>>> Hi again,
>>>> 
>>>> after playing a little, it seems there are more problems than just the zero recip_raw/ras, so it may be easier to just include a resetting of zero BU/BV at the end of the routine (although this may cover up unrecognized problems).
>>>> 
>>>> Since this all of this is sufficiently fragile, I’d like to add the new capabilitiies: RAS and strong implicit coupling in one of the test experiments. How about global_ocean.cs32x15/input.seaice ?
>>>> 
>>>> Martin
>>>> 
>>>> On Dec 2, 2014, at 11:55 AM, Martin Losch <Martin.Losch at awi.de> wrote:
>>>> 
>>>>> Hi Jean-Michel,
>>>>> 
>>>>> yesterday I implemented the option to compute the LSR solution on overlapping tiles, i.e. a restricted additive schwarz (RAS) method. imin/imax can be 1-SEAICE_OLx/sNx+SEAICE_OLx etc. This works fine for SEAICE_OLx <= OLx-2 on simple domains, but for cubed sphere experiments, I need to set SEAICE_OLx=OLx-3 for this to work. SEAICE_OLx = OLx-2 (=2) I get NaN in global_ocean.cs32x15.seaice (and others that use LSR), in the second linear iteration of the solver. 
>>>>> The reason seems to be BU=0 in some of the corner halos (34,18,2,1), and this is because recip_raw(34,18,2,1) seems to be zero.
>>>>> 
>>>>> Can you reproduce this problem (set SEAICE_OLx=2, and SEAICE_OLy = 2 in global_ocean.cs32x15/input.seaice/data.seaice and run ./testreport -t global_ocean.cs32x15)?
>>>>> Can we find a decent way to fix that? Initialise recip_raw=1 always? or do I have to check my BU/BV and reset them if they are zero?
>>>>> 
>>>>> Martin
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> MITgcm-devel mailing list
>>>>> MITgcm-devel at mitgcm.org
>>>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>>>> 
>>>> 
>>>> _______________________________________________
>>>> MITgcm-devel mailing list
>>>> MITgcm-devel at mitgcm.org
>>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>>> 
>>> _______________________________________________
>>> MITgcm-devel mailing list
>>> MITgcm-devel at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>> 
>> 
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel