[MITgcm-devel] stripes in adjoint along tile edges

[Martin Losch]

Hi Patrick,

I think you solved my immediate problem.

I am basically using the more or less unmodified configuration MITgcm_contrib/arctic/cs_36km/, but I have by now turned off a lot of things, especially seaice is turned off and the stripes stay.

I didn’t provide any details before, because all I want to know is, if this has been observed before. 

But it turns out that your comment about the wrong place seems to be the correct idea: I compared now netcdf (with stripes) with mds output. The mds output does not seem to have the stripes and is also generally a little different (see attached png-file). I hadn’t had a look at the addummy_in_stepping in a while, but appearently the mds-io is treated differently than the mnc output. Good to know, and thanks for putting me on the right track. Now I can really start debugging … (hopefully)

Martin

On Nov 27, 2014, at 4:46 PM, Patrick Heimbach <heimbach at mit.edu> wrote:

> Hi Martin,
> 
> possibly, adUvel, adVvel, are dumped at the wrong place within the time step loop with respect to when they’re exchanged. We’ve seen this type of issues in the past. Is the implication of using an Arctic setup that it comes from the seaice? I.e. is it still present when useSeaice=.F.? Or is that just a coincidence?
> 
> As usual, easier to take a look if you provide a reference setup (and code version; I assume the latest?) 
> so one can get an understanding of what is being done?
> 
> Cheers
> p.
> 
> On Nov 27, 2014, at 9:37 AM, Martin Losch <Martin.Losch at awi.de> wrote:
> 
>> Hi there,
>> 
>> since there hasn’t been any other statement: Can I assume that no-one on this list has ever seen stripes like this in the adjoint (would be good, because it would mean, that I am making a mistake, that I should be able to find)? Or does the low reply rate mean that no-one has ever looked closely enough to see the stripes? 
>> 
>> I am running a little out of ideas about where they might come from. I have a suspicion, that they are related to cg2d/solve_for_pressure, but no good pieces of evidence, yet.
>> 
>> Martin
>> 
>> On Nov 20, 2014, at 4:28 PM, Martin Losch <Martin.Losch at awi.de> wrote:
>> 
>>> Hi adjoint people,
>>> 
>>> while debugging, I stumbled over interesting stripes in the adjoint fields along tile edges. See attached files for an example (based on arctic_cs36 with 24 cpus, sNx = 35, sNy = 48), log10(sqrt(adU**2)) for layers 1 and 10 after 1 adjoint timestep (after evaluating the cost function). As a cost function I evaluate theta(80,30), at the spot with the highest sensitivity. The stripes have low magnitudes (1e-8), but still large compared to the surrounding values. Does that look familiar in any way? Should I be worried about this? (I am because the model later explodes along these stripes). 
>>> 
>>> Martin
>>> 
>>> PS. I am using cdTargetResidual=1e-13 and cg2dMaxIters=1000, and these stripes appear when the seaice package is turned off totally (forwards and backwards), with advection scheme 30 and 33 for T/S.
>>> 
>>> 
>>> <adU_k1_t1.png><adU_k10_t1.png>
>>> _______________________________________________
>>> 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
> 
> 
> ---
> Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
> MIT | EAPS 54-1420 | 77 Massachusetts Ave | Cambridge MA 02139 USA
> FON +1-617-253-5259 | FAX +1-617-253-4464 | SKYPE patrick.heimbach
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20141127/9f54bb92/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mnc-mds_adU_k1_t1.png
Type: image/png
Size: 72756 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20141127/9f54bb92/attachment-0001.png>