[MITgcm-support] Can't replicate verification/global_oce_cs32/ad.sens

[Christopher Pitt Wolfe]

The problem seems to be related to setting nSx > 1. Runs with nSx = 6 and nPx = 4 produce zero salinity sensitivities and other subtle differences in the adjoint, but with nSx = 1 and nPx = 6 (with 32 x 32 tiles), the adjoint runs work fine. The forward runs are nearly identical if nSx = 1 or nSx > 1, but something in the adjoint code (or TAF) gets screwed up if nSx > 1.

Caveat utilitor,
Christopher

> On Jun 28, 2018, at 10:08 AM, Christopher Pitt Wolfe <c.l.p.wolfe at icloud.com> wrote:
> 
> Thanks Gael. I’ll check this against my output to see if there’s anything different about my configuration.
> 
> Christopher
> 
>> On Jun 28, 2018, at 8:50 AM, gael forget <gforget at mit.edu <mailto:gforget at mit.edu>> wrote:
>> 
>> Hi Christopher,
>> I attach this morning’s STDOUT below. I just checked that adxx_salt.0000000000.data is not filled with zeros in the GLACIER3.MIT.EDU <http://glacier3.mit.edu/> run.
>> Cheers,
>> Gael
>> 
>> <STDOUT.0000.gz>
>> 
>> 
>>> On Jun 27, 2018, at 11:51 AM, Christopher Pitt Wolfe <c.l.p.wolfe at icloud.com <mailto:c.l.p.wolfe at icloud.com>> wrote:
>>> 
>>> Hi All:
>>> 
>>> I’m trying to replicate Gael’s global_oce_cs32/ad.sens verification example that makes use of the generic cost and control packages. 
>>> 
>>> I followed the instructions at https://mitgcm.readthedocs.io/en/latest/ocean_state_est/ocean_state_est.html#test-cases-for-estimation-package-capabilities <https://mitgcm.readthedocs.io/en/latest/ocean_state_est/ocean_state_est.html#test-cases-for-estimation-package-capabilities>, compiled (using TAF) and ran the ad.sens case, and compared my output to that given in results/output_adm.sens.txt. While there are some differences related to the different model versions (66c for the included results compared to my 67b setup), most of the AD results are similar to within a couple of decimal places. The only weird thing in my case is that all the salt sensitivities are ZERO: all the ad_dynstat_adsalt fields are zero and the ADJsalt files are zeros. The adxx_salt file is written, but is zeros as well. I have verified that salinity is specifically included as a generic control in data.ctrl and the xx_salt file is written at the beginning of the forward run, so I’m stumped.
>>> 
>>> This case is run as part of the daily testreport cases and appears to pass on glacier, but I can’t find any STDOUT files to check if the salt sensitivities remain nonzero.
>>> 
>>> I’m not sure where to look for the source of the problem, so any advice would be appreciated. If anyone has a recent STDOUT file from this test case, I’d be interested in that as well.
>>> 
>>> Thanks!
>>> Christopher
>>> 
>>> 
>>> 
>>> -----------------------------------------------------------
>>> Christopher L. Pitt Wolfe
>>> Assistant Professor
>>> School of Marine and Atmospheric Sciences
>>> Stony Brook University
>>> christopher.wolfe at stonybrook.edu <mailto:christopher.wolfe at stonybrook.edu>             631-632-3152
>>> -----------------------------------------------------------
>>> _______________________________________________
>>> MITgcm-support mailing list
>>> MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support <http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support>
>> 
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org <mailto:MITgcm-support at mitgcm.org>
>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-support

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.mitgcm.org/pipermail/mitgcm-support/attachments/20180629/5e5de82b/attachment.html>