[Mitgcm-support] Re: gmredi_slope_limit

mitgcm-support at dev.mitgcm.org mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:45:54 EDT 2003


Ralf,

I've solved the problem.
It was indeed a numerical problem related to
very large slope squares (probably related to
very small dSigmaDr's).

Sorry for the false alarm.
Cheers
-Patrick



Ralf Giering wrote:
> Hi Patrick,
> 
> The temporary should not make any difference.
> But I like to have a look at the adjoint differences.
> Can you send me the relevant diff output please.
> 
> Ralf
> 
> On Wednesday 08 January 2003 05:40 pm, you wrote:
> 
>>Hi Ralf,
>>
>>I have another problem in the gmredi_slope_limit routine.
>>I'm not sure whether it's a numerical issue or TAF-related:
>>
>>I am currently testing all tapering schemes.
>>Most of  them look good, but the option
>>        IF (GM_taper_scheme.EQ.'linear') THEN
>>is totally off.
>>
>>I can easily fix it by replacing the expression in this block
>>           ELSE IF ( SlopeSqr(i,j) .GT. maxSlopeSqr )  THEN
>>            taperFct(i,j) = sqrt(maxSlopeSqr / SlopeSqr(i,j))
>>           ENDIF
>>by
>>           ELSE IF ( SlopeSqr(i,j) .GT. maxSlopeSqr )  THEN
>>            tmpfld(i,j) = maxSlopeSqr / SlopeSqr(i,j)
>>            taperFct(i,j) = sqrt(tmpfld(i,j))
>>           ENDIF
>>The latter produces OK gradient checks.
>>
>>I quickly compared the adjoint codes, they seemed OK,
>>but I may have overlooked something.
>>
>>Just want to make sure, there's no bug.
>>
>>Thanks
>>-Patrick
> 
> 


-- 
_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
Patrick Heimbach     Massachusetts Institute of Technology
FON: +1/617/253-5259                    EAPS, Room 54-1518
FAX: +1/617/253-4464               77 Massachusetts Avenue
mailto:heimbach at mit.edu                 Cambridge MA 02139
http://www.mit.edu/~heimbach/                       U.S.A.




More information about the MITgcm-support mailing list