[MITgcm-devel] another bug in growth.F ?

[Dimitris Menemenlis]

Patrick, that bit is Jinlun's addition to remove snow from open water.  To make 
sure that this is the "only" troublesome addition, could you also try finite 
difference with attached routine.

I am returning discussion to MITgcm Devel to keep Martin and Jinlun in the loop.

Dimitris

> Dimitris,
> 
> sharp increase in gradient w.r.t. atemp comes in going from
> growth.F3 to growth.F4
> i.e. it seems, addition of block
>           IF(YNEG(I,J,bi,bj).GT.ZERO.AND.HSNOW(I,J,bi,bj).GT.ZERO) THEN
>            GHEFF(I,J)=MIN(HSNOW(I,J,bi,bj)/SDF,YNEG(I,J,bi,bj))
>            YNEG(I,J,bi,bj)=YNEG(I,J,bi,bj)-GHEFF(I,J)
>            HSNOW(I,J,bi,bj)=HSNOW(I,J,bi,bj)-GHEFF(I,J)*SDF
>            SEAICE_SALT(I,J,bi,bj)=SEAICE_SALT(I,J,bi,bj)
>      &                  -GHEFF(I,J)/0.920 _d 0
>           ENDIF
> 
> Funny thing is that adjoint gradients actually stay 
> same order of magn., it's really the finite difference grad.
> which increase by 4 orders of magn.
> I'll play with epsilon and also try sensit. to other controls.
> 
> -p.
> 
> 
> 
> On Mon, 2006-11-06 at 19:52 -0800, Dimitris Menemenlis wrote:
>> > Patrick, ignore previous modified growth.F file.
>> > I am attaching 6 variations of growth.F:
>> > growth.F0 is 1.29
>> > growth.F5 is 1.30
>> > It would be useful to know which of the incremental
>> > changes causes forward sensitivity to blow up.
>> > 
>> > D.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: growth.F6
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20061107/b9692885/attachment.el>