[MITgcm-devel] problem with 59c

[Martin Losch]

Hi Dimitris,

- the LSR boundary conditions were (dramatically) wrong in ckpt59c,  
my mistake, that is fixed and I am quite confident that this piece of  
code is now working properly (as you have seen from my "funnel"  
experiments), but if your problem occurs even without dynamics, then  
this is not the problem (I did not see any stability problems in my  
runs with the bug either, the solutions just looked very different)

- have you tried reverting the calc_r_star routine to pre 59c. The  
only thing that I changed in there is the way potiential errors are  
caught, but you never know ...

can this be the problem?
- add a cap on AREA after advecting AREA if seaice_growth is not called
probably not, because you have the thermodynamics turned on (by  
default), and then this piece of code is not used.

Martin

On 2 Jul 2007, at 02:48, Dimitris Menemenlis wrote:

> All, somewhere between 59b and 59c the CS510 configuration stopped  
> working. That is 59b works but 59c crashes after about 5 days of  
> integration.
>
> The crash appears to be caused by some sort of sea-ice-mixed-layer  
> depth interaction in the Southern Ocean.
>
> Crash happens with and without multicategory sea ice, with and  
> without sea ice dynamics, with AdvScheme=33 as well as with  
> AdvScheme=7, with Deltat=1200 as well as with Deltat=900.
>
> Looking at things that have changed between 59b and 59c: CS510 does  
> not use GMRedi, nor thsice, nor obcs, nor Visbeck's scheme.  CS510  
> does use pkg/seaice (but crash occurs even with dynamics turned  
> off), calc_r_star, and runoff.
>
> I am running out of ideas where to look next ... and tests are  
> cumbersome because I have not been able to reproduce problem in a  
> smaller configuration. So any suggestions would be much appreciated.
>
> A possible clue is that the Arctic CS510 domain runs fine with  
> latest code but it does not use the rStar.
>
> D.
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel