[MITgcm-devel] common block in seaice_dynsolver.F

[Martin Losch]

Hi Jean-Michel,

thanks for the clarification. I think this untested feature is a good idea and I should better keep it in mind

Martin

> On 20. Apr 2018, at 01:03, Jean-Michel Campin <jmc at mit.edu> wrote:
> 
> Hi Martin,
> 
> The main reason is to maintain an un-tested feature that consists in using
> a longer time-step for seaice-dynamics than for seaice thermo (and, I imagine, 
> the ocean time-step), thus solving for uIce,vIce less often.
> And with some option of ice/ocean stress, TAUX/TAUY are used in SEAICE_OCEAN_STRESS
> outside the if-block DIFFERENT_MULTIPLE(SEAICE_deltaTdyn, ...
> 
> Now this feature is not used very often, and I would recommand to disable it cleanly
> for the adjoint, since it makes the code more complicated (requires more storage dir;
> plus the adjoint of seaice-dynamics is tricky, so to simplify it could help.
> In addition, this feature is curently broken for the adjoint (lines 97-98,
> TAUX & TAUX are reset to zero).
> 
> If we were following this route (disable this with #ifdef ALLOW_AUTODIFF), we might
> be able to remove storage dir for stressDivergenceX/Y by resetting them to zero
> (like currently TAUX/TAUY) within #ifdef ALLOW_AUTODIFF
> 
> Cheers,
> Jean-Michel
> 
> On Thu, Apr 19, 2018 at 11:12:16AM +0200, Martin Losch wrote:
>> Hi Jean-Michel,
>> 
>> I probably should have participated more in the reviewing process of you pull-request about seaice wind stress and asked these unimportant questions then:
>> - Why do you need to put TAUX/TAUY into a local common block in seaice_dynsolver? 
>> - Why do these local variables need to be initialized for nIter0? They are initialized in each time step anyway in line 193/194 before seaice_get_dynforcing is called.
>> 
>> Martin
>> 
>> BTW. git (actually curl) on stan works again and today we an have up-to-date testreport.
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-devel
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mailman.mitgcm.org/mailman/listinfo/mitgcm-devel