[MITgcm-devel] more seaice blues for the adjoint

[chris hill]

Martin,

  Quick ice question.

Is it just a numerical issue i.e. is it clear what the underlying 
equations are, since ice is neither a classical fluid or a classical 
solid. What time scale does it spread on?

Chris
Martin Losch wrote:
> No, but it's a numerical issure. I did not pay any attention to this 
> before and maybe this was no problem in the original code that you 
> supplied but if you have zero strain, Delta = 0, and this is replaced by 
> some Delta_min (SEAICE_EPS in the code) in order to avoid division by 
> zero (and infinite viscosities). P however is still non-zero and 
> divergence(stress) end up being (dP/dx, dP/dy) .NE. 0, so that you have 
> a forcing of down the slope of P. This does not make sense (why would 
> ice spontaneously spread?), so that the pressure is replaced by P = 
> P*Delta/max(Delta,SEAICE_EPS), so that in the (rare) case of no strain P 
> = zeta = eta = 0 and thus no forcing by stress. Makes sense to me (and 
> is really only relevant in idealized test case, just like the spurious 
> motion of sigma coordinates vs. z-coordinates in ocean models).
> 
> Hope I did not misunderstand anything here.
> 
> Martin
> On 7 Jun 2007, at 13:40, Jinlun Zhang wrote:
> 
>> Hi Martin,
>> Would this be due to some kind of finite differencing problem?
>> Jinlun
>>
>> Martin Losch wrote:
>>> Hi Patrick,
>>>
>>> I have found (or rather, was pointed to) a problem with the seaice 
>>> solvers: The start to move spontaneously (in the absence of forcing), 
>>> if the sea ice distribution is NOT uniform.
>>> I have implemented a fix but this will cause problems with the 
>>> adjoint: I need terms like
>>> SQRT(deltaC), which used to be SQRT(MAX(deltaC,SEAICE_EPS_SQ)), so 
>>> that the derivate code will be involve 1/sqrt(deltaC). Should I put 
>>> this into #ifdefs?
>>>
>>> Martin
>>
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>