[MITgcm-support] Re: MITgcm files

Martin Losch Martin.Losch at awi.de
Mon May 26 09:49:13 EDT 2008


Renske,

the model horizontal domain is (1:nx,1:ny), but if you want to  
evaluate horizontal gradients along the boundaries you need at least  
one more grid point to do so. (otherwised you'd have to include code  
such as "if (1<i<nx) do something in the interior, else do something  
different at the boundary"), the second grid point you need for  
second derivatives for the laplacian (harmonic) viscosity and  
diffusivity. Olx/y>2 are needed for special advection schemes, etc.
So, yes the overlap ensures the exchange between tiles, but it's also  
necessary on only one tile in order to get the computational stencil  
along the domain boundaries right. Plus I don't know what happens  
with the "exchange" routines (which are always called regardless of  
domain decomposition, just to get potentially cyclic boundary  
conditions right) if they exchange stuff of length zero

I am pretty sure that your model crashes because of CFL-violations:  
Either your time step is too large for u*deltaT/deltax < (1,1/2, ...)  
or your viscosity is too large for the time step. Try smaller  
timesteps, for your resolution of 7.5km I recommend something like  
deltaT = 600s for a start.

Martin

On 26 May 2008, at 15:27, Renske Gelderloos wrote:

> Hi Martin,
>
> Thanks for the advice.  The part about the overlap in SIZE.h  I  
> don't really understand though.  Isn't that the amount of grid  
> cells that overlap between different tiles? So if you would use  
> only one tile, it doesn't make much sense to me to prescribe  
> overlap. Or is it not allowed to use only one tile? Furthermore, on  
> my computer the model crashes with nonzero overlap.. ("solution is  
> heading out of bounds (...), mon_solution: stopping calculation,  
> mon_solution: stopped due to extreme values of solution"). So what  
> is the idea behind the tiles and the overlap?
>
> Renske
>
> Martin Losch wrote:
>> Hi Renske,
>>
>> I am returning your files with my modficiations, but all of these  
>> are related to general stability of the code. I could not  
>> reproduce the problem that you described, namely that a non-zero  
>> heat flux does not change the surface temperature. For testing,  
>> try a non-uniform heat flux that introduces a horizontal pressure  
>> gradient and immediate flow. With a uniform heat flux and uniform  
>> initial conditions (in the horizontal) and a stable water column  
>> you should never get any flow. (as a side remark: I did get non- 
>> zero velocities after the second timestep, which should not  
>> happen, but that's not your problem. Is that a problem with step  
>> topography, and vertical diffusion? very worrying).
>>
>> a couple of problems:
>> SIZE.h: you set the overlaps to 0, that will probably not work  
>> ( and did not work for me, the model blows up immediately, I am  
>> surprised that we don't have a catch for that, maybe it is  
>> allowed?). I changed them to 4 (but 2 should be enough).
>>
>> data:
>> - deltaT = 86400 is much too large a time step and will also lead  
>> to the model blowing up immediately (once there is motion), but  
>> you probably knew that
>> - I put sref = 15*0.0, (although some compiler will probably take  
>> sref=0.0 as well)
>> - diffKzT = 20., is awfully large if implicitVerticalDiffusion is  
>> not turned on (default). I set it to 1.e-5 for now.
>>
>> topography:
>> - I usually try to set a minimum depth which should be at least  
>> the thickness of the top-most grid cell, can be set in "data" by  
>> hFacMinDz
>>
>>
>>
>>
>> On 23 May 2008, at 17:50, Gelderloos, Renske (KNMI) wrote:
>>
>>> Dear Martin,
>>>
>>> Here are all the files you requested. I would really appreciate  
>>> it if you would want to take a look at it, I'm pretty stuck...  
>>> Hope you can help me!
>>>
>>> Renske
>>>
>>> ________________________________
>>>
>>> From: Gelderloos, Renske (KNMI)
>>> Sent: Fri 5/23/2008 2:09 PM
>>> To: Gelderloos, Renske (KNMI)
>>> Subject: MITgcm files
>>>
>>>
>>>
>>> <data><data.mnc><data.pkg><eedata><Qnetuniform.bin><topogOBCmatlab.b 
>>> in><packages.conf><SIZE.h><LabradorOBC.m><heatuniform.m>
>> --------------------------------------------------------------------- 
>> ---
>>
>> _______________________________________________
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support




More information about the MITgcm-support mailing list