[MITgcm-support] model crashing & time step issues

[Martin Losch]

Neil,

the secret is probably the CFL-criterion (Courant-Friedrich-Levy) that puts a requirement on your time step (and you already found out by yourself):
u*dt/dx < 1 (or 1/2), see the tutorials for more details or any text book on numerics.
For pedagogical reasons (it's instructive to do that), I suggest that you set your monitorFreq = 1. and then have a look at the monitor variables advcfl_u/v/w_max, etc. They should increase exponentially for your 2deg simulation. You need to reduce your time step until that does not happen anymore. I once used this for a 2deg configuration.
 deltaTmom = 1200.0,
 deltaTtracer   = 43200.0,
 deltaTClock    = 43200.0,
 deltaTfreesurf = 43200.0,

Also there are constraint on the viscosity parameter (also some sort of CFL, again see the tutorials). I recommend to replace viscAh by an appropriate viscAhGrid (e.g. 0.1) to avoid the problem of adjusting the viscosity.

Surface temperature might not be a good diagnostic. The crash can also happen in the deep ocean somewhere.

Martin


On Jul 2, 2013, at 4:25 PM, Neil Patel <nigellius at gmail.com> wrote:

> Using the 4° resolution tutorial_global_oce_latlon as a starting point, I modified the input files to a higher, 2° resolution but similar values to the original. I changed the dxSpacing and dySpacing to 2 in input/data, and the number of grid cells to 180 and 80 for x and y (one processor and one thread configuration) respectively in code/SIZE.h. When I ran it I got this error:
> 
> terra tutorial_global_oce_latlon_newgrid/input> ../build/mitgcmuv > output.txt
> SOLUTION IS HEADING OUT OF BOUNDS: tMin,tMax= -3.088E+03  1.119E+04
> exceeds allowed range (monSolutionMaxRange=  1.000E+03)
> MON_SOLUTION: STOPPING CALCULATION at Iter=         6
> S/R ALL_PROC_DIE: ending the run
> ABNORMAL END: S/R MON_SOLUTION, stops due to EXTREME Pot.Temp
> 
> Mitgcm was able to run for 8 days (I told input/data to create an output file for each day) without crashing. I viewed the last temperature output file for day 8 and it appeared normal (sea surface temperatures maxed out around 30°C). Why was the simulation able to do 8 days without much error and then suddenly diverge in one day? Or are the problems being created in another parameter?
> 
> Then, I changed the length of all the time steps in input/data to 1/4 of the original. This time, it ran out to 20 days, then crashed in the same way as previously. Next, I changed the size of all the time steps in input/dat to 1/10 of the original. Then, it ran normally out to 200 days (which is when I set the simulation to end).
> 
> What's the reason for the sudden crashes? Is there a way to predict what time step is best for the simulation?
> 
> Thanks,
> 
> Neil
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support