Blowing up after 9966 timesteps

[kee]

Hi all,
I am very confused that the model blow up at the 9966 timestep. I am sure the CFL creteria is satisfied.
I think that if the DT is not small enough, the model can blow up shortly after the starting. Now it blowup at the 9966th timestep, it is more likely that waves shorter than 4*DX cause the results. But I am not sure.

Similar situations have occured times(like after around 190000 integrations), but I donot know how to deal with the problem.

I wonder that this maybe have something with the APPLICATION.h and Values set in the IN-file.

Both of the files are in the attachment.

Can anyone give some ideas?

Thanks in advance.

[kate]

We can't tell what the problem is from the setup without running it all over again. Instead, you need to build up tools for looking at the model output and learning to see what makes sense and what doesn't. Did it blow up because of a large velocity or because the tracer fields got weird? Where did it happen? At the surface, bottom or in the middle?

Blowing up at step 9966 could mean you were at the edge of instability and just crossed over or it could be something else entirely.

[ptimko]

You have what would appear to be a baroclinic instability. I would suggest checking you vertical mixing model. I have had problems in the past with ROMS by not specifying a vertical mixing model. The problem was solved by specifying a GLS mixing model in the model options during the build process. It may not be the probelm you have but it might be a starting point to look at...

[kee]

Did it blow up because of a large velocity or because the tracer fields got weird? Where did it happen? At the surface, bottom or in the middle?

Kate, thanks a lot!
I checked the output and some information is pasted below

Code: Select all

STEP   Day HH:MM:SS  KINETIC_ENRG   POTEN_ENRG    TOTAL_ENRG    NET_VOLUME
          C => (i,j,k)       Cu            Cv            Cw         Max Speed
... ... ... ...
   9965    20 18:15:00  1.879129E-02  2.160202E+04  2.160204E+04  1.049255E+16
          (741,047,11)  2.108176E-01  7.250023E-02  5.464230E-01  4.778940E+00
   9966    20 18:18:00  1.882934E-02           NaN           NaN  1.049255E+16
          (741,047,11)  2.095299E-01  7.145118E-02  5.451024E-01  4.739977E+00

It seems like that the break came from POTEN_ENRG and TOTAL_ENRG on the 11th level which is a middle level.
In my configuration the grid is 899*662*20 in Xi, Eta and Vertical respectively.

I donnot know how the POTEN_ENRG and TOTAL_ENRG are caculated in the model, either.

can you give more advice?

[kee]

... The problem was solved by specifying a GLS mixing model in the model options during the build process. It may not be the probelm you have but it might be a starting point to look at...

Hi ptimko,

Thanks for yr kindly reply.
I did specify a GLS mixing model in the APPLICATION.h, like

Code: Select all

#define GLS_MIXING
#undef MY25_MIXING
#if defined GLS_MIXING || defined MY25_MIXING
# define KANTHA_CLAYSON
# define N2S2_HORAVG
#endif

This can also be seen in the logfile

Code: Select all

... ...
 DOUBLE_PRECISION    Double precision arithmetic.
 GLS_MIXING          Generic Length-Scale turbulence closure.
 KANTHA_CLAYSON      Kantha and Clayson stability function formulation.
... ...

I am wondering if I can modify the values of the mixing parameters a little bigger to prevent instability, but I donot know.

is there any further suggestions?

[kate]

Code: Select all

STEP   Day HH:MM:SS  KINETIC_ENRG   POTEN_ENRG    TOTAL_ENRG    NET_VOLUME
          C => (i,j,k)       Cu            Cv            Cw         Max Speed
... ... ... ...
   9965    20 18:15:00  1.879129E-02  2.160202E+04  2.160204E+04  1.049255E+16
          (741,047,11)  2.108176E-01  7.250023E-02  5.464230E-01  4.778940E+00
   9966    20 18:18:00  1.882934E-02           NaN           NaN  1.049255E+16
          (741,047,11)  2.095299E-01  7.145118E-02  5.451024E-01  4.739977E+00

It seems like that the break came from POTEN_ENRG and TOTAL_ENRG on the 11th level which is a middle level.

You can look at diag.F and see how it computes the energies. They are globally integrated quantities, so the location you see listed there doesn't tell you where the NaN is coming from. The location is that of the largest velocity, one that is not alarmingly large. You need to be looking at the 3D fields in say the restart file if it created a record on blowup. I use ncview for a quick and dirty view of what's going on.

[kate]

I am wondering if I can modify the values of the mixing parameters a little bigger to prevent instability, but I donot know.

is there any further suggestions?

GLS has different behaviors depending on the parameters you give it. Some suggestions are in the ocean.in file. I have found some to behave better than others. I am using the K-omega option.

[ptimko]

I'm also using the K-omega option. Not sure what else to suggest right now, if I think I'll let you know.