[MITgcm-support] flux limiters and timestep advice...

Klymak Jody jklymak at uvic.ca
Thu Apr 9 14:27:49 EDT 2009


Hi all,

I'm running relatively high-resolution 2-D simulations of flow over  
topography.  I was relatively happy using 2-nd order centered  
advection. There was some gridscale noise, but the diffusivity tended  
to erase it and the major features look fine.  Now I am being a little  
more fussy and trying to balance energy with dissipation \epsilon =  
viscAz*(du/dz)^2 and find that my dissipation is scale dependent such  
that fine-scale models dissipate more than coarse scale.  My  
supposition is that this is due to the extra gridscale noise.

So I switched to a superbee flux limiting scheme (tempAdvScheme=77).   
It certainly removed much of the gridscale noise yet the resolved  
dissipative features remain the same.

However, after a while I start to get low-mode noise in the solution.   
When I was using centered 2nd-order I could run 12.5 s timesteps w/o  
difficulty.  If I run superbee I have to reduce this to 2 s, and still  
about 20 h I start getting low-mode grid-scale noise.  The noise seems  
largest near the edges of my topography, and extends through the whole  
water column.

Please see: http://web.uvic.ca/~jklymak/SuperbeeProblems.pdf

FWIW, these are not huge signals, but completely destroy my energy  
budgets.

Perhaps my solution is simply to go down to 1-s timesteps (that will  
be what I will do for now). However, I was curious if there were other  
suggestions. Should I decrease just the barotropic time step?   
Anything else I've set up incorrectly in the attached?

Thanks for any thoughts.

Cheers,  Jody

-------------- next part --------------
A non-text attachment was scrubbed...
Name: dataSBJmk
Type: application/octet-stream
Size: 3379 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20090409/e12c3288/attachment.obj>
-------------- next part --------------





--
Jody Klymak
http://web.uvic.ca/~jklymak/









More information about the MITgcm-support mailing list