[MITgcm-support] timestep

Jean-Michel Campin jmc at ocean.mit.edu
Mon Jun 27 09:13:40 EDT 2005


Hi,

On Mon, Jun 27, 2005 at 01:49:57PM +0200, G?ran Brostr?m wrote:
> Dear Support
> 
> In most model experiments I have considered I have always been surprised how 
> short time step I have to take to avoid the model to blow up. As far as I can 
> tell the fields are not even close to be at the point of being numerically i
> unstable (i.e., dt/dx<<1). Especially during long integration this has been a problem. 
For a 1.D advection problem, with the Adams-Bashforth-II, the CFL limitation
is only 1/2. And will generally be reduced for a 3.D advection problem.

> I have for some time suspected that it has something to do with the vertical 
> velocity but I have not been able to confirm that. Anyway, I saw that there 
> is an option for taking implicit vertical advection (Implvertadv). Is there 
> anyone who have had the same problem and solved it using the implicit 
> vertical advection? I have looked around at the model examples but I have 
> not seen that it is used so I am somewhat uncertain to use it.

Implicit vertical advection has not yet been used in a lot of different
set up, but I did some tests that look all right, and for some set-up,
manage to increase the time-step.

You will have to set:
 momImplVertAdv=.TRUE.,
 (for now, only available with vectorInvariantMomentum=.TRUE.)
and
 tempImplVertAdv=.TRUE.,
 saltImplVertAdv=.TRUE.,
 (but not yet implemented with all the tracer advection schemes,
  I think 30 & 33 are missing).
You can try this and see if it allows you to use a longer time step.

Cheers,

Jean-Michel 



More information about the MITgcm-support mailing list