[Mitgcm-support] Re: detatfreesurf == deltatracer

[mitgcm-support at dev.mitgcm.org]

I was more concerned that it seems like its an ad-hoc choice
numerically that I can't rationalize numerically. There must 
be some sort of stability criteria that underlies it. 

I can see that the free surface flux bc's need to have a 
correct notion of volumetric change but it doesn't feel like that
should be fixed by adjusting the dynamics. I would have 
expected to fix that by adjusting the solver source term
appropriately, but leaving the dynamical equations alone.
Fixing flux bc's by changing all the relaxation only
experiments seems weird without some sound numerical
reasoning.

Chris 

On Tue, 25 Jun 2002, Alistair Adcroft wrote:

> For natural boundary conditions it is clear it can only be
> deltaTtracer, right J-M? Aside from that, I don't know what
> you mean by ssh/H << 1 is violated? The variability in ssh
> leads to large "surface correction" tendancies which seem
> to be the source of the blow-up. Before the blow-up, the
> tracer fields were heading the wrong way well be large ssh
> anomalies appeared.
> 
> On the basis that asynchronous integration introduces
> two different time-steps for tracers and momentum,
> a third is clearly introducable for the barotropic mode.
> What it should be set to is probably quite arbitrary except
> for the limitation by natual boundary conditions.
> 
> Are you concerned about the extra cg2d iterations?
> 
> A.
> 
> Chris Hill wrote:
> > Its not obvious to me that setting dt_fs to tracer dt is
> > getting to the full answer. For the 4 degree runs
> > the equatorial grid spacing is about 400km and so the
> > dt=1200 secs we were using doesn't even allow time
> > for a free surface wave to travel 1 dx (which
> > would require roughly 2000 secs). The result will be
> > big ssh changes such that the relation ssh/H is small
> > is violated. This is presumably what leads to the blow up.
> > 
> > However, I can't see how the switch to dt = dttracer
> > is argued numerically. What is needed is a dt such
> > that ssh/H always stays small everywhere I think?
> > Should we add a ssh/H test that traps this error
> > for non NL freesurf cases? Based on that quantity
> > we could maybe actually choose deltatfreesurf
> > automatically and avoid having everyine ask what it
> > should be and why!
> > 
> > Chris
>