[Mitgcm-support] Re: Unstable 1deg Atl. model

[mitgcm-support at dev.mitgcm.org]

Tom,

First off, what release/version of the code is it?

Second, we can make lots of random suggestions (such as below) but
with a picture of how it's blowing up it's hard to guess what's happening.
If you can produce snap shots of the state a few steps before it dies and
an hour before it dies we can see where and guess what.

In the mean-time, some blind comments:

If the solver stops converging it usually means that the number
Sum(rhs) is non negligible and grows rapidly just before
the solver dies. Is this the case?

Your Munk layer is approx. 100 km thick and so just about resolved/stable.
Evidence that it isn't would be standing grid-scale waves on western
boundary. However, the decay time for the grid-scale due to viscosity
is 145 days. That is quite long and the most likely cause of trouble.
Switching to bi-harmonic and choosing a grid-scale decay time of
a few days (viscA4=1e15) or less will probably help but keep the
dissipation at larger scales moderately small. 

If you use steady forcing (not climatology but keep the forcing fixed
at the first day) does it still die?

The advection scheme for T/S is fourth order and so noisy and might
be better behaved with biharmonic diffusion rather than harmonic.
Try switching to scheme=77 (remove multiDimAdvection=.false.) and
if it becomes stable then we should iterate on the what advection scheme
is appropriate. 

Quadratic bottom drag might be too non-linear for the high frequency forcing.
Try linear drag.

A.