[MITgcm-devel] seaice code beyond checkoint61j blows for ECCO-GODAE

[Dimitris Menemenlis]

Martin,

The Arctic CS510 configuration crashes after a few days of integration  
for:
new code with CS510 defaults
and metric terms off
and clip_velocities true or false
and use_saltplume true or false

It runs OK with OLD_AND_BAD_DISRETIZATION
the test that we are running is with
use_saltplume=TRUE
clipvelocities=.FALSE.

curious why Hong's global CS510 worked.
the main difference between the setups is the open boundary conditions
(only oceanic here since there is no sea ice at the boundaries)
and the fact that global integration uses reafreshwaterforcing and
nonlinear free surface.

when we run the crashing new configuration with debuglevel=3
it stops at second time step with following error message:
ABNORMAL END: S/R SEAICE_LSR did not converge (uIce)

for good measure we are doing all of above integrations with deltat=120
even though the OLD_AND_SUPERBAD code run stably with deltat=1200

what else can we try?

An and Dimitris

On Apr 29, 2009, at 11:51 AM, Martin Losch wrote:

> Hi all,
> can I summarize, what has happened so far in terms of new sea ice
> code? I am a little confused as to what works and what not. Correct me
> if I am wrong:
>
> A. Martin's experiments of a lat/lon grid:
> 1. 0.25deg rotated regional grid of Arctic ocean with CORE forcing
> (1958-2004) work with new code for
> SEAICE_clipVelocities = .false.,
> SEAICE_zetaMin = 0.,
> SEAICE_advSnow=.true.
> both SEAICE_no_slip=.true. and .false.
> different types of open boundary conditions for sea ice and varying
> oceanic conditions
> 2. 2deg global ("isotropic") grid works
> 3. 1deg and 0.5deg regional ("isotropic") grid of the Weddell Sea
> works for at least 10years (Olaf Klatt's experiment)
>
> B. Patrick's ECCO-GODAE 1deg global runs (are they all iteration 75?)
> 1. blows up after transition from old to new seaice code in  
> iteration 75
> 2. runs (iteration 75?) with SEAICE_OLD_AND_BAD_DISCRETIZATION defined
> and old defaults (e.g. SEAICEadvSnow = .FALSE., SEAICE_clipVelocities
> = .TRUE.,, the rest shouldn't matter)
> 3. blows up with  with SEAICE_OLD_AND_BAD_DISCRETIZATION defined and
>      SEAICEadvSnow = .TRUE.,
>      SEAICEadvSalt = .TRUE.,
>      SEAICEadvAge  = .TRUE.,
>      SEAICE_clipVelocities = .TRUE.,
>
> C. Hong's CS510 experiments
> 1.a pure forward run works with SEAICE_OLD_AND_BAD_DISCRETIZATION
> undefined and use
>     SEAICEadvSnow      = .TRUE.,
>     SEAICEadvSalt      = .TRUE.,
>     SEAICEuseFlooding  = .TRUE.,
> 2. adjoint run breaks with new code, but works with old code (pre-
> checkpoint61j)
> 3. any news?
>
> So in summary, forward simulations seem to work, except for Patrick's
> ECCO-GODAE (which is probably the most important one). What's
> different for that configurations? The only thing I can think of is,
> that Patrick probably does not scale dy with cos(lat), so that there
> are elongated grid boxes near the poles. Could that be a problem
> numerically for the new code? I find it hard to believe, but it is
> true that my lat/lon-runs are all with nearly "quadratic" grid cells,
> and so are the CS-grid cells.
>
> How can we continue with this? How can I help?
>
> Martin
>
>
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel

Dimitris Menemenlis <menemenlis at jpl.nasa.gov>
Jet Propulsion Lab, California Institute of Technology
MS 300-323, 4800 Oak Grove Dr, Pasadena CA 91109-8099, USA
tel: 818-354-1656;  cell: 818-625-6498;  fax: 818-393-6720