[MITgcm-devel] verification/lab_sea/tr_run.salt_plume

[Dimitris Menemenlis]

Jean-Michel, Chris, and Martin, to follow-up on last week's threads and this 
morning's phone call, I ran the following verification/lab_sea tests and 
compared results between forge and a local workstation at JPL.  On forge the 
results alway give

> Y Y Y Y>16<16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16  .  .  .  .  .  .  .  . pass

for all the experiments, so below I only report results on the local workstation.

First, for the currently checked in configuration I get:

> Y Y Y Y>16<16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16  .  .  .  .  .  .  .  . pass  lab_sea
> Y Y Y Y>16<16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16  .  .  .  .  .  .  .  . pass  lab_sea.hb87
> Y Y Y Y>16<16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16  .  .  .  .  .  .  .  . pass  lab_sea.lsr
> Y Y Y Y> 6<12 12  9  8 13 14 10  7  8 10  7  6  8 10  7  6  .  .  .  .  .  .  .  . FAIL  lab_sea.salt_plume

Note that for above configuration, AdvScheme=7 and StaggerTimeStep=.TRUE.

Second for the previous lab_sea.salt_plume configuration, where AdvScheme=7 and 
StaggerTimeStep=.FALSE. I obtain:

> Y Y Y Y>12<13 16 16 16 16 16 16 16 16 16 13 16 13 13 13 16  .  .  .  .  .  .  .  . FAIL  lab_sea.salt_plume

A third test with AdvScheme=2 and StaggerTimeStep=.FALSE. gives:

> Y Y Y Y> 5<11 11  8  9 16 14 10  8  9  9  6  6  8  9  6  6  .  .  .  .  .  .  .  . FAIL  lab_sea.salt_plume

Note that in all test case I have set diffKhS/T=0.0, which may explains why 
AdvScheme=2 remains so sensitive.

What is puzzling is that, contrary to the GAD_CHECK warning, for AdvScheme=7, 
StaggerTimeStep=.FALSE. is much more stable than StaggerTimeStep=.TRUE.

Is it well established that AdvScheme=7 requires StaggerTimeStep=.TRUE. or is 
this only an educated guess?

Dimitris