[MITgcm-support] SSH drift and obcs again

[Thomas Haine]

Folks, 

I've been able to isolate the SSH drift/obc issue I reported a few weeks
ago to a specific part of the code. My expertise is dwindling as I get
deeper in, so maybe someone else can help:

Configuration: checkpoint57p_post, GFDL ic., linear free sfc, seaice
off, evap and precip off, obc sponge off, obc orlanski off. Realistic
Denmark Strait bathy. at 2km resolution and open bdys at (parts of) all
4 walls. I've corrected the OBCS_BALANCE code to zero the net transport
divergence so eta won't drift. (N.B., the existing OBCS_BALANCE code has
several bugs - when I get all this sorted out I'll provide a cleaned up
version).


i) What works: obcs switched on at N & S boundary with net barotropic
inflow at the north and outflow at the south. I see no drift in eta to
machine precision and small O(1e-14) cg2d Sum(rhs). Looks good.

ii) What doesn't work: as above but switch off northern obc (in
data.obcs). Now eta drifts, cg2d rhs is O(1) and w diverges after 2
steps. Looks bad!

iii) Also doesn't work (dies at step 2): 
as i) except: pkg/obcs FALSE,
as i) except: N obc closed and E or W obc open.

iv) Also works: As i) except with solid walls at all edges in my bathy
file and pkg/obcs off. Reassuring.

The pressure solver source is vastly different in i) and ii) at the
first solve. This leaves little room for the problem: it's somewhere in
cg2d and related code in relation to obcs. Any ideas where?

Thanks! Tom.