[Mitgcm-support] Re: possible problem in gmredi

[mitgcm-support at dev.mitgcm.org]

That sign error is exactly "the problem" we were alluding to
when you visited. The fix, I think, is to write code for the
unstable and stable cases. The bolus fluxes should switch
direction but the diffusive should still be down gradient so
doing it in gmredi_slope_limit isn't sufficient; we need to add
the the K13 and K23 components to do it propoerly.

I think the fix that NCAR use (speculation) is to consider
unstable stratification as very large slope ie. scale it all
to zero.

I have code that does this but I'm not at work 'til Wed.

See you then,

A.