LIM2 reproducibility problems

[charris]

I've just had a quick look at the LIM2 reproducibility issue and it seems to be the thermodynamics rather than the dynamics causing the problem. Problem comes from this bit of limthd:

      IF( fcor(1,1) < 0.e0 ) THEN
         DO jj = 1, njeqm1
            DO ji = 1, jpi
               frld(ji,jj) = frld(ji,jj) + 2.0
            END DO
         END DO
      ENDIF

For processors crossing the equator this is OK because njeqm1 is the last row in the Southern hemisphere (seem limmsh) but for processors entirely in the Southern hemisphere njeqm1=jpj-1 so one row is not updated. There are various ways this could be fixed but the simplest is just to add an lbclnk on frld after this loop.

This is definitely a bug but with the standard namelist (i.e. ice parameters used in the lateral accretion the same in both hemispheres) it shouldn't have been exposed. However it seems that the method by which 2.0 is added to frld in limthd and then stubtracted in limthd_lac can lead to differences ~1e-16 in addition to any genuine changes in frld from the lateral accretion.

With the above bug fix I have obtained identical results (with ln_limdyn=.true.) for ORCA1 NEMO-LIM2 DFS-forced 30 day runs on a (fairly limited) range of different processor decompositions on our IBM p6.

[charris]

Fix committed to trunk in [1924].

[nicolasmartin]

Milestone 2010 Stream 0: Coastal ocean ready deleted