[MITgcm-devel] [MITgcm-cvs] MITgcm/doc CVS Commit
Jean-Michel Campin
jmc at ocean.mit.edu
Sun Aug 7 17:34:28 EDT 2011
Patrick,
I don't disagree. Just that on 4 proc, the changes in data.grdchk
I was testing were giving the same cost & grad-check
(at truncation level) than the 1 proc (4 tiles) you made.
below are the gradient-check report.
Cheers,
Jean-Michel
Current test:
1 proc (4 tiles), iGloTile = jGloTile = 2, grdchkwhichproc = default :
grdchk output h.p: Id Itile Jtile LAYER bi bj X(Id) X(Id)+/-EPS
grdchk output h.c: Id FC FC1 FC2
grdchk output h.g: Id FC1-FC2/(2*EPS) ADJ GRAD(FC) 1-FDGRD/ADGRD
grdchk output (p): 1 31 7 1 2 2 0.000000000E+00 -1.000000000E-02
grdchk output (c): 1 -4.5678084888051E-02 -4.5678230030508E-02 -4.5677939737319E-02
grdchk output (g): 1 -1.4514659462186E-05 -1.4514692777338E-05 2.2952708652602E-06
grdchk output (p): 2 32 7 1 2 2 0.000000000E+00 -1.000000000E-02
grdchk output (c): 2 -4.5678084888051E-02 -4.5678220067362E-02 -4.5677949660060E-02
grdchk output (g): 2 -1.3520365083411E-05 -1.3520365078179E-05 -3.8693825921143E-10
grdchk output (p): 3 33 7 1 2 2 0.000000000E+00 -1.000000000E-02
grdchk output (c): 3 -4.5678084888051E-02 -4.5678177436933E-02 -4.5677992721924E-02
grdchk output (g): 3 -9.2357504487967E-06 -9.2361978162613E-06 4.8436323424994E-05
grdchk output (p): 4 34 7 1 2 2 0.000000000E+00 -1.000000000E-02
grdchk output (c): 4 -4.5678084888051E-02 -4.5678402417443E-02 -4.5677966121717E-02
grdchk output (g): 4 -2.1814786281896E-05 -2.8567598285188E-05 2.3638010923703E-01
with my modif, in STDOUT.0003:
4 proc (4 tiles), iGloTile = jGloTile = 1, grdchkwhichproc = 3 :
grdchk output h.p: Id Itile Jtile LAYER bi bj X(Id) X(Id)+/-EPS
grdchk output h.c: Id FC FC1 FC2
grdchk output h.g: Id FC1-FC2/(2*EPS) ADJ GRAD(FC) 1-FDGRD/ADGRD
grdchk output (p): 1 31 7 1 1 1 0.000000000E+00 -1.000000000E-02
grdchk output (c): 1 -4.5678084888051E-02 -4.5678230030508E-02 -4.5677939737319E-02
grdchk output (g): 1 -1.4514659459411E-05 -1.4514692777338E-05 2.2954620900739E-06
grdchk output (p): 2 32 7 1 1 1 0.000000000E+00 -1.000000000E-02
grdchk output (c): 2 -4.5678084888051E-02 -4.5678220067362E-02 -4.5677949660060E-02
grdchk output (g): 2 -1.3520365084799E-05 -1.3520365078179E-05 -4.8958370690855E-10
grdchk output (p): 3 33 7 1 1 1 0.000000000E+00 -1.000000000E-02
grdchk output (c): 3 -4.5678084888051E-02 -4.5678177436933E-02 -4.5677992721924E-02
grdchk output (g): 3 -9.2357504463680E-06 -9.2361978162614E-06 4.8436586378320E-05
grdchk output (p): 4 34 7 1 1 1 0.000000000E+00 -1.000000000E-02
grdchk output (c): 4 -4.5678084888051E-02 -4.5678402417443E-02 -4.5677966121717E-02
grdchk output (g): 4 -2.1814786280508E-05 -2.8567598285186E-05 2.3638010928556E-01
On Sun, Aug 07, 2011 at 04:23:34PM -0400, Patrick Heimbach wrote:
>
> Hi Jean-Michel,
>
> You can't change the tile index to (1,1), it needs to stay at (2,2).
> Reason is that the cost function is meridional heat transport
> across 26N in Atlantic, and over 10 timesteps it's not
> sensitive to changes in the South Pacific.
> If you move the test locations too far away from the cost location,
> you'll be testing numerical noise.
>
> -Patrick
>
> On Aug 7, 2011, at 4:06 PM, Jean-Michel Campin wrote:
>
> > Hi Patrick,
> >
> ...
>
> > There is only global_ocean.90x45 that might still have some issues with
> > gradient-check specification when using MPI.
> > I tried changing in data.grdchk:
> > iGloTile = 1, (instead of 2 in non-mpi run)
> > jgloTile = 1, (instead of 2 in non-mpi run)
> > grdchkwhichproc = 3, (instead of the default)
> > and got the right gradient check but in STDOUT.0003
> > (which is not tested by testreport).
> >
> > could see with you when you have time.
> >
> > Cheers,
> > Jean-Michel
> >
> > On Sun, Aug 07, 2011 at 03:53:51PM -0400, Jean-Michel Campin wrote:
> >> Update of /u/gcmpack/MITgcm/doc
> >> In directory forge:/tmp/cvs-serv1317
> >>
> >> Modified Files:
> >> tag-index
> >> Log Message:
> >> update adjoint verifications changes
> >>
> >>
> >> _______________________________________________
> >> MITgcm-cvs mailing list
> >> MITgcm-cvs at mitgcm.org
> >> http://mitgcm.org/mailman/listinfo/mitgcm-cvs
> >
> > _______________________________________________
> > MITgcm-devel mailing list
> > MITgcm-devel at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-devel
>
> ---
> Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
> MIT | EAPS 54-1420 | 77 Massachusetts Ave | Cambridge MA 02139 USA
> FON +1-617-253-5259 | FAX +1-617-253-4464 | SKYPE patrick.heimbach
>
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
More information about the MITgcm-devel
mailing list