[Mitgcm-support] Ecco with KPP on
mitgcm-support at dev.mitgcm.org
mitgcm-support at dev.mitgcm.org
Wed Jul 9 15:41:38 EDT 2003
HyperNews notification of new message. See:
http://escher.JPL.NASA.GOV:2000/HyperNews/get/forums/assim/105/2.html
I have a "working" version of ecco adjoint with kpp on, which ran for
3-days and gave the following gradient checks:
temperature:
original forward: sumfc = 5617.806407700504
pert. original forward: sumfc = 5617.806407723561
tangent linear forward: sumfc = 0.0000000000000000E+00
adoint forward: sumfc = 5617.806407700504
finite diff. sensitivity = 2.3057509679347272E-03
adjoint sensitivity = 2.2139507483762699E-03
linear sensitivity = 0.0000000000000000E+00
linear sensitivity = 0.0000000000000000E+00
taux:
original forward: sumfc = 5617.806407700504
pert. original forward: sumfc = 5617.806490207631
tangent linear forward: sumfc = 0.0000000000000000E+00
adoint forward: sumfc = 5617.806407700504
finite diff. sensitivity = 8250.712744484190
adjoint sensitivity = 5553.394531250000
linear sensitivity = 0.0000000000000000E+00
linear sensitivity = 0.0000000000000000E+00
There is no guarantee however that it will not blowup at a later time
though. The idea is to introduce a number (bcepsi, at Dimitri's
suggestion) to bound the values appearing in the denominator from below
so it won't be too small. bcepsi = 10 right now, which is a rather BIG
epsilon, but it can adjusted. CVS tree has been updated with the new kpp
code.
More information about the MITgcm-support
mailing list