You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Adjoint runs fail during the forward portion of the run (i.e. during calculation of the cost function) with CALC_R_STAR errors when I tried a run with salt/tempAdvScheme=33 and didn't specify the VertAdvScheme (set to 3 in ECCO setup). Not sure if this is expected behaviour. Cost function is a simple global temperature sum.
Background: I've spun up a version of the LLC grid forced by CORE2 normal year fields (setup verifications/global_oce_llc90/) which has various differences in the data namelists when compared to the ECCO standard setup. Adjoint runs fail with current options. By comparison with the ECCO parameters and several test runs I determined that it is the differing vertical advection schemes causing the failure. Specifying vertAdvScheme=3 leads to a successful run.
The text was updated successfully, but these errors were encountered:
p.s. also if this is the wrong place to raise this issue please let me know!
This and the @MITgcm issue tracker seem like the best place to get help on this. I cross referenced it in MITgcm/MITgcm#329 in case someone on that side might be able to jump in and might not follow this org
Apologies - this turns out not to be an Adjoint issue. Any forward run fails similarly with salt/tempAdvScheme=33 and salt/tempImplVertAdv=.TRUE. It runs successfully with vertAdvScheme=3. This may be a feature of the implicit vertical advection, there is no documentation I could find. Opened as MITgcm/MITgcm#331
Adjoint runs fail during the forward portion of the run (i.e. during calculation of the cost function) with CALC_R_STAR errors when I tried a run with salt/tempAdvScheme=33 and didn't specify the VertAdvScheme (set to 3 in ECCO setup). Not sure if this is expected behaviour. Cost function is a simple global temperature sum.
Background: I've spun up a version of the LLC grid forced by CORE2 normal year fields (setup verifications/global_oce_llc90/) which has various differences in the data namelists when compared to the ECCO standard setup. Adjoint runs fail with current options. By comparison with the ECCO parameters and several test runs I determined that it is the differing vertical advection schemes causing the failure. Specifying vertAdvScheme=3 leads to a successful run.
The text was updated successfully, but these errors were encountered: