[MITgcm-devel] ad-monitor and adjDumpFreq

[Jean-Michel Campin]

Hi Patrick,

I will check with you what to do about this.
But from my little investigations:
1) it does come from the ADEXCH calls in addummy_in_stepping.F
2) it's not specific to a particular set up
 (I see some diff when changing adjDumpFreq in standard global_ocean.cs32x15
  and also in global_ocean.90x40x15)
3) the problem is likely to be on both side (not accurate ad-monitor
 and side effect from adjDumpFreq coming from the ADEXCH calls)

Cheers,
Jean-Michel

On Tue, Jul 24, 2012 at 07:55:55PM -0400, Patrick Heimbach wrote:
> Hi Jean-Michel,
> 
> not sure.
> It could be related to the adjoint exchanges placed in addummy_in_stepping.F
> which I haven't checked in a while.
> Is it only for the adjoint velocity fields?
> (there could be an issue with where these exchanges are placed
> for tracer fields, compared to velocity fields).
> 
> -Patrick
> 
> On Jul 24, 2012, at 6:56 PM, Jean-Michel Campin wrote:
> 
> > Hi Patrick,
> > 
> > I found a curious behaviour in global_ocean.cs32x15/input_ad.thsice
> > when changing adjDumpFreq (was set to the time-step, i.e., 86400., 
> > and it is now set to 5.days = 5 iters):
> > The AD gradient change a little (15 digits of matching) but the ad-monitor
> > show much larger changes (monitorFreq and adjMonitorFreq are identical
> > and unchanged). 
> > 
> > For instance, with adjDumpFreq=1.day:
> > %MON ad_time_tsnumber             =                 72009
> > %MON ad_dynstat_aduvel_max        =   2.2450171119584E+01
> > %MON ad_dynstat_aduvel_min        =  -3.3833812610162E+01
> > %MON ad_dynstat_aduvel_mean       =   1.6255799062282E-02
> > %MON ad_dynstat_aduvel_sd         =   1.1963888239412E+00
> > %MON ad_dynstat_aduvel_del2       =   1.1141126706330E-02
> > %MON ad_dynstat_advvel_max        =   2.0745134286660E+01
> > %MON ad_dynstat_advvel_min        =  -1.6119016591924E+01
> > %MON ad_dynstat_advvel_mean       =  -4.7604298682127E-02
> > %MON ad_dynstat_advvel_sd         =   1.1114721035496E+00
> > %MON ad_dynstat_advvel_del2       =   9.3959189450550E-03
> > 
> > and with adjDumpFreq=5.days:
> > %MON ad_time_tsnumber             =                 72009
> > %MON ad_dynstat_aduvel_max        =   3.4749459081909E+01
> > %MON ad_dynstat_aduvel_min        =  -3.3833812610162E+01
> > %MON ad_dynstat_aduvel_mean       =  -3.3976996866385E-02
> > %MON ad_dynstat_aduvel_sd         =   1.3545133732673E+00
> > %MON ad_dynstat_aduvel_del2       =   1.4012211638183E-02
> > %MON ad_dynstat_advvel_max        =   2.2962319911420E+01
> > %MON ad_dynstat_advvel_min        =  -2.2903063571621E+01
> > %MON ad_dynstat_advvel_mean       =  -1.3326518558512E-01
> > %MON ad_dynstat_advvel_sd         =   1.3706312764493E+00
> > %MON ad_dynstat_advvel_del2       =   1.4279971249511E-02
> > 
> > I am getting the same ad-monitor (same problem) on both my laptop and 
> > on baudelaire, no MPI, gfortran optfile with -devel.
> > 
> > Does this means that there is a problem with adjDumpFreq ? or with ad-monitor ?
> > or it's just with this particular set-up ? 
> > 
> > Cheers,
> > Jean-Michel
> > 
> > _______________________________________________
> > 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