[Mitgcm-support] more Obcs ruminations.

[mitgcm-support at dev.mitgcm.org]

Hi Dan:  data file looks good. 

Would be interesting to know how the
adjoint results depend on cg2dTargetResidual. Veronique Bugnion once
needed a value as small as 1.e-17, although I've used 1.e-8 nowadays
without a visible problem. 

I've checked EXACTCONSERV very closely with Jean-Michel Campin. It makes
only a very small difference, so I've always used exactConserv = .FALSE. 

I've never checked useEnergyConservingCoriolis and I just use the
default, FALSE.

No idea about the impact of lopping schemes.

To my knowledge, nobody has ever used the adjoint of a higher order
advection scheme. We're scared because of the liberal use of intrinsic
functions.

Jake

On Wed, 12 Mar 2003, Daniel Lea wrote:

> 
> Here is my data file. Anything obviously wrong?
> 
> # ====================
> # | Model parameters |
> # ====================
> #
> # Continuous equation parameters
>  &PARM01
>  tRef=  7.0 ,  6.9 ,  6.1 ,  5.1 ,  4.1 ,
>         3.0 ,  2.0,
>  sRef= 35.04, 35.25, 35.23, 35.20, 35.10,
>        34.99, 34.90,
>  viscAz=1.E-3,
>  viscA4=2.5E10,
>  no_slip_sides=.FALSE.,
>  no_slip_bottom=.TRUE.,
>  diffK4T=2.5E10,
>  diffKzT=1.E-5,
>  diffK4S=2.5E10,
>  diffKzS=1.E-5,
>  beta=1.E-11,
>  f0=1.e-4,
>  tAlpha=2.E-4,
>  sBeta =7.4E-4,
>  gravity=9.81,
>  rigidLid=.FALSE.,
>  implicitFreeSurface=.TRUE.,
>  exactConserv=.TRUE.,
>  eosType='LINEAR',
>  readBinaryPrec=64,
>  writeBinaryPrec=64,
>  hFacMin=0.1,
> 
>  useEnergyConservingCoriolis=.TRUE.
> 
> # Implicit works on portland compilers but not gnu77
> 
>  implicitDiffusion=.TRUE.,
>  implicitViscosity=.TRUE.,
>  &
> 
> # Elliptic solver parameters
>  &PARM02
>  cg2dMaxIters=1000,
>  cg2dTargetResidual=1.E-13,
>  &
> 
> # Time stepping parameters
>  &PARM03
>  startTime=0,
>  nTimeSteps=65700,
>  deltaTmom=480.0,
>  tauCD=321428.,
>  deltaTClock =480.0,
>  cAdjFreq=0.,
>  abEps=0.1,
>  pChkptFreq=0.0,
>  chkptFreq=864000.0,
>  dumpFreq= 432000.0,
>  tauThetaClimRelax=2592000.0,
>  tauSaltClimRelax=2592000.0,
>  movieFreq=86400.0
>  &
> 
> # Gridding parameters
>  &PARM04
>  usingCartesianGrid=.FALSE.,
>  usingSphericalPolarGrid=.TRUE.,
>   delZ= 1.000000e+02, 2.000000e+02, 2.000000e+02, 5.000000e+02,
>         5.000000e+02, 1.000000e+03, 1.000000e+03,
>   phiMin=58.,
>  delY=144*0.08333333,
>  delX=156*0.166666667,
>  &
> 
> # Input datasets
>  &PARM05
>  bathyFile='bath3newopen126.bin',
>  hydrogThetaFile='T.final_y1.bin',
>  hydrogSaltFile='S.final_y1.bin',
>  uVelInitFile='U.final_y1.bin',
>  vVelInitFile='V.final_y1.bin',
>  pSurfInitFile='Eta.final_y1.bin',
>  &
> 
> 
> 
> On Wed, 12 Mar 2003, Chris Hill wrote:
> 
> > Dan,
> >
> >  I'll do a survey of people here and see what they suggest. Meantime,
> > any pictures or thoughts on whether noise is related to steep topo.,
> > thin shaved cells or at open boundaries or something else etc.... Also
> > the values you have in data and data. files might give a clue -
> > especially lopping thresholds ( hFacMin etc...).
> >
> >  Alistair - I remember Jake telling me that you had given him some fixes
> > to help with OBCS recently.
> >  Jake (who is skiing this week) any comments?
> >
> > Chris
> >
> > > -----Original Message-----
> > > From: Daniel Lea [mailto:daniel.lea at jhuapl.edu]
> > > Sent: Wednesday, March 12, 2003 12:47 PM
> > > To: Chris Hill
> > > Subject: grid scale oscillations
> > >
> > >
> > >
> > > Hi Chris,
> > >
> > > I spoke too soon last week! I was working with a non shaved
> > > cell version
> > > of the model I've switched to a shaved cell version and the grid scale
> > > oscillations are back. They are not strong and the model runs OK
> > > otherwise.
> > >
> > > As I said my resolution is 1/6 zonal by 1/12 meridional. I've tried
> > > playing around with the vertical and horizontal (biharmonic)
> > > diffusion and
> > > viscosity to no avail. Changing the advection scheme from 2nd
> > > order to 4th
> > > order is not helpful. I haven't tried a flux limited scheme but don't
> > > think that would be good for the adjoint...
> > >
> > > Should I smooth the topography somehow? (A very basic smoothing didn't
> > > help). What about conservation: E.g.  exactConserv or
> > > useEnergyConservingCoriolis?
> > >
> > > Thanks
> > >
> > > Dan
> > >
> > >
> > >
> > >
> >
> 
>