[Mitgcm-support] Re: planning

[mitgcm-support at dev.mitgcm.org]

Hi Jean-Michel,

 Will be around later, but here are a few comments for your
planning discussions.

 If 1) is critical to 2,3,4 then it would be great 
to have a working CN time-stepping scheme. If you
use the cube-sphere then you could ignore the FFT
issue, which might be quicker. You could still 
verify CN timestepping against the current AB setup
by using non-FFT experiements e.g. exp2, exp0,
hs94.zave, aim.5L_Equatorial_Channel etc....
 
 Not sure what 3) "does not follow topo." means
exactly?

 When you say "Atmosphere, linear free surface with topography"
this is still a general change i.e. the ocean code has this too,
but its free-surface topo. is fixed. Silly question I think!!

 6) can this be done in isolation? Isn't this tied up with
    implicit-gravity waves, XY advection etc....
    Seems like, apart from thinking about, this should
    be part of a later phase.

 Can you decide what you like doing re. maintaining a stable
main branch for the code. On the whole I prefer getting things 
tested in little steps and checked in and checkpointed often i.e. as 
they are done, but that means being responsive if the main-branch 
gets broken - that isn't always easy. 

Chris
 
Jean-Michel Campin wrote:
> 
> Hello Alistair,
> 
> I've thought a little about the next things I will to do after
> c35, and to discuss some points, I made this small list:
> 
> 1) Crank-Nickelson (incorporate modifs in c35 ; needs to see
> what's to do with FFT).
> 
> Atmosphere, linear free surface with topography :
> 2) change in cg2d :
>  - eta as a state variable and cg2d_x = eta * Bo(i,j)
>    for the ocean, Bo= gravity , cg2d_x = eta * g
>    for the atmosphere, Bo = alpha(Po_surf(x,y))
> 3) change in calc_gt & calc_gs: I think they are not dealing
> properly with the vertical advection part due to W_surface
> (= a kind of correction) because it does not follow the topo.
> Probably the easiest thing would be to use a mask and
> a new array ksurf(i,j) to compute separately this term
> (outside of the k loop) ?
> Advantages:
> * for the next step (4), this part will simply be ignored.
> * ksurf will be useful for all the surface related computation
>    (Evap, Cdrag and wind stress, ... )
> 
> 4) non-linear free surface : hfact(ksurf) time-varying.
> This part requires some changes in many routines.
> (--> development on a separated cvs branch ?)
> we need to agree on the variable names for initial (= static)
> and time dependent geometric factor : hfact, area, and recipr.
> 
> Concerning the Barotropic part, in a first step,
> (a) change only the right part (cg2d_b) of the 2-D solver.
> or
> (b) compute the matrix and preconditioner at each time
> step with hfact(current_iter).
> I am not sure that (a) is really simpler than (b) since
> all the terms that appears in the left part (A.x) with hfact_0
> have to be added in the right part (b) with (hfact-hfact_0).
> But if we adopt (b), "init_cg2d" will be called
> at each time step, and this might changes few things in the
> code (a new routine ?).
> 
> 5) non-linear free surface : pressure gradient, "b" time-varying
> due to changes in T,S and eta.
> (b= g*rho(T,S,z)/rho_0 in the ocean, and alpha(T,S,p) in atmosphere).
> again 2 solutions, like (a) or (b).
> 
> 6) implicit vertical advection: Pb of conservation if
> Adams-Bashforth time stepping is used ?
> 
> See you on Monday,
> 
> Jean-Michel