[Carpet] interpolator problem

[Erik Schnetter]

On Sep 27, 2006, at 15:07:25, Yosef Zlochower wrote:

> Hi,
>
>   I am having some difficulty interpolating analysis variables (
> specifically psi4).
> I give the variables global storage of 1 time level. I calculate
> the analysis quantities with the schedule option
> OPTIONS: GLOBAL loop-local
> I then determine which level is the finest level containing
> a given sphere of points (for a later spherical harmonic
> decomposition).
> I then call a routine which will interpolate these points with the
> the options
> OPTIONS: Global loop-level.
> Inside the routine I check if the refinement level is the one
> I found previously, and if it is, call CCTK_InterpGridArrays.
> To complicate the issue a little, the gfs are actually calculated
> and interpolated twice during the analysis step (because I use
> two different tetrads).
>
> This works for most spheres, but in the case of a sphere
> with radius 5, the global interpolator seems to pass the interpolation
> points to the wrong processor. The interpolation then fails at the
> local level with errors like:
>
> WARNING[L1,P0] (AEILocalInterp):
>    CCTK_InterpLocalUniform():
>         interpolation point is either outside the grid,
>         or inside but too close to the grid boundary!
>         (this may be caused by a global interpolation with
>          driver::ghost_size too small)
>         0-origin interpolation point number pt=4 of  
> N_interp_points=16137
>         interpolation point (x,y,z)=(0.294297,-0.00577926,4.99133)
>         grid x_min(delta_x)x_max = -5.87305(0.125)1.12695
>         grid y_min(delta_y)y_max = -5.87305(0.125)1.12695
>         grid z_min(delta_z)z_max = -5.99805(0.125)1.00195
>
> (note that the interpolation point is in the +z bitant while
> this processor owns the -z bitant)

I have looked at this error, and I don't know where it comes from.   
There are several possibilities:

- The grid point is not actually on this level, and there is an  
earlier warning which you didn't show, or Carpet "forgets" to output  
the corresponding warning
- Interpolation in level mode is not well tested; I think most people  
use global mode
- This error may have been corrected recently; did you try the most  
recent version?

Otherwise I'm afraid I cannot help you without further details.  As a  
first step, I would need to know what the routine map_points in  
CarpetInterp does, and how it gets things wrong.  You could add some  
print statements to this routine to see whether it does what it should.

-erik

-- 
Erik Schnetter <schnetter at cct.lsu.edu>

My email is as private as my paper mail.  I therefore support encrypting
and signing email messages.  Get my PGP key from www.keyserver.net.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : /archives/developers/attachments/20061011/f1233726/attachment.pgp