[Carpet] box boundaries

[Erik Schnetter]

On Nov 23, 2006, at 15:37:20, Ulrich Sperhake wrote:

> Hi,
>
> I have a brief question concerning the nature of a given processor's
> boundaries. In the uni-grid case, the cactus variable cctk_bbox  
> provided
> information on the nature of such a boundary (1 for outer boundary,  
> 0 for
> processor boundary). Is there any equivalent which also  
> distinguishes in
> Carpet between processor boundaries and refinement boundaries? The
> cctk_bbox variable gave me 0 for every boundary but the physical outer
> boundary.

No, there is no such variable.  Note that different parts of the same  
boundary may be filled either by inter-processor communication or by  
prolongation; consider e.g. an L-shaped refined region which is split  
into two rectangular components.

You can query Carpet -- the class "dh", to be exact -- about which  
grid points belong to what kind of boundary.  You can then e.g.  
iterate over all inter-processor or all prolongation boundaries.

These mixed boundaries are slightly less efficient, since they need  
to be as wide as the number of buffer zones dictates, i.e.,  
technically they then contain more ghost zones than other boundaries  
(only Cactus doesn't tell you this).  It would be possible to forbid  
this case, so that L-shaped refined regions need to be split into at  
least three components.

There is also a question about stability of the concave corners.  I  
think that Pretorious does not want to handle this case because he is  
worried about stability.

-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/20061123/f1af5261/attachment.pgp