[Carpet] buffer zones

[Ian Hinder]

On Fri, May 19, 2006 at 04:24:39PM +0200, Frank Loeffler wrote:
> Hi,
> 
> Ian Hawke wrote:
> > If using internal buffer zones
> 
> Yes, still. I know ...
> 
> > then you nullify the effect of the mesh
> > refinement (as interpolating from the coarse grid loses you the
> > additional accuracy that the fine grid should give you).
> 
> Ok, that is the loss of performance I was talking about. I am more 
> concerned of lower convergence than expected.
> 
> > With _really_
> > too many buffer zones you're just doing an expensive unigrid run with a
> > weird dissipation algorithm!
> 
> I have 6 instead of the needed 4. I think this does not count as _really_?

As far as I can see, for outer buffer zones it should not make any
difference to the output.  Carpet will fill the larger buffer zones
with interpolated data, then your finite difference scheme will update
the "physical" fine grid.  This physical grid should not see any
effect from the extra buffer zones as the points on the physical grid
should be "causally disconnected" from the buffer zones.  After a full
time integrator step, the buffer zones are overwritten again by
carpet.

However, if you are using inner buffer zones, then by increasing the
number of buffer zones and not changing the refinement boundary
location, you are effectively moving the real refinement boundary
inwards.  

-- 
Ian Hinder
hinder at gravity.psu.edu
http://www.gravity.psu.edu/~hinder