[Carpet] init_3_timelevels

[Scott Hawley]

Erik,
   Thanks for this clarification.  I'm happy to take the blame for this 
implementation error...although it would strange indeed since this 
error would have invalidated the whole point of implementing the 
scheme!  ;-)
       I was concerned that it was a design error --- the caption in 
Figure 5 of our paper agrees with your analysis of the way things 
SHOULD be.  Evolve forward & backward first, then interpolate.  It's 
essentially a recursive algorithm as far as setting up the "hourglass" 
structure, and then after that we do some additional evolution in order 
to "appease" Cactus.
      It's in these additional evolutions that the "fudge"/error arises, 
and this is a design decision --- as we say in the paper, one COULD do 
more work if one wanted to be "anal" about it, but at the time we 
didn't think it was necessary, and our convergence tests indicated that 
indeed more "work" wasn't necessary.
-Scott


--
Scott H. Hawley    Office RLM 9.206     
http://wwwrel.ph.utexas.edu/~shawley
Center for Relativity, Dept of Physics                            Tel:  
+1-512-471-5426
Univ. of Texas at Austin, Austin TX 78712 USA           Fax: 
+1-512-471-0890

On May 9, 2005, at 12:24 PM, Erik Schnetter wrote:

> On Monday 09 May 2005 18:40, Scott Hawley wrote:
>> Whoa whoa.  What do we mean "wrong order"?    There was considerable
>> thought put into this algorithm...
>> "Reverse order", yes.   If I recall correctly, there was a "trade
>> off" in the sense that certain boundary interfaces don't get handled
>> as consistently as one might like, but these errors converge... Can
>> you be more specific?
>
> The current algorithm evolves first the coarse grid forward in time,
> then the next finer grid.  During that, it prolongates the fine grid
> boundaries from the coarse to the fine grid.  This prolongation uses
> all three time levels of the coarse grid.  The last of the coarse grid
> time levels have not yet been initialised.
>
> The correct thing to do is to evolve the coarse grid first both 
> forwards
> and backwards in time before evolving the next finer grid.  This is
> e.g. shown in the figure "threelev_initdata.eps" in Carpet's
> documentation.
>
> It is quite possible that an older version of Carpet -- e.g. a version
> that was written by S. Hawley instead of "corrected" by E. Schnetter --
> already did this.  When I checked, the version from August 2004 was
> already wrong.
>
> This was unfortunately not the first time where I introduced an error
> into the 3TL initialisation.  I am now going to put some security
> checks in there, so that Carpet remembers which time levels of which
> refinement levels have been initialised and aborts on inconsistencies.
>
> (I am currently working on supporting refinement levels that are not
> refined in time; this requires some changes to the 3TL algorithm, and
> while thinking about these changes I noticed the above error.)
>
> Sorry, Scott.
>
> -erik
>
> -- 
> Erik Schnetter <schnetter at aei.mpg.de>   
> http://www.aei.mpg.de/~eschnett/
>
> 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.
> _______________________________________________
> developers mailing list
> developers at lists.carpetcode.org
> http://lists.carpetcode.org/listinfo/developers