[ET Trac] [Einstein Toolkit] #181: AEILocalInterp should not off-centre the interpolation stencil by default

Einstein Toolkit trac-noreply at einsteintoolkit.org
Fri Oct 21 03:20:01 CDT 2016

#181: AEILocalInterp should not off-centre the interpolation stencil by default
  Reporter:  hinder  |       Owner:                
      Type:  defect  |      Status:  confirmed     
  Priority:  major   |   Milestone:                
 Component:  Cactus  |     Version:                
Resolution:          |    Keywords:  AEILocalInterp

Comment (by hinder):

 We just ran into this issue again.  However, I don't understand in our
 case why there is a problem.  We are using order=4 Lagrange interpolation,
 which has a 5-point interpolation stencil, and we use 3 ghost and symmetry
 points.  Hence, the interpolator should never have to off-centre the
 stencil, because the interpolation point will always be in the 'default-
 centering-region', as defined in the documentation
 In fact, 2 ghost points should be sufficient.  We are using periodic
 boundary conditions, and the interpolation point is exactly on the
 boundary in one (maybe two) dimensions.  If CarpetInterp always gives the
 point to the component that "owns" it, and the symmetry interpolator the
 same, then setting these options (boundary_off_centering_tolerance=
 boundary_extrapolation_tolerance={0...}) should have no effect.  Instead,
 we see that without these parameters set, there is a symmetry violation (a
 geodesic which should evolve along the cell edge by symmetry diverges from
 it), and with them set, everything is fine.  Further, my understanding was
 that if AEILocalInterp previously decided to off-centre, and you set these
 options to 0, then it would return an error code, which we do not see,
 rather than giving "better" results.  If it could give "good" results
 without off-centering, then why is it deciding to off-center?

Ticket URL: <https://trac.einsteintoolkit.org/ticket/181#comment:5>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit

More information about the Trac mailing list