On Wed, May 11, 2011 at 12:53 PM, Bruno C. Mundim bcmsma@astro.rit.edu wrote:
Hi Roland:
Roland Haas wrote:
Hello Bruno,
a) Its evolution is stopped with Nans around 8.6M. Nans appear to come from the BSSN fields, but still needs further investigation.
Hmm I could evolve the qc0 example in McLachlan. Maybe there are differences in the parameter file that make the error "obvious".
A few obvious changes that I can think of:
- I reduced the number of spherical surfaces from 5 to 3. Now the
puncture tracker uses surfaces 0 and 1, AHFinderDirect track those surfaces and load AH related info there. QLM, CarpetTracker and CarpetMask uses those 2 surfaces then. I will go back to 5 surfaces as it was set before. Maybe I am missing some subtle scheduling order here. I don't know.
- I added a few general parameters I am used to use to CarpetInterp,
CarpetLib and CarpetRegrid2.
- From a discussion I had with Peter on the ET workshop, I decided
to change the AHFinderDirect interpolator to Hermite.
So I will switch back to what it was and avoid to do so many changes at once! :) However I find strange that these changes could be the cause of what I am seeing.
c) When CarpetReduce::weight is output, its values seem strange to me. Most of the domain has 1/2 set. A few points close to the boundary 1/4 and the boundaries 0 (physical and ghost). No mask out values inside AH is shown, ie they still are set to 1/2. These values seem to make more sense to AMR restriction operations.
Are you maybe looking at the z=0 plane? That plane has a weight of 1/2 since only half of the volume of each cell is used when eg. computing the simulation volume. I have never used CarpetMask myself (but have written a similar thorn of my own... would have paid to look into Carpet first I guess :-) )
Yes I was looking at the z=0 plane and had refection symmetry set up! Thanks for pointing this out. If we go out of the plane then the weight=1.0 values show up. A more careful look into the data, I noticed the zeros are set in the buffer zones indeed with the exception of the third point (ghostzone=3) in the domain (from the AMR boundary) that has its value set to 1/4. Do you know why?
For some reason that I currently cannot recall, the ghost zones in the mask may be set to 1 instead of 0. However, ghost zones are ignored anyway by a different mechanism when reducing (by using cctk_bbox and cctk_nghostzones).
The particular values indicate how much of the "control cell" surrounding this grid point should be taken into account when reducing. If the value is 1/4, then you may be looking at an edge point (e.g. and x-y edge in the z=0 plane), where 3/4 of the control cell are already covered by the corresponding coarse grid.
One important test is that the sum of all weights times the grid cell volume equals the total volume of the simulation domain. If you look into the simulation log output and search for "volume" there, you should find a corresponding statement.
-erik