Hi,
I have produced a setup which for me appears to cause some
rather severe energy conservation issues in FLASH. What I have done is
modify the standard Sedov test problem to simulate a run with a
mass-based refinement criteria. The central region of the box is fixed
to refinement level 5 and the exterior is fixed to refinement level 4.
The problem appears to evolve correctly until thew shock hits the
refinement boundary. At this point the solution goes all wacky. A
reflecting from the boundary of some sort ought to be expected, as FLASH
doesn't seem to have anything in its prolong and refine operators to
dampen this. A reflected wave is quite apparent in the solution.
However, inspection of the outputted flash.dat file also shows
significant energy loss.
I have included a tarball with the flash.dat file from my run and the
problem directory. I have reproduced this problem on a tru64/Compaq
Fortran Alpha cluster with my own patched version of FLASH and a clean
copy of FLASH2.1 on a Linux/Intel Fortran Intel box. The problem is run
in 2D The differences from the original distributed Sedov test are as
follows:
- central 1/8th of box is fixed at refinement level 5
- remainder of box at refinement level 4
- larger box size (2 by 2 instead of 1 by 1)
- longer run (to t=0.4)
The setup command used was:
./setup sedovref -auto -2d
I have observed that this issue does not appear when using truncation
level refinement criteria. Also, the reflected wave is expected when
using this forced refinement. However, non-conservation in a
conservative hydro scheme is worrying. Similar situations where shocks
must cross refinement boundaries will occur in astrophysical problems
were refinement must be determined by a mass criteria(ex: cosmology
problems).
Thanks,
Colin McNally
McMaster University
This archive was generated by hypermail 2b30 : Wed Aug 14 2002 - 09:48:33 CDT