[FLASH-BUGS] ppm/kurganov test differences

[Greg Weirs]

Hi Sean,

Actually, I am a bit suprised that the simulation ran at all. In general,
the change you specified in the Config file is not supposed to work. (BTW,
you should change it in the Modules file - Config's are not intended to be
edited by code "users", unless they are willing to dig into code. If you
used the Modules file to do it, the setup script would not allow you to
build the combination of modules you specified.)

The Kurganov module is designed for "alternative" drivers which are not
used by default. These alternative drivers allow you to specify the time
advancement algorithm independently of the spatial discretization (hydro
in this case), and currently you can choose between Euler explicit, a
third-order Runge-Kutta method, and two versions of Strang-splitting, both
of which are different (in software) than the Strang splitting method used
by PPM.

To use the kurganov hydro module, use one of the alternative
time-advancement methods, as follows - 

1.)   ./setup sod -auto

2.) Edit the Modules file:

Replace the hydro and driver lines by
INCLUDE hydro/explicit/delta_form/kurganov
INCLUDE driver/time_dep/delta_form/strang_state

Add a line for the formulation module
INCLUDE formulation/state_form

3.) ./setup sod
4.) gmake

That should build the same algorithm as you get by default, except with
the kurganov hydro instead of PPM. To use the other time advancement
algorithms, please refer to the User's Guide.

Other things you should be aware of:
In general the kurganov hydro requres a smaller time step than PPM; set
the cfl=0.4.

These modules will only work on "pure" hydro problem - no reactions, no
diffusion, no gravity, just the Euler equations. Tracking multiple
species (abundances) should work, but has not been rigorously 
tested. Also, just Cartesian coordinates. We are changing FLASH to
eliminate these restrictions, and to make the "alternative" modules
compatible with all the others, but that will not be available for a
while.

Greg

__________________

Greg Weirs
weirs at flash.uchicago.edu
773 834 3228
__________________

On Mon, 26 May 2003, Sean Matt wrote:

> Hello again,
> 
> 	We have recently been testing and comparing the ppm and kurganov
> hydro solvers, and we've encountered an interesting problem.  
> Specifically, we are running the standard sod problem in setups/sod.
> We setup with:
> 
> setup sod -auto -maxblocks=1000
> 
> And, we are running on Tru64 (OSF1 V5.1) using a Compaq fortran compiler
> (V5.5-1877-48BBF).  
> 
> 	We can run the standard run with ppm (without modifying
> "flash.par" or any other files) and get a result that is consistent with
> the true sod solution.  However, when we make only one change to the
> Config file to select the kurganov solver, the computed solution is wrong
> by exactly a factor of two.  The change we make to Config is to replace
> the line:
> 
> REQUIRES hydro
> 
> with:
> 
> REQUIRES hydro/explicit/delta_form/kurganov
> 
> Then, of course, we rerun the setup command and recompile.  We make no
> changes to the flash.par file (except to basenm for a different file
> name).  I've attached two gif's of 1-d slices in density of the solutions,
> so that you can see the error.  ppm.gif is the solution from ppm and
> kurg.gif is the solution from the kurganov solver, and each are from the
> final checkpoint file at a physical time of t=0.2.  As you may know, the
> true Sod problem solution looks the same at different times, but one
> simply has to rescale the x-axis proportional to time.  It is quite
> interesting and baffling that the kurganov solution is off by exactly a
> factor of two.  That is, if you rescale the x-axis of the kurganov
> solution by a factor of two, it would then match with the ppm.  In other
> words, the kurganov solution is behaving as though everthing is moving at
> half speed.
> 
> Have we done something wrong?  It is worth noting that we have also
> compared the two solvers for the standard sedov setup, and the shock speed
> is the same for both solvers (i.e., both solvers give the "right"  
> answer, in that setup).
> 
> 
> 		-Sean
> 
> 
>