[Carpet] gf contamination

[Erik Schnetter]

Hi Yosef,

I haven't seen this behaviour myself.  It could be caused by many  
things -- an error in Carpet's synchronisation, or an error in your  
code, etc.  I suggest that you fill the grid points with nan instead  
of 1e+4, and then run the NaNChecker to catch problems.  This should  
let you pinpoint the source of the problem.  Once you know where it  
comes from (prolongation, inter-processor synchronisiation,  
restriction, MoL, symmetries, boundary condition, wherever), we can  
correct it.

-erik

On Aug 15, 2006, at 16:20:09, Yosef Zlochower wrote:

> Hi,
>
>    I found that the problem goes away (at least for some
> tests) when I run on 1 processor. The problem also goes
> away if I set the number of ghost zones to be equal to
> the number of integration substeps times the unigrid
> ghost size (i.e. 9 ghost zones for RK3). The number
> of buffer points seems to have no effect.
>
> I have tried to use standard refinement in time,
> no refinement in time, 0th and 1st order time
> prolongation, 3rd, 5th, and 7th order
> spatial prolongation. In all cases I see the
> same contamination.
>
> Yosef
>
>
>
>
> Yosef Zlochower wrote:
>> Hi,
>>
>>    I am having some difficulty when evolving my code with carpet.   
>> I am
>> attempting to evolve my code with prolongation in space only.  To
>> better understand the problem I set up my code so that the RHS were
>> all filled with 1.0e4 prior to calculating them. My code then  
>> fills in
>> the RHS for all points on the grid except those on the three
>> gridplanes closest to the boundary (i.e. for (i=3; i<cctk_lsh[0]-3;
>> i++) etc ). In MoL_PostStep I first synchronize all the evolved  
>> fields
>> (but not the RHS) and then alter the values of \tilde g_{ij} and
>> \tilde A_{ij} so that the algebraic constraints are enforced. I am
>> working on a problem where gxy = 0. However, I found that after 1
>> timestep the value of gxy 1 point in from the buffer zones is
>> non-zero, and grows when I increase the initial value of the RHS to
>> 1.0e5. So it looks like some junk from the buffer zones has leaked
>> into the evolved fields.
>>
>> Here is the relevant segment of my parfile please note that my code
>> only needs 3 ghostzones but I choose 4. I also set
>> num_integrator_substeps = 4 even though 3 would be sufficient.
>> (setting it back to 3 does not remove the problem)
>>
>>
>> driver::ghost_size = 4
>> mol::ode_method                        = "RK3"
>> driver::global_nx                       = 81 #51 #227 #259
>> driver::global_ny                       = 81 #51 #227 #259
>> driver::global_nz                       = 81 #51 #116 #132
>> grid::dxyz                              = .25
>> grid::domain                            = "full"
>>
>> time::dtfac                             = 0.125
>> #Carpet::prolongation_order_space   = 7
>> Carpet::prolongation_order_space   = 1
>> Carpet::prolongation_order_time    = 0
>> Carpet::max_refinement_levels      = 3 #5
>> #Carpet::buffer_width               = 0
>> Carpet::time_refinement_factors = "[1,1,1]"
>> Carpet::num_integrator_substeps = 4
>> Carpet::use_outer_buffer_zones = "yes"
>> #Carpet::init_each_timelevel = "yes"
>>
>> CarpetRegrid::refinement_levels    = 3 #5
>> CarpetRegrid::regrid_every         = 0
>> CarpetRegrid::refined_regions = "centre"
>>
>> Is there something obviously wrong in the parfile?
>>
>> I examined gxy on the y axis and found that after the first time step
>> gxy on level 0 is zero except at y=+-5 and +-2.5 On level 1 gxy has
>> the same behavior but is also non-zero at the points directly above
>> and below +-5. On the finest level gxy is non-zero on +-2.5 and +-
>> 2.4375.
>>
>> Yosef
>> _______________________________________________
>> developers mailing list
>> developers at lists.carpetcode.org
>> http://lists.carpetcode.org/listinfo/developers
>
> _______________________________________________
> developers mailing list
> developers at lists.carpetcode.org
> http://lists.carpetcode.org/listinfo/developers
>


-- 
Erik Schnetter <schnetter at cct.lsu.edu>

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.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: PGP.sig
Type: application/pgp-signature
Size: 186 bytes
Desc: This is a digitally signed message part
Url : /archives/developers/attachments/20060927/f449f9c3/attachment.pgp