[Carpet] weight is initialised differently with use_grid_structure_from_checkpoint feature turned on
Erik Schnetter
schnetter at cct.lsu.edu
Wed Feb 21 16:07:22 CET 2007
On Feb 21, 2007, at 08:04:09, Thomas Radke wrote:
> Hi,
>
> various people had reported the problem of norm output being
> slightly different after recovery compared with the values from the
> original checkpoint run. Bruno created the attached WaveToy
> parameter file which can be used to easily reproduce the problem.
>
> I found that this happens only if
> IOHDF5::use_grid_structure_from_checkpoint is turned on, and buffer
> zones are being used. Disabling this feature does produce exactly
> the same values for all norms.
>
> Luckily the problem doesn't seem to be in the recovery code or in
> the Carpet scheduler, affecting the entire evolution. In addition
> to the norms, I also compared the full phi function as 3D ASCII
> data and found it to be bitwise-identical. This indicates that the
> problem is related to the calculation of the norms only. Indeed,
> comparing the weight function in 3D ASCII does show differences in
> the buffer zones. Somehow the weight is initialised differently
> after recovery.
>
> Is that enough information for you, Erik, to fix the problem ?
I looked at this problem yesterday, and I think I corrected it
yesterday with a patch to the development version. The problem was
indeed that the weights which are used for reduction operations were
not initialised correctly.
My patch changed things to that, after recovery, the postregrid bin
is run for all time levels. This should initialise the weight (and
all other such grid functions with time levels) correctly. I checked
with Bruno's parameter file, and it works now correctly for me.
I asked Bruno to check whether this solves his problem. Is there is
a problem left over which I did not catch?
-erik
--
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/20070221/03835ab2/attachment.pgp
More information about the developers
mailing list