Hello all,
I am wondering if there would be anything wrong with running ML_ADMConstraints with just a single time level for the constraints as well as using no SYNCs for it as long as I set it's calc_every parameter to the frequency of the coarsest level.
Obviously this will give my "nonssense" values in the ghost zones, buffer zones and boundary (though on the boudary I can just set it to zero).
However since the constraints are pure output and nothing ever depends on them the bad values only matter if they were to eg show up in the norms. It is my recollection that CarpetReduce actually does not use any values from the ghost zones (certainly that would be wrong) or buffer zones (this may be more recent).
Is there any any other way the "bad" values could make it into the norms? Note that I do not worry about 1d, 2d, 3d output for the constraints.
Yours, Roland
On Thu, Sep 22, 2016 at 08:14:42PM -0500, Roland Haas wrote:
I am wondering if there would be anything wrong with running ML_ADMConstraints with just a single time level for the constraints as well as using no SYNCs for it as long as I set it's calc_every parameter to the frequency of the coarsest level.
Obviously this will give my "nonssense" values in the ghost zones, buffer zones and boundary (though on the boudary I can just set it to zero).
However since the constraints are pure output and nothing ever depends on them the bad values only matter if they were to eg show up in the norms. It is my recollection that CarpetReduce actually does not use any values from the ghost zones (certainly that would be wrong) or buffer zones (this may be more recent).
Is there any any other way the "bad" values could make it into the norms? Note that I do not worry about 1d, 2d, 3d output for the constraints.
I think this should be fine, provided you not only set calc_every to your coarse step, but also your output frequency for the norms (or adjust your post-processing for that). The former shouldn't even be necessary, but why would you calculate them more often then you output them...
Frank
The ghost and buffer zones are also used for time interpolation, but since you suggest to only evaluate the constraints when the coarse level is active, you are circumventing this.
The largest issue I see is that e.g. a regular L2 norm over the simulation domain is not very useful. It emphasizes very much the coarse grid, which is very large, and where the boundary conditions might do strange things to the constraints. Similarly, you will need to be careful to exclude the interior of the horizons (see CarpetMask) to avoid looking near the punctures. CarpetMask is not smooth in time, so looking at the time evolution of the constraint norm is then also problematic.
You might be more happy if you look e.g. at the "isum" or "inorm2" output. This output is not weighted by coordinate volume, but by grid cells, so that the coarse grid is deemphasized.
-erik
On Thu, Sep 22, 2016 at 9:14 PM, Roland Haas <roland.haas@physics.gatech.edu
wrote:
Hello all,
I am wondering if there would be anything wrong with running ML_ADMConstraints with just a single time level for the constraints as well as using no SYNCs for it as long as I set it's calc_every parameter to the frequency of the coarsest level.
Obviously this will give my "nonssense" values in the ghost zones, buffer zones and boundary (though on the boudary I can just set it to zero).
However since the constraints are pure output and nothing ever depends on them the bad values only matter if they were to eg show up in the norms. It is my recollection that CarpetReduce actually does not use any values from the ghost zones (certainly that would be wrong) or buffer zones (this may be more recent).
Is there any any other way the "bad" values could make it into the norms? Note that I do not worry about 1d, 2d, 3d output for the constraints.
Yours, Roland
-- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net.
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Hello Erik, Frank, all,
yes, I would only compute and output the constraints on the coarse level timestep.
The largest issue I see is that e.g. a regular L2 norm over the simulation domain is not very useful. It emphasizes very much the coarse grid, which is very large, and where the boundary conditions might do strange things to the constraints. Similarly, you will need to be careful to exclude the interior of the horizons (see CarpetMask) to avoid looking near the punctures. CarpetMask is not smooth in time, so looking at the time evolution of the constraint norm is then also problematic.
CarpetMask is active and set up (Ian had thankfully already pointed this out).
You might be more happy if you look e.g. at the "isum" or "inorm2" output. This output is not weighted by coordinate volume, but by grid cells, so that the coarse grid is deemphasized.
Ok, I will add isum and inorm2 to the output.Currently I had norm_inf, norm1 and norm2. Really these are only intended to check for obvious errors in the parameter file the make the contraints grow.
Yours, Roland
users@lists.einsteintoolkit.org
-
Erik Schnetter -
Frank Loeffler -
Roland Haas