Hi,
Recently, I found the multipole moment calculations in QuasiLocalMeasures show large errors, e.g. 20% for the second mass moment of my project.
The error exists even in the test data of the thorn but it's around a few percent. But it couldn't be improved even with increased resolution. From a rough investigation, I found this is because the one in coordinate construction which was supposed to be equal to cos(theta) for a stationary black hole deviates a bit, making erroneous value. I didn't go further. I'd like to know your opinion or any experiences on this.
Cheers,
Hee Il p.s. I haven't checked whether there's an update of the thorn in the latest version of ET.
On 24 Jun 2015, at 03:15, Hee Il Kim heeilkim@gmail.com wrote:
Hi,
Recently, I found the multipole moment calculations in QuasiLocalMeasures show large errors, e.g. 20% for the second mass moment of my project.
The error exists even in the test data of the thorn but it's around a few percent.
Note that the "test" directory contains regression tests, which are designed to check that the results haven't changed unintentionally when the code changes. They are not meant to test correctness. As such, there is no claim of accuracy for their results. In fact, they should be as small and fast as possible, which usually means that they are NOT accurate. I'm not saying that there is no problem; just that the parameter files in the "test" directory should not be considered as tests of the correctness of the code.
But it couldn't be improved even with increased resolution. From a rough investigation, I found this is because the one in coordinate construction which was supposed to be equal to cos(theta) for a stationary black hole deviates a bit, making erroneous value. I didn't go further. I'd like to know your opinion or any experiences on this.
Cheers,
Hee Il p.s. I haven't checked whether there's an update of the thorn in the latest version of ET.
Which version of the code are you using?
It has not changed much recently.
$ svn log https://svn.cct.lsu.edu/repos/numrel/LSUThorns/QuasiLocalMeasures/trunk/
------------------------------------------------------------------------ r57 | eschnett | 2015-04-27 19:05:14 +0200 (Mon, 27 Apr 2015) | 1 line
Handle different CCTK_INT kinds ------------------------------------------------------------------------ r53 | eschnett | 2014-10-23 18:41:40 +0200 (Thu, 23 Oct 2014) | 1 line
Use IOUtil parameter for out_group_separator ------------------------------------------------------------------------ r52 | eschnett | 2014-10-23 17:41:45 +0200 (Thu, 23 Oct 2014) | 1 line
Use hyphen as group name separator ------------------------------------------------------------------------ r51 | eschnett | 2014-10-23 16:58:55 +0200 (Thu, 23 Oct 2014) | 1 line
Use hyphen as group name separator ------------------------------------------------------------------------ r50 | eschnett | 2014-10-23 16:40:31 +0200 (Thu, 23 Oct 2014) | 1 line
Use hyphen as group name separator ------------------------------------------------------------------------ r47 | rhaas | 2014-04-18 06:26:51 +0200 (Fri, 18 Apr 2014) | 2 lines
remove commented out code, remove unused variable
------------------------------------------------------------------------ r40 | diener | 2012-10-24 22:04:25 +0200 (Wed, 24 Oct 2012) | 5 lines
An EinsteinExact version of qlm-ks.par. Generates NaN's inside the horizon, but those get's masked by the carpet mask and so does not affect the norms. Passes the test when compared with the data produced by qlm-ks.par (at least on my laptop).
Hi,
2015-06-24 16:59 GMT+09:00 Ian Hinder ian.hinder@aei.mpg.de:
On 24 Jun 2015, at 03:15, Hee Il Kim heeilkim@gmail.com wrote:
Hi,
Recently, I found the multipole moment calculations in QuasiLocalMeasures show large errors, e.g. 20% for the second mass moment of my project.
The error exists even in the test data of the thorn but it's around a few percent.
Note that the "test" directory contains regression tests, which are designed to check that the results haven't changed unintentionally when the code changes. They are not meant to test correctness. As such, there is no claim of accuracy for their results. In fact, they should be as small and fast as possible, which usually means that they are NOT accurate. I'm not saying that there is no problem; just that the parameter files in the "test" directory should not be considered as tests of the correctness of the code.
I understand. But the issue is, even though I'm using sufficiently high resolution for my collapsing star study, the code calculates inaccurately the higher multipole moments for n >0, where n is the order of the Legendre polynomial. So it calculates the mass and the spin very accurately.
Which version of the code are you using?
It has not changed much recently.
$ svn log https://svn.cct.lsu.edu/repos/numrel/LSUThorns/QuasiLocalMeasures/trunk/
r57 | eschnett | 2015-04-27 19:05:14 +0200 (Mon, 27 Apr 2015) | 1 line
I'm using the one in ET branch, /repos/numrel/LSUThorns/QuasiLocalMeasures/!svn/ver/54/branches/ET_2014_11
Thanks,
Hee Il
On 25/06/15 05:07, Hee Il Kim wrote:
Hi, Recently, I found the multipole moment calculations in QuasiLocalMeasures show large errors, e.g. 20% for the second mass moment of my project.[...]
I understand. But the issue is, even though I'm using sufficiently high resolution for my collapsing star study, the code calculates inaccurately the higher multipole moments for n >0, where n is the order of the Legendre polynomial. So it calculates the mass and the spin very accurately.
Dear Hee Il,
thanks for reporting the issue. May I ask: how are you testing the code? Which quantities are you comparing to the output of QLM, and how did you obtain them (a formula, another code, etc.)?
I see that you mention a collapsing star: what multipoles are you calculating in that case? QLM only computes the geometric multipoles of 2D surfaces (and not, for instance, of the matter density). Does an apparent horizon form in your simulation, and in that case, is this what you are calculating the multipoles of?
Best, Eloisa
Thanks Eloisa,
2015. 6. 26. 오후 8:50에 "Eloisa Bentivegna" eloisa.bentivegna@ct.infn.it님이 작성:
On 25/06/15 05:07, Hee Il Kim wrote:
Hi, Recently, I found the multipole moment calculations in QuasiLocalMeasures show large errors, e.g. 20% for the second mass moment of my project.[...]
I understand. But the issue is, even though I'm using sufficiently high resolution for my collapsing star study, the code calculates inaccurately the higher multipole moments for n >0, where n is the order of the Legendre polynomial. So it calculates the mass and the spin very accurately.
Dear Hee Il,
thanks for reporting the issue. May I ask: how are you testing the code? Which quantities are you comparing to the output of QLM, and how did you obtain them (a formula, another code, etc.)?
What I have checked is actually to see whether the QLM results satisfy the no-hair theorem of Kerr black holes, i.e., to see whether,
M_0 = M_bh, M_2 = -(J_bh)^2/M_bh, ...
J_1= J_bh, J_3 = -(J_bh)^3/(M_bh)^2, ... , when the bh formed in the simulation becomes stationary.
E.g.,
plot "quasilocalmeasures::qlm_multipole_moments..asc" u 9:(-$23*$23/$13) replot "quasilocalmeasures::qlm_multipole_moments..asc" u 9:15
where $23=M_bh, $13=J_bh, and the 15th column of the data represents the 2nd mass multipole moment of the bh.
I see that you mention a collapsing star: what multipoles are you calculating in that case? QLM only computes the geometric multipoles of 2D surfaces (and not, for instance, of the matter density). Does an apparent horizon form in your simulation, and in that case, is this what you are calculating the multipoles of?
Sorry for the confusion. The multipole, here, means the mutipole moment of the black hole found by the horizon finder (AHFinderDirect), not the multipole moment of matter.
Cheers,
Hee Il
users@lists.einsteintoolkit.org
-
Eloisa Bentivegna -
Hee Il Kim -
Ian Hinder