Hi,
At least Tanja, Roland, myself and Erik cannot join the ET call tomorrow, so let's cancel the meeting for this week.
However: in order not to lose focus in the ET paper it would be nice to have some progress report for the largest missing sections, e.g., the examples, on this mailing list.
For the TOVs:
I ran a couple of resolutions of a standard TOV model and can easily see the fundamental mode. It is pretty clear that in order to get a good measurement of the frequency the resolution cannot be laptop or workstation-friendly, but that is to be expected. I cannot easily see or measure any overtones though - it's a nice, damped sine. I will finish to run the high-res simulations and then include that frequency measurement in the paper, comparing with the expected result.
Frank
For the bbh simulation:
I still have problems with NaNs around 8-10M of evolution. The parameter file I prepared BBHLowRes.par quits always around 8.5M of evolution time (I have tried a few parameter combinations when trying to nail down this issue). The Nan mask indicates large values coming from the boundary closer to the Pi symmetry boundary for the second finest level. The same problem occurs for the bbh.par parameter file (it quits at 10.5M). Interestingly that doesn't seem to be the case for qc0-mclachlan.par. It runs smoothly until 12M (so far). The most obvious difference is that AHFinderDirect uses PunctureTracker::pt_loc_[xyz] to track origin in the bbh.par case but it doesn't for qc0-mclachlan.par. However I don't see why this would be the cause of these Nans. Maybe PunctureTracker doesn't understand Pi symmetry? In any case, I will need to investigate closely the differences between these two parameter files to find the culprit.
Roland: which Carpet version did you use when you ran qc0-mclachlan.par?
Cheers, Bruno.
Frank Loeffler wrote:
Hi,
At least Tanja, Roland, myself and Erik cannot join the ET call tomorrow, so let's cancel the meeting for this week.
However: in order not to lose focus in the ET paper it would be nice to have some progress report for the largest missing sections, e.g., the examples, on this mailing list.
For the TOVs:
I ran a couple of resolutions of a standard TOV model and can easily see the fundamental mode. It is pretty clear that in order to get a good measurement of the frequency the resolution cannot be laptop or workstation-friendly, but that is to be expected. I cannot easily see or measure any overtones though - it's a nice, damped sine. I will finish to run the high-res simulations and then include that frequency measurement in the paper, comparing with the expected result.
Frank
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Ok, something is broken in the development version. I am running BBHLowRes.par with the Curie release version and I haven't seen any Nans so far.
Cheers, Bruno
Roland Haas wrote:
Hello Bruno,
Roland: which Carpet version did you use when you ran qc0-mclachlan.par?
I used the release version (Curie).
Yours, Roland
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
Hi,
For the distorted rotating black hole case I thought I had the runs completed. However, when I took a look at the waveforms, I saw only 2nd order convergence (very nice 2nd order convergence, though). I had unfortunately not noticed that the default values for WeylScal4::fd_order and Multipole::integration_method resulted in only 2nd order accurate wave extraction. I'm currently rerunning with modified parameter settings.
Is there any reason why the default values for those parameters could not be set to give higher accuracy?
Cheers,
Peter
On Sun, 15 May 2011, Frank Loeffler wrote:
Hi,
At least Tanja, Roland, myself and Erik cannot join the ET call tomorrow, so let's cancel the meeting for this week.
However: in order not to lose focus in the ET paper it would be nice to have some progress report for the largest missing sections, e.g., the examples, on this mailing list.
For the TOVs:
I ran a couple of resolutions of a standard TOV model and can easily see the fundamental mode. It is pretty clear that in order to get a good measurement of the frequency the resolution cannot be laptop or workstation-friendly, but that is to be expected. I cannot easily see or measure any overtones though - it's a nice, damped sine. I will finish to run the high-res simulations and then include that frequency measurement in the paper, comparing with the expected result.
Frank
On 17 May 2011, at 16:39, Peter Diener wrote:
Hi,
For the distorted rotating black hole case I thought I had the runs completed. However, when I took a look at the waveforms, I saw only 2nd order convergence (very nice 2nd order convergence, though). I had unfortunately not noticed that the default values for WeylScal4::fd_order and Multipole::integration_method resulted in only 2nd order accurate wave extraction. I'm currently rerunning with modified parameter settings.
Is there any reason why the default values for those parameters could not be set to give higher accuracy?
It seemed that the simplest and lowest order of accuracy made a sensible default, and would work with any number of ghost zones. I agree that it is a little surprising for someone who always works with higher order methods, but then SummationByParts also defaults to second order, presumably for the same reasons.
For WeylScal4, this would require a larger number of ghost zones than 1. Kranc (unfortunately) does not check that you are using a sufficient number of ghost-zones for the finite difference method in use, so this could lead to hard-to-detect problems for new users. I will see how easy it is to add this check in (https://github.com/ianhinder/Kranc/issues/29).
For Multipole, I am unsure as to the effects of using higher order integration on BBH waveforms. I tested that the integration methods are correct and converge at the required order for test functions, but high order methods often lead to high frequency errors if the data is not smooth, so out of an abundance of caution, I always use the low order method with a very high resolution angular grid (this is not very expensive in a production simulation).
Cheers,
Peter
On Sun, 15 May 2011, Frank Loeffler wrote:
Hi,
At least Tanja, Roland, myself and Erik cannot join the ET call tomorrow, so let's cancel the meeting for this week.
However: in order not to lose focus in the ET paper it would be nice to have some progress report for the largest missing sections, e.g., the examples, on this mailing list.
For the TOVs:
I ran a couple of resolutions of a standard TOV model and can easily see the fundamental mode. It is pretty clear that in order to get a good measurement of the frequency the resolution cannot be laptop or workstation-friendly, but that is to be expected. I cannot easily see or measure any overtones though - it's a nice, damped sine. I will finish to run the high-res simulations and then include that frequency measurement in the paper, comparing with the expected result.
Frank
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
users@lists.einsteintoolkit.org
-
Bruno C. Mundim -
Bruno Coutinho Mundim
-
Frank Loeffler -
Ian Hinder -
Peter Diener -
Roland Haas