Updated BBH example for higher accuracy waveforms #267
Conversation
There was a problem hiding this comment.
This all looks great, just some minor stylistic changes.
Whilst we are here and making this example efficient, can we please define constraint_calculation_level=0 and wrap the constraint calculation into a at_level_timestep_multiple(constraint_calculation_level) as we do for the Weyl data since this is otherwise calculating at every post timestep, and it only outputs every coarse timestep so I can't see why that is needed.
Also it would be good to update the usual params.txt file for the better params - especially to rescale the sigma value.
Another unrelated point that we could fix is that the KerrBH example has an error in the params.txt - the lower boundary should be 2 2 2 not 0 0 0, and the max level should probably be 6 not 3.
We will also need to update the wiki for these changes, and having the speeds there for each set of params will be very useful.
Examples/BinaryBH/BinaryBHLevel.cpp
Outdated
| #include "SixthOrderDerivatives.hpp" | ||
| #include "SmallDataIO.hpp" | ||
| #include "TraceARemoval.hpp" | ||
| #include "TwoPuncturesBoxExtractionTaggingCriterion.hpp" |
There was a problem hiding this comment.
Can we rename this MovingBoxesAndExtractionTaggingCriterion.hpp as it is not a TwoPunctures specific method.
Source/CCZ4/CCZ4RHS.hpp
Outdated
| double a_dx, //!< The grid spacing | ||
| double a_sigma, //!< Kreiss-Oliger dissipation coefficient | ||
| int a_formulation = USE_CCZ4, //!< Switches between CCZ4, BSSN,... | ||
| int a_rescale_sigma = 0, //!< Allows a space dependent KO coefficient |
There was a problem hiding this comment.
This should be a bool, rather than an int, unless we plan to provide multiple settings?
| if (m_rescale_sigma == 1) | ||
| { | ||
| // rescale KO coefficient with lapse so that it is zero near puncture | ||
| sigma = this->m_sigma * pow(vars.lapse, 6.0); |
| public: | ||
| double sigma; // Kreiss-Oliger dissipation parameter | ||
|
|
||
| int rescale_sigma; |
| // of it, which means inner \pm L/4 | ||
| // we want each level to be double the innermost one in size | ||
| const double factor = pow(2.0, m_max_level - m_level - 1); | ||
| // const int m_horizon_max_levels = 9; |
There was a problem hiding this comment.
Remove old commented code here
| #include "SphericalExtraction.hpp" | ||
| #include "Tensor.hpp" | ||
|
|
||
| class TwoPuncturesBoxExtractionTaggingCriterion |
| auto regrid = simd_compare_lt( | ||
| max_abs_xyz, 2.5 * factor * m_puncture_masses[ipuncture]); | ||
| criterion = simd_conditional(regrid, 100.0, criterion); | ||
| // if (m_level < m_max_level){ |
| } | ||
| else | ||
| { | ||
| // where am i? |
| // } | ||
| } | ||
| } | ||
| else |
There was a problem hiding this comment.
Can we add a comment here about why this is needed - ie for when they are merging/merged we need a larger region.
| m_puncture_coords[0][i] - m_puncture_coords[1][i]; | ||
| puncture_separation_squared += displacement * displacement; | ||
| } | ||
| // make sure the inner part is regridded around the horizon |
| #TP_swap_xz = false | ||
| TP_momentum_plus = -0.000510846 0.0841746 0.0 | ||
| TP_momentum_minus = 0.000510846 -0.0841746 0.0 | ||
| TP_spin_plus = 0.0 0.0 0.0 |
There was a problem hiding this comment.
@JCAurre as long as you leave this PR open I will keep adding things! Ha ha ha.
It would be a great service to future people if you added a comment here as follows:
# Note that these are the spin values J = chi * M^2 where chi is the dimensionless spin # So e.g. for a dimensionless spin of 0.6, and a target mass of 0.5, the value should be 0.15
KAClough
added
the
enhancement
AreefW
force-pushed
the
bbh_example
branch
2 times, most recently
from
da7b9e5 to
85b25ad
Compare
|
We should also remove the TwoPunctures README.md file in the BBH example, as this is now duplicated in the wiki here: https://github.com/GRTLCollaboration/GRChombo/wiki/TwoPunctures-Initial-Data. |
5 participants
This PR updates the BinaryBH example, whose accuracy has always been highly sensitive to the adaptive mesh refinement. The primary cause of the inaccurate phase evolution of the binary was an artificial growth of the black hole masses, which only converges to zero at very high resolutions. Here we have identified that this mass growth is very sensitive to the Kreiss-Oliger (KO). For example, the growth rate is worse for larger
sigmacoefficients and the waveforms are more accurate for small values ofsigma. However, choosing smallsigmadoes not remove high frequency errors near the extraction spheres, which result in noisy waveforms. Given the dependence of the KO on the timestep Δt, we now introduce a space-dependent coefficient that turns off the KO near the punctures. For now, we have found thatlapse^6 * sigmayields optimal results when starting with a pre-collapsed lapse. However, one might want to update this in the future tolapse^4 * chi * sigmaif the lapse starts from 1. This PR also implements a tagging criterion closer to the standard 'Moving Boxes' approach, and introduces default params.txt files at two resolutions: N=48, which runs at 100 M/hr on 800 CPUs, and N=64, which runs at 50 M/hr on 1500 CPUs. Hence, these setups generate relatively accurate waveforms in 1 and 2 days, with errors of approximately 1% and 0.5%, respectively. These updates seem to allow us to achieve similar accuracies to the waveforms in arXiv:2112.10567, but with lower resolution (and thus faster) simulations.