[Users] Simulating hyperbolic encounters
schnetter at cct.lsu.edu
Wed Feb 24 11:35:12 CST 2021
A few general comments:
- Whether you have sufficient resolution or not for the calculation to
survive should be clear in the first 10 M of a calculation. If you
monitor horizon masses and constraints, you'll see whether things are
roughly working out or not. Of course, a detailed convergence test
comparing horizon tracks, waveforms etc. will be necessary in the end,
but it seems you're still in the exploratory stage of your setup.
- The exploratory stage of a setup, where one has to discover good
grid structures, resolutions, etc., is the most difficult part. Most
people would get stuck here, and it helps to have experience with
black hole simulations.
- You mentioned that you disabled symmetries. That might be necessary
in the end, but in the current stage, symmetries will reduce computing
time while you experiment to find a good setup.
- You mentioned that you disabled HDF5 output. That means that the
simulation is a black box for you. I would enable output, at least for
some scalar quantities (lapse, trace K, Hamiltonian constraint,
conformal factor) every so often. This way you can look at what
happens in the simulation. If things look boring – good. If there's an
interesting feature – that might explain what is going wrong. Look
especially for things you're not expecting to see (e.g. problems
caused by outer boundaries, by refinement boundaries, weird
reflections, etc.). It'll be necessary to examine the calculation
output in 3d, not just in the equatorial plane.
- If things go wrong at a time, try restarting from a checkpoint
before that time and increase frequency and amount of output. This
will let you examine the calculation in more detail.
- The fact that you reach periastron is a good sign. This means that,
if you compare the state of the calculation when things go wrong to an
earlier state, you might find pointers as to what went wrong.
Finally, nothing beats experience. Myself, I'm a bit crusty when it
comes to running simulations myself, but I'd be happy to meet you on
Zoom to look at some details. We can do that before you have HDF5
output, but it would be best if you tried to produce some output.
All the best,
On Wed, Feb 24, 2021 at 10:51 AM Santiago Jaraba Gómez
<santiago.jaraba at uam.es> wrote:
> Hi everyone,
> I am a PhD student at the IFT in Madrid. I am trying to adapt the par/arxiv-1111.3344 /bbh.par parameter file to a hyperbolic encounter with an initial separation of 100M. In fact, my problem is very similar to the one here: http://lists.einsteintoolkit.org/pipermail/users/2020-February/007292.html
> Learning from the GW150914.rpar parameter file, I have turned the bbh.par into a .rpar file, so that I can automatically set some parameters such as the initial positions, momenta, masses, etc. In addition, I have set the boundaries to 768 to accommodate the new initial positions x = ±50 and increased the refinement levels from 7 to 10, increasing a bit the resolution of the finest grid (from 2/2^6=0.03125 to 12.8/2^9=0.025). I have also disabled all the HDF5 output as well as the 180º rotating symmetry, since I will later want to use it for different masses, and enabled the QuasiLocalMeasures thorn. Finally, by following the advice for the mentioned doubt of February 2020, I have increased the parameters TwoPunctures::npoints_A,B,phi to 50, 50 and 20, respectively.
> I thought that I had it, since the simulation was successfully running, even past the point of closest approach. However, shortly after that, the parameter "ML_BSSN::trK norm2" that the output showed abruptly turned to -nan and the simulation crashed. Could anyone help me find the problem? I thought it might be the resolution, but the bbh.par has a bit worse finest grid resolution and it runs smoothly, even with stronger fields involved. Could the problem be in the initial conditions generated by TwoPunctures at such large distances?
> I attach the .rpar file in case it helps. I am very new to the Einstein Toolkit, so I might very well have overlooked something relevant.
> Thank you in advance.
> Best regards,
> Users mailing list
> Users at einsteintoolkit.org
Erik Schnetter <schnetter at cct.lsu.edu>
More information about the Users