Tested on 2018-07-31 using my Linux workstation and gcc 8 and 1 core (1 rank, 1 thread). Updated figure on website, provided tarball of results. Results look visually close to what was there before though there was no numerical data provided.
I do however see failures if I ran the test with 12 MPI ranks (1 thread each) which is a bug and I am following up on it.
following up on your email, I would suggest to use option 2 (static_tov.par) for the gallery example. If you agree I will ask Giuseppe to run it and upload the corresponding plots.
Currently we have two different setups in the tutorial and gallery. Newbies like my students may find that confusing, especially because the results don't match. Therefore, I suggest to either use the same parameter file or adding a note in both.
This means that he results not (qualitatively match), thankfully getting rid of the confusion though it still persists with respect to the data in the original GRHydro paper. If we desire to remove that confusion as well we should change both the tutorial and the gallery to 2d_Poltytrope.
After some more discussion, we decided the following:
- update the gallery example to evolve a polytropic eos (2d_Polytrope), i.e., as close as possible to the 2011 paper. This will have a large grid + resolution (similar to static_tov.par)
- for consistency: update the tutorial with that same physical setup, but with lower resolution and smaller grid size.
Giuseppe is looking into this now.
on behalf of G. Ficarra:
In order to get the results obtained by https://arxiv.org/abs/1111.3344 (Figure 15), I have modified the parameter file static_tov.par (Cactus/par/static_tov.par) to make it as similar as possible to the original parameter file tov.rpar (Cactus/par/arXiv-1111.3344/tov/tov.rpar) which should give the results of the paper. At a first glance both parameter files describe the same physical system. The differences between them are given by the use of the dissipation and the poison value used in the process.
First, I have run simulations to test if using dissipation led to different results. Using both resolutions dx=8 (first plot) and dx=16 (second plot), I find no relevant differences between implementing or not the dissipation, while the late time evolution of the system is strongly influenced by the choice of the resolution; compare
In both cases I don’t get anything like Figure 15 of the paper for the hydrobase density (HydroBase::rho). At this point I also used the same poison value from the original parameter file for a sanity check and got as output also the grhydro density (GRHydro::dens).
Using a resolution dx=16 I always get the same profiles for both the hydrobase and grhydro densities; see
Also in this case a significant discrepancy is obtained between my results and those presented in the paper.
PS: what is the preferred praxis to upload parameter files to the report?
For reference, this is the plot for the three resolutions that Frank ran it for the paper (dx = 2,4,8):
The plot in the paper is for dx=2.
These results do depend on resolution since we are not adding an explicit perturbation and thus the only perturbation away from the stable TOV solution is due to numerical errors when interpolating the 1D spherical data onto the Cartesian grid. This means that at high enough resolution the oscillations converge away as we approach the analytic solution which is stationary.
we did compare our data to the raw data corresponding to that plot for dx=8 and find large deviations (compare to the first plot in my message from 2019-04-19).
I have started to re-run the dx=8 parfile with older versions of the code. Maxwell (2011_10) reproduces the plot in the paper.
I will try and find the release where a change is visible.
The change in behaviour happens in between the Herschel (ET_2014_11) which produces results identical to the paper and Hilbert (ET_2015_05) which produces results identical to Proca releases.
Hilbert is the release where GRHydro's default switched from F90 to C++ code. However that is not it as even running Hilbert with use_cxx_code = no does not recover Herschel's behaviour.
There is also some code in #2120 that would avoid this from happening again, and which I will commit.
This is the result of adding:
GRHydro::sources_spatial_order = 4
thanks for tracking this down, and I agree with your change.
I asked Giuseppe to repeat the gallery example with the updated parameter setting for the Proca release.
With the change GRHydro::sources_spatial_order=4 we do indeed recover the behaviour of the release paper (https://arxiv.org/abs/1111.3344 (Figure 15)) as shown below.
If you agree I will upload the data and a brief description to the gallery. Do we want to use the same setup with a smaller grid for the tutorial so that they are consistent?
Yes, please upload the new data to the website and the tarball to bitbucket. Yes, please, if you can do provide a lower resolution / smaller gird setup for use in the tutorial. You may want to have a look at the current one provided by Bruno which is very short and runs in 10 minutes or so (it has very low resolution and very small grid).
Done with the gallery example, please check. Thanks to Giuseppe!
OK, we’ll set up a small test for the tutorial.
Looks fine with one exception: I would not include the date of run in the parfile (mostly b/c none of the other examples do, partially b/c it may look odd in the next release if the parfile needs no change so should not change the name [it also makes change tracking in git a bit easier but not by much]).