Neutron star gallery example produces output different from what is shown on website in ET_Tesla

Issue #2118 closed
Roland Haas created an issue

The gallery example https://einsteintoolkit.org/gallery/ns/index.html produces output significantly different from what is shown in the gallery. Namely the maximum density drops to ~88% of the correct value compared to only ~97% in the plot in the gallery.

Attached please find stdout, data file and the plot.

| Release | Date | ρmax |

Gallery 2013-05 rho_max.png
HerschelCXX 2014-11 rho-HerschelCXX.png
Herschel 2014-11 rho-Herschel.png
HilbertF90 2015-05 rho-HilbertF90.png
Hilbert 2015-05 rho-Hilbert.png
Sommerville 2015-11 rho-Sommerville.png
PayneGaposhkin 2016-12 rho-PayneGaposhkin.png
masterF90 2017-02 rho-masterF90.png
master 2017-02 rho.png

Keyword: Gallery
Keyword: GRHydro
Keyword: regression

Comments (17)

  1. Roland Haas reporter
    • removed comment

    I will try and see if I can find out if there is any ET release that reproduces the plot on the website.

    Note that until #2117 is fixed one cannot run the parfile with more than one MPI rank and the proposed release code.

  2. Roland Haas reporter
    • removed comment

    The difference shows up between the Herschel (good) and Hilbert (bad) releases:

    || Herschel || 2014-11 || [[Image(rho-Herschel.png, 50%)]] || || Hilbert || 2015-05 || [[Image(rho-Hilbert.png, 50%)]] ||

    which happens to be when the 'default' code in GRHydro switched from F90 (Herschel) to CXX (Hilbert). However flipping the use_cxx_code parameter does not affect the output so that it does not seem to be a difference between CXX and F90 code but something else that changed in between the releases.

  3. Roland Haas reporter
    • removed comment

    I ran git bisect which eventually reports:

    commit cef0ee829c329d5edbfb288a94c8379b2fa32d62
    Author: Frank Loeffler <knarf@cct.lsu.edu>
    Date:   Mon Nov 24 09:25:42 2014 -0600
    
        GRHydro: remove dependency on ADMMacros::spatial_order
    
        Introduce GRHydro::sources_spatial_order in favor of
        ADMMacros::spatial_order. Only the location (and name) of the parameter
        changed.
    

    so I will try what happens if I use the new parameter GRHydro::sources_spatial_order.

  4. Roland Haas reporter
    • changed status to resolved
    • removed comment

    Yup, that was it, adding "GRHydro::sources_spatial_order = 4" restores the old plot:

    [[Image(rho-source_spatial_order4.png, 50%)]]

    I have updated the parfile in the gallery.

    Lesson for the future:

    Though shalt not silently change default behaviour, for backwards compatibility is a fickle god and will be punishing the maintainers for the sin of the developers to the third and fourth generation of those who hate it.

    or so.

  5. Ian Hinder
    • removed comment

    We should probably tell people about this on the mailing list, in case they have some of their own parfiles which need fixing. This will be good even if #2120 is implemented.

  6. Log in to comment