merge bugfixes from Zelmani into main branch

Create issue
Issue #827 closed
Roland Haas created an issue

A while ago we moved GRMHD development (for various reasons) from the main repository into a git repository (Zelmani). Since then we have accumulated a number of (important) bugfixes (in particular for MHD but also for other subsystems) in that repository.

Should we mere them back (this soon before the release)?


Comments (16)

  1. Roland Haas reporter
    • removed comment

    Attached please find the proposed patches. They are a subset of the changes in Zelmani. We will need to wait for Philipp Moesta to ok patch 10 which is the major GRMHD fixup patch. Currently Philipp is out-of-country.

    For those who use Llama: MHD does not yet work with multipatch (only half the changes for that are in).

  2. Roland Haas reporter
    • removed comment

    One issue with patch no 5 is that, while it makes MHD work with arbitrary EOS the way the "normal" RHD does, it also makes MHD much slower (factor eight I believe).

  3. Roland Haas reporter
    • removed comment

    I just discussed with Philipp in a bit more details and the status is that in fact the 3d inversion code is not used at all right now since the patch in question only modifies a routine that is not yet used. And even were it used it contains "if" statements on eos_key to switch back to the faster 2d scheme for polytrope and gamma law eos. So the status right now is that we have a fast inversion routine that is used but only works for polytrope and gamma law and a possibly slow routine (with fast fallback) that is designed to work with any eos and which does not obviously fail with a hybrid eos.

  4. Roland Haas reporter
    • removed comment

    There has been no reply on any of these from anyone outside of Caltech. Unless I here otherwise I'll assume that the silent majority is ok with me committing these on Sunday. There are some more outstanding OpenMP changes for MHD that Philipp has collected since this ticket was created and that will also be committed at that point.

  5. anonymous
    • removed comment

    I think patches 0001 to 0009 and 0011 can be applied immediately. They fix a variety of bugs from incorrect treatment of a static atmosphere (for the magnetized case) to race conditions in the more recent implementation of constraint transport, among other minor bug fixes.

    Patch 0010 needs a more careful inspection. Since this part of the code is still in development and not used in production, I would recommend to apply that patch tomorrow evening (to give us all a last chance to look at it) and if problems arise we fix it later. I don't think we can claim the GRMHD code is totally ready for production yet. So in this release we will only have a much improved version.

  6. Roland Haas reporter
    • removed comment

    I applied patches 1-9 and 11. Patch 10 is the most important one so I would really like to see it included in the release. The MHD code is technically "released" in the Maxwell release so we should treat it as if it was used for production I think.

  7. Bruno Mundim
    • changed status to open
    • removed comment

    Ok, patch 10 looks good to me with two caveats:

    1) I don't fully understand assigning psidc as a pseudoscalar yet. However a flat space analysis seems to indicate it is actually a pseudoscalar. divB is clearly a pseudoscalar.

    2) divB seems to be calculated twice with this patch in place.

    I recommend to apply this patch. We can then fix 2) in the next commit.

  8. Log in to comment