[PATCH]McLachlan_BSSN: fix the Hamiltonian constraint for CCZ4

Create issue
Issue #1199 closed
anonymous created an issue

A typo in a8794cd2227c87c7457ba51d040e863be2f981e7 has caused the additional CCZ4 terms to overwrite the BSSN value instead of adding to it.

Tested to work on TwoPunctures initial data:

h = h5py.File('/tmp/testbssn/H.h5') # BSSN reference h1 = h5py.File('/tmp/testccz4/H.h5') # with the patch h2 = h5py.File('/tmp/testccz42/H.h5') # without the patch

In [89]: numpy.sum(numpy.abs(h.values()[0][:] - h1.values()[0][:])) Out[89]: 8.0082674998918897e-15

In [90]: numpy.sum(numpy.abs(h.values()[0][:] - h2.values()[0][:])) Out[90]: 7.9536720382719803


Comments (5)

  1. Barry Wardell
    • removed comment

    Thanks! This was not picked up by the testsuites as they did not include output for the Hamiltonian constraint.

    I have committed the patch to the master branch along with improved tests which include the missing constraints. Should these also be backported to other branches (eg. the most recent release branch)?

  2. Barry Wardell
    • removed comment

    This bug has just been rediscovered by a user using the Ørsted release, so I think we should backport it to avoid others being affected. It is very clear that the patch is correcting the previously wrong behaviour. Thoughts?

  3. Roland Haas
    • removed comment

    Obvious bug fixes should be backported in particular if they are of the "silent" kind that produce wrong results but don't obviously kill the run. So my vote is to port the patch back into Ørsted ML .

  4. Log in to comment