NaNs in the AMR boundary closest to Pi symmetry boundary around 10M of evolution

Create issue
Issue #436 closed
Bruno Mundim created an issue

The parameter file attached runs smoothly with the ET curie release version. However when I use the current ET development version, NaNs show up in the AMR boundary closest to Pi symmetry boundary around 10M of evolution. This happens for the second finest grid level.

I am repeating the experiment without Pi symmetry and I will report on its result shortly.


Comments (4)

  1. Bruno Mundim reporter
    • removed comment

    Running the same parameter file without RotatingSymmetry180 works just fine: no NaNs observed until around 16M. Some strange interaction between RotatingSymmetry180 and Carpet mercurial version may be causing these NaNs.

  2. Bruno Mundim reporter
    • removed comment

    I changed VECTORISE_ALIGNED_ARRAYS to "no" according to Erik's comment below:

    "In the light of recent events I would like to point out again that you should not be using the configuration option VECTORISE_ALIGNED_ARRAYS. Some time ago, I accidentally committed (and later undid) a corresponding change to SimFactory. This option enables experimental changes in the Mercurial version of Carpet and is likely to lead to nans or other inconsistencies in your simulations. Please check your configuration options..."

    It didn't solve the problem, but I thought to keep it noted after this test. Note that the Carpet hg changeset was still an old one for this test: tip at 3287. I will update the tree for the next tests.

  3. Bruno Mundim reporter
    • changed status to resolved
    • removed comment

    On Sun Aug 14 I re-run this parameter file with Pi symmetry and the Carpet/hg and the simulation went on without any problem. I stopped it around 22M. So I am closing this ticket. Please re-open if necessary.

  4. Log in to comment