Metric normalization error in petsc4py test
I realised that plex.metricNormalize
is not tested in test_dmplex.py::testMetric
, so made this commit. However, the test fails in each case because the global normalisation factor computed by DMPlexComputeIntegralFEM
is erroneously zero, even though the metric is positive-definite at all nodes.
On closer inspection, I think there is an issue with orientations. The metric is constant, so the integral contribution from each element should be the same positive value. The values come out the same, except they have oscillating signs. Could this be an issue with the way the mesh is defined in the test @knepley? Or am I missing something?