Tests using Exact should use EinsteinExact

Create issue
Issue #878 closed
Ian Hinder created an issue

The EinsteinExact arrangement provides initial data accurate to roundoff, in contrast to the Exact thorn which does not. The initial data from the Exact thorn is also highly sensitive to roundoff level differences.

Hence, tests should be converted to using the EinsteinExact arrangement where possible. In some cases, EinsteinExact will not support the required metrics or parameters. These could either be added to EinsteinExact, or the Exact thorn could continue to be used.

Keyword:

Comments (6)

  1. Barry Wardell
    • removed comment

    Below is a list of the tests which currently use Exact:

    RotatingSymmetry180/Kerr-rotating-180-staggered RotatingSymmetry180/Kerr-rotating-180 RotatingSymmetry180/Kerr-staggered RotatingSymmetry180/Kerr RotatingSymmetry180/KerrSchild-rotating-180 RotatingSymmetry90/Kerr-rotating-90-staggered RotatingSymmetry90/Kerr-rotating-90 RotatingSymmetry90/KerrSchild-rotating-90 AHFinderDirect/checkpointML AHFinderDirect/Kerr-Cartoon AHFinderDirect/Kerr-definition-expansion-product AHFinderDirect/Kerr-definition-expansion AHFinderDirect/Kerr-definition-inner-expansion AHFinderDirect/Kerr-definition-mean-curvature AHFinderDirect/Kerr-modification-radius AHFinderDirect/Kerr-rotating-180 AHFinderDirect/Kerr-rotating-90 AHFinderDirect/Kerr-selection-areal-radius-definition-expansion-product AHFinderDirect/Kerr-selection-areal-radius AHFinderDirect/Kerr-selection-mean-coordinate-radius AHFinderDirect/Kerr AHFinderDirect/recover AHFinderDirect/recoverML QuasiLocalMeasures/qlm-bl QuasiLocalMeasures/qlm-ks-boosted QuasiLocalMeasures/qlm-ks-shifted QuasiLocalMeasures/qlm-ks-tilted QuasiLocalMeasures/qlm-ks McLachlan/ML_BSSN_Test/ML_BSSN_NewRad McLachlan/ML_BSSN_Test/ML_BSSN_O8_sgw3d McLachlan/ML_BSSN_Test/ML_BSSN_sgw3d McLachlan/ML_BSSN_Test/ML_BSSN_sgw3d_harmonic McLachlan/ML_BSSN_Test/ML_BSSN_sgw3d_rhs McLachlan/ML_CCZ4_Test/ML_CCZ4_sgw3d McLachlan/ML_CCZ4_Test/ML_CCZ4_sgw3d_rhs

    For each of these, we can create a new test which uses EinsteinExact instead. If we want to be extra careful, we can check that the Exact version converges to the EinsteinExact version at the appropriate order. Is that necessary? I have already checked that the Kerr-Schild and ShiftedGaugeWave solutions (used by most tests) in Exact converge to EinsteinExact (see the EinsteinExact/tests directory).

    Of these tests, the only one which uses a spacetime which is not currently provided by EinsteinExact is QuasiLocalMeasures/qlm-bl, which uses the Schwarzschild metric in Brill-Lindquist coordinates. We have almost the same thing in EinsteinExact/ModifiedSchwarzschildBL, except for the fact that the mass has a radial dependence inside the horizon. In principal, is it the case that this shouldn't matter if QuasiLocalMeasures is only looking on a surface at/outside the horizon? Is it likely to be the case in reality?

    In terms of missing parameters, the only one I can see is support for a boost. This is something which would be nice to have in EinsteinExact anyway, so we should add it.

  2. Erik Schnetter
    • removed comment

    The Brill-Lindquist coordinates are really simple -- the conformal factor is 1+m/(2r), everything else is Minkowski. We should add this to EinsteinExact. Could you add a parameter to change the matching radius to ModifiedSchwarzschildBL? In experiments we learned that one needs to have about ten unmodified grid points inside the event horizon to avoid things propagating to the outside.

    On a different note: many exact solutions will have singularities. Would it make sense to have a generic way of handling them, i.e. a generic way to replace r=0 in analytic expressions?

    I don't know this particular test case, but in principle QLM is also used inside the horizon.

  3. Log in to comment