GRHydro test suite failures (the ones using McLachlan)

[Bruno Mundim]

Hi,

I am facing the following problems with the GRHydro test suites that uses McLachlan:

INFO (GenericFD): The stencil for ML_BSSN_O2_convertFromADMBaseGamma requires 2 points, but the lower y boundary has only 1 points. INFO (GenericFD): The stencil for ML_BSSN_O2_convertFromADMBaseGamma requires 2 points, but the upper y boundary has only 1 points.

any ideas of a simple fix?

Keyword:

[Erik Schnetter]

Ian Hinder replied on the mailing list:

"Set the upper and lower boundary sizes to 2. This is a check which was added to Kranc quite recently to catch cases where people have not used enough boundary or ghost points for the evolution stencil of their code. The "convert from ADMBase" routine uses an advection stencil (which requires two adjacent points) to compute the A and B gauge variables.

"Note that tests which use McLachlan and dissipation will have to be regenerated anyway because the dissipation was corrected. Before commit ec32fb2e34d6a2e8b7534195c16bbc2008835ec3, the dissipation term for X was being added spuriously to the B equation. This is probably harmless but fixing it changes the numerical results. I haven't gotten around to doing this yet. If someone else wants to do it, you should check that the McLachlan test passes before that commit, update to the commit after that one, where the code is regenerated, and check that the test fails. Copy the test output into the test directory, and check that the latest version still passes. Then commit the grhydro test."

[Ian Hinder]

I fixed this a while ago. Closing the ticket.

User: hinder Date: 2011/07/25 09:47 AM

Modified: /trunk/test/ GRHydro_test_tov_ppm_ML.par, GRHydro_test_tov_ppm_ML_disable_internal_excision.par, GRHydro_test_tov_ppm_no_trp_ML.par

Log: Use 2 CoordBase boundary points instead of 1 in McLachlan tests

McLachlan (ML_BSSN_O2) requires two ghost/boundary points due to the dissipation and upwinded advection operators. Kranc-generated thorns now check that the correct number of ghost and boundary points are available, and this check causes three GRHydro tests to fail as they use a boundary size of 1. It turns out that if you disable the check, the original GRHydro test suite still passes on Intel with Linux (Damiana) but segfaulted on GCC with Mac OS (my laptop). The dissipation and unwinding operators will be reading off the edge of the grid, so the behavior is undefined. Dissipation is set to 0 in this test, but a nonzero shift is used. This commit sets the CoordBase boundary size to 2 in the tests, and this requires the test data to be regenerated (the following commit).

<<<