WeylScal4 fails tests with development version of Carpet

Create issue
Issue #496 closed
Ian Hinder created an issue

Both WeylScal4 tests fail with the development version of Carpet but not with the stable version. The report is

Psi4i.d.asc: substantial differences significant differences on 16 (out of 62) lines maximum absolute difference in column 13 is 0.0042979716483804 maximum relative difference in column 13 is 0.221177389498685 (insignificant differences on 20 lines) Psi4i.z.asc: substantial differences significant differences on 24 (out of 108) lines maximum absolute difference in column 13 is 0.0123048458581252 maximum relative difference in column 13 is 1.43992798290495 (insignificant differences on 54 lines) Psi4r.d.asc: substantial differences significant differences on 16 (out of 62) lines maximum absolute difference in column 13 is 2.99002416115686e-05 maximum relative difference in column 13 is 0.995857600242238 (insignificant differences on 26 lines) Psi4r.x.asc: substantial differences significant differences on 42 (out of 146) lines maximum absolute difference in column 13 is 0.0063512431519604 maximum relative difference in column 13 is 1.35438090371356 (insignificant differences on 84 lines) Psi4r.y.asc: substantial differences significant differences on 42 (out of 146) lines maximum absolute difference in column 13 is 0.0064343032695815 maximum relative difference in column 13 is 1.1458862046889 (insignificant differences on 84 lines) Psi4r.z.asc: substantial differences significant differences on 24 (out of 108) lines maximum absolute difference in column 13 is 0.00010846878486765 maximum relative difference in column 13 is 2.89015233246968 (insignificant differences on 64 lines)

One possibility is that the grids in this test are so small that the grid structure is being affected by the Carpet regridding algorithm which ensures proper nesting, and this algorithm might have changed unavoidably between the stable and development versions of Carpet. If we confirm that this is the case, we could either use a larger grid to prevent this from happening, or just regenerate the data. It would be useful to add output of the coordinate gridfunctions to catch the cases where the grids are a different shape.

(Reporting against Carpet so it can be fixed as part of the move to the development version)

Keyword:

Comments (3)

  1. Roland Haas
    • removed comment

    Working on it :-). This might be an issue with the usage of the grid but not with the grid being actually different. Carpet also reports that the sum of the weights is not equal to the simulation domain. I am not sure yet if this is due to a domain that is too small or not. As far as I can tell the grid created is the same for both versions (needed min_distance = 0). At least 3D ASCII output contains the same points (but different values).

  2. Log in to comment