The test complains that the domain::physical_separation is too large. That might just be an issue with tolerances. However, the bigger issue is that tolerances like this are also used in production code when determining corners. Make the following change:
Now the maps are used to determine block neighbors (like in the BBH domain). The test above will now fail with an incorrect number of boundary conditions. That's pretty catastrophic for the simulation :)
Now the Wedge3D test fails with very wrong Jacobians. In particular, when the Inverse radial distribution is selected (note: the choice of radial distribution is randomized in the test). This looks like an issue with numerical cancellations, and it improves a bit with the following change:
Bug reports:
Expected behavior:
I can create a
Sphere
orBinaryCompactObject
domain with an outer radius of 1e11.Current behavior:
Relevant code: Wedge coordinate map https://github.com/sxs-collaboration/spectre/blob/develop/src/Domain/CoordinateMaps/Wedge.hpp
The following test fails:
The test complains that the
domain::physical_separation
is too large. That might just be an issue with tolerances. However, the bigger issue is that tolerances like this are also used in production code when determining corners. Make the following change:Now the maps are used to determine block neighbors (like in the BBH domain). The test above will now fail with an incorrect number of boundary conditions. That's pretty catastrophic for the simulation :)
Finally, try this change:
Now the
Wedge3D
test fails with very wrong Jacobians. In particular, when theInverse
radial distribution is selected (note: the choice of radial distribution is randomized in the test). This looks like an issue with numerical cancellations, and it improves a bit with the following change: