Closed kellertuer closed 7 months ago
Attention: 26 lines
in your changes are missing coverage. Please review.
Comparison is base (
81a43d4
) 99.57% compared to head (618ffe4
) 99.33%.:exclamation: Current head 618ffe4 differs from pull request most recent head 31f4707. Consider uploading reports for the commit 31f4707 to get more accurate results
Files | Patch % | Lines |
---|---|---|
src/manifolds/Sphere.jl | 0.00% | 26 Missing :warning: |
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
- complex coefficients in a real-valued basis, which is sometimes, especially for the sphere here a bit strange, since it would mean that in the North Pole the first component is only allowed to be purely imaniary for the spheres
I think we should restrict complex-coefficient bases to complex manifolds, while the complex sphere is actually a CR manifold but not a complex manifold.
Or, at least, I wouldn't add complex-coefficients bases to non-complex CR manifold without some understanding of the theory behind CR-manifolds.
That would again be something breaking and different to what we use those currently, but sure that is something to discuss.
I fixed the one allocation bug I had but now I notice that my idea might not work as intended (doing the same rotation as in the real case and taking the complex value of the first component as the last one). So I fear I have to close this since I have no clue how to compute an ONB or its coordinates here.
This should (up to specifying the right basis) resolve https://github.com/JuliaManifolds/Manopt.jl/issues/340
For now there is a few things to check though
get_coordinates
(real-valued case) are not precise enough, the inner product for $\lambda$ seems wrong and the used $x$ is not defined. Will check. The same holds for the inverseget_vector
Here is a bit of test code that I try to verify the new method with
a) real check for comparison
The same should hold now if we perform that on
M2 = Sphere(2, ℂ)