[Spglib-users] Inconsistency in `get_symmetry_dataset`

[Gaël D. <gg...@ba...>]

Hi,

I'm using spglib 1.15.1's python interface to identify the sets of
transformations needed to go back-and-forth between original supercells
and spglib's standard and primitive representations in non-orthorhombic 
systems.

It *works* for 3 (non-ortho) systems but not on the 4th one (attached).
Note that you need to relax the symprec tolerance to 2e-4 for it to
work. Hall number should be 15.

To make things very simple, I have reasons to believe the
transformation matrix provided by `get_symmetry_dataset` is wrong. It
is different from P by directly solving:

(a, b, c) P = (a', b', c').

Note that the origin shift is zero and that by "different" I really
mean "different": `transformation_matrix` is diagonal (or can be made
so by simple axis swaps) and P is not. This is not a transpose or
numerical accuracy problem.

So is it a bug or did I forget an important part of the calculation?

Thanks.

Best regards,
Gaël