ABINIT Discussion Forums

Dear forum,
I'm working on bulk Nickel (a magnetic system) with spin-orbit interaction and spinors.

Abinit does not use the inversion symmetry.
I found out that the reason is the following line in the ingeo.F90 subroutine :
use_inversion=1 ;if (nspden==4.or.pawspnorb>0) use_inversion=0

My understanding is that in my system the pure time-reversal symmetry is broken, but the spatial inversion should still be ok.
Is there a fundamental reason why use_inversion=0 when nspden=4 or maybe it is just a "coding" reason?
(i.e. the spatial inversion flips the spin/magnetization, and this is not coded with nspden=4)

I attach the input file.
(The parameters are not at convergence...)

Thanks in advice. Regards,
Davide

Dear Davide,

A quick answer... trying to remember what has been done several years ago.

In Abinit, the symetry set is common to k-point generation, reciprocal space symetrization and real space symetrization; it is not possible to deactivate a symetry for the generation of k-points only while activating it in real space...

* If you include spin-orbit coupling, I the inversion has to be removed from the list of symetries; with spin-orbit, there is a spin-space coupling and you cannot separate spatial inversion and time-reversal symetry easily.In the reciprocal space, the inversion is the TR-symetry. The latter is broken when spin-orbit is activated. So, you have to exclude inversion from the symetry.

* If you don't include spin-obit and use non-collinear magnetism... this is different.TR-symetry is not broken... so, I don't remember why the "nspden==4" statement appears in the line you cited. We probably faced issues when implemented this part of the code and, for safety, we decided to exclude the inversion from the set of symetries. or, Another possible explanation, there are perhaps some parts of the code where it is difficult to discriminate between pure non-collinear magnetism and spin-orbit.
Anyway, it is better to suppress inversion than wrongly keeping it.

There is a simple test you can do: suppress "nspden==4" from the line and run Abinit with nsym=1 and nsym=0. If you system is not too simple (really non-collinear), you could find a difference.You can also do the same test with spin orbit (forcing use_inversion=1).

Marc