[Wien] spin + SOC slab issue

[pluto]

Dear All,

I am calculating a slab with 44 atoms (22 inequivalent atoms). I have 
two equivelent manganese atoms in the slab.

Space group discovered by sgroup is 164, it has inversion.

I run "instgen_lapw -ask" and I set Mn spin to u, and all other atoms to 
"n" (other atoms are non-magnetic). I monitor this during the SCF cycle 
by "grep :MM case.scf", and I am getting around 4.33 for the Mn moment, 
a reasonable number.

The calculation is spin-polarized with SOC. During init_so_lapw I accept 
changes made by symmetso.

I run two separate calculations with magnetization 001 and with 00-1. 
This is set in case.inso. Energy (0.0001) and charge (0.001) is 
converged.

The two calculations give different eigenvalues, also at the Gamma 
point. Differences are not tiny, they are of the order of 50 meV and 
bands look slightly different. Correct me if I am wrong, but I think 
this is not physical in this setup.

Could you please suggest what can be wrong and what can be checked.

Am I allowed to converge with M = 001, save_lapw, then re-run 
init_so_lapw with M=00-1, and then re-run the SCF cycle with -NI? Or 
should I always run M=00-1 from the scratch, that is on top of converged 
non-SOC calculation?

Best,
Lukasz