ABINIT Discussion Forums

Dear all,

We have noticed a difference in the value of the magnetization for NiO when using PAW and non-colinear magnetism (nspden=4) as compare when non-PAW pseudopotentials are used.

NiO is a collinear spin system. When using non-PAW pseudopotentials, we obtain the same difference in the integrated atomic density meaning the same local magnetic moment, a value of 1.35.

When we use PAW pseudopotentials, the colinear calculation (nspden=2) gives the same value of local moment as the non-PAW calculations. The difference between the non-PAW and PAW is only 3% which can be explained by the different pseudopotentials.

When the calculation is done using non-colinear (nspden=4) and PAW, the local magnetization is lower. We obtain 0.85 which is very different than the 1.35 of the other methods. We note also that the convergence is a lot more difficult in nspden=4 with PAW as compare to non-PAW and PAW in colinear.

The outputs files are attached to this post. We can provide the input and log files if needed. At the moment, the Forum allows for only 3 files to be attached to each post. All the calculations are with the 6.2.2 version of ABINIT. We had the same problem with 6.0.0.

We are wondering if there might be some problems in the implementation of the non-colinear magnetism in PAW? We are willing to help find the problem but we wanted to have the opinion of other users on this issue first.

Michel

Hi Michel,

A quick answer... to go further I have to go deeper into your files.
Concerning the magnetic moment differences:
- the difference between non-PAW and PAW (1.35 vs 1.30) is probably due to the difference in the radius used to integrate the density (2.0 for non-PAW, 2.3 for PAW); you could try to impose ratsph...
- if I compare paw_nspden2 and paw_nspden4, I find that the total energies are similar, as well as the occupation matrices rhoij (and psp strengths Dij).
So, the two calculations seem to be the same... we could suspect a problem in the integration; to be confirmed...

Marc