should magetisation always be an integer number?

[rgc]

Hi,

Could anyone help me understand this basic issue?

When I calculate systems with an electronic band gap, i always obtain an integer value for the total difference between alpha and beta electrons in the cell (magnetisation), even when I don't fix NUPDOWN.

But this does not happen for bulk metal calculations when I don't fix NUPDOWN... in this cases I obtain fractional values sometimes. My question is: Is this because of the smearing? Or even in the limit sigma-> 0 the difference between alpha and beta electrons can be non-integer for a conducting system?

Many thanks for any explanation!

[admin]

the magnetization is calculated from the differences of the spin up and down charges in your cell. Hence, it may differ from integer numbers, depending on the quality of your k-mesh, the BZ-integration you chose and the smearing parameter.
Please also check the charge density congvergence of your calculation in OSZICAR (rms(c)-column). If this convergence is bad, please decrease the charge- and spin density mixing parameters (AMIX, BMIX, AMIX_MAG, BMIX_MAG)

[hra]

If one the bands for one spin component are fully occupied you always get an integer magnetic moment (M = Nup - Ndown; N = Nup + Ndown; because N is an integer, if Nup or Ndown is an integer also M is an integer.) If you're calculating an insulator or semiconductor, it is likely that either Nup or Ndown must be an integer because you have a fully occ band. In the case of metals both up and down bands are partially filled, so there's no reason for either of them to have integer occupation. I don't see what this has to do with BZ integration.