NEDOS and Spin

[aina2412]

Dear Admin,

This is the input in my INCAR file.
PREC = Accurate
ICHARG = 2
IBRION = 2
NSW = 0
LNONCOLLINEAR = .TRUE.
MAGMOM = 144*0 0 0 1 0 0 -1 0 0 -1 0 0 1
LREAL = Auto
LVTOT = .TRUE.
LELF = .TRUE.
LAECHG = .TRUE.
LCHARG = .TRUE.
LORBIT = 11
EMIN = -20
EMAX = 15
NEDOS = 5250
ISMEAR = 0
SIGMA = 0.01
LVHAR = .TRUE.
ALGO = FAST

I am calculating for CeRu2Al10 to know the DOS and the spin direction after relaxation.
There are 4 Cerium in the unitcell. After the calculation has finished, I analyze the result and found that
by running NEDOS = 3500, the spin direction was at c-axis, 0.06 -0.06 -0.06 0.06.
But, when I change to NEDOS = 5250, the spin change to -0.072 0.072 0.072 -0.07 (180 degrees flip)
But, it changed again to 0.06 -0.06 -0.06 0.06 after I increased the NEDOS to be 7000. *again 180 degrees flip*

Would you suggest me any idea why it is happening?
Why the NEDOS could change the spin direction and the magnetic moment value?
I though NEDOS only change the DOS resolution.
Sorry for my bad english.

Thank you.

[admin]

The system is quite large, the calculated spin values are small, and if you use just
the default precision this fluctuation of spin values is nothing but numerical noise.