The result of vasp static calculation is different when NSW= 0 and ≠ 0

Question

I'm doing a static calculation. IBRION is set to-1, but NSW forgot to set it to 0. I found that the energy of the second ion step is not the same as that of the first ion step. The structure has not changed, so why is the energy different? Which result is reliable?

INCAR:

ALGO = Fast
EDIFF = 1E-5
EDIFFG = -2E-2
ENCUT = 520
IBRION = -1
ICHARG = 1
ISIF = 2
ISMEAR = 0
POTIM = 0.05
ISPIN = 2
ISTART = 1
LORBIT = 11
LREAL = Auto
#ADDGRID = True
LWAVE = False
LCHARG = False
#MAGMOM = 30*0.6
NELM = 400
NSW = 000
PREC = Accurate
SIGMA = 0.05
#SYMPREC = 1E-4
#ISYM = 0

IVDW = 1

NCORE = 4

LMAXMIX = 4
# DFT+U
LDAU = True
LDAUTYPE = 2
LDAUL = 0 0 2   0
LDAUU = 0 0 5.3 0
LDAUJ = 0 0 0   0
LDAUPRINT = 1

The result is:

   1 F= -.12046558E+04 E0= -.12046422E+04  d E =-.272572E-01  mag=     2.0115
   2 F= -.12047406E+04 E0= -.12047334E+04  d E =-.145517E-01  mag=     1.7560
   3 F= -.12047407E+04 E0= -.12047334E+04  d E =-.145535E-01  mag=     1.7565
   4 F= -.12047407E+04 E0= -.12047334E+04  d E =-.146082E-01  mag=     1.7569
   5 F= -.12047407E+04 E0= -.12047333E+04  d E =-.146615E-01  mag=     1.7573
   6 F= -.12047407E+04 E0= -.12047334E+04  d E =-.145626E-01  mag=     1.7575
   7 F= -.12047407E+04 E0= -.12047334E+04  d E =-.145611E-01  mag=     1.7575
   8 F= -.12047407E+04 E0= -.12047334E+04  d E =-.145244E-01  mag=     1.7579

I guess the result may be related to ICHARG=1, which causes the subsequent calculation to read the previous CHGCAR. If so, the energy of static calculation should be set to NSW=0 or NSW>1, and the convergent energy should be used.

Answer 1

My guess is that electronic convergence (to the desired accuracy) has not been achieved in the first step.

Looking at the EDIFF setting from the VASP Wiki (link), my impression is that the SCF terminates at the first instance when the energy difference is below EDIFF = 1E-5, but it is not guaranteed that the next SCF iteration must have an energy difference below EDIFF (this is my understanding, but it would be good for someone more experienced / familiar with VASP's underlying code to weigh in).

You didn't seem to have set NELMIN (link), so it takes the default value of NELMIN = 2, which means that 2 SCF steps will be taken in each subsequent "geometry optimization" step (where the structure actually stays constant). But this means that the orbitals and charge density are likely to continue being optimized at each step, which also explains how you can see the mag parameter getting converged over the fake "geometry optimization" steps.

You could try doing a series of static calculations with EDIFF = 1E-5, EDIFF = 1E-6, EDIFF = 1E-7 and so on to check how the E0 and mag values converge with respect to EDIFF.