Bader charges not converged with increasing NGX(/Y/Z)F grid

[jianmin]

Dear all,

I have built a CeO2(111) 3x3x9 surface with 27 Ce atoms and 54 O atoms and a 15-Angstrom vaccuum, and got the total charges of Ce atoms and O atoms with respect to NGXF as follows (|e| means number of charges):
NGFX 154 179 205 230 256
|e| of Ce atoms 62.36 61.44 60.41 61.21 62.06
|e| of O atoms -62.36 -61.44 -60.41 -61.21 -62.06

Obviously, charges were not converged with respect to increasing NGXF grid. How do we handle this problem? Replies are greatly appreciated.

VASP settings are as follows:
1. Parameters in the INCAR file
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
ENCUT = 500.00 eV
PREC = Normal
LREAL = Auto
EDIFF = 1E-5
NSW = 0
IBRION = 2
ISIF = 2
ISMEAR = 0
SIGMA = 0.05
ISPIN = 2
# FFT grid
NGXF = 154
NGYF = 154
NGZF = 307
# DFT+U settings
LDAU = .TRUE. # switch on the +U corrections
LDAUTYPE = 2 # Dudarev's approach
LDAUL = -1 3 # -1 no U; 1 U to p electrons; 2 d and 3 f.
LDAUU = 0.0 4.5 # U values
LDAUJ = 0.0 0.0 # do not specify them for Dudarev's approach
# dipole corrections
IDIPOL = 3
LCORR = .TRUE.
# output controls
LAECHG = .TRUE.
LCHARG = .TRUE.
LWAVE = .FALSE.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

2. A 3x3x1 k-point mesh was used. PAW potential files for Ce and O were used.

[graeme]

I don't understand the magnitude of the charge values. How do values of ~60 make any sense?

If you are trying to integrate the total (including the core) charge, this will be very difficult. Instead (if possible) you can do the analysis on the total charge and integrate just the valence charge using "bader charge_valence -ref charge_total

[jianmin]

Dear Prof. Graeme,
Take NGXF=154 for an example, the |e| value for all 27 Ce atoms is 62.36, meaning 27 Ce atoms lose 62.36 electrons while 54 O atoms gain 62.36 electrons.
What I am asking is that: should it be that the |e| value should not vary but rather converge to a value with increasing NGX(Y,Z)F grid?

[graeme]

Oh I see; these values are the sum over all Ce/O atoms.

For faster convergence with respect to the grid try the new "-b weight" method. This has quadratic rather than linear convergence with respect to the grid density.

[jianmin]

Dear Prof. Graeme,
Thanks for the reply.
Instead of issuing the command: bader CHGCAR -ref CHGCAR_sum
I shoud issue a command like: bader -b weight CHGCAR -ref CHGCAR_sum
Is this what you meant?

[graeme]

yes