UT theoretical chemistry code forum

Hello,

I have calculated total electron density of Aluminium with GGA-PBE in VASP. Both the valence electron and core electron does not seem right, even the calculation is converged (in terms of energy)...Wondering where went wrong. Has anyone run bader on metallic system before that might have any suggestion?

In INCAR setting (I attached in the end in details), LAECHG = .TRUE. and NGX(YZ)F is dense (240), the CHGCAR_sum is get with chgsum.pl nicely provided. The k-points and Encut is converged better than 1 meV for four atoms in the cell.

I am aware that it needs sufficient dense grid to converge the electron, at least to get the number right.((13*4=52 in total and 3*4=12 for valence) However, the NGXF is already quite dense and CHGCAR file generated is considerably large. Should I keep increasing grid?

Thanks a lot,
Zezhong

./bader CHGCAR_sum
-------------Bader Analysis Result for total electron in ACF.dat--------------
# X Y Z CHARGE MIN DIST ATOMIC VOL
--------------------------------------------------------------------------------
1 0.0000000 0.0000000 0.0000000 10.8211516 1.4047073 16.4904430
2 0.0000000 2.0202265 2.0202265 10.8211878 1.4047073 16.4907675
3 2.0202265 2.0202265 0.0000000 10.8211292 1.4047073 16.4902712
4 2.0202265 0.0000000 2.0202265 10.8210926 1.4047073 16.4899468
--------------------------------------------------------------------------------
VACUUM CHARGE: -0.0000
VACUUM VOLUME: 0.0000
NUMBER OF ELECTRONS: 43.2846

./bader CHGCAR -ref CHGCAR_sum
-------------Bader Analysis Result for valence electron in ACF.dat--------------
# X Y Z CHARGE MIN DIST ATOMIC VOL
--------------------------------------------------------------------------------
1 0.0000000 0.0000000 0.0000000 2.8432735 1.4047073 15.1453143
2 0.0000000 2.0202265 2.0202265 2.8433114 1.4047073 15.1456483
3 2.0202265 2.0202265 0.0000000 2.8432535 1.4047073 15.1451377
4 2.0202265 0.0000000 2.0202265 2.8432166 1.4047073 15.1448132
--------------------------------------------------------------------------------
VACUUM CHARGE: -0.0001
VACUUM VOLUME: 0.0383
NUMBER OF ELECTRONS: 11.3729

-------------INCAR--------------
SYSTEM = Al

# Starting parameters for this run:
ISTART = 0 job : 0-new, 1-cont, 2-samecut
# ICHARG = 2 charge: 0-wave, 1-file, 2-atom, >10-const
# INIWAV = 1 electr: 0-lowe 1-rand 2-diag

# Electronic Relaxation:
PREC = Accurate
ADDGRID = .TRUE. reduce the noise in the forces
SYMPREC = 1E-10
ENCUT = 500
LAECHG = .TRUE.
# NGX = 120
# NGY = 240
# NGZ = 240
NGXF= 240
NGYF= 240
NGZF= 240

NELMDL = -8 number of delayed ELM steps
# NELM = 101 number of ELM steps
EDIFF = 1E-04 energy stopping-criterion for ELM
LREAL = .FALSE. real-space projection (.FALSE.--for very small cells/accurate charge density, .TRUE., On, Auto)
ALGO = normal algorithm for electronic minimisation (normal: DAV, veryfast: RMM, fast: DAV+RMM)
# WEIMIN = 0 stabilises RMM algo

# Writing files:
LCHARG = .TRUE. write electronic charge density
LWAVE = .FALSE. write WAVECAR
LVTOT = .TRUE.
LORBIT = 11

# Ionic Relaxation:
# NSW = 50 max number of geometry steps
# IBRION = 2 ionic relax: 0-MD, 1-quasi-Newton, 2-CG, 3-Damped MD
# EDIFFG = -0.01 force (eV/A) stopping-criterion for geometry steps
# ISIF = 3 (force|stress|ions|shape|vol 0:ynynn 1:yyynn 2:yyynn 3:yyyyy)
# ISYM = 0 (1-use symmetry, 0-no symmetry)

# DOS related values:
# NEDOS = 1201
ISMEAR = -5 (-1-Fermi, 1-Methfessel/Paxton, 0:Gaussian, -5:Blochl tetrahedron)
SIGMA = 0.05 broadening in eV

# Spin-polarized calculations:
# ISPIN = 2
# MAGMOM = 71*0 1

#Dipole moment
# IDIPOL = 3
# LDIPOL = .TRUE.

#Van Der Waals interaction
# LVDW=.TRUE.

# Parallelization flags:
NPAR = 4 ( =1: all nodes work on each band, = else: only one node will work on each band)
LPLANE = .TRUE. (parallelization of plane wave coefficients)

I have now get converged valence electron with increased NGXF. The core electron, however, still not converging with 360 grids in each direction. I guess core charge distribution is not the aim of VASP with PP or PAW methods used. Thus, a full potential method like Wien2K can be for the core in this purpose...

I think that it is just fundamentally difficult to accurately represent functions with cusps on a regular cartesian grid. An additional problem is that while the valence density is normalized so that the sum of the density values (really the charge in each cell) add up to the total, the same is not true for the total charge. Our approach is to do the Bader partitioning based upon the total charge and then integrate the valence charge to calculate atomic properties. We find that this approach converges smoothly with increasing grid points.