noncollinear spin-orbit calculation not converging

[alpinnovianus]

Hello,

I am having problems with non-collinear spin-orbit calculation that does not converge after many iterations.
Would you please let me know if I made a mistake in the input files below? I use VASP 6.2.1.

The OSZICAR is showing that the dE fulfils the EDIFF but the d eps is stuck. Thus, the electronic loop never finishes. This happens regardless the SAXIS value being is 1 0 0 or 0 0 1.
I also attached the OUTCAR.

Also, could you tell me the reason why in non-collinear calculation the NBANDS value has to be doubled from the collinear one?


INCAR

System = General
LREAL = .FALSE.
NPAR = 4
KPAR = 4
ISIF = 2
LWAVE = .FALSE.
LCHARG = .FALSE.
GGA = PE
METAGGA = SCAN
LASPH = .TRUE.
LMIXTAU = .TRUE.
ADDGRID = .TRUE.
ALGO = Normal
NELM = 2000
PREC = A
IBRION = -1
NSW = 0
ISYM = -1
EDIFF = 1E-05
ENCUT = 550
ENAUG = 650
ISMEAR = 0
SIGMA = 0.01
LORBIT = 11
LMAXMIX = 6
LMAXTAU = 8
AMIX = 0.2
AMIX_MAG = 0.8
BMIX = 0.0001
BMIX_MAG = 0.0001
LSORBIT = .TRUE.
SAXIS = 0 0 1
MAGMOM = 12*0 0 0 5 0 0 -5 6*0 24*0
NBANDS = 176

POSCAR

Ba4 Cu2 Hg2 O8
1.00000000000000
5.5110943570964475 -0.0000000000000000 -0.0000000000000000
-0.0000000000000000 5.5110943570964475 -0.0000000000000000
-0.0000000000000000 0.0000000000000000 9.6241899892571894
Ba Cu Hg O
4 2 2 8
Direct
-0.0000000000000000 0.5000000000000000 0.3027935213623231
0.0000000000000000 0.5000000000000000 0.6972064786376768
0.5000000000000000 0.0000000000000000 0.3027935213623231
0.5000000000000000 -0.0000000000000000 0.6972064786376768
0.0000000000000000 0.0000000000000000 0.5000000000000000
0.5000000000000000 0.5000000000000000 0.5000000000000000
0.0000000000000000 0.0000000000000000 -0.0000000000000000
0.5000000000000000 0.5000000000000000 -0.0000000000000000
-0.0000000000000000 -0.0000000000000000 0.2061066482903998
0.0000000000000000 0.0000000000000000 0.7938933367096025
0.5000000000000000 0.5000000000000000 0.2061066482903998
0.5000000000000000 0.5000000000000000 0.7938933367096025
0.7500000000000000 0.2500000000000000 0.5000000000000000
0.2500000000000000 0.7500000000000000 0.5000000000000000
0.7500000000000000 0.7500000000000000 0.5000000000000000
0.2500000000000000 0.2500000000000000 0.5000000000000000

KPOINTS
Automatic mesh
0
Gamma
8 8 4
0 0 0

OSZICAR

OSZICAR.txt.zip

OUTCAR

OUTCAR.zip

[alexey.tal]

Dear alpinnovianus,

It is recommended to use damped MD (ALGO = Damped) or conjugate gradient (ALGO = All) algorithms in combination with METAGGA functionals (see wiki). So perhaps you can try these algorithms and see if the convergence improves.

Also, could you tell me the reason why in non-collinear calculation the NBANDS value has to be doubled from the collinear one?

In non-collinear calculations, the orbitals are represented by spinors with up and down components, hence the number of required bands should be doubled.

[alpinnovianus]

About changing the ALGO/TIME tag and changing the AMIX, AMIX_MAG tag, may I ask a question?
Can they be changed together (say reduce AMIX, AMIX_MAG to 0.05 and also setting ALGO=A with TIME=0.1)?

[alexey.tal]

By default the direct optimization algorithms (ALGO = All and ALGO = Damped) do not use density mixing and should not be affected by the density mixing parameters. However, if the flag LSUBROT = .TRUE. is set, the code uses the density mixing in the subspace rotations.