Convergence problem in electronic relaxation of surface slab

[cwandtj]

Hello, VASP master and VASP users.

I have question on convergence problem in electronic optimization of surface slab.

I had no problem in convergence when I use relatively smaller slab and bigger slab with "weak polarity and simple geometry".

However, if I want to simulate "big and complex" surface systems that have more than like 5 different ion species, point defect, and strong surface perturbation to have stable charge distribution, I can't have electornic convergence easily.

I thought about options that I can choose.

1) Just wait until electronic strucuture is converged.
2) Use ALGO=Normal not Fast
3) Play with Mixing parameters ( I am using default values)

Please let me know what will be the good ways to solve this problem.

FYI, I am using VASP.4.6.31.

Thank you for your help in advance.

CW


Starting parameters for this run:
NWRITE = 2 write-flag
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 = Normal low | medium | high
ENCUT = 600 Cutoff E is 600eV
NELMDL = -7 number of delayed ELM steps
NELM = 80 number of ELM steps
NELMIN = 10
EDIFF = 1E-05 energy stopping-criterion for ELM
LREAL = AUTO real-space projection (.FALSE., .TRUE., On, Auto)
WEIMIN = 0.001
ALGO = Fast
### IALGO = 48 algorithm (8=CG for small, 48=RMM for big systems)
NSIM = 1 NSIM bands are optimized simultaneously (IALGO=48 only)
ADDGRID =.TRUE.
LMAXMIX = 6
VOSKOWN = 1
ISPIN = 2
MAGMOM = 19*0.0 4.0 -4.0 -4.0 4.0 4.0 -4.0 -4.0 4.0 4.0 -4.0 -4.0 4.0 4.0 -4.0 -4.0 4.0 4.0 55*0.0 0.0 -4.0
IDIPOL = 3

LCHARG = .TRUE.
LWAVE = .TRUE.


Ionic Relaxation:
NSW = 100 max number of geometry steps
IBRION = 3 ionic relax: 0-MD, 1-quasi-Newton, 2-CG, 3-Damped MD
SMASS = 0.4
EDIFFG = -0.02 force (eV/A) stopping-criterion for geometry steps
ISIF = 1 (1:force=y stress=trace only ions=y shape=n volume=n)
ISYM = 2 (1-use symmetry, 0-no symmetry)
POTIM = 0.2 initial time step for geo-opt (increase for soft sys)
NFREE = 2

DOS related values:
### RWIGS = 1.0 0.5 1.2
LORBIT = 12
NEDOS = 3000
EMIN = -15.0
EMAX = 15.0
ISMEAR = 0 (-1-Fermi, 1-Methfessel/Paxton)
SIGMA = 0.02 broadening in eV

Parallelization flags:
NPAR = 2
LPLANE = .TRUE.

[admin]

I would suggest that you set ALGO=Normal (especially if the eigenvalue apectrum is very dense) and decrease the charge and spin density mixing parameters. Please also check whether this poor convergence may be due to oszillations in the magnetic moments.

[cwandtj]

OK, master. I am running my job with ALGO=Normal.

Following questions will be

1) Changing mixing parameters will not change total energy and other relevant values?

2) I check oscillations of magnetic moments.

For my system (LaSrCoFeO3 with oxygen vacancy), I have following values by grep “magnetization� OUTCAR. They are extracted from third ionic relaxation step of my previous calculation.

If I have poor convergence due to magnetic moments, what will be the solution? Do I need to turn off calculation of collinear magnetic moments and only turn on ISPIN?

Thank you~

CW

.
.

augmentation part 151.3052926 magnetization -1.7037021
number of electron 796.0000019 magnetization 0.1346938
augmentation part 151.1814342 magnetization -0.3205949
number of electron 796.0000019 magnetization 0.1387488
augmentation part 151.1674853 magnetization -0.9308089
number of electron 796.0000019 magnetization 0.1413456
augmentation part 150.8690327 magnetization -0.1032669
number of electron 796.0000019 magnetization 0.1450278
augmentation part 151.1196480 magnetization -1.5373301
number of electron 796.0000019 magnetization 0.1514250
augmentation part 151.0919565 magnetization -1.1796085
number of electron 796.0000019 magnetization 0.1512879
augmentation part 151.1552807 magnetization -0.7156791
number of electron 796.0000019 magnetization 0.1573001
augmentation part 151.1633173 magnetization -0.9592272
number of electron 796.0000019 magnetization 0.1578711
augmentation part 151.2394533 magnetization -0.4516363
number of electron 796.0000019 magnetization 0.1585066
augmentation part 151.1019688 magnetization -0.2709689


Following values are from different system which also have problem in convergence.

.
.

number of electron 1176.0000008 magnetization 0.0003615
augmentation part 222.8464046 magnetization -0.1075454
number of electron 1176.0000008 magnetization 0.0003157
augmentation part 222.8448712 magnetization -0.1075871
number of electron 1176.0000008 magnetization 0.0002651
augmentation part 222.8451344 magnetization -0.1079324
number of electron 1176.0000008 magnetization 0.0001993
augmentation part 222.8450869 magnetization -0.1085909
number of electron 1176.0000008 magnetization 0.0001140
augmentation part 222.8443133 magnetization -0.1087549
number of electron 1176.0000008 magnetization 0.0000440
augmentation part 222.8442661 magnetization -0.1091210
number of electron 1176.0000008 magnetization -0.0000001
augmentation part 222.8440713 magnetization -0.1092318
number of electron 1176.0000008 magnetization -0.0000452
augmentation part 222.8435546 magnetization -0.1093096

[admin]

1) it does not look to me as if the magnetic convergence is particularly bad. (please dont compare the moments stemming from the augmentation to the total moments).
have you decreased AMIX,BMIX, AMIX_MAG and BMIX_MAG for this run?
2)the mixing parameters must not have any influence on the converged total energies.
3) if your system has a magnetic moment, you have to set ISPIN.
unless you set LNONCOLLINEAR explicitely , collinear magnetism is assumed by default, there is nothing to be specified in extra (except from starting with FM or AFM configuration by choosing the MAGMOMs accordingly)
4) please in any case check if the convergence of ALL ionic steps is bad. (consider that it may be possible that you relaxed into an unreasonable geometry which does not converge electronically).