Unstable magnetic states and non-reproducible energies in DFT+U

[abdelali_elomrani]

Dear VASP Community,

I am facing a strange problem in my spin-polarized DFT+U calculations. I perform a full structural relaxation, and the calculation converges normally. However, when I copy the CONTCAR to POSCAR and restart the calculation, the system continues to iterate further, and the total energy changes significantly compared to the previously converged result.

I repeated the restart several times, and the same behavior occurs each time. As a result, I cannot reproduce the same final energy consistently from one run to another.

I also noticed that the magnetic moments fluctuate between atoms in the system, and the magnetic configuration changes after restarting the calculation.

I tested lower mixing parameters (AMIX and AMIX_MAG), down to:

AMIX = 0.1
AMIX_MAG = 0.2

but the same problem still occurs.

In addition, at the beginning of the study, I tested several MAGMOM initializations, including low-spin, intermediate-spin, and high-spin configurations for the transition-metal elements in the structure, and selected the configuration that minimized the total energy. Despite this, the calculations still converge to different magnetic states after restarting.

Please take a look at my INCAR file (vasp 6.5.0 version).

NCORE = 7
LWAVE = .FALSE.
LCHARG = .FALSE.
ISTART = 1
ISMEAR = 0
SIGMA = 0.04
ICHARG = 2
IBRION = 2
POTIM = 0.2
ISIF = 3
NSW = 3000
NELM = 3000
NELMIN = 6
ISYM = 0
ISPIN = 2
LMAXMIX = 6
ADDGRID = .TRUE.
ALGO = All
TIME= 0.1
ISEARCH = 1
NELMDL = -20
LREAL = Auto
PREC = Accurate
EDIFF = 1E-06
EDIFFG = -0.02
ENCUT = 520.0
GGA = PE
####Magnetization####
LORBIT = 11
MAGMOM = 81*0.1 27*2 5*0.01 5*0.01 6*0.01 6*0.01 5*0.01
NUPDOWN = 76
AMIX = 0.1
BMIX = 0.00001
AMIX_MAG = 0.2
BMIX_MAG = 0.00001

LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = -1 2 -1 -1 -1 -1 -1
LDAUU = 0.0 3.32 0.0 0.0 0.0 0.0 0.0
LDAUJ = 0.0 0.0 0.0 0.0 0.0 0.0 0.0
LDAUPRINT = 0
LASPH = .TRUE.

[martin.schlipf]

If you want to prevent the magnetic configuration not to change you need to keep the WAVECAR or at least the CHGCAR between restarts. When the magnetic configuration changes it is expected that forces change as well.

Code: Select all

MAGMOM = 81*0.1 27*2 5*0.01 5*0.01 6*0.01 6*0.01 5*0.01
NUPDOWN = 76

Please check that this is really the setup you want. Note that NUPDOWN will have precedence so you should get exactly a magnetization of 76 and VASP is only free to decide where to put the electrons not to change the spin polarization. You should only set this flag if you know that 76 is definitely correct. Otherwise you may prevent discovery of a lower lying state. Anyway, the MAGMOM (total 62.3) is inconsistent with the 76 electrons so you will see large steps in the first few iterations as the density shifts from the initialization (MAGMOM) to the biased ground state (NUPDOWN).