how to keep the spin states during the relaxation

[zongtan_fang1]

Hi,

I am having an issue of spin state change on a specific Ni atom in my NMC (Ni, Mn, Co) system during relaxation in my GGA+U calculations.

The initial ionic step gives Ni3+, then gradually lost spin to Ni4+ during the relax.

Also my GGA PBE calculation without U is able to keep Ni3+ unchanged.

I am wondering if there is a way to keep the spin state for GGA+U?

I also tried reading WAVECAR from PBE to do PBE+U, still converged to a different state.
NUPDOWN also didn't help.

I am using the following INCAR for GGA+U; any comment or suggestion is welcomed.

Thanks.


=========================================
Electronic relaxation:
ENCUT = 480.0 ! Cutoff energy for plane waves basis set in eV
ALGO = Fast ! A Davidson blocked iteration scheme
NELMIN = 4 ! Minimum number of eletronic selfconsistency (SC) steps
NELM = 500 ! Maximum number of electronic SC steps
NELMDL = -8 ! Number of NON-selfconsistency steps
EDIFF = 1.0E-6 ! Global-break condition for the electronic SC-loop (ELM)
AMIX = 0.4
BMIX = 1.000

Calculation mode:
PREC = Accurate ! Calculation level (Changes FFT-grids)
ISPIN = 2 ! spin-polarized calculations
ADDGRID = .TRUE. ! PAW flag

LREAL=.FALSE. ! Should not be used for final total energy calculations
LWAVE=.FALSE.
LCHARG=.FALSE.

Ionic relaxation:
NSW = 500 ! Number of ionic steps
EDIFFG = -0.03 ! stop if all forces are smaller than |EDIFFG|
IBRION = 2 !
ISIF = 3 ! Force, stress, relax ions, keep cell shape, keep cell volume
POTIM = 0.02

Integration over the Brillouin zone (BZ):
ISMEAR = 0 !
SIGMA = 0.05

DOS calculation:
LORBIT = 11 ! Calculate the DOS without providing the Wigner Seitz radius
NEDOS = 1001 ! Number of points to calculate the DOS

MAGMOM=120*0 60*0 36*2 12*4 12*0

GGA+U Calculations:
LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = -1 -1 2 2 2
LDAUU = 0 0 5.96 5.1 5.0
LDAUJ = 0 0 0 0 0
LMAXMIX = 4

INIWAV = 1 ! initial guess
ISTART = 0 ! from scratch
ICHARG = 2 ! Take superposition of atomic charge densities

VOSKOWN = 1 ! Vosko-Wilk-Nusair interpolation is used
NBLOCK = 1
NWRITE = 1

NPAR=8
=======================