Converging defect supercells containing Europium 3+

[mbkumar]

Hello All,
I am trying to minimize energies of O-Eu defect cluster in BaBrF at different charge states. Eu defect is on Ba sublattice and O defect can be on either of Br or F sublattices. The defect supercells with overall net charge of -1 converged without any problems. But I am finding it hard to converge the neutral cells, where Eu and O have 3+ and 2- oxidation states respectively. I played with different settings such as AMIX, BMIX, NSIM with no success.
I am giving below my current INCAR and the head of the POSCAR. Any help is appreciated.

Thanks,

INCAR:
ALGO = Normal
PREC = Normal
LREAL = Auto
ISYM = 0
#MAXMIX = 100
#AMIX = 0.2
#AMIN = 0.01
#BMIX = 0.01
#AMIX_MAG = 0.2
#AMIN_MAG = 0.01
#BMIX_MAG = 0.001
NELMDL = 8
NSIM = 2

IBRION = 2
ISIF = 2
NELM = 300
NSW = 10
EDIFFG = -0.05
EDIFF = 1e-05

ENCUT = 400
SIGMA = 0.02
ISPIN = 2
ISMEAR = -2
FERWE = 438*1 138*0
FERDO = 432*1 144*0
#MAGMOM = 6 107*0
#NUPDOWN = 6

LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = 3 -1 -1 -1 -1
LDAUU = 2.2 0.0 0.0 0.0 0.0
LDAUJ = 0.0 0.0 0.0 0.0 0.0
LASPH = .TRUE.
LMAXMIX = 6

LORBIT = 11
LWAVE = False

The head of the POSCAR is:
Eu1 Ba35 Br35 O1 F36
1.00000000000000
13.7314930000000004 0.0000000000000000 0.0000000000000000
0.0000000000000000 13.7314930000000004 0.0000000000000000
0.0000000000000000 0.0000000000000000 15.1371839999999995
Eu Ba Br O F
1 35 35 1 36

[admin]

Have you tried it without FERWE and FERDO defined? This is another degree of freedom which could improve the convergence
(4f6 spin up is automatically formed on Eu3+).

[alex]

Hi mbkumar,

your Eu(III) in the state f6 is a notoriously hard case to converge. But I think you figured that out already. ;-)

I might add some suggestions for INCAR:
NELMDL = 30 ! You should converge the wavefunction to sth. like 1e-10 before you start calculating and mixing with charges
NUPDOWN = 6 ! Keep it. Help the program by giving it the spin state. Otherwise you'll start fully metallic.
Holds for MAGMOM, too.
NELM = 1000 ! I've seen cases with converge after 800 cycles only
ISMEAR = 0 ! Gaussian smearing for large cells
Loose FERWE and FERDO.

In a second set I'd suggest in addition:
AMIX = 0.2 ! linear mixing parameter
BMIX = 0.0001 ! cutoff wave vector for Kerker mixing scheme
AMIX_MAG = 0.8
BMIX_MAG = 0.0001

Cheers,

alex

[mbkumar]

Have you tried it without FERWE and FERDO defined? This is another degree of freedom which could improve the convergence
(4f6 spin up is automatically formed on Eu3+).

Thank you for your suggestion and you are right.
After posting the question, I was able to get the convergence. To get the convergence, I removed FERWE and FERDO definitions and increased the smearing factor. In my opinion, the parameter that really helped in convergence is the smearing factor. When I increased SIGMA from 0.02 to 0.1 without defining FERWE and FERDO, all the troubling systems converged.

For BaBrF:O-Eu defect complex with neutral cell, the 4f states of Eu3+ are found to be energetically very close to O2- states, which are nearly 1 eV above VBM. When I increased SIGMA, multiple states in the vicinity of Fermi level are fractionally occupied, leading to convergence.

[mbkumar]

Hello Alex,

Thanks for the suggestions. I am posting the INCAR that helped me to get convergence quickly for future reference. I used gamma k-point only.

ALGO = Normal
PREC = Normal
LREAL = Auto
ISYM = 0
NELMDL = 8

IBRION = 2
ISIF = 2
NELM = 200
NSW = 30
EDIFFG = -0.03
EDIFF = 1e-05

ENCUT = 400
ISMEAR = 0
SIGMA = 0.08 # helps to have it around 0.1
ISPIN = 2
MAGMOM = 6 107*0
NUPDOWN = 6

LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = 3 -1 -1 -1 -1
LDAUU = 2.2 0.0 0.0 0.0 0.0
LDAUJ = 0.0 0.0 0.0 0.0 0.0
LASPH = .TRUE.
LMAXMIX = 6

The first ionic step took 182 cycles to converge, but the following steps took much fewer cycles.

I also got convergence when I used
ALGO=Fast
NSIM = 2
NELMDL=8
while keeping the rest of the parameters same as in the INCAR I posted first (i.e. with defined FERWE and FERDO).