relaxation of water/ammonia ice slab not converging with B3LYP

[heisen]

Dear Users,
I created a "slab" of ammonia dihydrate containing 8 water molecules and 4 ammonia molecules. I first relaxed the system without a vacuum to replicate the "bulk" system. This converged fine. I then added a vacuum , fixed the bottom layer of 4 water moles and 2 ammonia molecules to the bulk positions and let the top layer relax. The scf cycle now doesn't converge (the ionic relaxation does converge with the unconverged scf cycles). The d_eps values reach around 0.1*10^(-3) but don't reduce further and the dE values oscillate around +-0.1*10^(-4). I am using B3LYP with the dDsC dispersion correction method. When I performed this same calculation with PBE and the dDsC dispersion correction method everything converged fine (the scf and the ionic relaxation).
I tried:
increasing NELM to 200
changing BMIX=3.0 AMIN=0.01
using the atomic positions from the relaxed slab obtained using PBE and the dDsC dispersion correction method

I have a vacuum of 14A. I have a k-point grid of 2x2x1.
This is my INCAR file:
ISPIN=2
NCORE=6
LREAL=Auto
#ALGO=Fast
ISTART = 0
# ICHARG = 2
ENCUT = 400
ISMEAR = 0 ; SIGMA = 0.05
EDIFF = 1E-5
IBRION=2
POTIM=0.5
NSW=500
#ISIF=3
IVDW=4
#LVDW_EWALD=.TRUE.
LHFCALC = .TRUE.
GGA = B3
AEXX = 0.2
AGGAX = 0.72
AGGAC = 0.81
ALDAC = 0.19
VDW_SR=1.78
VDW_S6=4.89

I would be grateful for any advice on how to make the scf cycle converge for my system,

Dr. Helen Eisenberg
Research Associate
Fritz Haber Center for Molecular Dynamics
Hebrew University of Jerusalem

[ferenc_karsai]

Could you please post the neccessary files of your calculation according to the forum guidelines?

[heisen]

Here are my input files and output files in the attached zip file. Note these are the files for the B3LYP relaxation done starting from the relaxed PBE water/ammonia ice atomic positions. Here the scf also does not converge, but d_eps decreases to around 0.3*10^(-4) eV.
Dr. Helen Eisenberg
Research Associate
Fritz Haber Center for Molecular Dynamics
Hebrew University of Jerusalem

[heisen]

Dear Users,
I managed to fix the problem of the B3LYP scf not converging by:
1) Performing the relaxation using PBE
2) Starting the B3LYP relaxation from the relaxed PBE atomic positions AND wavefunction
3) SETTING ALGO=All. This selects an "all band simultaneous update of orbitals" (the conjugate gradient algorithm is used). [This was the crucial step] Just using steps (1) and (2) didn't solve the problem (I'd already unsuccessfully tried steps (1) and (2))

I found this solution in the vaspwiki page on metaGGA functionals https://www.vasp.at/wiki/index.php/METAGGA
"If convergence problems are encountered, it is recommended to preconverge the calculations using the PBE functional, to read the PBE WAVECAR file. Furthermore, ALGO = A (conjugate gradient algorithm for orbitals) is often more stable than charge density mixing, in particular, if the system contains vacuum regions."
I hope this will help other users who have a scf convergence problem with hybrid functionals and vacuums,
Dr. Helen Eisenberg
Fritz Haber Center for Molecular Dynamics
Hebrew University of Jerusalem

[ferenc_karsai]

Well done!