Clarification on GW method in VASP

[askhetan]

I have been trying to do ScGW0 calculation for analysis of band gaps on an Li2O2 of 48 atoms (108 electrons) having 10 Angst slab thickness + 15 Angst vacuum gap. I do all simulations at 11x11x1 kpts on a gamma centered mesh.
I performed the hexagonal slab relaxation at ENCUT=640 and EDIFF = 1E-8 and NBANDS = 324 (3 times the number of occupied bands). Next I used ALGO=Exact to diagonalize the hamiltonian at ENCUT=600 and NBANDS=216 (2 times the number of occupied bands.) Finally my ScGW0 script looks like:

ALGO = SCGW0
ENCUTGW = 200
NOMEGA = 100
ISMEAR = -5
ISPIN = 2
SIGMA = 0.01
LREAL = .FALSE.
NELM = 5
LORBIT = 11
PRECFOCK = FAST
MAXMEM = 7850
KPAR = 4
NBANDS = 216

Upon doing this, my calculations always stop at the point:
---------------------------------------------------------------------------------------
LDA part: xc-table for Pade appr. of Perdew
found WAVECAR, reading the header
WAVECAR: different cutoff or change in lattice found
POSCAR, INCAR and KPOINTS ok, starting setup
WARNING: small aliasing (wrap around) errors must be expected
FFT: planning ...
reading WAVECAR
the WAVECAR file was read successfully
initial charge from wavefunction

The Fermi energy was updated, please check that it is located mid-gap
values below the HOMO (VB) or above the LUMO (CB) will cause erroneous energies
E-fermi : -5.7589

calculate exact exchange contribution
---------------------------------------------------------------------------------------

And there are no further results ever after a couple of days. Could someone tell me please, what is going wrong here? ALso, why does the Fermi level need to be updated ? I have a couple of other questions as well.
1) I know there is a mistake here that instead of NBANDS, I should use NBANDSGW=216. I checked out the default value of NBANDSGW in OUTCAR was -1. What does this value stand for - twice of the number of occupied bands (as mentioned on the vasp website)? This would make it automatically as 216, which is what I want to calculate, so that would not be a problem. However, is the default NBANDSGW enough ?
2) If I use NBANDSGW = 1.5 the number of total bands, for eg 162, then which are the bands that are updated - are they the lowest energy bands or the highest energy bands (including the empty ones) ? Is there a way to post process these analyses to get bands belonging to certain atoms ? Could you please direct me to it.

Thanks in advance for your answer.

MfG

[admin]

Have you tried the same calculation with a simple 3D structure (to exclude problems originating in the vacuum layer)?

[askhetan]

I tried the bulk system. And the error is exactly the same. For the bulk system, here is my POSCAR file:
===========================================================
Li24 O24
1.00000000000000
6.3607239722999998 0.0000000000000000 0.0000000000000000
-3.1803619861999999 5.5085485465000001 0.0000000000000000
0.0000000000000000 0.0000000000000000 11.5552835464000001
Li O
24 24
Selective dynamics
Direct
0.0000000000000000 0.0000000000000000 0.0000000000000000 T T T
0.9999999999930296 0.4999999999546176 0.0000000000000000 T T T
0.4999999999764171 0.0000000000000000 0.0000000000000000 T T T
0.4999999999694467 0.4999999999546176 0.0000000000000000 T T T
0.1666660010957486 0.3333329857221017 0.1711953871351994 T T T
0.1666660010393315 0.8333330154486305 0.1711953871351994 T T T
0.6666669846080069 0.3333329857221017 0.1711953871351994 T T T
0.6666669845515898 0.8333330154486305 0.1711953871351994 T T T
0.0000000000000000 0.0000000000000000 0.3354601808605580 T T T
0.9999999999930296 0.4999999999546176 0.3354601808605580 T T T
0.4999999999764171 0.0000000000000000 0.3354601808605580 T T T
0.4999999999694467 0.4999999999546176 0.3354601808605580 T T T
0.3333329855940974 0.1666660010799603 0.4999999999826912 T T T
0.3333329856002933 0.6666669845967590 0.4999999999826912 T T T
0.8333330154413261 0.1666660010799603 0.4999999999826912 T T T
0.8333330154475220 0.6666669845967590 0.4999999999826912 T T T
0.0000000000000000 0.0000000000000000 0.6645398489612973 T T T
0.9999999999930296 0.4999999999546176 0.6645398489612973 T T T
0.4999999999764171 0.0000000000000000 0.6645398489612973 T T T
0.4999999999694467 0.4999999999546176 0.6645398489612973 T T T
0.1666660010957486 0.3333329857221017 0.8288045979452150 T T T
0.1666660010393315 0.8333330154486305 0.8288045979452150 T T T
0.6666669846080069 0.3333329857221017 0.8288045979452150 T T T
0.6666669845515898 0.8333330154486305 0.8288045979452150 T T T
0.3333329855940974 0.1666660010799603 0.1046348481557757 T T T
0.3333329856002933 0.6666669845967590 0.1046348481557757 T T T
0.8333330154413261 0.1666660010799603 0.1046348481557757 T T T
0.8333330154475220 0.6666669845967590 0.1046348481557757 T T T
0.3333329855940974 0.1666660010799603 0.2377056122073498 T T T
0.3333329856002933 0.6666669845967590 0.2377056122073498 T T T
0.8333330154413261 0.1666660010799603 0.2377056122073498 T T T
0.8333330154475220 0.6666669845967590 0.2377056122073498 T T T
0.1666660010957486 0.3333329857221017 0.4333554499444361 T T T
0.1666660010393315 0.8333330154486305 0.4333554499444361 T T T
0.6666669846080069 0.3333329857221017 0.4333554499444361 T T T
0.6666669845515898 0.8333330154486305 0.4333554499444361 T T T
0.1666660010957486 0.3333329857221017 0.5666445798774191 T T T
0.1666660010393315 0.8333330154486305 0.5666445798774191 T T T
0.6666669846080069 0.3333329857221017 0.5666445798774191 T T T
0.6666669845515898 0.8333330154486305 0.5666445798774191 T T T
0.3333329855940974 0.1666660010799603 0.7622944027295446 T T T
0.3333329856002933 0.6666669845967590 0.7622944027295446 T T T
0.8333330154413261 0.1666660010799603 0.7622944027295446 T T T
0.8333330154475220 0.6666669845967590 0.7622944027295446 T T T
0.3333329855940974 0.1666660010799603 0.8953651518961507 T T T
0.3333329856002933 0.6666669845967590 0.8953651518961507 T T T
0.8333330154413261 0.1666660010799603 0.8953651518961507 T T T
0.8333330154475220 0.6666669845967590 0.8953651518961507 T T T
===========================================================

The INCAR file that I use for the wavefunction-diagonalization part is part is:
===========================================================
ALGO = Exact
IBRION = 2
ICHARG = 1
ISIF = 2
ISMEAR = -5
ISPIN = 2
LREAL = .FALSE.
NELM = 1
PREC = Accurate
SIGMA = 0.01
IDIPOL = 3
LDIPOL = .TRUE.
LDAU = .TRUE.
LDAUJ = 0 0 # Li O
LDAUL = 0 2 # Li O
LDAUU = 0 5 # Li O
LDAUPRINT = 1
LDAUTYPE = 2
ADDGRID = .TRUE.
LMAXMIX = 6
LORBIT = 11
LOPTICS = .TRUE.
EDIFF = 1E-8
MAXMEM = 7850
ENCUT = 600
NBANDS = 216
KPAR = 4
===========================================================

In the system there are 108 filled bands. I relax and optimise the geometry of the system with NBANDS=324 (3x108). I diagonalise with NBANDS=216 (only 2x108 because of memory limitations) bands and then also do the GW with NBANDSGW=216. What do you think about the choices of these values. WHat is the default value of NBANDSGW ? In the OUTCAR, I see it as -1. WHat does that imply?

Thanks in advance for your answers.

MfG,

[admin]

At a high memory requirements the calculation ends with a segmentation fault or can simply stay hanged
(this is most probably your case). Make tests, decrease e.g. the k-points set and check the memory requirements.
To your setup: ENCUT very high, KPOINTS set very high. The high number of NBANDS in the DFT calculation
is not necessary at all (316). The increase of NBANDS could be important in the GW step, but increases
the cost of the calculation.

[askhetan]

http://s15.postimg.org/ossd63a5n/kpoints.png
As you can see in this convergence test where the upper lines are lumo and the lower lines are homo, convergence is fist achived only at 11x11x1 kpts. its is equal to 8x8x1 and 10x10x1, however, this is a direct band gap and I need to always include the gamma point. therefore I cannot use the even numbered grid points. 9x9x1 is way off the converged ones.
http://s14.postimg.org/n364y5dg1/bands.png
As you see here, for a smaller system with 72 instead of 108 occuupied bands - although the lumo seems to convege at 144 bands, the homo is still not converged at 180 bands which is > 144 (2x72). This is why I have chosen such high values.