How to perform GW calculations in VASP?

[sandeep.kumar]

I am trying to calculate band structure of a system containing two atoms per unit cell (graphene) using GW calculations. It is taking a lot of time and after 5 days it shows wall time exceeded.
I have done the self consistent procedure in 3 steps.
In 1st step i have used the INCAR file given as

# Basic setup:
SYSTEM = Graphene
ISMEAR = 0
SIGMA=0.05
EDIFF = 0.00000001

and used the POTCAR file of PBE potentials give as some few lines...

PAW_PBE C_h 06Feb2004
4.00000000000000
parameters from PSCTR are:
VRHFIN =C: s2p2
LEXCH = PE
EATOM = 147.1560 eV, 10.8157 Ry

TITEL = PAW_PBE C_h 06Feb2004
LULTRA = F use ultrasoft PP ?
IUNSCR = 1 unscreen: 0-lin 1-nonlin 2-no
RPACOR = 0.800 partial core radius
POMASS = 12.011; ZVAL = 4.000 mass and valenz
RCORE = 1.100 outmost cutoff radius
RWIGS = 1.630; RWIGS = 0.863 wigner-seitz radius (au A)


Then i copy the CONTCAR file of first step and repeat the calculations using INCAR file as

# Basic setup:
SYSTEM = Graphene unoccupied orbitals
ALGO=Exact
NELM=1
NBANDS=96
ISMEAR = 0
SIGMA=0.05
LOPTICS=.TRUE.

and same POTCAR file as given above.

Then i copied the CONTCAR, WAVECAR and WAVEDER file of second step and repeated the calculations using INCAR as

# Basic setup:
SYSTEM = Graphene GW
ALGO=GW0
LSPECTRAL=.TRUE.
NOMEGA=50
LRPA=.FALSE.
NBANDS=96


and same POTCAR file as above....

and 2nd and 3rd steps are taking a lot of time, 2nd step completed in 4 days on 1 processor and 3rd step even could not get completed after 4 days.
I have also done the calculations on 20 processors, even then it could not get completed, error was wall time exceeded.

Can someone please tell me, where i am doing the mistake. Its taking a huge time even on a very small system

[askhetan]

how big is your system ? How many ions ? How many electrons ? How many kpoints ? Can you share your POSCAR ?
GW calculations usually take a huge amount of memory and time and a lot of empty bands to get any meaningful values. Its virtually impossible to say anything about your system unless you mention your exact specifications.

[sandeep.kumar]

Here is my POSCAR file.

vasp
2.467
0.500 0.866 0.000
-0.500 0.866 0.000
0.000 0.000 10.000
2
Direct
0.333 0.333 0.000
0.666 0.666 0.000

and i have used Monkhorst pack for sampling K-space with a mesh of (25*25*1) in KPOINTS file.

[kaynat_kalvi]

Hi, If you successfully solved the problem can you please guide me as well. or any developer please help here, as I am having the same issue.

[merzuk.kaltak]

Dear kaynat_kalvi,

The attached tutorial provides guidance on GW methodology, including procedures for calculating quasi-particle band structures using Wannier interpolation:
https://www.vasp.at/wiki/index.php/GW_a ... -_Tutorial

For optimal workflow efficiency:

• Begin GW calculations with coarse k-point sampling (e.g., 2×2×1) and systematically increase grid density until convergence criteria are met.

• Use the LDOWNSAMPLE parameter to select subsets of k-points from existing WAVECAR files, eliminating redundant DFT recalculations.

Example:

• Generate initial Kohn-Sham states with 24×24×1 k-point sampling (stored in WAVECAR)

• Perform subsequent GW calculations using coarser k-point grids:
  2×2×1 | 3×3×1 | 4×4×1 | 8×8×1 | 12×12×1

[kaynat_kalvi]

Hi, Thanks for a quick response, Is it not possible to do band structure plots out of a GW cal without using Wannier interpolation?

[kaynat_kalvi]

following this tutorial (https://www-.vasp.at/wiki/index.php/Bandgap_of_Si_in_GW), I got the QP energies for all k-points. the command (./gap_GW.sh OUTCAR), results in
HOMO: band: 8 E= 21.9372
LUMO: band: 9 E= 7.6669
, which is strange, what does this means ?

I am doing this for 2 atom SiGe system, having 8x8x8 G-centered kpoints.

And please guide me plotting the GW band structure with out using Wannier90

[merzuk.kaltak]

Hi, Thanks for a quick response, Is it not possible to do band structure plots out of a GW cal without using Wannier interpolation?

The interpolation of quasiparticle (QP) energies presents inherent limitations compared to Kohn-Sham density functional theory. Specifically, QP band structure visualization requires Wannier interpolation methods, as conventional energy interpolation techniques applicable to Kohn-Sham systems prove inadequate for quasiparticle spectra. This fundamental difference arises from the non-local nature of self-energy corrections in many-body perturbation theory, which prevents direct application of standard band interpolation approaches.

[merzuk.kaltak]

following this tutorial (https://www-.vasp.at/wiki/index.php/Bandgap_of_Si_in_GW), I got the QP energies for all k-points. the command (./gap_GW.sh OUTCAR), results in
HOMO: band: 8 E= 21.9372
LUMO: band: 9 E= 7.6669
, which is strange, what does this means ?

I am doing this for 2 atom SiGe system, having 8x8x8 G-centered kpoints.

And please guide me plotting the GW band structure with out using Wannier90

gap_GW.sh has been taylored to the system in the tutorial. It might not work for any possible system.
I would check if band 8 is really the HOMO and band 9 the corresponding LUMO.
To identify the HOMO and LUMO manually, inspect the OUTCAR as described in the tutorial.