about the computation time of GW

[zsjan]

Dear all:
I recently want to calculate the GW corrections for KCl. However, I was told that my computation is too time consuming according to one's experience. I do not know the reason for this. I have compared the time using the tutorial files but found no big differences. Therefore, I post here my input file and procedure to ask you that is it really problematic in my run? Thank you!

pw.in:
scf:
&control
calculation = 'scf'
restart_mode='from_scratch'
prefix='KCL'
verbosity = 'high'
pseudo_dir = './'
outdir = './'
/
&system
ibrav = 2
a = 6.29
nat = 2
ntyp = 2
ecutwfc = 60
nbnd = 500
occupations = 'smearing'
smearing = 'fermi-dirac'
degauss = 0.001
/
&electrons
/
ATOMIC_SPECIES
Cl 35.4527 Cl.pbe-mt_fhi.UPF
K 39.0983 K.pbe-mt_fhi.UPF

ATOMIC_POSITIONS {crystal}
K 0.00000 0.00000 0.00000
Cl 0.50000 0.50000 0.50000

K_POINTS automatic
4 4 4 0 0 0

nscf:
&control
calculation = 'nscf'
prefix='KCL'
verbosity = 'high'
wf_collect = .true.
pseudo_dir = './'
outdir = './'
/
&system
ibrav = 2
a = 6.29
nat = 2
ntyp = 2
ecutwfc = 60
nbnd = 500
occupations = 'smearing'
smearing = 'fermi-dirac'
degauss = 0.001
/
&electrons
/
ATOMIC_SPECIES
Cl 35.4527 Cl.pbe-mt_fhi.UPF
K 39.0983 K.pbe-mt_fhi.UPF

ATOMIC_POSITIONS {crystal}
K 0.00000 0.00000 0.00000
Cl 0.50000 0.50000 0.50000

K_POINTS automatic
4 4 4 0 0 0

and the GW input:
em1d # [R Xd] Dynamical Inverse Dielectric Matrix
gw0 # [R GW] GoWo Quasiparticle energy levels
ppa # [R Xp] Plasmon Pole Approximation
HF_and_locXC # [R XX] Hartree-Fock Self-energy and Vxc
EXXRLvcs= 3991 RL # [XX] Exchange RL components
Chimod= "Hartree" # [X] IP/Hartree/ALDA/LRC/BSfxc
% BndsRnXp
1 | 500 | # [Xp] Polarization function bands
%
NGsBlkXp= 400 RL # [Xp] Response block size
% LongDrXp
1.000000 | 0.000000 | 0.000000 | # [Xp] [cc] Electric Field
%
PPAPntXp= 27.21138 eV # [Xp] PPA imaginary energy
% GbndRnge
1 | 500 | # [GW] G[W] bands range
%
GDamping= 0.10000 eV # [GW] G[W] damping
dScStep= 0.10000 eV # [GW] Energy step to evalute Z factors
DysSolver= "n" # [GW] Dyson Equation solver (`n`,`s`,`g`)
%QPkrange # [GW] QP generalized Kpoint/Band indices
1| 8| 1|500|
%
%QPerange # [GW] QP generalized Kpoint/Energy indices
1| 8| 0.0|-1.0|
%
The run cost four days using 12 cores:
<04d-12h-50m-53s> [07] Game Over & Game summary

So is it really some problem in my calculations?

[Daniele Varsano]

Dear zsjan,
please fill your signature with the complete affiliation. This is a rule of the forum.
I do not know if I well understood your question: anyway if you ask for the reason of the long duration of your run, this is due to the fact you are calculating quasiparticle correction for 500 bands and 8 kpoints, so 4000 states!!!

Code: Select all

%QPkrange # [GW] QP generalized Kpoint/Band indices
1| 8| 1|500|
%

Why are you interested on such a big amount of states?, why do you want calculate GW correction for high energies unoccupied states? This is useless,
If you are interested to a band structure, surely you can stop to few states above the Fermi energy, saving a lot of time.
Best,
Daniele

[zsjan]

Dear zsjan,
please fill your signature with the complete affiliation. This is a rule of the forum.
I do not know if I well understood your question: anyway if you ask for the reason of the long duration of your run, this is due to the fact you are calculating quasiparticle correction for 500 bands and 8 kpoints, so 4000 states!!!

Code: Select all

%QPkrange # [GW] QP generalized Kpoint/Band indices
1| 8| 1|500|
%

Why are you interested on such a big amount of states?, why do you want calculate GW correction for high energies unoccupied states? This is useless,
If you are interested to a band structure, surely you can stop to few states above the Fermi energy, saving a lot of time.
Best,
Daniele

Thank you!
Yes, I am concerned with the duration of the run since it is too computational cost!
About the states, my previous understanding is that we need enough empty states to converge the screening interactions in the GW calculations, so I have used many states up to 500. Is it a mistake?

[Daniele Varsano]

Yes,
check the difference between GbndRnge (empty states you include to converge your calculation) and QPkrange (states you want to calculate the GW correction).
If you want calculate for instance a direct gap only you need to include just two states in QPkrange, of course including a lot of emtpy states in GbndRnge to get convergences. Now you are calculating GW correction for 4000 states, considering states you are not interested in.

Best,
Daniele