Large GW gap

[sdwang]

Dear SdWang,

the problem of your calculation is that you are treating a 1d system,
and in this cases there are problem of divergences in the integrals of
the coulomb potential in the Bz. With your sampling you are doing
a 1d integral instead that a 3d integral.The error in this case is particularly
big for the exchange part of the self energy that tends to explode
increasing the k-point sampling. The problem can be overcome
by using the random integration method (RIM). You can activate it
by adding the yambo -c option and activating these two lines:

Code: Select all

RandQpts = 1000000		#	(RIM)	 Number of random q-points in the BZ
RandGvec	RL = 1 	#	(RIM)	Coulomb interaction RS components

Check these values, looking at the exchange part that is the less expensive.
I invite you to read the documentation about the RIM.
This part is also documented in the Yambo paper

Dear,
I wonder about the coulomb cutoff. You said the problem can be overcome
by using the random integration method (RIM). So, is the coulomb cutoff neccesarry in 0D,1D,2D system? Can I overcome the problem ONLY use RIM.
Another question: the coulomb cutoff is only used to speed convergence???

[Daniele Varsano]

Dear Shudogn Wang,
the RIM eliminate the divergencies in the integration of the Coulomb potential with non 3d sampling.
The cutoff coulomb potential speed-up convergencies with respect the cell size volume.
For GW calculation the convergence with the volume of the cell could be extremely slow,
please read the paper I suggested you in previous post. In order to apply the coulomb cutoff,
some care is needed in the choice of the cutoff parameter with respect the cell volume.

Cheers,

Daniele

[sdwang]

Dear Daniele:
I have test my system with no coulomb cutoff,and the GW gap is 5.97 eV.but when applied coulomb cutoff,the gap is enlarged ,my supercell with constant a=4.7 au,b=47au,c=27au.and periodic is along a(x direction),my system within b direction is 20au,so the vaccum =47-20=27au,along c direction the vacuum is 27au.I test with
CUTGeo= "box yz" # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere
% CUTBox
0.000 | cuty.000 | cutz.000 |
where
cuty=22,24,25,26,27,28,30,30,33,35,40 and corresponding
cutz=14,14,14,14,14,14,14,25,14,25,25,and GW
gap=9.23/6.77/6.24/6.71/6.31/7.31/6.87/7.88/5.21/7.83/8.66
The gap is very differen without appling coulomb cutoff. So I wanna kown which result is more "sense", with or without the coulomb cutoff???

Thanks!

[Daniele Varsano]

Dear Shudong,
you have very strong oscillation of your gap that looks me strange.
Be sure you switched on the RIM while constructing the box cutoff.
In the box cutoff the length of the side is the total side, when using the cylinder
it is the half, i.e. you cut from -L/2 to L/2.

When using not cubic supercell as in you case, the choice of the cutoff parameter
could be difficult. You can try to repeat your calculations without cutoff, but using
the RIM, with bigger supercell and looking at the behavior of the gap toward
convergences. You can also compare your values using the cylinder, in this case only the
radius have to be tested (but also a convergence with you k-point sampling is required).

Moreover, what is the cause of your strong oscillations, the HF or the correlation part of the
self energy? Be sure you are not mixing parts of the calculation calculated with some parameter
with other files previously calculated with different parameter, it should be written in the report.

Anyway, as your cell is not that big, I suggest you to repeat calculations without using cutoff
but bigger cell, in this way you have an idea of your converged value.

Best,

Daniele

[sdwang]

Thank you very much!I wiil try it.

[sdwang]

Dear Shudong,
you have very strong oscillation of your gap that looks me strange.
Be sure you switched on the RIM while constructing the box cutoff.
In the box cutoff the length of the side is the total side, when using the cylinder
it is the half, i.e. you cut from -L/2 to L/2.

I give a picture to demonstrate my problem:
...|...====...|...====......====......
_____|--L-|
__|----D------|
where the '====' is ribbon with width L, and '......' is vaccum with thickness D-L, D is the lattice prameter in nonperiodic direction y. My question is :
the cutoff in input file is L<Zcut<D-L? or L<Zcut<D and or L<Zcut<L+(D-L)/2? e.g. in( I take box like cutoff)
% CUTBox
0.000 | Zcut | 0.000 | # [CUT] [au] Box sides?
Thanks!

[Daniele Varsano]

Dear SD Wang,
this has been discussed in another post.
Please have a look here.
This is the condition you have to fulfill: L<Zcut<D-L, please read the reference article, and take
in mind that there is a factor 2 when constructing a box shaped cutoff, so zcut > 2L

Best

Daniele

[sdwang]

Dear SD Wang,
this has been discussed in another post.
Please have a look here.
This is the condition you have to fulfill: L<Zcut<D-L, please read the reference article, and take
in mind that there is a factor 2 when constructing a box shaped cutoff, so zcut > 2L

Best

Daniele

I am confused.For my unitcell, L=20au, D=47au, so 20<zcut<27 au? and the factor 2 how to considered? I set zut in
CutOff Geometry :box y
Box sides length [au]: 0.00000 23.00000 0.000
or 23/2=11.5 au or 23*2=46au?

[Daniele Varsano]

In this case zcut should be at least 40, 46 is better.
This means an interaction range of 20/23 that is what you need.
Than may be you should consider unit cells little bit bigger,
but could be enough, the density decay exponentially, so the error you
commit can be negligible.

Cheers,

Daniele

[sdwang]

Dear developers:
I have the GW corrections about my 2D rectangle unitcell including 2O and 4 atoms.I chose the cueoff along Z direction(40 au) as following:
Zcut =18 /19 / 20 / 21 / 22 / 23 / 24 / 25 / 26 / 27 / 28, and corresponding GW gap are:
GWg=5.52/5.63/4.80/5.96/5.61/5.12/4.99/4.96/5.87/6.35/5.69 ,the GW gap does not converge.( I test the cutoff value, it is reasionable between 18 au-26au because of in output file of cutoff test ,the third column and the 4th column resemble each other )
My input file are:
>yambo -i
MaxGvecs= 89195 RL # [INI] Max number of G-vectors planned to use
and I didnot changge this value in all calculation.
>yambo -c -x -v 2
rim_cut # [R RIM CUT] Coulomb interaction
xxvxc # [R XX] Hartree-Fock Self-energy and Vxc
FFTGvecs= 14257 RL # [FFT] Plane-waves
RandQpts= 1000000 # [RIM] Number of random q-points in the BZ
RandGvec= 11 RL # [RIM] Coulomb interaction RS components
#QpgFull # [F RIM] Coulomb interaction: Full matrix
% Em1Anys
0.000 | 0.000 | 0.000 | # [RIM] X Y Z Static Inverse dielectric matrix
%
IDEm1Ref=0 # [RIM] Dielectric matrix reference component 1(x)/2(y)/3(z)
CUTGeo= "box z" # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere
% CUTBox
0.00000 | 0.00000 | Zcut.00000 | # [CUT] [au] Box sides (Zcut is 18-28)%
CUTRadius= 0.000000 # [CUT] [au] Sphere/Cylinder radius
CUTCylLen= 0.000000 # [CUT] [au] Cylinder length
#CUTCol_test # [CUT] Perform a cutoff test in R-space
EXXRLvcs= 89195 RL # [XX] Exchange RL components
%QPkrange # [GW] QP generalized Kpoint/Band indices
1| 3| 11|18|
%
%QPerange # [GW] QP generalized Kpoint/Energy indices
1| 3| 0.0|-1.0|
>yambo -c -g n -p p -V 2
rim_cut # [R RIM CUT] Coulomb interaction
gw0 # [R GW] GoWo Quasiparticle energy levels
ppa # [R Xp] Plasmon Pole Approximation
xxvxc # [R XX] Hartree-Fock Self-energy and Vxc
em1d # [R Xd] Dynamical Inverse Dielectric Matrix
FFTGvecs= 14257 RL # [FFT] Plane-waves
RandQpts= 1000000 # [RIM] Number of random q-points in the BZ
RandGvec= 11 RL # [RIM] Coulomb interaction RS components
#QpgFull # [F RIM] Coulomb interaction: Full matrix
% Em1Anys
0.000 | 0.000 | 0.000 | # [RIM] X Y Z Static Inverse dielectric matrix
%
IDEm1Ref=0 # [RIM] Dielectric matrix reference component 1(x)/2(y)/3(z)
CUTGeo= "box z" # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere
% CUTBox
0.00000 | 0.00000 | Zcut.00000 | # [CUT] [au] Box sides (same as above input)
%
CUTRadius= 0.000000 # [CUT] [au] Sphere/Cylinder radius
CUTCylLen= 0.000000 # [CUT] [au] Cylinder length
#CUTCol_test # [CUT] Perform a cutoff test in R-space
EXXRLvcs= 89195 RL # [XX] Exchange RL components
XfnQPdb= "none" # [EXTQP Xd] Database
XfnQP_N= 1 # [EXTQP Xd] Interpolation neighbours
% XfnQP_E
0.000000 | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v)
%
% QpntsRXp
1 | 32 | # [Xp] Transferred momenta
%
% BndsRnXp
1 | 200 | # [Xp] Polarization function bands
%
NGsBlkXp= 1000 RL # [Xp] Response block size
CGrdSpXp= 100.0000 # [Xp] [o/o] Coarse grid controller
% EhEngyXp
-1.000000 |-1.000000 | eV # [Xp] Electron-hole energy range
%
% LongDrXp
1.000000 | 0.000000 | 0.000000 | # [Xp] [cc] Electric Field
%
PPAPntXp= 27.21138 eV # [Xp] PPA imaginary energy
GfnQPdb= "none" # [EXTQP G] Database
GfnQP_N= 1 # [EXTQP G] Interpolation neighbours
% GfnQP_E
0.000000 | 1.000000 | 1.000000 | # [EXTQP G] E parameters (c/v)
%
% GbndRnge
1 | 200 | # [GW] G[W] bands range
%
GDamping= 0.10000 eV # [GW] G[W] damping
QPreport= "kpbne0ees0" # [GW] QP info. Keys: kp/bn/xx/xc/s0/sq/e0/eq/ee/zf/ds/lm/lf
%QPkrange # [GW] QP generalized Kpoint/Band indices
1| 3| 11| 18|
%
%QPerange # [GW] QP generalized Kpoint/Energy indices
1| 3| 0.0|-1.0|
%
The NGsBlkXp= 1500 converge the GW value.

My question is why the GW gap do not show convergence as the cutoff value? Where is my mistake?
S. D. Wang