Example of a** yambo input file for GW calculation in a solid**. In order to generate the following input file you can use the command “`yambo -g n -p p -V par"` with yambo 4.0.2 (and with yambo 3.4.2 “yambo -g n -p p”).

gw0 # [R GW] GoWo Quasiparticle energy levels ppa # [R Xp] Plasmon Pole Approximation rim_cut # [R RIM CUT] Coulomb potential HF_and_locXC # [R XX] Hartree-Fock Self-energy and Vxc em1d # [R Xd] Dynamical Inverse Dielectric Matrix NLogCPUs=0 # [PARALLEL] Live-timing CPU`s (0 for all) X_all_q_CPU= "2 2 2 2" # [PARALLEL] CPUs for each role X_all_q_ROLEs= "q k c v" # [PARALLEL] CPUs roles (q,k,c,v) Parallelization of the dieletric constant the product of each role in X_all_q_CPU should be equal to the total number of processors. In this example I use 16 processors 2x2x2x2. Notice that the parallelization in q is more efficient but use more memory. X_all_q_nCPU_invert=4 # [PARALLEL] CPUs for matrix inversion Number of processors to invert the dielectric constant. In general matrix inversion is very inefficiently parallelized, therefore do not use many processor for this part, I advice about 4. X_Threads= 1 # [OPENMP/X] Number of threads for response functions DIP_Threads= 1 # [OPENMP/X] Number of threads for dipoles SE_Threads= 1 # [OPENMP/GW] Number of threads for self-energy If your calculation uses to much memory you can run with openmp and increase here the number of Threads for the dipoles, xhi and the self-energy. Use the same number in all of them. SE_CPU= "2 4 2" # [PARALLEL] CPUs for each role SE_ROLEs= "q qp b" # [PARALLEL] CPUs roles (q,qp,b) Similar to the previous parallelization. The total product should equal to the number of processors. The parallelization in qp is the more efficient while the one on b uses less memory. FFTGvecs= 20065 RL # [FFT] Plane-waves Total number of G-vectors used in the calculation. Decrease this number to use less memory. RandQpts= 3000000 # [RIM] Number of random q-points in the BZ RandGvec= 1 RL # [RIM] Coulomb interaction RS components Parameters for numerical integration of the Coulomb potentianl at q=0. Do not change them. EXXRLvcs= 20065 RL # [XX] Exchange RL components Number of G-vectors in the exchange, Eq. 104 in Ref[1].This number should be less than the total number of G-vectors FFTGvecs otherwise yambo automatically decreases it to FFTGvecs. Chimod= "" # [X] IP/Hartree/ALDA/LRC/BSfxc % BndsRnXp 1 | 1600 | # [Xp] Polarization function bands % Number of bands in the dielectric constant calculation. See eq. 102 in Ref. [1] NGsBlkXp= 3002 mHa # [Xp] Response block size Number of G-vectors in the dielectric constant, see Eq. 26 and 27 in Ref. [1] % LongDrXp 1.000000 | 0.000000 | 0.000000 | # [Xp] [cc] Electric Field % Direction of the electric field for the calculation of the q=0 component of the dielectric constant e(q,w) PPAPntXp= 27.21138 eV # [Xp] PPA imaginary energy Second frequency used to fit the Godby-Needs plasmon-pole model (PPM). If results change by changing this frequency, the PPM is not adequate for your calculation. % GbndRnge 1 | 1600 | # [GW] G[W] bands range % Number of bands used to expand the Green's function, Eq. 110 in Ref. [1] This number is usually larger than the number of bands used to calculated the dielectric constant. You can use the same number for both. GDamping= 0.10000 eV # [GW] G[W] damping Damping in the Green's function definition, the delta parameter in Eq. 49 in Ref. [2]. Few people investigate the effect of this parameters on the final results, usually everybody assume it equal to 0.1 eV. dScStep= 0.10000 eV # [GW] Energy step to evaluate Z factors DysSolver= "n" # [GW] Dyson Equation solver ("n","s","g") Parameters related to the linearized solution of the Dyson equation, see Eq. 58 in Ref.[2]. GTermKind= "BRS" # [GW] GW terminator ("none","BG" Bruneval-Gonze,"BRS" Berger-Reining-Sottile) Terminator for the self-energy and the dielectric constant. This flag speeds-up the convergence with the number of conduction bands, see Phys. Rev. B82, 041103(R). %QPkrange # [GW] QP generalized Kpoint/Band indices 1| 2|144|432| % K-points and band range where you want to calculate the GW correction. The syntax is first kpoint | last kpoint | first band | last band

If you are working with** isolated molecules** consider the following tricks and suggestions in your input file.

1) Add the Coulomb cut-off to your input by adding the -r (-c in yambo 3.4.1) flag in the yambo comand line

```
CUTGeo= "box XYZ" # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere X/Y/Z/XY..
% CUTBox
10.00 | 10.00 | 20.00 | # [CUT] [au] Box sides
%
CUTRadius= 0.000000 # [CUT] [au] Sphere/Cylinder radius
CUTCylLen= 0.000000 # [CUT] [au] Cylinder length
These parameters specify a cut-off for the Coulomb interaction
in such a way to speed-up convergence with the supercell. My advise is to use
the box cut-off with sides slightly smaller than the cell. In this
example I suppose to have a supercell (12x12x22 a.u.) and I choose
a cutoff 10x10x20 a.u. (see PRB
```**73**, 205119)

2) Parallelization for isolated systems:

when you a single-kpoint (for an isolated molecule for example) you cannot parallelize in q and k you have to set the number of processors for these roles equal to 1.

References:

[1] Quasiparticle calculations in solids

AULBUR W. G., JÖNSSON L., WILKINS J. W.

http://www.physics.ohio-state.edu/~wilkins/vita/gw_review.ps

[2] The GW method

F. Aryasetiawan and O. Gunnarsson.

http://xxx.lanl.gov/pdf/cond-mat/9712013v1

Pingback: Yambo input file for BSE calculations explained | Claudio Attaccalite's Homepage

SimilHi,

I have a system with an organic molecule on top of Au surface. Is it possible to get Projected Density Of States (PDOS) of the molecule after GW calculations in Yambo?

attaccPost authorDear Simil

as fas as I know, it is not possible. You can try to look if you can force

quantum-espresso to read the quasi-particle correction before calculation the PDOS,

but I don’t know how much it can be complicated

regards

Claudio

FabioHi, i have a question regarding the GW convergence.

BndsRnXp, NGsBlkXp, GbndRnge cannot be converged independently, right?

For what i have saw in my results, these quantities are inter-dependent.

Am i doing something wrong? If not, is there a way to reduce the number of calculations we have to do

for reach convergence? Aren’t BndsRnXp and NGsBlkXp related somehow?

Fabio

attaccPost authorDear Fabio

regarding BndsRnXp, NGsBlkXp you are right, they are inter-dependent,

but GbndRnge should be independent from the other two.

Try to converge BndsRnXp, NGsBlkXp together and then converge GbndRnge.

Usually the gap converges faster than the single band values.

Let me know

Claudio

FabioDear Claudio,

I converged BndsRnXp, NGsBlkXp together with GbndRnge (100) fixed.

I found that BndsRnXp=500 , and NGsBlkXp=8 Ry is enough to reach convergence.

Then I converged GbndRnge, with BndsRnXp=500, NGsBlkXp=8, 10, 12, 14 Ry.

I found that the convergence is achieved for GbndRnge=500, BndsRnXp=500, NGsBlkXp=12 Ry and not NGsBlkXp=8 Ry.

I considered a convergence criterion within 0.05 eV.

This shows that there is an inter-dependence between

GbndRnge and NGsBlkXp. Is that normal?

Thanks you,

Fabio

attaccPost authorDear Fabio

yes this is normal, in the sense that usually the inter-dependence between GbndRnge and NGsBlkXp

is small but not zero.

Notice that in general number of bands to converge the Green’s functions

is larger than the one for the dielectric constant GbndRnge >= BndsRnXp.

Did you converge the absolute value of the bands or their energy difference (as the gap)?

Because in general the energy difference converge fast, while the absolute value is slower.

Claudio

Claudio

FabioDear Claudio,

I converged the energy difference. But the inter-dependence

between GbndRnge and NGsBlkXp is indeed small.

And my convergence criterion is <= 0.05 eV. If i choose <= 0.1 eV a smaller

GbndRnge and NGsBlkXp would be enough.

Thank you,

Fabio

ShiHi Claudio,

I have a general question regarding GW and BSE.

Say I have finished those heavy calculations, which files/folder should I back up in order to reproduce those calculations in the future while reducing the computational time? Is that even possible?

It is hard to back up all the files (in many cases > 10Gb).

I used yambo -J job_string_dir when running yambo. So some files are stored in job_string_dir as well besides SAVE directory.

Thanks!

Shi

attaccPost authorDear Shi

it depends from your calculation. If you performed GW calculation, the imporat results

are in the ndb.QP that contains all information about the quasi-particle corrections.

If you performed BSE, the files are the ndb.BS*, that contain the BSE matrix.

Another important file could be the dielectric constant ndb.pp*

best

Claudio