Skip to content

Second lesson on GW

Treatment of metals.

This lesson aims at showing how to obtain self-energy corrections to the DFT Kohn-Sham eigenvalues within the GW approximation, in the metallic case, without the use of a plasmon-pole model. The band width and Fermi energy of Aluminum will be computed.

The user may read the paper

  • F. Bruneval, N. Vast, and L. Reining, Phys. Rev. B 74 , 045102 (2006), for some information and results about the GW treatment of Aluminum. He will also find there an analysis of the effect of self-consistency on quasiparticles in solids (not present in this tutorial, however available in Abinit). The description of the contour deformation technique that bypasses the use of a plasmon-pole model to calculate the frequency convolution of G and W can be found in

  • S. Lebegue, S. Arnaud, M. Alouani, P. Bloechl, Phys. Rev. B 67, 155208 (2003),

with the relevant formulas. We will refer to these papers as the [Bruneval2006] and the [Lebegue2003] papers.

A brief description of the equations implemented in the code can be found in the GW notes Also, it is suggested to acknowledge the efforts of developers of the GW part of ABINIT, by citing the 2005 ABINIT publication.

The user should be familiarized with the four basic lessons of ABINIT, see the tutorial home page as well as the first GW lesson.

This lesson should take about one hour to be completed (also including the reading of [Bruneval2006] and [Lebegue2003].

1 The preliminary Kohn-Sham band structure calculation

Before beginning, you might consider to work in a different subdirectory as for the other lessons. Why not “Work_gw2”?

In lesson 4, we have computed different properties of Aluminum within the LDA. Unlike for silicon, in this approximation, there is no outstanding problem in the computed band structure. Nevertheless, as you will see, the agreement of the band structure with experiment can be improved significantly if one relies on the GW approximation.

In the directory ~abinit/tests/tutorial/Input/Work_gw2, copy the files ~abinit/tests/tutorial/Input/tgw2_x.files and, and modify the tgw2_x.files file as usual (see base1 lesson).
Then (supposing abinit is the proper alias), issue:

abinit < tgw2_x.files >& tgw2_1.log &

This run generates the WFK file for the subsequent GW computation and also provides the band width of Aluminum. Note that the simple Fermi-Dirac smearing functional is used (occopt=3), with a large smearing (tsmear=0.05 Ha). The k point grid is quite rough, an unshifted 4x4x4 Monkhorst-Pack grid (64 k points in the full Brillouin Zone, folding to 8 k points in the Irreducible wedge, ngkpt=4 4 4). Converged results would need a 4x4x4 grid with 4 shifts (256 k points in the full Brillouin zone). This grid contains the Gamma point, at which the valence band structure reaches its minimum.

The output file presents the Fermi energy

 Fermi (or HOMO) energy (eV) =   7.14774   Average Vxc (eV)=  -9.35982

as well as the lowest energy, at the Gamma point

 Eigenvalues (   eV  ) for nkpt=   8  k points:
 kpt#   1, nband=  6, wtk=  0.01563, kpt=  0.0000  0.0000  0.0000 (reduced coord)
  -3.76175  19.92114  19.92114  19.92114  21.00078  21.00078

So, the occupied band width is 10.90 eV. More converged calculations would give 11.06 eV (see [Bruneval2006]).
This is to be compared to the experimental value of 10.6 eV (see references in [Bruneval2006]).

2 Calculation of the screening file

In order not to lose time, let us start the calculation of the screening file before the examination of the corresponding input file. So, copy the file, and modify the tgw2_x.files file as usual (replace occurrences of twg2_x by tgw2_2). Also, copy the WFK file (tgw2_1o_WFK) to tgw2_2i_WFK. Then run the calculation (it should take about 30 seconds on a 3 GHz PC).

We now have to consider starting a GW calculation. However, unlike in the case of Silicon in the previous GW tutorial, where we were focussing on quantities close to the Fermi energy (spanning a range of a few eV), here we need to consider a much wider range of energy: the bottom of the valence band lies around -11 eV below the Fermi level. Unfortunately, this energy is of the same order of magnitude as the plasmon excitations. With a rough evaluation, the classical plasma frequency for a homogeneous electron gas with a density equal to the average valence density of Aluminum is 15.77 eV. Hence, using plasmon-pole models may be not really appropriate.

In what follows, one will compute the GW band structure without a plasmon-pole model, by performing explicitly the numerical frequency convolution. In practice, it is convenient to extend all the functions of frequency to the full complex plane. And then, making use of the residue theorem, the integration path can be deformed: one transforms an integral along the real axis into an integral along the imaginary axis plus residues enclosed in the new contour of integration. The method is extensively described in [Lebegue2003].

Examine the input file The ten first lines contain the important information. There, you find some input variables that you are already familiarized with, like optdriver, ecuteps, ecutwfn, but also new input variables: gwcalctyp, nfreqim, nfreqre, and freqremax. The purpose of this run is simply to generate the screening matrices. Unlike for the plasmon-pole models, one needs to compute these at many different frequencies. This is the purpose of the new input variables. The main variable gwcalctyp is set to 2 in order to specify a non plasmon-pole model calculation. Note that the number of frequencies along the imaginary axis governed by nfreqim can be chosen quite small, since all functions are smooth in this direction. In contrast, the number of frequencies needed along the real axis set with the variable nfreqre is usually larger.

3 Finding the Fermi energy and the bottom of the valence band

In order not to lose time, let us start the calculation of the band width before the study of the input file. So, copy the file, and modify the tgw3_x.files file as usual (replace occurrences of twg2_x by tgw2_3). Also, copy the WFK file (tgw2_1o_WFK) to tgw2_3i_WFK, and the screening file (tgw2_2o_SCR) to tgw2_3i_SCR. Then run the calculation (it should take about 2 minutes on a 3 GHz PC).

The computation of the GW quasi-particle energy at the Gamma point of Aluminum does not differ from the one of quasi-particle in Silicon. However, the determination of the Fermi energy raises a completely new problem: one should sample the Brillouin Zone, to get new energies (quasi-particle energies) and then determine the Fermi energy. This is actually the first step towards a self-consistency!

Examine the input file

The first thirty lines contain the important information. There, you find some input variables with values that you are already familiarized with, like optdriver, ecutsigx, ecutwfn. Then, comes the input variable gwcalctyp=12. The value x2 corresponds to a contour integration. The value 1x corresponds to a self-consistent calculation with update of the energies only. Then, one finds the list of k points and bands for which a quasi-particle correction will be computed: nkptgw, kptgw, and bdgw. The number and list of k points is simply the same as nkpt and kpt. One might have specified less k points, though (only those needing an update). The list of band ranges bdgw has been generated on the basis of the LDA eigenenergies. We considered only the bands in the vicinity of the Fermi level: bands much below or much above are likely to remain much or much above the Fermi region. In the present run, we are just interested in the states that may cross the Fermi level, when going from LDA to GW. For commodity, one could have selected an homogeneous range for the whole Brillouin zone, e.g. from 1 to 5, but this would have been more time-consuming.

In the output file, one finds the quasi-particle energy at Gamma, for the lowest band:

k =    0.000   0.000   0.000
 Band     E_lda   <Vxclda>   E(N-1)  <Hhartree>   SigX  SigC[E(N-1)]    Z     dSigC/dE  Sig[E(N)]  DeltaE  E(N)_pert E(N)_diago
    1    -3.762    -9.451    -3.762     5.689   -15.049     5.676     0.777    -0.287    -9.390     0.060    -3.701    -3.684

(the last column is the relevant quantity). The updated Fermi energy is also mentioned:

 New Fermi energy:     2.469501E-01 Ha ,    6.719854E+00 eV

The last information is not printed in case of gwcalctyp lower than 10.

Combining the quasi-particle energy at Gamma and the Fermi energy, gives the band width, 10.404 eV. Using converged parameters, the band width will be 10.54 eV (see Bruneval[2006]). This is in excellent agreement with the experimental value of 10.6 eV.

4 Computing a GW spectral function, and the plasmon satellite of Aluminum

The access to the non-plasmon-pole-model self-energy (real and imaginary part) has additional benefit, e.g. an accurate spectral function can be computed, see [Lebegue2003]. You may be interested to see the plasmon satellite of Aluminum, which can be accounted for within the GW approximation.

Remember that the spectral function is almost (except some matrix elements) the spectrum which is measured in photoemission spectroscopy (PES). In PES, a photon impinges the sample and extracts an electron from the material. The difference of energy between the incoming photon and the obtained electron gives the binding energy of the electron in the solid, or in other words the quasiparticle energy or the band structure. In simple metals, an additional process can take place easily: the impinging photon can extract an electron together with a global charge oscillation in the sample. The extracted electron will have a kinetic energy lower than in the direct process, because a part of the energy has gone to the plasmon. The electron will appear to have a larger binding energy… You will see that the spectral function of Aluminum consists of a main peak which corresponds to the quasiparticle excitation and some additional peaks which correspond to quasiparticle and plasmon excitations together.

In order not to lose time, this calculation can be started before the examination of the input file. So, copy the file, and modify the tgw4_x.files file as usual (replace occurrences of twg2_x by tgw2_4). Also, copy the WFK file (tgw2_1o_WFK) to tgw2_4i_WFK, and the screening file (tgw2_2o_SCR) to tgw2_4i_SCR. Then run the calculation (it should take about 2 minutes on a 3 GHz PC).

Compared to the previous file (, the input file contains two additional keywords: nfreqsp, and freqspmax. Also, the computation of the GW self-energy is done only at the Gamma point.

The spectral function is written in the file tgw2_4o_SIG. It is a simple text file. It contains, as a function of the frequency (eV), the real part of the self-energy, the imaginary part of the self-energy, and the spectral function. You can visualize it using your preferred software. For instance, issue

$ gnuplot
gnuplot>  p'tgw2_4o_SIG' u 1:4 w l

You should be able to distinguish the main quasiparticle peak located at the GW energy (-3.7 eV) and some additional features in the vicinity of the GW eigenvalue minus a plasmon energy (-3.7 eV - 15.8 eV = -19.5 eV).

Another file, tgw2_4o_GW, is worth to mention: it contains information to be used for the subsequent calculation of excitonic effects by the EXC code (usually available at; if not, see the ETSF software page and further links).