.

ICTP/Psi-k/CECAM School on Electron-Phonon Physics from First Principles

Electron-phonon coupling with EPW

Hands-on Session (Wed.4)

In this session we will learn to use the core capabilities of EPW.

First copy the tutorial input files and go in the first exercise:

$ wget http://epw.org.uk/uploads/School2018/Wed.4.Verdi.tar
$ tar -xvf Wed.4.Verdi.tar ; cd tuto_Wed4/exercise1

Exercise 1
In this exercise we will repeat the calculation of the electron-phonon coupling strength and of the
Eliashberg spectral function of lead that you performed yesterday using Quantum Espresso and the
PHonon code. This time we will use EPW, which will allow us to calculate these quantities using much
denser grids at small additional cost.

I Run a self-consistent calculation for lead as in Exercise 1 of tutorial Tue.5.

Note: The energy cutoff ecutwfc needed for convergence should be 90 Ry.

$ mkdir exercise1; cd exercise1; cp ../../tuto_Tues5/exercise1/pb.scf.in .

$ cp ../../tuto_Tues5/exercise1/pb_s.UPF .
$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < pb.scf.in > pb.scf.out

I Run a phonon calculation on a homogeneous 3 × 3 × 3 q-point grid using the following input:
-- pb.ph.in
&inputph
prefix = 'lead',
fildyn = 'lead.dyn',
fildvscf = 'dvscf',
ldisp = .true.,
nq1 = 3,
nq2 = 3,
nq3 = 3,
tr2_ph = 1.0d-12
/

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/ph.x -npool 4 < pb.ph.in > pb.ph.out

The keyword fildvscf tells the code to write on file the change of the self-consistent potential due to
phonon perturbations, ∂qν V scf , that is needed to compute the electron-phonon matrix elements.
In the output file, locate the list of 4 irreducible q points in the Brillouin Zone (IBZ):

Dynamical matrices for ( 3, 3, 3) uniform grid of q-points

( 4q-points):
N xq(1) xq(2) xq(3)
1 0.000000000 0.000000000 0.000000000
2 -0.333333333 0.333333333 -0.333333333
3 0.000000000 0.666666667 0.000000000
4 0.666666667 -0.000000000 0.666666667
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 1 of 20
The list of irreducible q points is also written in the lead.dyn0 file. If you type ls, you can see a
lead.dynX file containing the dynamical matrix has been produced for each irreducible q point. The
dvscf files are all named lead.dvscf1 and are located inside the _ph0/lead.q_X/ folders, except
for the one corresponding to the first q point (Γ) that is located in _ph0/.

I Gather the .dyn and .dvscf files into a new save/ directory which EPW will read. The files in
_ph0/lead.phsave/ containing the displacement patterns are also needed. This can easily be done
using the pp.py python script which is already included in the EPW distribution:

$ python /home/nfs3/smr3191/q-e/EPW/bin/pp.py

The script will ask you to prompt the prefix of your calculation (lead).

Note: if this doesn’t work, you need to install numpy on your desktop. Just type “pip install numpy” and run the
script again.

I Run a non self-consistent calculation on a homogeneous 6 × 6 × 6 k grid. You cannot use the
K_POINTS automatic feature of pw.x, but you need to generate a positive definite grid between
0 and 1 and specify K_POINTS crystal. To do so, first copy the input file for the self-consistent
calculation into a new one:

$ cp pb.scf.in pb.nscf.in

then using vi or another text editor modify the input variable calculation to nscf, remove
wf_collect=.true., set the number of bands to nbnd=10, and delete the last two lines speci-
fying the k-point grid. Then you can use, for example, a simple script provided in the wannier90
distribution to generate the homogeneous k grid, kmesh.pl:

$ /home/nfs3/smr3191/q-e/wannier90-2.1.0/utility/kmesh.pl 6 6 6 >> pb.nscf.in

Now your input file should look like this:

&control pb.nscf.in
calculation='nscf'
restart_mode='from_scratch',
prefix='lead',
pseudo_dir = './',
outdir='./'
/
&system
ibrav= 2,
celldm(1) = 9.2225583816,
nat= 1,
ntyp= 1,
ecutwfc = 30.0
occupations='smearing',
smearing='marzari-vanderbilt',
degauss=0.05
nbnd=10
/
&electrons
conv_thr = 1.0d-10
mixing_beta = 0.7
/
ATOMIC_SPECIES
Pb 207.2 pb_s.UPF
ATOMIC_POSITIONS

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 2 of 20

 Pb 0.00 0.00 0.00
K_POINTS crystal
216
0.00000000 0.00000000 0.00000000 4.629630e-03
0.00000000 0.00000000 0.16666667 4.629630e-03
0.00000000 0.00000000 0.33333333 4.629630e-03
...

Note: in general you should add nosym=.true. in the &system namelist to make sure pw.x calculates each k point
provided in the list, without using any symmetry operations. Also note that the k and q grids need to be commensu-
rate, with the k grid at least of the size of the q grid. Since we chose a 6 × 6 × 6 k grid, the 3 × 3 × 3 grid used in the
phonon calculation is appropriate, however a 6 × 6 × 6 q grid would be needed in order to interpolate more accurately
the dynamical matrix and the electron-phonon matrix elements.

Now you can run:

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < pb.nscf.in > pb.nscf.out
Since EPW does not yet support G-vector parallelization, we use k-point parallelization, which means
that np needs to be always equal to npool. If you forget this the code will crash, asking to use only
one processor per pool.

I We are now ready to perform a calculation using EPW. Prepare the following input file and run EPW:
-- pb.epw.in
&inputepw
prefix = 'lead',
amass(1) = 207.2
outdir = './'
dvscf_dir = './save'

elph = .true.
kmaps = .false.
epbwrite = .true.
epbread = .false.
epwwrite = .true.
epwread = .false.

wannierize = .true.
nbndsub = 4
nbndskip = 5
num_iter = 300
dis_win_max = 21
dis_win_min = -3
dis_froz_min= -3
dis_froz_max= 13.5
proj(1) = 'Pb:sp3'
wdata(1) = 'bands_plot = .true.'
wdata(2) = 'begin kpoint_path'
wdata(3) = 'G 0.00 0.00 0.00 X 0.00 0.50 0.50'
wdata(4) = 'X 0.00 0.50 0.50 W 0.25 0.50 0.75'
wdata(5) = 'W 0.25 0.50 0.75 L 0.50 0.50 0.50'
wdata(6) = 'L 0.50 0.50 0.50 G 0.00 0.00 0.00'
wdata(7) = 'G 0.00 0.00 0.00 K 0.375 0.375 0.75'
wdata(8) = 'end kpoint_path'
wdata(9) = 'bands_num_points = 10'

elecselfen = .false.
phonselfen = .false.

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 3 of 20

 a2f = .false.

parallel_k = .true.
parallel_q = .false.

fsthick = 1 ! eV
eptemp = 0.075 ! K
degaussw = 0.1 ! eV

nkf1 = 6
nkf2 = 6
nkf3 = 6
nqf1 = 3
nqf2 = 3
nqf3 = 3

nk1 = 6
nk2 = 6
nk3 = 6
nq1 = 3
nq2 = 3
nq3 = 3
/
4 cartesian
0.000000000000000E+00 0.000000000000000E+00 0.000000000000000E+00
-0.333333333333333E+00 0.333333333333333E+00 -0.333333333333333E+00
0.000000000000000E+00 0.666666666666667E+00 0.000000000000000E+00
0.666666666666667E+00 -0.555111512312578E-16 0.666666666666667E+00

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/epw.x -npool 4 < pb.epw.in > pb.epw.out

Note: Note: The list of q points given at the end of the input file should be exactly the same as the list contained in
the file prefix.dyn0. In dvscf_dir = './save' we specify the directory where the .dyn, .dvscf and patterns files
are stored.

With this input, we are not instructing EPW to calculate any physical quantities; instead, EPW will
perform the following main steps (see the technical lecture Wed.3 summarizing each step):

• Wannierizing the band structure using wannier90 as an internal library. In the output you can
look at the Wannier function centers and spreads obtained:

Wannier Function centers (cartesian, alat) and spreads (ang):

( 0.07779 0.07779 0.07779) : 2.22268

( 0.07779 -0.07779 -0.07779) : 2.22268
( -0.07779 0.07779 -0.07779) : 2.22268
( -0.07779 -0.07779 0.07779) : 2.22268

while the full output from the wannier90 run is in the file lead.wout. The input variables
for the wannierization are in the block following wannierize = .true.. nbndsub corresponds to
the number of Wannier functions (4, starting from Pb sp3 orbitals as the initial guess), while
nbndskip is the number of valence bands not wannierized (generally a set of bands lying at lower
energies, such as semicore states in this example). For the other input variables you can refer to
the technical lecture Tue.3, and to the EPW website http://epw.org.uk/Documentation/Inputs.
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 4 of 20
 It is always possible to add extra variables that are read by wannier90 by using the input
wdata(index), with increasing index number. Here we use these extra variables in order to plot
the interpolated band structure.

• Calculating the electron-phonon matrix elements on the initial (k, q) grid for each irreducible
q-point in the Brillouin zone, and unfold to the full Brillouin zone using symmetries. This is
the most expensive part of the run. In the output you can see:

...
===================================================================
irreducible q point # 1
===================================================================

Symmetries of small group of q: 48

in addition sym. q -> -q+G:

Number of q in the star = 1

List of q in the star:
1 0.000000000 0.000000000 0.000000000
...

• Writing on disk the files lead.epbX (one per CPU) containing the Hamiltonian, the dynamical
matrices and the electron-phonon matrix elements on the initial (coarse) k and q grids in the
full Brillouin zone:

Writing epmatq on .epb files

The .epb files have been correctly written

• Transform all quantities from reciprocal (Bloch) space to real (Wannier) space, and store
on file the resulting matrices and additional information needed for restarting a calculation
(lead.epmatwp1, crystal.fmt, dmedata.fmt, and epwdata.fmt files):

Writing Hamiltonian, Dynamical matrix and EP vertex in Wann rep to file

I We should always check that the Wannier-interpolated electron and phonon band structure corre-
spond to the ones calculated using pw.x and ph.x, respectively. To do so, we use the restart feature
by reading all quantities written at the end of the previous run (epwread = .true.). Copy the input
file into a new one:

$ cp pb.epw.in pb.epw.in2

and make the following modifications:

... pb.epw.in2
kmaps = .true.
epbwrite = .false.
epbread = .false.
epwwrite = .false.

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 5 of 20

 epwread = .true.

wannierize = .false.
...
band_plot = .true.
...

filqf = 'lead_band.kpt'
filkf = 'lead_band.kpt'
!nkf1 = 6
...
!nqf3 = 3
...

Here we instructed EPW to save on file the interpolated bands (band_plot = .true.), and we com-
mented the input variables nkf1 ... nqf3 defining homogeneous k and q grids for the interpolated
quantities; instead, we choose to interpolate them onto a Brillouin-zone path defined by filqf, filkf and
read from the file lead_band.kpt. This file has been conveniently generated for us by wannier90,
and the coordinates are in units of the reciprocal lattice vectors (crystal). EPW always reads coor-
dinates from file in crystal units. If you now run EPW again:

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/epw.x -npool 4 < pb.epw.in2 > pb.epw.out2

the files band.eig and phband.freq are produced. To extract easily the data to plot, you can
simply run /home/nfs3/smr3191/q-e/bin/plotband.x and prompt the input file (band.eig or
phband.freq), the energy range (-5,20 for example), the output file with the data to plot (band.dat
or phband.dat); the other inputs are not important, just put 1 when asked.
You can now compare the phonon dispersions with the results from the tutorial Tue.5 (note: we have
a denser q path here). You can also create a new folder and run a non self-consistent calculation
to obtain the electronic band structure with pw.x as you learned in the tutorial Mon.4. If you
compare it with the wannierized band structure you should obtain this plot (with the wannierized
bands superimposed in black):

10

5
Energy (eV)

-5

-10

-15

Γ X W L Γ K

I The Wannier-Fourier interpolation technique is based on the decay properties of the Wannier
functions and of the phonon perturbation in real space. To check how each quantity is decaying within
the supercell corresponding to our initial k and q grids, we can plot the files decay.H (Hamiltonian),
decay.D (dynamical matrix) and decay.epmat_wanep (matrix elements; you can plot the data using
the first and last column). You should obtain the plots reported here (note the log scale on the y
axis).

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 6 of 20

From these plots we see clearly that we need a larger supercell (denser q grid) in order to interpolate
more accurately the phonon properties, whereas the electronic part decays well (remember that we
are using a 6 × 6 × 6 k grid, and only a 3 × 3 × 3 q grid).

I We now calculate the phonon linewidths γqν in the commonly used double-delta approximation
(Eq (1), tutorial Tue.5) along the Brillouin zone, as well as the electron-phonon coupling strength
λqν (Eq. (2), tutorial Tue.5) by setting selfen_phon=.true. and delta_approx=.true.. Since
calculating γqν and λqν requires an integration over k, we interpolate the matrix elements onto a
denser 20 × 20 × 20 k grid.

Copy the file pb.epw.in2, which we already used to restart an EPW calculation, to pb.epw.in3,
modify it as follows:
... pb.epw.in3
phonselfen = .true.
a2f = .false.
band_plot = .false.
delta_approx= .true.

...
filqf = 'lead_band.kpt'
!filkf = 'lead_band.kpt'
nkf1 = 20
nkf2 = 20
nkf3 = 20
!nqf1 = 3
!nqf2 = 3
!nqf3 = 3
...

and run EPW:

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/epw.x -npool 4 < pb.epw.in3 > pb.epw.out3
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 7 of 20
Note: The parameter fsthick determines the energy window around the Fermi level for which the electron-phonon matrix
elements are interpolated. This can reduce significantly the cost of calculations: for example, only electronic states within
a few phonon energies from the Fermi level will contribute to the phonon linewidth, therefore we can use fsthick = 1 (eV).

You can see that now the phonon linewidths and coupling strengths are printed in output for
each phonon wavevector q and mode ν (lambda___ and gamma___). The sum of λqν over all
phonon modes, λq , is also written (lambda___( tot )). γqν and λqν are stored in the files
linewidth.phself and lambda.phself, respectively. Inspect those files to familiarize yourself with
the format and learn how to plot these quantities. For example, to plot the q-dependent linewidth
of the third phonon, you can type in gnuplot:
gnuplot> plot "linewidth.phself" u 1:4 every 3::2 w l lw 2
and you should be able to produce a plot like this:

0.15

0.1
γqν (meV)

0.05

0
Γ X W L Γ K

Although these linewidths are still unconverged, are you able to explain the main features?

I Let’s calculate the total (integrated) electron-phonon coupling strength λ and the isotropic Eliash-
berg spectral function α2 F (ω) (see Eq. (3) of tutorial Tue.3) by setting a2f = .true. and modifying
the input as follows:
$ cp pb.epw.in3 pb.epw.in4
... pb.epw.in4
a2f = .true.

...
!filqf = 'lead_band.kpt'
!filkf = 'lead_band.kpt'
nkf1 = 20
nkf2 = 20
nkf3 = 20
nqf1 = 12
nqf2 = 12
nqf3 = 12
...

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/epw.x -npool 4 < pb.epw.in4 > pb.epw.out4

Note that now we need to perform a Brillouin-zone integration also over q, hence we interpolate the
matrix elements onto a 20 × 20 × 20 k grid and 12 × 12 × 12 q grid. At the end of the calculation
we should get λ:
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 8 of 20
 ===================================================================
Eliashberg Spectral Function in the Migdal Approximation
===================================================================

lambda : 1.7886205

Note that the converged value for λ should be around 1.1.

You can plot α2 F (ω) from the file lead.a2f.01. This file contains 11 columns: the first one is the
frequency grid, and the remainder 10 columns correspond to α2 F (ω) calculated using different values
of broadening, starting from degaussq = 0.05 meV, with steps of delta_qsmear = 0.05 meV. If
you plot this function for two values of broadening (0.1 meV in blue and 0.5 meV in red, for example)
you should obtain:

1.5

1
α2F( (meV)

0.5

0
0 2 4 6 8 10
ω (meV)

I Try increasing the coarse q grid to better converge the phonons. If you use a 4 × 4 × 4 q grid,
what k grids should you use?

I Try using denser k and q meshes for the interpolation (increased nqf1,...,nkf3), and investigate
the convergence of λ and α2 F (ω) with respect to the BZ sampling and to the smearing degaussw.
You should note that the sampling over the electron wavevectors k is more critical than the sampling
over the phonon wavevectors q.
Is the convergence faster if you use randomly generated grids? The input parameters are rand_q = .true.
and rand_nq = 2000, for example.
For the converged results and parameters, you can refer to the figure in tutorial Tue.5 and to the
paper https://www.sciencedirect.com/science/article/pii/S0010465516302260

Note: during the first EPW run, the maximum number of CPUs you can use is limited by the total
number of k points in the irreducible BZ (since there is no G parallelization). However, when restart-
ing from epwread = .true. you can also use a larger number of CPUs than in the first run. Try
running with 8 CPUs and observe the speedup of the calculation.

I Explore the effect of spin-orbit coupling in lead by re-running all calculations including SOC. You
need to:

• add noncolin = .true. and lspinorb = .true. in the &system namelist of the scf and
nscf calculations;

• change the lattice parameter to the relaxed value including SOC, celldm(1) = 9.269771512;

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 9 of 20

 • increase the number of bands in the nscf calculation to nbnd=26;
• double the number of bands nbndsub and nbndskip in the EPW input file.

Exercise 2
In this exercise we will examine the electron-phonon interactions in SiC. As you learned in tutorial
Tue.5, SiC is a polar material, where the long-range polar coupling results in a 1/|q| divergence of
the matrix elements near Γ. Here we will see how to correctly interpolate the matrix elements in this
case, and we will calculate the electron linewidths.

I Run a self-consistent calculation for SiC as in Exercise 2 of tutorial Tue.5.

$ cd ../ ; mkdir exercise2; cd exercise2
$ cp ../../tuto_Tues5/exercise2/scf.in sic.scf.in
$ cp ../../tuto_Tues5/exercise2/*.UPF .
$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < sic.scf.in > sic.scf.out

I Run a phonon calculation on a homogeneous 3 × 3 × 3 q-point grid.

-- sic.ph.in
&inputph
prefix = 'sic'
fildvscf = 'dvscf'
ldisp = .true
fildyn = 'sic.dyn'
nq1=3,
nq2=3,
nq3=3,
tr2_ph = 1.0d-12
/

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/ph.x -npool 4 < sic.ph.in > sic.ph.out

Note that the output now contains also the Born effective charges Z ∗ and the electronic dielectric
constant ∞ :

Electric Fields Calculation

....

End of electric fields calculation

Dielectric constant in cartesian axis

( 7.214167210 0.000000000 0.000000000 )

....

Effective charges (d Force / dE) in cartesian axis

atom 1 Si
Ex ( 2.67035 0.00000 0.00000 )
....
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 10 of 20
These quantities are automatically calculated for an insulating system, using the finite response to
an electric field, and determine the energy splitting of the TO and LO phonons.

You should now know well how to plot the phonon dispersions using q2r.x and matdyn.x. The plot
should look like this:

120

100

80
ω (meV)

60

40

20

0
Γ X W L Γ K

Note that now there are 3 acoustic and 3 optical branches, separated by an energy gap, and that
there is the LO-TO splitting at Γ.

I Gather the .dyn, .dvscf and patterns files into a save/ directory as in Exercise 1 using the
pp.py script:

$ python /home/nfs3/smr3191/q-e/EPW/bin/pp.py

I Run a non self-consistent calculation on a homogeneous 6 × 6 × 6 k grid. To do so, first copy

the input file for the self-consistent calculation into a new file pb.nscf.in; then modify the input
variable calculation to nscf, set the number of bands to nbnd=4 (we will look only at the valence
bands here), and delete the last two lines specifying the k-point grid. Then you can use again the
script kmesh.pl to generate the homogeneous k grid:

$ /home/nfs3/smr3191/q-e/wannier90-2.1.0/utility/kmesh.pl 6 6 6 >> sic.nscf.in

Now your input file should look like this:

&control sic.nscf.in
calculation = 'nscf'
prefix = 'sic'
wf_collect = .false.
pseudo_dir = './'
outdir = './'
/
&system
ibrav = 2
celldm(1) = 8.237
nat = 2
ntyp = 2
ecutwfc = 30.0
nbnd = 4
/
&electrons

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 11 of 20

 diagonalization = 'david'
mixing_beta = 0.7
conv_thr = 1.0d-10
/
ATOMIC_SPECIES
Si 28.0855 Si.pz-vbc.UPF
C 12.01078 C.UPF
ATOMIC_POSITIONS alat
Si 0.00 0.00 0.00
C 0.25 0.25 0.25
K_POINTS crystal
216
0.00000000 0.00000000 0.00000000 4.629630e-03
0.00000000 0.00000000 0.16666667 4.629630e-03
...

Now you can run:

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < sic.nscf.in > sic.nscf.out

I Calculate the interpolated electron-phonon matrix elements along some Brillouin-zone directions
(input variable prtgkk = .true.).
-- sic.epw.in
&inputepw
prefix = 'sic'
amass(1) = 28.0855
amass(2) = 12.0107
outdir = './'
dvscf_dir = './save'

elph = .true.
kmaps = .false.
epbwrite = .true.
epbread = .false.
epwwrite = .true.
epwread = .false.
lpolar = .true.

wannierize = .true.
nbndsub = 4
nbndskip = 0
num_iter = 300
proj(1) = 'Si:sp3'

elecselfen = .false.
phonselfen = .false.
a2f = .false.
prtgkk = .true.

parallel_k = .true.
parallel_q = .false.

fsthick = 7.0
eptemp = 20
degaussw = 0.05

filqf = 'path.dat'
nkf1 = 1
nkf2 = 1
nkf3 = 1

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 12 of 20

 nk1 = 6
nk2 = 6
nk3 = 6
nq1 = 3
nq2 = 3
nq3 = 3
/
4
0.000000000000000E+00 0.000000000000000E+00 0.000000000000000E+00
-0.333333333333333E+00 0.333333333333333E+00 -0.333333333333333E+00
0.000000000000000E+00 0.666666666666667E+00 0.000000000000000E+00
0.666666666666667E+00 -0.555111512312578E-16 0.666666666666667E+00

Since the unit cell of SiC is the same as Pb, we can re-use the same BZ path, after copying it from
Exercise 1:
$ cp ../exercise1/lead_band.kpt path.dat
Note also that in order to speed up the calculation, we chose to only print the matrix elements for
the initial electronic states at k = Γ (nkf1=nkf2=nkf3=1).

Since SiC is a polar material (non-zero Born effective charges), we set lpolar = .true. in order
to correctly treat the long-range interaction in bulk crystals. The strategy consists in subtracting the
long-range component g L from the full matrix element g before interpolation and adding it back after
interpolation (see lectures Wed.3 and Thu.1). An analogous strategy is implemented to correctly
interpolate the dynamical matrix including the long-range dipole-dipole interactions which result in
the LO-TO splitting.

The matrix elements for each q along the BZ path are written in the output, with columns corre-
sponding to each band and phonon mode. An average over degenerate bands and modes has already
been performed. If you want to plot the matrix elements for the valence-band top (n = m = 4) and
the highest-energy phonon branch (ν = 6), for example, you can type:
$ grep '4 4 6' sic.epw.out | awk '{print $7}' > matel.dat
and the plot should look like:

1000

800
|g|(q) (meV)

600

400

200

0
Γ X’ X W L Γ K

If you use a denser q-point path, the matrix elements will keep diverging near Γ, in agreement with
the direct calculations using ph.x.
Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 13 of 20
The matrix elements calculated in tutorial Tue.5 are reproduced as green squares. Note that the
interpolated ones are not fully accurate, since a larger coarse q-grid is needed in order to accurately
interpolate the electron-phonon matrix elements.

Note: if you want to inspect the results of the interpolation when the long-range contribution to the matrix elements
is not taken into account, you can run a new calculation using lpolar=.false.. Remember to run this test in a new
directory, or to run again an EPW calculation from scratch using lpolar=.true..

I Plot the interpolated electron and phonon band structure and compare them with the results from
pw.x (you will need to perform a band structure calculation) and from matdyn.x, respectively. Note
that we are only wannierizing the 4 valence bands. You can also look at the decays, as done in
Exercise 1.

I We now calculate the linewidths of the valence states in SiC along high-symmetry lines, which
correspond to twice the imaginary part of the electron self-energy. To do so, we need to use the input
variable elecselfen = .true., the k-point path, and a homogeneous q grid for the integration.
Copy the EPW input into a new one:

$ cp sic.epw.in sic.epw.in2
and modify it as follows:
... sic.epw.in2
elph = .true.
kmaps = .true.
epbwrite = .false.
epbread = .false.
epwwrite = .false.
epwread = .true.
lpolar = .true.

wannierize = .false.
...

elecselfen = .true.
phonselfen = .false.
a2f = .false.
prtgkk = .false.
efermi_read = .true.
fermi_energy= 9.6
...

!filqf = 'path.dat'
filkf = 'path.dat'
nqf1 = 20
nqf2 = 20
nqf3 = 20
...

Note that since we use a k path, the Fermi energy calculated from the fine (interpolated) grid will
not be accurate. To overcome this problem, we provide the Fermi energy in the input by using
efermi_read = .true. and fermi_energy=9.6 (just above the valence-band top).

In the output you can monitor the progression of the q integration, before reading the electron
self-energy for each k point:

Progression iq (fine) = 50/ 8000

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 14 of 20
 Progression iq (fine) = 100/ 8000
...
...
Average over degenerate eigenstates is performed
WARNING: only the eigenstates within the Fermi window are meaningful

ik = 1 coord.: 0.0000000 0.0000000 0.0000000

-------------------------------------------------------------------
E( 2 )= -0.2405 eV Re[Sigma]= 95.950765 meV Im[Sigma]= 0.392034 meV
E( 3 )= -0.2405 eV Re[Sigma]= 95.950765 meV Im[Sigma]= 0.392034 meV
E( 4 )= -0.2405 eV Re[Sigma]= 95.950765 meV Im[Sigma]= 0.392034 meV
-------------------------------------------------------------------
...

Note that the electron energies are now reported with respect to EF . Moreover, the self-energy for
the first valence band is not computed since fsthick is 7 eV. To plot the linewidths you can use
the file linewidth.elself that has been created. For example, ImΣ for the highest valence band
should look like:

gnuplot> plot "linewidth.elself" u 1:4 every 3::2 w lp

100

80
ImΣ(meV)

60

40

20

0
Γ X W L Γ K

If you want to plot the electron lifetimes, these are given by τnk = ~/(2ImΣnk ).

Can you understand the behavior of the linewidths along the Brillouin-zone path? It is useful to look
also at the electronic band structure.

Note: You can also look at the contribution of each phonon mode by using iverbosity = 3 in the input. The file
linewidth.elself will then contain the mode-resolved linewidths. Can you tell which phonon has the largest contri-
bution and why?

I Try increasing the q grid, and also using a random set of q points: you will see the linewidths
are not well converged yet. For polar materials it is indeed more difficult to converge Brillouin-zone
integrals, due to the 1/|q| divergence of the Fröhlich matrix elements.

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 15 of 20

Exercise 3
In this exercise we will calculate the spectral function of n-doped MgO.

I Run a self-consistent calculation for MgO. First, create a new folder and download the pseudopo-
tentials: here we will use pseudopotentials from the ONCV library.

$ cd ../ ; mkdir exercise3; cd exercise3

$ wget http://www.quantum-simulation.org/potentials/sg15_oncv/upf/O_ONCV_PBE-1.0.upf
$ wget http://www.quantum-simulation.org/potentials/sg15_oncv/upf/Mg_ONCV_PBE-1.0.upf
Prepare the scf input and run the calculation:
&control mgo.scf.in
calculation='scf',
prefix='mgo',
pseudo_dir= './',
outdir = './',
/
&system
ibrav=2,
celldm(1) = 7.9595
nat=2,
ntyp=2,
ecutwfc = 30,
/
&electrons
conv_thr=1.d-10,
/
ATOMIC_SPECIES
Mg 24.305 Mg_ONCV_PBE-1.0.upf
O 15.999 O_ONCV_PBE-1.0.upf
ATOMIC_POSITIONS crystal
Mg 0.0 0.0 0.0
O 0.5 0.5 0.5
K_POINTS {automatic}
6 6 6 0 0 0

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < mgo.scf.in > mgo.scf.out

I Run a phonon calculation on a homogeneous 4 × 4 × 4 q-point grid.

-- mgo.ph.in
&inputph
prefix = 'mgo'
fildvscf = 'dvscf'
ldisp = .true
fildyn = 'mgo.dyn'
nq1=4,
nq2=4,
nq3=4,
tr2_ph = 1.0d-12
/

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/ph.x -npool 4 < mgo.ph.in > mgo.ph.out

The Born effective charges Z ∗ and the electronic dielectric constant ∞ are also computed, since
MgO is a polar insulator.

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 16 of 20

Plot the phonon dispersions using q2r.x and matdyn.x (note that for real calculations you should
relax the structure using a converged ecutwfc, and use at least a 4 × 4 × 4 q grid for the phonons).

I Gather the .dyn, .dvscf and patterns files into a save/ directory as in Exercise 1 and 2, using
the script /home/nfs3/smr3191/q-e-/EPW/bin/pp.py.

I Run a non self-consistent calculation on a homogeneous 4 × 4 × 4 k grid, preparing the input as

in Esercise 1 and 2, with the difference that now you will prompt:
$ /home/nfs3/smr3191/q-e/wannier90-2.1.0/utility/kmesh.pl 4 4 4 >> mgo.nscf.in
Use nbnd=12.

$ mpirun -np 4 /home/nfs3/smr3191/q-e/bin/pw.x -npool 4 < mgo.nscf.in > mgo.nscf.out

I Run EPW: wannierize the electronic band structure and check the interpolation using band_plot=.true..
-- mgo.epw.in
&inputepw
prefix = 'mgo'
amass(1) = 24.305
amass(2) = 15.999
outdir = './'
dvscf_dir = './save'

elph = .true.
kmaps = .false.
epbwrite = .true.
epbread = .false.
epwwrite = .true.
epwread = .false.
lpolar = .true.

wannierize = .true.
nbndsub = 4
nbndskip = 5
num_iter = 300
dis_win_min = 0
dis_froz_max= 11.0
dis_win_max = 21.8
proj(1) = 'O:p'
proj(2) = 'Mg:s'

elecselfen = .false.
phonselfen = .false.
a2f = .false.
band_plot = .true.

parallel_k = .true.
parallel_q = .false.

fsthick = 2.0
eptemp = 20
degaussw = 0.05

filqf = 'path.dat'
filkf = 'path.dat'

nk1 = 4
nk2 = 4
nk3 = 4

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 17 of 20

 nq1 = 4
nq2 = 4
nq3 = 4
/
8
0.000000000000000E+00 0.000000000000000E+00 0.000000000000000E+00
-0.250000000000000E+00 0.250000000000000E+00 -0.250000000000000E+00
0.500000000000000E+00 -0.500000000000000E+00 0.500000000000000E+00
0.000000000000000E+00 0.500000000000000E+00 0.000000000000000E+00
0.750000000000000E+00 -0.250000000000000E+00 0.750000000000000E+00
0.500000000000000E+00 0.000000000000000E+00 0.500000000000000E+00
0.000000000000000E+00 -0.100000000000000E+01 0.000000000000000E+00
-0.500000000000000E+00 -0.100000000000000E+01 0.000000000000000E+00

You can use again the Brillouin-zone path from Exercise 1 and 2.

I Calculate the spectral function A(k, ω) in the Fan-Migdal approximation (see lectures Thu.1 and
Thu.2) for n-doped MgO. Doping will be treated in the rigid band approximation, that is by a rigid
shift of the Fermi level. First, introduce a small doping charge of 0.00037 electrons/unit cell, which
corresponds to a density of about 2 × 1019 cm−3 : open the file crystal.fmt and modify line 3, from
16 (number of electrons) to 16.00037. Then calculate the Fermi level using EPW modifying the input
as follows:

$ cp mgo.epw.in mgo.epw.in2

... mgo.epw.in2
kmaps = .true.
epbwrite = .false.
epbread = .false.
epwwrite = .false.
epwread = .true.
lpolar = .true.

wannierize = .false.
...
band_plot = .false.
...

degaussw = 0.002

!filqf = 'path.dat'
!filkf = 'path.dat'
nkf1 = 100
nkf2 = 100
nkf3 = 100
nqf1 = 1
nqf2 = 1
nqf3 = 1
...

Note the very dense k grid and small smearing degaussw: these parameters are needed in order to
converge the Fermi energy at very small dopings. Stop the calculation just after the Fermi energy
has been written in the output. You should obtain:

Skipping the first 5 bands:

The Fermi level will be determined with 6.00037 electrons

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 18 of 20
 Fermi energy is calculated from the fine k-mesh: Ef = 13.381852 eV

hence the Fermi energy is shifted about 80 meV above the conduction band bottom.

For calculating the electron spectral function you need to specify specfun_el = .true. and an
energy range, as referred to the Fermi level. The latter is read from the input file, and corresponds
to the value that you just calculated. Modify the input as follows:

$ cp mgo.epw.in2 mgo.epw.in3

... mgo.epw.in3
specfun_el = .true.
wmax_specfun = 0.1
wmin_specfun = -0.5
nw_specfun = 300
efermi_read = .true.
fermi_energy= 13.381852

fsthick = 2.0
eptemp = 20
degaussw = 0.01

!filqf = 'path.dat'
filkf = 'path2.dat'
nqf1 = 40
nqf2 = 40
nqf3 = 40
...

The file path2.dat contains a dense mesh of k points only along the ΓX direction (to speed up the
calculation, we only look at this BZ direction). This path is provided in the exercise3/ folder, and
it can be generated for example using the following script:
cat > kpath.py << EOF
import numpy as np

nk=0.2/0.005
print int(nk)

for k in np.arange(0.0,0.2,0.005):
print ' 0.0 '+str(k)+' '+str(k)+' 1.0'
EOF

$ python kpath.py > path2.dat

After running EPW, the calculated electron spectral function will be saved in the file specfun.elself,
whereas the file specfun_sup.elself contains the frequency-dependent self-energy Σnk (ω). You
can plot the spectral function for example by using gnuplot, with the command:

gnuplot> set view map ; splot "specfun.elself" w pm3d

You should see a satellite band below the conduction band bottom, however its position and intensity
are not converged. Try using a finer q mesh and/or random q points, and observe the speed of
convergence. A more converged plot is reported below.

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 19 of 20

 0.1

-0.1

Energy (eV) -0.2

-0.3

-0.4

-0.5
Γ 0.2 X

Trieste, 19-23 March 2018 C. Verdi Hands-on Session | 20 of 20