Questions about the Si example

[tzhu11]

Dear Dr. Ponce,

Thank you very much for your earlier reply on the Wannier90 projection. I'm asking the current question here, as I think this might also be interesting to other EPW beginners who are benchmarking with si.
I'm also attaching my EPW input below, with which I met some questions and difficulties, and I'll appreciate greatly your help.

1) What baffles me the most now are the electron energies in "linewidth.elself",which systematically shift up for unknown reasons.
I did the example in Wannier90, the same as http://www.physics.rutgers.edu/pythtb/examples/w90_quick.png
All these band curves are well reproduced in EPW, except that all the EPW bands (electron energies) shifts up by ~6eV as a whole.

In the EPW output, I found

Fermi energy coarse grid = 6.506759 eV
Fermi energy is calculated from the fine k-mesh: Ef = -5.767008 eV

I feel this can explain my problem, but I don't know why I got this unacceptable difference.

2) The other problem I met, which may well be related to the above, is that "Im(Sgima)" are orders larger than the results in your Figure 8.

3) I'm not allowed to set nbndsub=12 even I calculated 12 bands in the nscf step, is it because the Winner windows are not properly set please?

(since I'm trying to reproduce the Figure 8 in your recent work [S. Ponce et al., Comp. Phys. Comm. 2016], as others may also do, is it possible that you could post your EPW input file for the Si case please? It will be really helpful if I can study your input.) http://www.sciencedirect.com/science/article/pii/S0010465516302260



Thanks a lot,
Tai

Code: Select all

&inputepw
  prefix      = 'Si'
  amass(1)    = 28.085500
  outdir      = './'

  iverbosity  = 3

  elph        = .true.
  epbwrite    = .true.
  epbread     = .false.

  etf_mem     = .false. !.true.

  epwwrite    = .true.
  epwread     = .false.

  nbndsub     =  8
  nbndskip    =  0

  wannierize  = .true.
  num_iter    = 3000
  iprint      = 2
  dis_win_max = 17
  dis_froz_max= 6.5
  proj(1)     = 'Si:sp3'

  elecselfen  = .true.
  nest_fn     = .true. 
  phonselfen  = .true.
  a2f         = .true.

  specfun      = .true.
  wmin_specfun = -4
  wmax_specfun = 12
  nw_specfun   = 100

  parallel_k  = .true.
  parallel_q  = .false.

  fsthick     = 20 ! eV
  eptemp      = 300 ! K
  degaussw    = 0.001 ! eV

  dvscf_dir   = './save/'
  filukk      = './Si.ukk'

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

rand_nk=30000
!  nkf1        = 40
!  nkf2        = 40
!  nkf3        = 40

rand_nq=150000
!  nqf1        = 40
!  nqf2        = 40
!  nqf3        = 40
 
  nk1         = 6
  nk2         = 6
  nk3         = 6

  nq1         = 6
  nq2         = 6
  nq3         = 6
 /
      16 cartesian
     0.000000000   0.000000000   0.000000000 0.08
    -0.166666667   0.166666667  -0.166666667 0.08
    -0.333333333   0.333333333  -0.333333333 0.08
     0.500000000  -0.500000000   0.500000000 0.08
     0.000000000   0.333333333   0.000000000 0.08
    -0.166666667   0.500000000  -0.166666667 0.08
     0.666666667  -0.333333333   0.666666667 0.08
     0.500000000  -0.166666667   0.500000000 0.08
     0.333333333   0.000000000   0.333333333 0.08
     0.000000000   0.666666667   0.000000000 0.08
     0.833333333  -0.166666667   0.833333333 0.08
     0.666666667   0.000000000   0.666666667 0.08
     0.000000000  -1.000000000   0.000000000 0.08
     0.666666667  -0.333333333   1.000000000 0.08
     0.500000000  -0.166666667   0.833333333 0.08
    -0.333333333  -1.000000000   0.000000000 0.08

[carla.verdi]

Dear Tai,

The electron energies in linewidth.elself are shifted by the Fermi energy.

I suspect your linewidths might be wrong because you are using a path for the q sampling (filqf = 'path.dat'), whereas you should input a fine uniform grid or a large number of random points (in Fig.8 of the paper 150,000 random q points are used).

Regarding your last question - yes, try modifying the dis window in the wannierization.

Best
Carla

[tzhu11]

Dear Carla,

Thank you very much for your reply, which helped a lot.

To be clear, is my procedure correct please? First I used both dense k mesh and dense q mesh, then I read the epb files for electron/phonon linewidths.
I calculated the phonon linewidth without giving filkf, but with fine k mesh a filqf. When I calculate electron linewidth, I dont give filqf, but use a fine q mesh and a filkf.
If this is correct, I'm still getting a lot "INFINITY" in the phonon linewidth:

Code: Select all

# Phonon frequency and phonon lifetime in meV 
# Q-point  Mode   Phonon freq (meV)   Phonon linewidth (meV)
        1     1      0.12613452E+02              Infinity
        1     2      0.13944123E+02      0.1424561421E+43
        1     3      0.46076707E+02      0.2313555549E+37
        1     4      0.49912501E+02              Infinity
        1     5      0.59294073E+02              Infinity
        1     6      0.59590967E+02              Infinity
        2     1      0.12614058E+02              Infinity

Meanwhile, the electron linewidth is still too large.

Code: Select all

# Electron lifetime (meV)
#      ik       ibnd                 E(ibnd)      imode          Im(Sgima)(meV)
        1          1   -0.97407444321053E+01          1    0.15711864648894E+64
        1          1   -0.97407444321053E+01          2    0.16171103400851E+64
        1          1   -0.97407444321053E+01          3    0.57346605047599E+64
        1          1   -0.97407444321053E+01          4    0.15700045522368E+64
        1          1   -0.97407444321053E+01          5    0.52901398899141E+64
        1          1   -0.97407444321053E+01          6    0.44891611481372E+64
        1          2   -0.71395640547082E+01          1    0.61918595455640E+63

any suggestions please?

Thanks a lot,
Tai

[gbrunin]

Dear Tai,

Did you manage to solve your problem ? I face the same issue : infinite phonon linewidth and Im(Sigma) on the order of 10^100... Here is my input file :

Code: Select all

EPW computation for Si
&inputepw
  prefix      = 'Si'
  dvscf_dir   = '../ph/save/'
  amass(1)    = 28.086
  outdir      = './'

  elph        = .true.
  kmaps       = .false.

  !epbread     = .false.
  !epbwrite    = .true.

  epbread     = .true.
  epbwrite    = .false.

  epwwrite    = .true.
  epwread     = .false.

  etf_mem     = .false.

  nbndsub     =  8
  nbndskip    =  0

  wannierize  = .false.
  !wannierize  = .true.
  num_iter    = 2000
  iprint      = 2
  dis_win_max  = 17
  dis_froz_max = 6.5
  proj(1)     = 'Si:sp3'

  wdata(1) = 'bands_plot = .true.'
  wdata(2) = 'begin kpoint_path'
  wdata(3) = 'G 0.000 0.000 0.000 X 0.000 0.500 0.500'
  wdata(4) = 'X 0.000 0.500 0.500 W 0.250 0.500 0.750'
  wdata(5) = 'W 0.250 0.500 0.750 K 0.375 0.375 0.750'
  wdata(6) = 'K 0.375 0.375 0.750 G 0.000 0.000 0.000'
  wdata(7) = 'G 0.000 0.000 0.000 L 0.500 0.500 0.500'
  wdata(8) = 'L 0.500 0.500 0.500 U 0.250 0.625 0.625'
  wdata(9) = 'U 0.250 0.625 0.625 W 0.250 0.500 0.750'
  wdata(10) = 'W 0.25 0.500 0.750 L 0.500 0.500 0.500'
  wdata(11) = 'L 0.50 0.500 0.500 K 0.375 0.375 0.750'
  wdata(12) = 'U 0.25 0.626 0.625 X 0.000 0.500 0.500'
  wdata(13) = 'end kpoint_path'
  wdata(14) = 'bands_plot_format = gnuplot'

  iverbosity  = 0

  elecselfen  = .true.
  phonselfen  = .true.
  a2f         = .true.

  parallel_k  = .true.
  parallel_q  = .false.

  fsthick     = 10  ! eV
  eptemp      = 300  ! K
  degaussw    = 0.01 ! eV
  degaussq    = 0.05 ! meV

  filukk      = './Si.ukk'

  rand_nk      = 20000
  rand_nq      = 80000

  rand_q      = .true.
  rand_k      = .true.

  nk1         = 12
  nk2         = 12
  nk3         = 12

  nq1         = 6
  nq2         = 6
  nq3         = 6
 /
      16 cartesian
  0.000000000   0.000000000   0.000000000    0.004629629629629629
 -0.166666667   0.166666667  -0.166666667    0.037037037037037035
 -0.333333333   0.333333333  -0.333333333    0.037037037037037035
  0.500000000  -0.500000000   0.500000000    0.018518518518518517
  0.000000000   0.333333333   0.000000000    0.027777777777777776
 -0.166666667   0.500000000  -0.166666667    0.111111111111111111
  0.666666667  -0.333333333   0.666666667    0.111111111111111111
  0.500000000  -0.166666667   0.500000000    0.111111111111111111
  0.333333333   0.000000000   0.333333333    0.055555555555555556
  0.000000000   0.666666667   0.000000000    0.027777777777777776
  0.833333333  -0.166666667   0.833333333    0.111111111111111111
  0.666666667  -0.000000000   0.666666667    0.055555555555555556
  0.000000000  -1.000000000   0.000000000    0.013888888888888888
  0.666666667  -0.333333333   1.000000000    0.111111111111111111
  0.500000000  -0.166666667   0.833333333    0.111111111111111111
 -0.333333333  -1.000000000   0.000000000    0.055555555555555556

Would increasing the number of k/q-points solve the problem ?
Thanks for any suggestion,
Guillaume

[sponce]

Dear tzhu11,

Sure, all my inputs, psp etc are publicly available on our group GitHub page: https://github.com/mmdg-oxford/papers/t ... e-CPC-2016

Note however that (1) those input are for an old version of EPW and you will need to adapt them and (2) I usually provide only 1 representative input but I might have used different one or a lot of other one to produce the results of the paper. Nonetheless, you could start from there if you want.

@Guillaume: Can you try with an homogeneous k and q grid to start with (to see if you get reasonable linewidth). Have you printed your electron and phonon bandstructure witin EPW to make sure that the Wannierization is ok? You need band_plot variable and filfk and filfq with the high symmetry path.

Best,
Samuel

[gbrunin]

Dear Samuel,

Thank you for your reply. I had a good electron band structure and did not check the phonon one. However, I saw in other posts that sometimes GGA was a problem, and I was using GGA. I switched to LDA, reran everything and now I seem to have much better results (Im(Sigma) on the order of 10, 100 meV).

Regards,
Guillaume

[sponce]

Hello Guillaume,

Glad to hear it. Usually LDA gives better phonons.

However, I have used GGA on silicon and it worked perfectly for me. There should be no reason for GGA to not work.
I'm suspecting something else (maybe negative phonons or wrong Wannierziation etc).

Best,
Samuel