Electron self-energy in polar materials - sanity check?

[chrisewolf]

Dear all,

I am facing an "interesting" behavior for MgO. The self-energy (Im(sigma)) gets fairly high (~10^7 meV) and I cannot figure out if this is a convergence problem or a "result".

- phonon interpolation works well when reading the force constants from q2r and imposing crystal asr.
https://drive.google.com/file/d/1UBmSN6 ... 6Lvm3/view


- I have increased the fine grid to 50x50x50 but there is not much change above 25x25x25
- I am currently running it with random points to see if that improves the sampling near k=0 because there seems to be some noise:

The electron-band interpolation itself from wannier and epw looks otherwise good;
I offset the wannier band by 0.2 eV to make the plot easier to read

https://drive.google.com/open?id=1TKVSm ... yYU9HC1Y0O

here is the Im(Sgima)(meV) from linewidth.elself as colormap on the highest VB

https://drive.google.com/open?id=1POEdU ... OvC5Hi-Wl4

I ran the examples and I think my qe 6.2.1 and epw are otherwise fine;

Any idea how to check my results for sanity?

Have a great Sunday everyone!

Best,
Chris

[sponce]

Hello Chris,

What is the value of your dielectric constant and Born effective charges for MgO ?
Are they both converged (at the ph.x level). The k-grid needs to be large (like 20x20x20) to converge those.

Best,
Samuel

[chrisewolf]

Hi Samuel,

thank you for your swift reply! I have converged the calculation in the "SCF" sense at a k-grid of ~12 and a PW cutoff of about 120 (the pseudos are of ONCV type with PZ functional, similar to the GaN example).

The physical properties are then a diel. constant of ~3.0 (iirc the experiment is around 2.95) and the born charge on the Mg is a solid +2.

I plotted all quantities here: https://drive.google.com/open?id=15TmRl ... fVXY-C8vPy

Best,
Chris

[sponce]

Hello Chris,

Fair enough.

I'm afraid I see no obvious issue here. It could be a problem of restarting the calculation.

What you can try is to do "prtgkk" (look the input variable). This will print the electron-phonon matrix elements.

That way you will know if it is an issue of the "g" that becomes very large or because of the bandstructure or phonon frequencies.

If it is the g that becomes very large, then you need to make sure that the interpolation works well ( you can check the associated decay.epmat* files).

Best,
Samuel

[carla.verdi]

Dear Chris,

It seems indeed there is some problem with the interpolation, already at the phonon dispersions level, so most likely also with the matrix elements. Looking at your phonons plots https://drive.google.com/file/d/1UBmSN6 ... 6Lvm3/view
the dispersion looks good if you use the force constants from q2r, but the fact that it looks so different when you read the dyn mat directly within epw indicates there is probably a problem there. Have you checked that the save folder contains all the data? Are the dyn mat read correctly?

Best
Carla

[chrisewolf]

Thanks for the hint Samuel, I will check it tomorrow!

Dear Carla, in both cases the code does not throw an obvious error.

(1) the "wiggly phonons":

Code: Select all

  !lifc        = .true.
  !asr_typ     = 'crystal'
  lpolar      = .true.

shows

Code: Select all

        Imposing acoustic sum rule on the dynamical matrix
        Read dielectric tensor and effective charges
     Frequencies of the matrix for the current q in the star
    -0.00000    -0.00000    -0.00000   371.52240   371.52240   371.52240

(2) the "good interpolation"

Code: Select all

  lifc        = .true.
  asr_typ     = 'crystal'
  lpolar      = .true.

Code: Select all

        Imposing acoustic sum rule on the dynamical matrix
        Read dielectric tensor and effective charges
  Dyn mat calculated from ifcs
     Frequencies of the matrix for the current q in the star
    -0.00001    -0.00001     0.00001   372.64354   372.64354   372.64354

        q(    1 ) = (   0.0000000   0.0000000   0.0000000 )

seems that in both cases at least at gamma everything is fine. I then checked to see if any file is missing but the dyn1-dyn16 and dvscf files look ok.

I am currently running the GaN example to see if there is a problem there as well and I will update you when the results trickle in...

Thank you both for your time and help!

Best,
Chris


PS I would post the output files but I am sure no one wants to read through 120 MB of raw text. If there is anything else that helps diagnosing please let me know

[chrisewolf]

PPS: can someone help me out with the interpretation of the matrix elements

Code: Select all

# Spatial decay of e-p matrix elements in Wannier basis
   0.0000000000000000        3.7810338992837471E-002
   2.9818693093930255       0.13268456398053011     
   2.9818693093930255        4.5142482588624487E-002
   2.9818693093930255        4.9374750837164888E-002
   2.9818693093930255        3.2249215113515924E-002
   2.9818693093930255        4.0332192873203332E-002
   2.9818693093930255        3.3389574161240831E-002
   2.9818693093930255        4.8507229542031988E-002
   2.9818693093930255        6.5257687145065441E-002
   2.9818693093930255        3.7867830065880313E-002
   2.9818693093930255        2.9275825643998412E-002
   2.9818693093930255        5.2683211960116119E-002
   2.9818693093930255        4.6127278458422266E-002
   4.2170000185677115        3.2055293952381755E-002
   4.2170000185677115        3.2915778145332503E-002
   4.2170000185677115        3.3632182918114775E-002
   4.2170000185677115        3.2100841448966994E-002
   4.2170000185677115        4.6947594938131255E-002
   4.2170000185677115        7.2676172768031705E-002
   5.1647491453990408        3.6194763418870735E-002
   5.1647491453990408        3.2875466371838344E-002
   5.1647491453990408        4.2539295509953848E-002
   5.1647491453990408        5.7612813250484590E-002
......

my wannier basis comes from projection on the O:p orbitals

Code: Select all

  wannierize  = .true.
  nbndsub     =  3
  nbndskip    =  5
  num_iter    = 300
  iprint      = 2
  dis_win_max = 12
  dis_froz_min= 0
  dis_froz_max= 10
  proj(1)     = 'O:p' 
  wdata(1) = 'bands_plot = .true.'
  wdata(2) = 'begin kpoint_path'
  wdata(3) = 'K1 0.50  0.50  0.50   K2  0.0  0.0   0.0'
  wdata(4) = 'K2 0.00  0.00  0.00   K3  0.75 0.375  0.375'
  wdata(5) = 'end kpoint_path' 
  wdata(6) = 'write_hr=.true.'
  wdata(7) = 'write_hr_diag = .true.'

and gives

Code: Select all

 On-site Hamiltonian matrix elements
     n        <0n|H|0n> (eV)
   -------------------------
     1         6.358981
     2         6.358981
     3         6.358981

[sponce]

Hello,

Code: Select all

# Spatial decay of e-p matrix elements in Wannier basis
   0.0000000000000000        3.7810338992837471E-002
   2.9818693093930255       0.13268456398053011     
   2.9818693093930255        4.5142482588624487E-002

Seems wrong that the second one is larger than the first one.

Can you try the "simple" asr using matdyn.x ?
Do you also get such bad interpolation ?

Can you post the full scf.in and epw.in input files ?

Best,
Samuel

[chrisewolf]

Hi Samuel,

thanks for your fast reply (again)!

all inputs are here: https://drive.google.com/open?id=1zg4N4 ... 0NowwNVDW1

I will try to see if the asr type changes things, I will keep you posted!

Best,
Chris

[chrisewolf]

I am wondering if I made some mistake with the k point units in one of the steps as well;

As far as I can see the filkf (filqf) uses "crystal" units (indicated by 200 crystal in the file header) and all weights equal summing up to 1 ;

Code: Select all

201 crystal
0.5 0.0 0.0 0.00497512437811
0.495 0.0 0.0 0.00497512437811
0.49 0.0 0.0 0.00497512437811
0.485 0.0 0.0 0.00497512437811
0.48 0.0 0.0 0.00497512437811
0.475 0.0 0.0 0.00497512437811

I used xcryden to get the path which also gives "crystal" coordinates (and subsequently changed the weights to add up to 1)

Code: Select all

The content of the generated PWscf K_POINTS file is the following:
----------------------------------------------------------------
 K_POINTS crystal
         100
        0.5000000000    0.0000000000    0.0000000000    1.0
        0.4949494949    0.0000000000    0.0000000000    1.0
        0.4898989899    0.0000000000    0.0000000000    1.0
        0.4848484848    0.0000000000    0.0000000000    1.0
        0.4797979798    0.0000000000    0.0000000000    1.0
        0.4747474747    0.0000000000    0.0000000000    1.0
        0.4696969697    0.0000000000    0.0000000000    1.0
        0.4646464646    0.0000000000    0.0000000000    1.0
        0.4595959596    0.0000000000    0.0000000000    1.0
        0.4545454545    0.0000000000    0.0000000000    1.0

this "should work", right?

Best,
Chris