Wrong VBM and CBM in mobility calculation

[sdwang]

Dear all,
I ran the mobility calculation and found the VBM and CBM in the epw2.out are not consistent with nscf.out values.
In the epw2.out I found that:

...skipping...
Fermi energy coarse grid = 0.000000 eV

===================================================================

Fermi energy is read from the input file: Ef = 4.539000 eV

===================================================================

ibndmin = 10 ebndmin = 0.040
ibndmax = 17 ebndmax = 0.628


Number of ep-matrix elements per pool : 2239488 ~= 17.09 Mb (@ 8 bytes/ DP)
Applying a scissor shift of 1.50000 eV to the conduction states

Temperature 300.000 K
Valence band maximum = 8.848697 eV
Conduction band minimum = 8.359119 eV
Warning: too many iterations in bisection
Ef = 8.359119
Mobility VB Fermi level 8.359119 eV
Valence band maximum = 8.848697 eV
Conduction band minimum = 8.359119 eV
Warning: too many iterations in bisection
Ef = 8.848697
Mobility CB Fermi level 8.848697 eV

===================================================================
Scattering rate
===================================================================
where the VBM and CBM are far larger than in the nscf.out , where they are 4.539 and 6.6407.
My in put of epw2.in is:
&inputepw
prefix = 'tc'
amass(1) = 12.01078,
!amass(2) = 32.07,
outdir = './'

iverbosity = 0

elph = .true.
kmaps = .true.
epbwrite = .false.
epbread = .false.
epwwrite = .false.
epwread = .true.
etf_mem = 1

scattering = .true.
scattering_serta = .true.
int_mob = .true.
carrier = .true.
ncarrier = 1E13


scissor = 1.5
nstemp = 1
tempsmin = 300
tempsmax = 300

restart = .true.
restart_freq = 100



nbndsub = 20
nbndskip = 0
lifc = .true.
asr_typ = 'crystal'


wannierize = .false.
fsthick = 4.0 !
eptemp = 300.00 ! K (same as PRB 76, 165108)
degaussw = 0.015 ! eV
efermi_read = .true.
fermi_energy= 4.5390
dvscf_dir = '../phonon/save'
!filkf = './inp.kpt'
!filqf = './inp.kpt'


nkf1 = 36
nkf2 = 36
nkf3 = 36

nqf1 = 36
nqf2 = 36
nqf3 = 36

nk1 = 12
nk2 = 12
nk3 = 12

nq1 = 6
nq2 = 6
nq3 = 6
/
32 cartesian
...

Thanks!

SD

[sponce]

Dear sdwang,

You should put the Fermi energy in the middle of your gap not at the VBM. So around 5.5 eV
Now if the gap is too large, it might be difficult for the bisection method to converge (deals with huge exponentials). Therefore I would remove the scissor = 1.5 from the calculation.
Indeed the mobility will not be affected. The scissor is typically used if the DFT bandgap is very small.

Note that the bisection algo has been improved. You can try the most recent GitLab developer version. In any case there will be a new release of QE and EPW at the end of Feb 2019.

Best wishes,
Samuel

[sdwang]

Dear Samuel,
Thank you for the reply.
I have removed the scissor and fixed the Fermi level in the middle o the gap, and the CBM seems OK but the
VBM is still high than the VBM(like below). The band structure of the interpolated with EPW agrees well with DFT one.

Temperature 300.000 K
Valence band maximum = 8.846493 eV
Conduction band minimum = 6.859119 eV
Warning: too many iterations in bisection
Ef = 6.859119
Mobility VB Fermi level 6.859119 eV
Valence band maximum = 8.846493 eV
Conduction band minimum = 6.859119 eV
Warning: too many iterations in bisection
Ef = 8.846493
Mobility CB Fermi level 8.846493 eV

===================================================================
Scattering rate
===================================================================


Fermi Surface thickness = 4.500000 eV
This is computed with respect to the fine Fermi level 5.558700 eV
Only states between 1.058700 eV and 10.058700 eV will be included
...
Thanks in advance!

Best

SD

[sponce]

Dear SD,

If you still get the warning in the bissection method, then you cannot expect correct results.

Have you tried the development version https://gitlab.com/QEF/q-e

Best wishes,
Samuel

[sdwang]

Dear Samuel,
I'v installed the dev version of QE, and I will try the new EPW to see waht's goning on then...
I noticed in the test-suite there is a direotry with a USPP for EPW. Does this mean we can deal the system with USPP in the
new version of EPW?
Thanks!

Shudong

[sponce]

Dear sdwang,

It is very new and therefore not very well tested but yes, in principle USPP should
now be supported

Best,
Samuel

[piotr]

Dear, all
I've got similar problem, but in my case increasing number of q-points (from 6x6x6 to 8x8x8) helps. If plot interpolated bands on the coarser grid than there is an artifact near the VBM that causes the issue. Here mesh is 12x12x12, so probably it is not the case, but plotting interpolated bands may be helpful anyway.

Best wishes,
Petr

[sdwang]

Dear Petr,
Thanks for the advice.
My interpolated bands agree well with the DFT one for 12x12x12 k-point. Anyway, I would increase the q-point to check
the results.

Best,

Shudong

[sponce]

Are you sure that "nbndskip = 0" is correct in your case.

How many electrons do you have in your nscf.in calculations ?
With respect to your nscf calculation, are you skipping bands or not ? If so you have to tell EPW. This typically will lead to wrong VBM and CBM determination.

Best wishes,
Samuel

[sdwang]

Dear Samuel,
My nscf file contains 32 electrons and nbnd=32 in the input, and I set nbndsub = 20 , nbndskip = 0 in the epw.in.
In this case, How do I tell EPW the 'real' band setting in epw.in file?
Does it matter with fsthick = 4.0 since I included 20 nbndsub?
Thanks!

Best

Shudong