EPW phonon linewidth inconsistent with QE results

General discussion around the EPW software

Moderator: stiwari

hlee
Posts: 415
Joined: Thu Aug 03, 2017 12:24 pm
Affiliation: The University of Texas at Austin

Re: EPW phonon linewidth inconsistent with QE results

Post by hlee »

Dear Zishen:
I note that another thing different from epw and qe:
The prints from EPW/src/ephwann_shuffle.f90 doesn't mention the square of the electron-phonon matrix, and in EPW/src/selfen.f90:
In the comments of the subroutine you mentioned, that expression is for the case that the delta approximation is not used and in that expression, the part including the electron-phonon (e-ph) vertex is just omitted [Please pay your attention to the "sans"]; the e-ph vertex is correctly included in the actual calculation in EPW.

Since you are using the delta approximation (you set delta_approx= .true. in your EPW input) [See J. Noffsinger et al, Comput. Phys. Comm. 181, 2140 (2010)], the expressions used in the subroutine of selfen_phon_q in EPW/src and that of elphsum in PHonon/PH are the same. Both of them consider the spin degeneracy.
As I said, the only difference is that in EPW, the thing printed has an additional factor of 2 [The printing of phonon linewidth is done in EPW/src/ephwann_shuffle.f90] according to the expression of \Gamma = - 2 * Im \Pi^R .
And I compare curves between phonon dispersion and linewidth, the epw linewidth equal to zero points is exactly the imaginary phonon dispersion points. I think this is the reason for difference and it may affect other electron-phonon coupling calculations (electron-phonon matrix? electron-phonon constant?).
This is related to the different thing.
In EPW, when the phonon frequency is less than eps_acustic (See https://docs.epw-code.org/doc/Inputs.html#eps-acustic), some properties are ignored (set to zero). In your case, this leads to the zero phonon linewidth for the q values where the phonon frequency is less than eps_acustic.

(Excerpt from the subroutine of selfen_phon_q in EPW/src/selfen.f90)

Code: Select all

      wq(:) = zero
      DO imode = 1, nmodes
        wq(imode) = wf(imode, iq)
        IF (wq(imode) > eps_acustic) THEN
          g2_tmp(imode) = one
          inv_wq(imode) = one / (two * wq(imode))
        ELSE
          g2_tmp(imode) = zero
          inv_wq(imode) = zero
        ENDIF
      ENDDO
Sincerely,

H. Lee

zz11ss11zz
Posts: 18
Joined: Sat Jul 11, 2020 1:59 am
Affiliation: National University of Singapore

Re: EPW phonon linewidth inconsistent with QE results

Post by zz11ss11zz »

Dear H. Lee,

Thank you!
The tag eps_acustic works for phonon linewidth. After considering the factor of 2: the linewidths between qe and epw are very similar except for small difference:
epw&qe linewidth.png
epw&qe linewidth.png (57.95 KiB) Viewed 1784 times
I also calculated electron-phonon coupling matrix g (lowest phonon - the band which cross fermi surface) and electron-phonon coupling constant lamda, and I find the imaginary frequency also induce problem in them (I have added eps_acustic):
matrix & lamda.png
matrix & lamda.png (17.3 KiB) Viewed 1784 times
I was wondering if there is a relevant tag to solve this problem? And for the case of lamda, it is said it typically range between 0~2. The unreasonable large lamda may also because: lamda=phonon-linewidth/(pi*hbar*N(0)*w^2)?
N(0): density of fermi surface
w: phonon frequency.
As our structure has imaginary phonon frequency, the phonon curve will have very small w, this will induce large lamda as observed? I am not sure if I mkae mistake here. And in previous paper (prb 99, 161119 (2019)), they used large degauss to overcome the problem of imaginary frequency, then obtaining lamda.

Sincerely,
Zishen

Post Reply