epwdata.fmt format

[zahra]

Dear all,

I have run test jobs and am mostly interested in electron-phonon coupling matrix elements. I know that we can have these elements in output for coarse grid Bloch representation (prefix.epb files which are binary) and also in wannier representation (epwdata.fmt).
I'd like to know how to have this matrix for fine grid in Bloch representation (numbered as task 6 in following paper: http://www.sciencedirect.com/science/ar ... 5510003218 in section 3.3)
and
what is the format of epwdata.fmt? what each line stands for? How the matrix elements are sorted?
Thanks in advance.

Best,
Zahra

[sponce]

Dear Zahra,

The epwdata.fmt file does not contain the electron-phonon matrix element in Wannier representation.

It contains the eigenenergies, nb of bands, nb kpt, nb of modes, nb of qpoints and the Hamiltonian in real-space.

The electron-phonon matrix element in real space is contain in the binary file XX.epmatwp1

The code does not output the el-ph matrix in bloch space as it would be a huge file. I just wrote in the code the element I was interested by hacking the code for the paper (the new EPW paper).

The matrix elements are stored in the variable epf17 in ephwann_shuffle.f90

Best,

Samuel

[zahra]

Hi Samuel,

Thank you so much for your reply and nice clarification.
In fig.15(c) of your paper (http://www.sciencedirect.com/science/ar ... 5516302260) what is the horizontal axes? k+q or q?
I am a bit confused. Variable epf17 has 4 indices which seems to me as if the first two are bands indices which appear in Fermi energy window, the third is the phonon mode and the last is counter over q (for case parallel_k). I can have it in the output but I'd rather to have it as a function of initial and final particle wave vectors for each phonon mode (same as the aforementioned figure of yours and the fig.2 in http://journals.aps.org/prb/abstract/10 ... .87.115418). In fact I need to know the wave vectors for ibnd and jbnd which appear to be the initial and final states respectively. Are the variables labeled xkk (initial?!) and xkq (final?!) the wave vectors? If so, in which unit are they?
And the last question is that for finding
abs(epmatf (jbnd, ibnd, imode))**two / (two * wq)
which you addressed earlier in forum, should I be worried about the acoustic phonon frequencies which are zero in \Gamma point?
Thanks.

Regards,
Zahra

[sponce]

Dear Zahra,

For Fig 15(a) it is the k-point
For Fig 15(b) it is the q-point
For Fig 15(c) it is the q-point

The g output for Fig 15(c) should be g at k=Gamma, iband = jband = VBM (the red line in Fig 15(a) ) and omega= max phonon freq (red line in Fig 15(b)). What is shown is the change of that g for different q-points. Note that I have performed average whenever the band or phonon modes are degenerate for that q-point.

This should answer your question on epf17.

For epmatf, you need indeed to worry about zero phonon freq. In the code we neglect phonon mode with very small phonon frequency.

Best,

Samuel

[zahra]

Hi Samuel,

Thank you very much for your time. This was so helpful.
Sorry for asking the question again but since the variable xkq is addressed in ephwann_shuffle.f90 as 'Current k+q point on the fine grid' and is addressed in elph2.f90 as 'local k+q grid, coarse (3, nks)' and with different dimensions, I am at lost in here. I need the true wave vectors for initial and final states for 'fine' grid and I need to be sure that they are correctly paired to epmatf when I try to print it out.

Best,
Zahra

[carla.verdi]

Dear Zahra,

You are right there are two different variables with the same name, however xkq in elph2 is a global variable that is used for k+q on the coarse grid, whereas xkq in ephwann_shuffle is a local variable that stores the coordinates of each k+q point while looping over each k and q in the fine grid. Note that there is no conflict because when using the module 'elph2' in ephwann_shuffle, xkq is not included in the 'only' list.

Best
Carla

[zahra]

Hi Carla

Thanks a lot. Then I will use xkk and xkq.

Best,
Zahra

[zahra]

Hi,

I've noticed that the variable wf(imode,iq) which stands for phonon frequencies mixes the phonon bands as they crosses each other. It seems that at every q point in the reciprocal lattice the phonon bands are counted and labeled as from smallest value to largest one. This then means that there is always a band which is labeled differently at diverse q points since it might crossed other bands. I plotted phonon dispersion both as an output of both ph.x and epw.x and I found out in both cases bands are mislabeled through out the path of high symmetry points. I first noticed when I had the map plots of g matrix where a trace of phonon band mixture can obviously be seen. It seems that even dvscf files from ph.x calculation are also affected.
Now, I'm asking you how to avoid this feature in materials having band crossing in phonon dispersion?

Regards,
Zahra

[sponce]

Hello Zahra,

This is correct.

This is a common problem. You have 2 possibilities to solve this:
1) The hard way: You have to compute wavefunctions overlap to know automatically the real bands dispersion. In the case of phonons it will be phonon wavefunctions.
2) The easy way: You know your system and when the bands are crossing. You manually hack the code so that it follow your path. For example from the X point to the Y points you take band I and then from Y to Z you take the band J.

Best,

Samuel

[zahra]

Hi Samuel,

I see. Thank you very much.

Regards,
Zahra