EPW Forum

Dear all,

I am interested in calculating electron-phonon coupling (EPC) with EPW 4. Currently, I have a question about EPC calculation with wannierize=.false..
As mentioned in http://epw.org.uk/Documentation/Inputs, the simple solution to determine MLWFs is proj(1)='random'.

In order to do convergence test, I designed two runs with different fine electron grids, namely run-1 and run-2, respectively. In these two runs,
the MLWFs are calculated separately by using proj(1)='random'. And I noticed these two sets of MLWFs are different. I worry about the differences in
MLWFs will affect the convergence test.

So I want to use the same set of MLWFs in these two runs. For run-1, the EPC calculation was done as usual. For run-2, I copy the **.ukk file generated by run-1,
and set wannierize=.false., without other changes in the job script. But the calculation (run-2) was interrupted, without 'CRASH' file in the working directory. I found
some clue in the standard output of my computers, which said 'forrtl: severe (47): write to READONLY file, unit 6, file ../../../phonons/save/Chi3.phsave/patterns.1.xml.'

So my question is how to use previously generated MLWFs to do EPC calculation. Should I set epwread=.true. or anything else? And which files are needed? Thanks in advances!

Best regards,
Miao Gao
Ningbo University, Zhejiang, China.

Dear Miao Gao,

The electron fine grid should not affect the MLWFs at all.

Using random value for your initial projection might be a tricky business as the Wannier code might not be able to find the "global" minimum (i.e. minimum spread) if the initial solution is too far from the optimal one.

I suggest to use chemically based initial projection like "sp" etc..

If you still want to use random projection, you need to make sure that the Wannier code found the correct minimum.

To check this you can use the EPW "band_plot" input variable to get the Wannier interpolated electronic bandstructure. You can then check that this interpolated bandstructure compares well with a bandstructure generate with an nscf calculations and a lot of points.

For your other question, you might need to copy the .ukk file but also maybe the .chk file (I need to check this).
Moreover, can you access your "../../../phonons/save/Chi3.phsave/patterns.1.xml." file from where the calculation is run?

Best,

Samuel

Dear Dr. Samuel,

Thank you very much for your quick response! You are right, the electron fine grid should not affect the MLWFs. I just want to use different electron fine grid
to check the convergence of lambda. As you said, the random projection method is not robust. The same input parameters could give two different sets of MLWFs,
when using random projection. For example,
run-1: Final Spread (Ang^2) Omega Total = 13.812596145
run-2: Final Spread (Ang^2) Omega Total = 23.039443459

Thank you for your advice on checking the correct minimum of MLWFs. But it seems to me, the right interpolated bandstructure does not correspond to
the correct minimum of MLWFs. Although the omega_totals are evidently different, I found the interpolated bandstructures given by above two sets of MLWFs
are the same.

I have also examined the decay.H files in both runs, and found decay.H files are ok in the two cases. The maximal values of Hamiltonian in Wannier basis are
about 1.2E-5 and 4.6E-5, respectively, when the hopping distance is 26.7 angstrom. Does this kinds of decay in Hamiltonian means above two sets of MLWFs
are both reasonable to do following epw calculation?

I can access the file "../../../phonons/save/Chi3.phsave/patterns.1.xml.". But in my understanding, epw code should just read these files produced by the
phonon codes, and does not write into them, am I right? So the message write to READONLY file is strange. By the way, besides ukk and chk files,
does we need epwdata.fmt, dmedata.fmt, and .epb files, when setting wannierize= .false.?

Best regards,
Miao Gao