EPW Forum

Hello everyone, and thanks in advance for any advice you may be able to offer me.

I am attempting to use the ZG method to evaluate temperature-dependent properties for a variety of phonon calculation methods. I have force constant (FC) files for several method x temperature combinations for my material. I'm fairly confident the FC files are correct as I double checked phonon spectra from my scripts and the output from ph.x and they matched. Here's my ZG settings:

&input
asr='crystal',
amass terms here in AMU
flscf='cubic.scf.in'
flfrc='qe444.fc'
atm_zg terms here
T=300
dim1=4, dim2=4, dim3=4, incl_qA=.false.
compute_error=.true., synch=.true., error_thresh=0.05, niters = 50000
/

I tried lower error_thresh initially, but it has issues even converging to 0.05. I set incl_qA=.false. as recommended for larger supercells. The issue is, I am seeing less subtle property shifts that I was expecting (e.g., 1.5 eV bandgap to 2 eV bandgap for the above ZG calculation settings). I think this shift should be <0.1 eV for our system.

I'm getting a quite large "Sum of diagonal terms per q-point". Or, it seems large to me at 98 compared to the <1 outputs I saw in the tutorial calculations:

Optimum configuration found !

Sum of diagonal terms per q-point: 97.605602
Error and niter index: 0.049835 3819

In addition, we have some (two, each around 1E-022) imaginary frequencies. Could this be the source of the error? I saw in ZG's source code that the magnitude is simply used for imaginary frequencies. Could the above issue be remedied if I instead pass over these cases? Otherwise, could some of the other ZG settings that I've left as defaults be relevant? Thanks for the insights.

Best,
Sean Nations

Hello,

Thank you for posting this question All questions are reasonable. Below I provide a point-to-point response:

>> I have force constant (FC) files for several method ...

What do you imply by external force constants in the title? Are the force constants calculated by density functional perturbation theory (ph.x) or finite differences using other codes/methods?

>> I tried lower error_thresh initially, but it has issues even converging to 0.05.

A value of 0.05 is a tight threshold. Any value below 0.1 for a 4x4x4 supercell should be okay.

>> I set incl_qA=.false. as recommended for larger supercells.

You should set incl_qA=.false., if your system has degenerate band edges in the equilibrium structure and you want to minimize the degeneracy splitting effect after applying ZG displacements. So my suggestion is to use first the equilibrium configuration (equil_pos.dat) in a supercell and check which bands are degenerate (if any). This will also help you to determine a correct value for the band gap later; in case of degeneracy splitting you should take the average over the number of the initially degenerated states (both in the CBM and VBM).

>> The issue is, I am seeing less subtle property shifts that I was expecting (e.g., 1.5 eV bandgap to 2 eV bandgap for the above ZG calculation settings). I think this shift should be <0.1 eV for our system.

So to be clear: with the ZG calculation, you get a band gap opening of 0.5 eV? I didn't quite understand.

Another important point that you would like to check: since you are using a supercell, you need to identify the correct K-point(s) that map to the k-point(s) of the VBM and CBM in the unit cell. An example of how to do it is provided in the exercise1 here: https://docs.epw-code.org/doc/TutorialZG.html. You can also try band structure unfolding (more expensive) as in exercise2 in order to make sure you probe the correct VBM and CBM.

>> I'm getting a quite large "Sum of diagonal terms per q-point". Or, it seems large to me at 98 compared to the <1 outputs I saw in the tutorial calculations:

You shouldn't worry about this value; it depends on the phonons of the system and the number of branches (modes).

>> In addition, we have some (two, each around 1E-022) imaginary frequencies. Could this be the source of the error?

Those modes are probably two of the three acoustic modes at Gamma whose frequencies are essentially zero. They do not affect the generation of special ZG displacements. If you get more WARNING messages, then it means your phonons are not stable ...

>> Otherwise, could some of the other ZG settings that I've left as defaults be relevant?

It seems you have all the necessary info for your calculation.

So bottom line you could check (i) for degeneracies and their correct treatment, and (ii) whether you are using the correct K-points in the supercell to calculate the band gap. Another suggestion here is to start from a 3x3x3 supercell and check for convergence.

Please let me know if you have more questions

With best wishes,

Marios Zacharias