Regarding work flow

[aseemrk]

Dear all,

I have a small idea which might help in reducing computational expense. The work flow that I have followed is as follows:

1. SCF and the Phonon calculations using all symmetries (dvscf is saved)
2. Carrying out NSCF calculation (nosym = .true.) to generate wavefunctions over entire BZ
3. Using wavefunctions generated at NSCF to compute electron-phonon matrix elements (same input as phonons in step 1 but set trans =.false. to reuse the dvscf computed in step 1)

I think that this would reduce the expense of computing phonons without symmetries, i.e. using k-point sampling in entire BZ.

Warm regards,
Aseem

[nguyen-group]

Hi Aseem,

Thank you very much for your great suggestion.

Previously, I tested with the workflow: SCF --> NSCF --> PH, in which the k-point of NSCF > SCF. However, it seems that the electron-phonon calculation is linked to only the wavefunction of the SCF step. Thus, the electron-phonon and electron-photon do not match.

Your ideal is: SCF --> PH (with symmetries) --> NSCF --> PH (without symmetries). It might be a good trick. Does it work with you? I will test this idea.

Best,
Nguyen

[aseemrk]

Hi Nguyen

We had tested this for our in-house code for resonant Raman (it is not public) and it had worked very well for Si, GaAs etc. I have not tested this for examples in QERaman. Please let me know if you find it to work well.

Input ph.inp after scf (with all symmetries):

&INPUTPH
prefix = 'test',
outdir = './tmp'
tr2_ph = 1.0d-16,
verbosity='high',
fildvscf = 'dvscf'
electron_phonon = 'epc'
/
0.0 0.0 0.0

Input elph.inp after nscf

&INPUTPH
prefix = 'test',
outdir = './tmp'
tr2_ph = 1.0d-16,
verbosity='high',
fildvscf = 'dvscf'
electron_phonon = 'epc'
trans=.false.,
/
0.0 0.0 0.0

Warm regards,
Aseem