Question about ecut test, SCF run length, preconditioning

[hqzhou]

Hi,

I'm a beginner of qbox, thank you for your fast code. I'm running a series of tests for
deepening my understanding of usage of the software. Unfortunate, I'm quite confused with
the results. For example, in test 1, I failed to reach convergence of energy with increasing
ecut, and the results vary just too big; In test 2, increasing ecutprec gave influence to
the energy; In test 3, number of SCF iterations made big difference in energy. So, I want
to ask following questions:

1. What's wrong or missing in my input file? How big is big enough for ecut?
Do we have any guideline?
2. How to set ecutprec? I have tried several values between its lower limit and upper limit,
but the energies show big difference.
3. How long is long enough for SCF iteration? Usually, the same system converged within
30 SCF iterations using VASP or PWscf.

Thank you.

zhou huiqun
@earth sciences, nanjing university, china

Below are my input file and the results:
================================================
pv.sys
set ecut 60
set xc PBE
set wf_dyn PSDA
set ecutprec 20
randomize_wf
run 0 100

Test 1: energy convergence with increasing ecut
ecut/ecutprec etotal
40/15: <etotal> 133538.06879833 </etotal>
50/17: <etotal> 126959.71015589 </etotal>
60/17: <etotal> 132046.86765584 </etotal>
70/21: <etotal> 117297.12296868 </etotal>
80/30: <etotal> 108257.72309249 </etotal>

Test 2: fixed ecut with different ecutprec
ecut/ecutprec etotal
60/15: <etotal> 145589.95367565 </etotal>
60/17: <etotal> 132046.86765584 </etotal>
60/20: <etotal> 121048.71808568 </etotal>
60/25: <etotal> 117796.60714825 </etotal>
60/30: <etotal> 112451.42932916 </etotal>

Test 3: different number of SCF iteration
ecut=60, ecutprec=20
50 SCF: <etotal> 121048.71808568 </etotal>
100 SCF: <etotal> 127690.57294852 </etotal>

[fgygi]

The results you show seem to indicate that the energy is not converged at all. Can you post the contents of the file pv.sys? Also, please consider getting an account on http://estest.ucdavis.edu where you can then upload input and output files. This will make it easier to discuss these issues.

[hqzhou]

Sorry to everyone. I rechecked the geometry of my system, and realized that
the structure of my supercell may be wrong. I used elk's spacegroup to create
the supercell, but now I just couldn't reproduce the wrong structure with the
same input file. It's really funny.

Anyway, I'll try again and report the result. Before that, I still have one question
to ask, even though a structure is wrong, shouldn't a series of ecut tests lead
to convergent result?

zhq

[fgygi]

Yes, running with increasing ecut should show convergence. However, each run must show reasonable convergence to a ground state energy if you want to make meaningful comparisons.

[hqzhou]

Hi, Francois,

This time the results are a little bit too good and unbelievable, I never saw such a match with other code.
For each ecut, I set ecutprec to ecut/3. My system involves Mg, Si and O. When I calculated it using
PWscf, I must set kinetic energy cut to 40 Ry for 1 mRy of energy difference even though I used USPP.
This result tells me I can use even shorter ecut (<30 Ry) for norm-conserving PP, is it correct?

Huiqun

ec30: <etotal> 110958.62382151 </etotal>
ec40: <etotal> 110958.62382151 </etotal>
ec50: <etotal> 110958.62382151 </etotal>
ec60: <etotal> 110958.62382151 </etotal>
ec70: <etotal> 110958.62382151 </etotal>
ec80: <etotal> 110958.62382151 </etotal>

[fgygi]

These energies look like they are not converged. It is impossible to help you debug your input if you don't post it.