[Pw_forum] Some questions on Electron Phonon Calculations.

Vikas Varshney vv0210 at gmail.com
Thu Feb 2 20:48:53 CET 2012


Dear Users,
I am willing to run electron-phonon calculations for my Ni system of
interest and have few queries to clarify and questions as well. I will
highly appreciate your answers towards better understanding of the el-ph
calculations.

Based on Al example, I have gathered that after scf-relaxation (for dense k
points), the el-ph calculations is performed this way.
1. Estimate # of q points (based on provided nqs)
2. For each q point, calculate the dynamical matrix
     2a) First estimate # of irreducible representations (which are the
branches in phonon dispersion curve) for that q
     2b) For each irreducible representation, run the convergence test
3. Once 2a) and 2b) are done, diagonalize the dynamical matrix.
4. Calculate the frequencies (n) for that q.
5. Calculate lambda (q,n) and gamma (q,n) with 10 different gaussian
broadening (0.005 to 0.05)
6. Proceed to step 2 for next q.

Please guide if my understanding is wrong. The queries I have are follows

*A. What is gamma (q,n) in point 5 which shows in elph output file as
follows. *
  lambda( 2)=  0.0841   gamma=   72.85 GHz
  lambda( 3)=  0.0306   gamma=   97.74 GHz
**
*B. The lambda and gamma are calculated for 10 different gaussian
broadening values. This value is irrespective of what is used in scf.fit
calculations prior to elph calculations. How should we interpret these 10
values? How should one decide which gaussian broadening is appropriate? *

Now coming to my problem of interest. I am planning to calculate
cummulative lambda for single layer of Ni atoms. So, I start by relaxing a
bulk Ni crystal, cut out a fcc plane from (relaxed crystal) that with 4
atoms and create a supercell (with empty vaccum space). Now, I run scf
calculation with dense grid (24,24,8) followed by second scf run with
(8,8,2) grid. Then I run my elph-calculation. Here is the input file for
elph.

 &INPUTPH
                      outdir = './' ,
                      prefix = '1slab' ,
                      fildyn = '1slab.dyn' ,
                    fildvscf = '1slab' ,
                       ldisp = .true.,
                       tr2_ph=1.0d-10,
                         nq1 = 8 ,
                         nq2 = 8 ,
                         nq3 = 2 ,
                        alpha_mix(1) = 0.4 ,
                       epsil = .false.,
                        elph = .true.,
                        fpol = .false. ,
                     recover = .false. ,
                    amass(1) = 58.69340 ,
                  iverbosity = 1 ,
 /
*I have two more questions for this particular system. *
*1. For q=(0,0,0) convergence is achieved successfully. However, I see 2
negative frequencies. Similarly for q=(0,0,-0.12), I see one neg frequency.
Does that mean that*
*   a) My atoms are not fully relaxed. *
*   b) My SCF grid is not dense enough?*
*   c) erho, ecut values are not big enough? I used erhowfc of 55 and
erhocut of 550?*
**
*Please guide.*
**
*2. For the next q point, I do not get convergence acheived in 100
(default) interations for certain polarization. This is what I get at the
end*
**
      iter #  89 total cpu time :  8505.8 secs   av.it.:  34.3
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.439E+23
      iter #  90 total cpu time :  8648.0 secs   av.it.:  36.0
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.592E+23
      iter #  91 total cpu time :  8790.0 secs   av.it.:  35.9
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.195E+25
      iter #  92 total cpu time :  8931.3 secs   av.it.:  35.7
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.199E+25
      iter #  93 total cpu time :  9079.5 secs   av.it.:  37.7
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.197E+25
      iter #  94 total cpu time :  9230.5 secs   av.it.:  38.4
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.295E+27
      iter #  95 total cpu time :  9382.6 secs   av.it.:  38.7
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.426E+26
      iter #  96 total cpu time :  9538.3 secs   av.it.:  39.8
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.130E+27
      iter #  97 total cpu time :  9697.1 secs   av.it.:  40.4
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.182E+29
      iter #  98 total cpu time :  9855.5 secs   av.it.:  40.4
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.216E+28
      iter #  99 total cpu time : 10019.0 secs   av.it.:  41.8
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.732E+28
      iter # 100 total cpu time : 10183.3 secs   av.it.:  42.0
      thresh= 0.100E-01 alpha_mix =  0.400 |ddv_scf|^2 =  0.123E+31
     End of self-consistent calculation
     No convergence has been achieved

*It can be seen that   |ddv_scf|^2 is exponentially increasing instead of
decreasing. What could be the possible reason behind it? *

Any help will be highly appreaciated.
Thanks in advance and best Regards,
Vikas

Computational Reasearcher,
Wright Patterson Air Force Base, Dayton OH.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20120202/ed91a999/attachment.html>


More information about the users mailing list