Dear QE Community,

I have compiled qe-6.0 with intel-14 and openmpi-1.6.5. In attempting to speed 
up the phonon calculation (as suggested in other posts), I have tried two 
different alpha_mix(1) values- the default value of 0.7 and 0.3. For Si, the 
two alpha_mix values give almost identical phonon frequencies for all q-points. 
For GaAs, I get different frequencies. I suspect it may be due to the LO-TO 
splitting at the zone center but the differences exist for other q-points as 
well particularly for the acoustic modes.

For example, for q=(0,0,0), I get the following frequencies in wavenumber in 
the .dyn files:
alpha_mix(1)=0.7 : 5.936331, 5.936331, 5.936331, 
249.858731,249.858731,249.858731
alpha_mix(1)=0.3 : -17.197390, -17.197390,-17.197390, 247.538697, 247.538697, 
247.538697

For q=(0.125,-0.125,0.125) I get:
alpha_mix(1)=0.7 : 31.681603, 31.681603, 62.388768, 246.723626, 246.723626, 
266.652186
alpha_mix(1)=0.3 : 34.440636, 34.440636, 62.925154, 247.034644, 247.034644, 
266.813369


For q=(-0.25,0.25,-0.25) I get:
alpha_mix(1)=0.7 : 49.997159, 49.997159, 120.235286, 242.448112, 242.448112, 
255.746902
alpha_mix(1)=0.3 : 52.457875, 52.457875, 121.042662, 242.645430, 242.645430, 
255.836026

The issue of negative phonon frequencies has been exhaustively discussed on the 
forum and is not the point of my post. I assume that  alpha_mix is not a 
convergence parameter and  should not change the frequencies but it does. I 
have also run on SnSe where I do not get negative frequencies at all for either 
alpha_mix values but the frequencies differ by 10-20 wavenumbers between the 
two sets. I should add that all self-consistent calculations for the individual 
modes reached convergence. Convergence tests were done for the cutoff energy 
and K-grid as well.

Here are my input files for the initial self-consistent run followed by the 
phonon calculation:

============================
 &control
    calculation     = 'scf'
    prefix          = 'gaas'
    restart_mode    = 'from_scratch'
    wf_collect      = .false.
    pseudo_dir      = './'
    outdir          = './'
    verbosity       = 'high'
    tprnfor         = .true.
    tstress         = .true.
 /
 &system
    ibrav=2,celldm(1)=10.838928,
    nat= 2, ntyp= 2 , nbnd=16
    ecutwfc = 70, ecutrho=280,
    occupations = 'fixed'
 /
 &electrons
    diagonalization = 'david'
    mixing_beta     = 0.7
    conv_thr        = 1.0d-10
 /
ATOMIC_SPECIES
Ga   69.723   Ga.pbe-hgh.UPF
As   74.92160 As.pbe-hgh.UPF
ATOMIC_POSITIONS  crystal
  Ga 0.0   0.0   0.0
  As -0.25  0.75  -0.25
K_POINTS   automatic
   12 12 12 0 0 0

==============================

GaAs
&inputph
  tr2_ph   =  1.0d-12
  prefix   = 'gaas'
  amass(1) = 69.723
  amass(2) = 74.92160
  ldisp = .true.
  epsil = .true.
  nq1=8, nq2=8, nq3=8
  outdir   = './'
  alpha_mix(1)=0.3
  fildyn   = 'gaas.dyn'
  fildvscf = 'dvscf'
 /
===============================

I would appreciate any input as to why alpha_mix alters the results for polar 
materials such as GaAs and SnSe.

Thank you,

Vahid

Vahid Askarpour
Department of Physics and Atmospheric Science
Dalhousie University,
Halifax, NS, Canada
_______________________________________________
Pw_forum mailing list
Pw_forum@pwscf.org
http://pwscf.org/mailman/listinfo/pw_forum

Reply via email to