[Pw_forum] ph.x cannot reach convergence within 100 steps

GAO Zhe flux_ray12 at 163.com
Sun Jul 17 09:33:56 CEST 2011


Thank you very much, Prof. Isaev.
I tried all of 'scf', 'relax' and 'vc-relax' before ph.x, but got the same problem in each case~
I was not sure whether this problem caused by empty-band number, so I increase nbnd. It may too large for my model~ Now, I am trying to change the values of alpha_mix(1) and nmix_ph for a better convergent.
Thank you, again ^_^

--
GAO Zhe
CMC Lab, MSE, SNU, Seoul, S.Korea


At 2011-07-16 16:39:40,"Eyvaz Isaev" <eyvaz_isaev at yahoo.com> wrote:

Hi,



>Recently, I am trying to calculate phonon dispersion via ph.x. However, in some cases, ph.x cannot reach convergence within 100 steps and stop >automaticly.
 
1. You should  perform 'scf', not 'vc-relax' before phonon calculations.
2. You have a rather large system (nbnd=100), but used alpha_mix(1) by default (0.7). Try it around 0.1.
3. Try changing nmix_ph (by default 4).
4. If this does not help change 'maxter=100' in phcom.f90 and recompile.

Bests,
Eyvaz.


-------------------------------------------------------------------
Prof. Eyvaz Isaev,
Department of Physics, Chemistry, and Biology (IFM), Linkoping University, Sweden
Theoretical Physics Department, Moscow State Institute of Steel & Alloys, Russia,
isaev at ifm.liu.se,eyvaz_isaev at yahoo.com

From: GAO Zhe <flux_ray12 at 163.com>
To: PWSCF Forum <pw_forum at pwscf.org>
Sent: Saturday, July 16, 2011 7:40 AM
Subject: [Pw_forum] ph.x cannot reach convergence within 100 steps


Dear QE developer and users:
Recently, I am trying to calculate phonon dispersion via ph.x. However, in some cases, ph.x cannot reach convergence within 100 steps and stop automaticly.
I have tested the convergent lattice parameter, cut-off energy and k-points (M-P grid), and the thresholds of force and stress were also very low. Even if I increase the tr2_ph from 1.0d-12 (default) to 1.0d-10, the |ddv_scf|^2  values was around 10^(-6) or 10^(-7) magnitude. The problem usually occurred at the second q-point, but sometimes, even if lucky enough, it would definitely occur at the third q-point (4 q-points in 3x3x3 case).
Are there any methods to solve this problem? Any suggestion will be welcome.
Calculations were used GGA-PBE USPP.
This is my input file:
cat >W25.relax.in <<EOF
&control
     calculation = 'vc-relax' ,
     disk_io = 'low' ,
     prefix = 'W' ,
     outdir = '$TempDIR/' ,
     pseudo_dir = '$PseudoDIR/' ,
     etot_conv_thr = 1.0d-12 ,
     forc_conv_thr = 1.0d-4 ,
/
&system
     ibrav = 1 ,
     celldm(1) = ******* ,
     nat = ** ,
     ntyp = 3 ,
     nbnd = 100 ,
     ecutwfc = 47.5 ,
     ecutrho = 450 ,
     occupations = 'smearing' ,
     smearing = 'mp' ,
     degauss = 0.003 ,
/
&electrons
     conv_thr = 1.0d-12 ,
     startingpot = 'atomic' ,
     startingwfc = 'atomic+random' ,
     diagonalization = 'cg' ,
     electron_maxstep = 150 ,
/
&ions
     ion_dynamics = 'bfgs' ,
/
&cell
     cell_dynamics = 'bfgs' ,
     press_conv_thr = 0.01 ,
/
ATOMIC_SPECIES
  V   50.9415   V.pbe-n-van.UPF
  W   183.800   W.pbe-nsp-van.UPF
  C   12.0110   C.pbe-van_ak.UPF
ATOMIC_POSITIONS crystal
  V   0.50  0.50  0.00
...........
  C   0.50  0.50  0.50
K_POINTS automatic
13  13  13   0  0  0
EOF
echo -e "  Relaxing Structure...\c"
$MPIBIN/mpirun -n 6 pw.x -npool 2 <W.relax.in >W.relax.out
echo -e " DONE"

cat >W.ph.in <<EOF
Phonon Calculation
&inputph
     outdir = '$TempDIR/' ,
     prefix = 'W' ,
     ldisp = .true. ,
     nq1 = 3 ,
     nq2 = 3 ,
     nq3 = 3 ,
     tr2_ph = 1.0d-10 ,
     fildyn = 'W.dyn' ,
/
EOF
echo -e "  Calculating Dynamic Matrix...\c"
$MPIBIN/mpirun -n 6 ph.x -npool 2 <W.ph.in >W.ph.out
echo -e " DONE"



--
GAO Zhe
CMC Lab, MSE, SNU, Seoul, S.Korea





_______________________________________________
Pw_forum mailing list
Pw_forum at pwscf.org
http://www.democritos.it/mailman/listinfo/pw_forum

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.democritos.it/pipermail/pw_forum/attachments/20110717/c0f015c0/attachment.htm 


More information about the Pw_forum mailing list