[Pw_forum] a convergency problem

Yi Wang wangwy13 at gmail.com
Wed Oct 19 05:32:04 CEST 2016


Dear everyone,

I have a convergency problem facing the pw.x after version 5.0.2

The example input is given in the last section.

The structure is a bcc-like supercell containing Fe and Cu atoms (relaxed)
(with GBRV uspp).
The structure can be run properly with smearing, the problem comes from
"tetrahedra" occupation.

If I use a non-offset k point mesh, as given in the input, the convergency
would get stuck around 1E-6~1E-8 Ry, never get better like most other
supercells of Fe and Cu.

The problem can be solved by offset the k mesh. However, my local mentors
do not suggest me to do so. Because we have lots of different supercells,
the offseted k mesh would give very different centering k point in
different supercells, it would appear to be trouble if I compare their
energies. (btw, what is your opinions on this?)

The problem can also be solved by increasing the size of k mesh, but still,
not cheap and absolute necessary for the calculation of lots supercells.

However, in version 5.0.2, there is no such problem, and convergency can be
achieved quickly (if I rerun [in post 5.0.2 versions] starting from the
converged charge density [by v 5.0.2], convergency still cannot be achieved
). That's why I'm asking for help here, before I turn to other solutions.

So far, I tried to change number of bands, mixing beta, and parallel
running parameter...all those appear to be irrelevant. Only k mesh would
change the result.

Your suggestions and help is appreciated.

Yi Wang
Nanjing university of Science and Technology

&control
   calculation = 'scf' ,
   prefix = 'pwscf' ,
   outdir = './temp-p_t/' ,
   pseudo_dir = '/home/sim/pot/pwscf/' ,
   restart_mode = from_scratch ,
   disk_io= 'none'
/
&system
   ibrav = 0 ,
   ecutwfc = 40 ,
   ecutrho = 400 ,
   occupations = 'tetrahedra' ,
   nspin=2,
   nbnd= 120,
   nat = 12 ,
   ntyp = 2 ,
   starting_magnetization( 1 )= 0.01,
   starting_magnetization( 2 )= 0.26,
   use_all_frac=.true.
/
&electrons
   conv_thr = 1d-9 ,
   diagonalization = 'david' ,
   mixing_mode = 'local-TF' ,
   startingpot = 'atomic' ,
   startingwfc = 'atomic+random'
   mixing_beta = 0.12 ,
   mixing_ndim =  16,
   diago_david_ndim=4
/
CELL_PARAMETERS bohr
5.689675841  0.000000000  0.000000000
0.000000000  -10.558955237  0.000000000
0.000000000  0.000000000  -15.996363568
ATOMIC_SPECIES
  Cu 63.546 cu_pbe_v1.2.uspp.F.UPF
  Fe 55.845 fe_pbe_v1.5.uspp.F.UPF
ATOMIC_POSITIONS crystal
Fe       1.000000000   0.498189104   0.666054422
Fe       0.500000000   0.251812190   0.500611021
Fe       1.000000000   0.003971761   0.664566351
Fe       0.500000000   0.746028153   0.502102015
Fe       1.000000000   0.500799770   0.333333330
Fe       0.500000000   0.249203627   0.833333320
Cu       1.000000000   0.998064252   0.333333150
Cu       0.500000000   0.751929936   0.833333540
Fe       0.500000000   0.251812380   0.166055619
Fe       0.500000000   0.746028003   0.164564485
Fe       1.000000000   0.498188874   0.000612278
Fe       1.000000000   0.003971951   0.002100469
K_POINTS automatic
   12 7 4 0 0 0
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20161019/34b17b54/attachment.html>


More information about the users mailing list