[Pw_forum] why does band calculation takes too long

Eduardo Menendez eariel99 at gmail.com
Mon Apr 4 23:00:41 CEST 2016


Hi,
I wonder why is the band calculation much longer than a scf calcuation.
I did a calculation for a 64-atoms supercell using 20 cpu cores.
In summary, I did
1) SCF calculation with only gamma point, run in two minutes.
2) Non-self-consistent band calculation with 27 k-points.

In the nscf calculation each k-point takes 15-20 minutes and the total
calculation lasts 7 hours. I understand that for the k-points along the
line the wfc are complex, the symmetry is smaller, but even so the time
scale is quite different.
The machine has to keep all the wfc in memory, but it was not swapping.
Even the gamma point took 15 minutes, why does it takes longer than the scf
calculation?

Please, let me know if  I am doing something wrong. Below are the inputs
for the scf and the bnd calculations.

SCF:

&CONTROL

calculation = 'scf'

restart_mode='from_scratch',

pseudo_dir = "./",

outdir="./tmp/",

tstress = .true.,

tprnfor = .true. ,

prefix= 'cdte64',

disk_io = 'low',

etot_conv_thr = 1.0D-3,

wf_collect=.TRUE.,

verbosity='high'

/

&SYSTEM

ibrav= 1,

celldm(1) = 25.02,

nat = 64,

ntyp = 2,

ecutwfc = 28.67,

ecutrho = 180,

degauss = 0.001D0,

occupations = "smearing",

smearing = "gauss",

tot_charge = 0,

nspin = 1 , ! set to 2 for spin polarized

! tot_magnetization = 0 , ! change to 1

/

&ELECTRONS

diagonalization='david'

conv_thr = 1.D-6,

mixing_beta = 0.4D0,

diago_david_ndim= 2,

/

&IONS

pot_extrapolation='second_order',

trust_radius_max = 0.7,

trust_radius_ini = 0.7,

/

ATOMIC_SPECIES

Cd 112.411 cd_pbe_v1.uspp.F.UPF

Te 127.600 te_pbe_v1.uspp.F.UPF


ATOMIC_POSITIONS (alat)

Cd 0.5000000000000000 0.5000000000000000 0.5000000000000000

Cd 0.0000000000000000 0.0000000000000000 0.0000000000000000

Cd 0.2500000000000000 0.2500000000000000 0.0000000000000000

….


K_POINTS gamma


The NSCF band calculation input is:

&CONTROL

calculation = 'bands'

! restart_mode='from_scratch',

pseudo_dir = "./",

outdir="./tmp/",

tstress = .true.,

tprnfor = .true. ,

prefix= 'cdte64',

disk_io = 'low',

etot_conv_thr = 1.0D-3,

verbosity='high'

/

&SYSTEM

ibrav= 1,

celldm(1) = 25.02,

nat = 64,

ntyp = 2,

ecutwfc = 28.67,

ecutrho = 180,

degauss = 0.001D0,

occupations = "smearing",

smearing = "gauss",

tot_charge = 0,

nspin = 1 , ! set to 2 for spin polarized

! tot_magnetization = 0 , ! change to 1

/

&ELECTRONS

diagonalization='david'

conv_thr = 1.D-6,

mixing_beta = 0.4D0,

diago_david_ndim= 2,

/

&IONS

pot_extrapolation='second_order',

trust_radius_max = 0.7,

trust_radius_ini = 0.7,

/

ATOMIC_SPECIES

Cd 112.411 cd_pbe_v1.uspp.F.UPF

Te 127.600 te_pbe_v1.uspp.F.UPF


ATOMIC_POSITIONS (alat)

Cd 0.5000000000000000 0.5000000000000000 0.5000000000000000

Cd 0.0000000000000000 0.0000000000000000 0.0000000000000000

Cd 0.2500000000000000 0.2500000000000000 0.0000000000000000

….

K_POINTS crystal_b

3

0.500 0.500 0.500 17

0.000 0.000 0.000 10

0.500 0.000 0.000 1


Thank you.

Eduardo Menendez Proupin
Departamento de Fisica, Facultad de Ciencias, Universidad de Chile
URL: http://www.gnm.cl/emenendez
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20160404/b4fab8b1/attachment.html>


More information about the users mailing list