Hello together!
I started using QE to calculate some total energies of molecular
crystals. It was said, that using rpb-pseudopotentials together with
input_dft = 'vdW-DF'
should lead to good results.
So I created the potentials using the PSlibrary-code. Some elements
didn't converge, but the ones, which generated a UPF-file should be fine
- aren't they?
To find the ECUTWFC it was said to be sufficient, to just calculate one
atom of the type you are interested in, so I took the following input:
&control
calculation = 'scf'
restart_mode='from_scratch',
prefix='',
tstress = .false.
tprnfor = .false.
pseudo_dir = '/gpfs/home/aasfu/espresso-4.3.1/pseudo/',
outdir='/gpfs/home/aasfu/tmp/'
verbosity = 'high'
/
&system
ibrav=0,
celldm(1)=2,
nat = 1,
ntyp = 1,
ecutwfc = 30,
ecutrho = 120,
input_dft = 'vdW-DF',
occupations='smearing',
smearing='gaussian',
degauss=0.01,
force_pairing = .true.
/
&electrons
conv_thr = 1.0d-8
electron_maxstep= 1000
/
CELL_PARAMETERS cubic
10.0 0.0 0.0
0 10.0 0
0.0 0.0 10.0
ATOMIC_SPECIES
O 1.0 O.rpb-rrkjus.UPF
ATOMIC_POSITIONS angstrom
O 0 0 0
K_POINTS automatic
1 1 1 0 0 0
for the elements of interest (H, C, N, O, P, Sn) with increasing
cutoff-value and different multiplicators to get ecutrho (4 times
ecutwfc and 10 times ecutwfc (ultrasoft?) ).
Some of them converge (e.g, C at 36 / 140 and 37 / 370 for rpb-rrkjus),
but others do oscillate quite large (O, Sn).
What is a good convergence criterium?
I found numbers like 1mRy but also 5mRy per atom
(http://users.aims.ac.za/~daniel/hands_on/Brandon_workshop_handout.pdf)!
Does this mean, the more atoms I have in my structure (will be ~200),
the bigger the allowed deviation (1Ry in my case)?
I always took the difference between n Ry and n+1 Ry in ecutwfc as a
criteria - is this right?
Or is it better to test convergence with small molecules, which contain
the atoms in the configuration of my molecule (sp2-carbons, carboxylate
functions etc.)?
Thanks a lot for clearification,
Guntram
