[Pw_forum] vc-md question

[Miguel Martínez Canales]

Hello everybody,

I'm trying to relax a GeH4 structure at some 76GPa, for which I am doing a 
vc-md calculation. The problem is that the run always ends at 50 
iterations. If I use the last structure in the input and re-run it, it 
still doesn't end after 50 iterations. And that has happened four times.

This might point to a low dt. So, I have changed dt from 20 to 200 (50 
iterations seemed too much, so I thought it was an order-of-magnitude 
question). But then, the structure went crazy... which I expected if dt was 
way too large.

Just in case this is relevant, doing a scf run with the last structure will 
give me small forces and the stress tensor is nearly diag(a,a,a). The 
forces on the ions that should move more are relatively small, and a relax 
run would easily take care of them.

1) Is this normal? When I was using VASP, I could set a fixed volume and 
relax the structure, and it would converge pretty fast if the starting 
input wasn't too far off. In this case, the starting structure shouldn't be 
far off, as it has been modified from a VASP output at 72GPa (the change in 
volume from 72 to 76 GPa's is pretty small).

2) On another ground, why are the stresses shown in a scf run and in a 
vc-md run different? Aren't they calculated using the Feynman-Hellman 
theorem with the knwon wavefunctions? How are they calculated in each case?

3) Would a cp or fpmd run be better suited for my needs (good relax for a 
future phonon calculation)?

4) VASP featured fixing some atoms. As far as I know, espresso doesn't 
allow this (at least for pw.x runs). Am I totally wrong?

Side note: I'm running ESPRESSO v3.0. Under espresso 2.1.5, the Wentzkovitz 
lagrangian seemed to not evolve, while the Parrinelo-Raman lagrangian would 
not run unless nosym=.true. was stated.

The pw.x input is the following:

cat > geh4.scf.in << EOF
  &control
     calculation='vc-md'
     title='Germane at 75GPa',
     restart_mode='from_scratch',
     tstress = .true.
     tprnfor = .true.
     prefix='h',
     dt = 2.0d1,
     etot_conv_thr = 1.0d-5
     forc_conv_thr = 1.0d-4
     pseudo_dir = '$PSEUDO_DIR/',
     outdir='$TMP_DIR/'
  /
  &system
     ibrav=  7, celldm(1) =6.162, celldm(3)=1.28604, nat=  5, ntyp= 2,
     ecutwfc =50.0, ecutrho=400.0,
     occupations='smearing', smearing='methfessel-paxton', degauss=0.02
  /
  &electrons
     conv_thr =  1.0d-8
     mixing_beta = 0.7
  /
  &ions
     ion_dynamics='beeman'
  /
  &cell
     cell_dynamics='w'
     press =760.0D0
     cell_factor =2.0
  /
ATOMIC_SPECIES
  Ge  72.612  Ge.pw91-n-van.UPF
  H   1.00797  H.pw91-van_ak.UPF
ATOMIC_POSITIONS
Ge   0.00   0.00   0.00
H    0.50   0.00   0.00
H    0.00   0.50   0.00
H    0.00   0.00   0.4691
H    0.00   0.00  -0.4691
K_POINTS {automatic}
  10 10 10  1 1 1
EOF

Thanks in advance,

Miguel

PS: I'm running this on an Itanium 2 cluster, having compiled espresso with 
  Intel Fortran Compiler v8 and mkl v8, just in case it matters.


-- 
----------------------------------------
Miguel Martínez Canales
    Dto. Física de la Materia Condensada
    UPV/EHU
    Facultad de Ciencia y Tecnología
    Apdo. 644
    48080 Bilbao (Spain)
Fax:  +34 94 601 3500
Tlf:  +34 94 601 5437
----------------------------------------

"I don't think that Debian can really compete with
Gentoo. Sure it might be okay, but when it comes to
dependencies, you probably are still going to have
to get them all on your own. Or is there something
like portage in the Debian world as well?"

Annonymous