The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi,

- 20 angströms of vacuum seems really gigantic. It may be usefull to perform a convergence study on this parameter.
- Your k-point mesh is apparently not optimized (is it ?) for such a large cell along z axis. You may want  to use prtkpt in order to get some more sensible results.
- Do not change your ecut if it is the "converged" value
-  Since you have some forces in your cell you can use toldff instead of toldfe. 
-  You don't need to go down to tolmxf 5e-6 for your atomic position to be converged
-  you may reduce your number of scf steps by using iscf 17
- you will decrease the preconditionning time by using "diecut 0.5".

regards

PMA

On Fri, Apr 11, 2008 at 12:56 AM, <chona00@gmail.com> wrote:

Hi,

  I'm trying to perform PAW calculations on the rutile TiO2 110 surface using
a slab consisting of a 2x1 surface unit cell (along [001] and [-110]) four
layers thick.  I chose my vacuum layer as 20 angstroms in keeping with other
calculations in the literature.

  My question has to do with the amount of time it takes to calculate the
energy of my Ti16O32 slab. Using the above parameters takes 31000+ seconds of
wall time on 4 CPUs.  My concern is with how long the slab optimization will
take given this fact.

1) Is there a way I can cut down on the run time besides reducing ecut?

2) I've considered using more processors since my calculations use 16 k-points,
but I don't want to tie up our cluster (which is on a GigE switch fabric.)  Do
you think Abinip will scale  well given the collective MPI calls and our
cluster's slow interconnect?

For brevity's sake, here is my input file without xred:

# Use PAW
ecut 20.0
pawecutdg 40.0

# Definition of occupation numbers
occopt 3
tsmear 0.04 eV

# Use PBE exchange-correlation
ixc 11

# SCF convergence
iscf 13
iprcel 45
nstep 500
toldfe  1.0d-10
tolmxf 5.d-6    # Will stop when, twice in a row, the difference
               # between two consecutive evaluations of total force
               # differ by less than toldfe (in Hartree)

# Definition of the k-point grids
kptopt 1
prtkpt 0
ngkpt 4 4 2
nshiftk 1
shiftk  0.5 0.5 0.5

# Definition of the unit cell
acell  5.92166 6.57158 33.1432 angstrom
vacuum 0 0 1
rprim
 1.0 0.0 0.0
 0.0 1.0 0.0
 0.0 0.0 1.0
ntypat 2
znucl 22 8
typat
 2 2 1 1 1 2 1 1 2 2 2 1 2 2 1 2 2 2 1 2 2 2 2 2
 2 2 1 1 1 2 1 1 2 2 2 1 2 2 1 2 2 2 1 2 2 2 2 2
natom 48

I used the PAW atomic data for Ti and O from the Abinit webpage.  My choice of
ecut and pawecutdg were based on convergence studies of the energy of the bulk
rutile crystal.  These   also gave unit cell parameters within 1% of the
experimental data.

Thanks,
Chona

-- 
Pierre-Matthieu Anglade