[CPMD-list] concerning slab calculations

[Axel Kohlmeyer]

On Tue, 15 Jun 2004, John Kofi Dogbe wrote:

JD> On Tue, 15 Jun 2004, Axel Kohlmeyer wrote:
JD> 
JD> > and write them out in what format?
JD> 
JD> I write them out in xyz. In fact I just found out that I don't have to go
JD> through this anymore. Maybe others already know about this. You can
JD> actually write out your coordinates in a CPMD format from
JD> MOLDEN. It writes out a CPMD.inp file with some 'default' keywords. One
JD> can edit this file and tailor it to the job at hand. Give it a try (if
JD> you haven't already).
JD> 
JD> > i don't think you should need such a large cutoff here.
JD> > if i am not toally mistaken 30.0 should be enough here.
JD> 
JD> Hmm ... I see. I got a cutoff of 30.0eV when MOLDEN created the cpmd input
JD> file.

the plane wave energy cutoff in cpmd is specified in rydberg,
30.0eV would be about 2.2 rydberg which is not very useful.

perhaps a few remarks are in place here. finding the best cutoff
for a calculation is not always easy. what works well for one system
may be far to low for another. you always have to make
a compromise between accuracy and speed (which also applies to
supercell size, number of k-points and so on).
generally the 'hardest' pseudopotential determines the cutoff.
to get a feeling for what kind of accuracy can be achieved,
i like to start with a series of single point calculations
with only a few atoms and increasing cutoff, PRINT FORCES ON
and CONVERGENCE ORBITALS set to 1.0e-6 or smaller,
and check how well the forces are converged with respect to
the cutoff (you can also look at the single point energy).

also you have to be aware of the fact that some of the available
pseudopotentials are optimized to be 'soft' at the expense of
transferability. 

JD> 
JD> >
JD> > also, this is not a slab but more like an isolated cluster,
JD> > there are at least 5 angstrom between one end of the si atoms
JD> > and the other end of the si atoms in the next image.
JD> 
JD> Yes, I needed to test the calculations on isolated clusters since I'm
JD> already familiar with what to look for and to know that I'm doing things
JD> right before moving on to slabs. Slabs are my ultimate goal.
JD> 
JD> > compiling on an SGI is a bit trick currently. to the best of my knowledge
JD> > you have to make sure, that wrener.F is compiled without optimizations.
JD> 
JD> Thank you very much for this bit. Will it just be enough to remove the
JD> optimization flags in the Makefile? In that case nothing gets optimized,
JD> but that is not what you are saying here, right?

true. the best way is to hack the makefile internals, but that is
a bit tricky. the fastest way is to compile cpmd with full optimization,
then delete wrener.o and type 'make wrener.o'. you then see the command
that is used to compile this file. you cut-n-paste the command
line and replace -O3 with -O0 and hit enter. now you type 'make'
again and then you should have a new and hopefully working cpmd
executable.

JD> 
JD> I did convert all the coordinates I got from MOLDEN to au and got a job
JD> completed. It converged well and the results compared well with results
JD> from other packages. The only thing I did different from that job was to
JD> remove the keyword ODIIS. I set this to 15 in the job that is still
JD> running.

that should not be the reason. ODIIS with 10 steps is the default,
IIRC. the problem is, that your wavefunction does not converge
when your atoms are too close.

axel.

JD> 
JD> Thanks,
JD> John
JD> --
JD> **The first is not necessarily the Leader**
JD> 
JD> 	"If I have spoken evil, bear witness of the evil:
JD> 		but if well, why smitest thou me?"
JD> 				-- Jesus Christ (John, 18:23)
JD> _______________________________________________
JD> CPMD-list mailing list
JD> CPMD-list at cpmd.org
JD> http://cpmd.org/mailman/listinfo/cpmd-list
JD> 
JD> 

-- 


=======================================================================
Dr. Axel Kohlmeyer                        e-mail: axel.kohlmeyer at rub.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================