[CPMD-list] unknown error message

[Ari P Seitsonen]

Dear all,

  Adding some comments to Axel's and Abraham's helping remarks on a recent
topic.

On Fri, 18 Feb 2005, Axel Kohlmeyer wrote:

> >>> "AA" == Abraham Alaka <chokosabe at yahoo.co.uk> writes:
> 
> AA> Your CUTOFF value might be too small (~70 might be
> AA> better) and your CONVERGENCE ORBITALS a bit too large
> AA> (1.0D-7 might do).
> 
> in fact, for an oxide and your choice of pseudopotentials
> you may need an even higher cutoff (~120ry) for reasonably
> converged forces.

Yes, this depends on the pseudo potentials one uses. 70/80 Ry is probably
fine, if one has a Troullier-Martins pseudo for oxygen with LMAX=D and the
cut-off radii are large enough - unless the pseudo for Ti requires even a
higher cut-off energy (please see below).

> well, lowering CONVERGENCE ORBITALS will not help,
> since it does not converge at all.

If one increases the cut-off energy, the calculation might converge
better.

> AA> The problem you are referring to might be fixed in
> AA> v3.9. Unless you are doing a band structure
> AA> calculation, you may find it easier not to use kpoints
> AA> and just do a gamma point calculation instead (this is
> AA> default and you may need to include more atoms in your
> AA> simulation cell).
> 
> i agree. you should do some tests with gamma point 
> only first (e.g. plane wave cutoff).

Yes, usually CPMD converges much better using only the Gamma point,
however you probably shouldn't rely on the results obtained in any
physical or chemical sense, if the 

> AA> Another thing is that you are using an nlcc
> AA> pseudopotential; this is normally most useful when
> AA> doing spin polarised calculations (i.e you include LSD
> AA> in &CPMD and MULTIPLICITY in &SYSTEM).
> 
> to the best of my knowledge the logic is the other way
> around. for many spin polarized calculations you need
> pseudopotentials with NLCC. the reason being, that without
> NLCC you don't have much electron density close to the 
> atom core, especially if you don't use a semi-core pseudopotential.

Eeee, I think that you both agree on this, no? ;)

> AA>  --- Jin Zhao <zhaojin4576 at gmail.com> wrote: 
> >> &CPMD
> >> OPTIMIZE WAVEFUNCTION
> >> CONVERGENCE ORBITALS
> >> 1.0e-5
> >> LANCZOS DIAGONALISATION  
> 
> for the gamma point / cutoff tests you could try:
> 
>   ODIIS NO_RESET=30
>    5
>   LANCZOS DIAGONALIZATION RESET=2
>   CONVERGENCE ORBITALS
>     1.0e-6
>   PRINT FORCES ON

Even if this is an insulating material I would add some mixing of the
charge density, sometimes it's necessary to achieve convergence even in
semi-conductors and insulators.

> >> &DFT
> >> FUNCTIONAL BLYP
> 
> adding
> 
> GC-CUTOFF
>    1.0d-6
> 
> here should improve convergence. the BLYP gradient correction
> diverges for low electron density, so it is better not 
> calculate it there. adjusting this parameter is always a
> tradeoff between stability and accuracy of the calculation.
> so as a rule of the thumb it should be as low as possible
> but high enough to avoid convergence problems. it also
> depends a bit on the plane wave cutoff.

Just to be accurate here: the BLYP functional probably doesn't diverge,
however due to numerical incompleteness it might do so within a computer
program, especially if the density and its gradients are represented on a
grid. Again here a higher cut-off energy might help, i.e. the GC-CUTOFF
can be kept at a more strict value then.

> >> &SYSTEM
> >> ANGSTROM
> >> CELL VECTORS
> >> 5.916 0.0000  0.0000
> >> 0.000 6.4970  0.0000
> >> 0.000 0.0000  18.7144
> 
> better use:
> SYMMETRY
>   8
> CELL ABSOLUTE
> 5.916 6.4970 18.7144 0.0 0.0 0.0
> 
> this allows CPMD to take advantage 
> of the cell symmetry.

Having symmetries is very useful when employing k points, indeed!

> >> KPOINTS MONKHORST-PACK
> >> 2  2  1
> 
> with the above change in the cell symmetry,
> this setting will result in a single k-point,
> so not much improvement over gamma point.

Actually this is not true, since that k points would be 1/4,1/4,0, right?
That would correspond to a density of four k points in the whole 1st
Brillouin zone instead of one as in the case of having only Gamma point.

> >> &ATOMS
> >> 
> >> *Ti_MT_BLYP_NLCC.psp KLEINMAN-BYLANDER nonlinear
> >> core correction 
> >> LMAX=D
> 
> hmmm, most d-metals have ghost states unless
> you use LOCAL=S here. i don't know whether titanium
> is an exception. please check the available literature.
> there should be some plane wave pseudopotential DFT 
> calculations of ti compounds published...

This depends strongly on the pseudo potential used. I have had problems
getting good results for TiO2 with Troullier-Martins pseudo potentials
unless I included the semi-core states into the valence. Well using a
Goedecker pseudo would be even better, however this would increase the
cut-off energy required even further.

    Greetings,

       apsi

-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-
 Ari Paavo Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
 Tel +41 1 635 44 97 / Fax +41 1 635 68 38 / GSM +41 79 719 09 35      
 Anschrift: Physikalisch Chemisches Institut (PCI), Universität Zürich (UniZh)
 Indirizzo: Winterthurerstrasse 190, CH-8057 Zürich
 Address:   Schweiz / Svizzera / Suisse / Svizra / Switzerland