[CPMD-list] WAVEFUNCTION OPTIMIZATION

Juerg Hutter hutter at pci.unizh.ch
Mon Oct 7 21:40:20 CEST 2002


Hi

I would like to give some general comments to
wavefunction optimizations.

- any optimization that takes more than 100 steps
  should be considered slow
- optimizations using ODIIS that have repeated resets
  (more than one) will probably never converge
- convergence for LSD is normaly slower and more
  difficult than for unpolarized cases
- if the ODIIS converger gets stuck (more than one reset)
  stop the run and restart using
  PCG MINIMIZE
  TIMESTEP
   20
  The conjugate gradient minimizer with line search is
  much more robust. For LSD it should be used from the
  start.
  A typical behavior will be that after the restart the
  energy goes down and the gradient increases. This means
  that we are in a region where there are negative
  curvatures. In such regions the DIIS minimizer moves
  in the wrong direction.
  After some iterations we will be back to normal behavior,
  energy and gradient get smaller. At this point it may be
  save to switch back to ODIIS.
- Wavefunction optimizations for geometries that are far
  from equilibrium are often difficult. If you are not
  really interested in this geometry (e.g. at the
  beginning of a geometry optimization or this is just the
  start of a MD) you can relax the convergence criteria
  to 10**-3 or 10**-4 and do some geometry steps.
  After that optimization will be easier.

Now for USPP:

The implementation of uspp's in CPMD has two major problems.
(ok more than two, but these are relevant here)
1) augmentation charges are calculated in G-space
2) there is no efficient way of increasing the density
   cutoff

If you use the default settings, you will have
small unphysical modulations of the density. These
lead to "strange" effects especially with gradient
corrected functionals causing the optimization
to fail.
What can you do in the current version to avoid this.
- try to set in the &DFT section the value of
  GC-CUTOFF
   1.D-6     or 1.D-5
- increase the cutoff and/or the number of mesh points
  MESH keyword in &SYSTEM
  use this with care, it is expensive!

I hope these tips help to save some iteration cycles.

Juerg



----------------------------------------------------------
Juerg Hutter                   Phone : ++41 1 635 4491
Physical Chemistry Institute   FAX   : ++41 1 635 6838
University of Zurich           E-mail: hutter at pci.unizh.ch
Winterthurerstrasse 190
CH-8057 Zurich, Switzerland
----------------------------------------------------------


On Mon, 7 Oct 2002, Veronique Van Speybroeck wrote:

>
> Dear all,
>
> I am studying a system that contains 32 atoms, among which are O, N, P, H,
> and use therefore Vanderbilt pseudopotentials.  I'm doing now the initial
> optimization of the wavefunction.  How many steps are common for the
> initial wavefunction optimization.  I have already done 1500 steps still
> the convergence is not reached.  Is this normal.  The total energy does
> not systematically decrease.  As an illustration I give the TOTAL ENERGY
> quantity for some of the last steps :
> (K+E1+L+N+X)           TOTAL ENERGY =         -137.29507284 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29556795 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29535927 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29511531 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29560496 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29543130 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29523925 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29551439 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29553778 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29524706 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29508326 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29423680 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29556416 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29524258 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29297862 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29506422 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29504717 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29551214 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29504128 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29497465 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29156757 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29534837 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29456003 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29543937 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29530715 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29538352 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29561316 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29531235 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29488259 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29496710 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29371573 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29536463 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29503137 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29426423 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29537266 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29501795 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29543533 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29546353 A.U.
>  (K+E1+L+N+X)           TOTAL ENERGY =         -137.29495343 A.U.
> Does this seem reasonable or is there something wrong?
>
> Thanks to all in advance
>
> Veronique
>
> -----------------------------------------------------------------------
> Dr. ir. Van Speybroeck Veronique
> Laboratorium voor Theoretische Fysica
> Universiteit Gent
> Proeftuinstraat 86
> 9000 Gent
> Tel +32-9-264.65.58  GSM : +32/474/259767
> Fax +32-9-264.65.60
> email : veronique.vanspeybroeck at rug.ac.be
> http://inwfaxp2.rug.ac.be/~web/onderzoek/proj_3.html
> -----------------------------------------------------------------------
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://www.cpmd.org/mailman/listinfo/cpmd-list
>




More information about the CPMD-list mailing list