[CPMD-list] ODIIS with k points

[Juerg Hutter]

Hi

when asking for eigenvalues the orbitals are rotated into
canonical form. This causes changes in the route for the
DIIS optimizer. If the convergence criteria is not very strict
this may then lead to slightly different final energies.
You can easily test this by setting the convergence
criteria 1 or 2 orders of magnitude smaller.

regards

Juerg Hutter

----------------------------------------------------------
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 Wed, 4 Aug 2004, Eung-Gun Kim wrote:

> Dear List Subscribers,
>
> I am running density optimization on a hydrocarbon crystal using v3.9.1 on IBM
> SP4 (p655).  Following Axel's warning on a combined use of ODIIS and k points, I
> ran a couple of runs to make sure results are reasonable in comparison with
> those from a Lanczos calculation.  What puzzles me is that when I "PRINT ON
> FORCES EIGENVALUES," energies change from those without these key
> words---although not significant.  LANCZOS DIAGO seems very robust and stable,
> but the calculation takes 10 times longer than ODIIS, at least in this
> particular case.
>
> Another question is that ELECTROSTATIC ENERGY from LANCZOS is very different
> than that from ODIIS.  The fact that TOTAL ENERGY is very similar in both cases
> only suggests that this term may refer to a different quantity, depending on
> which optimization method you use.
>
> Regards,
>
> EG Kim
>
>
> ODIIS with PRINT ON FORCES EIGENVALUES -----------------------------------------
>
>  (K+E1+L+N+X)           TOTAL ENERGY =         -122.64331840 A.U.
>  (K)                  KINETIC ENERGY =           14.43425982 A.U.
>  (E1=A-S+R)     ELECTROSTATIC ENERGY =         -125.12750514 A.U.
>  (S)                           ESELF =          134.04460621 A.U.
>  (R)                             ESR =            3.42570289 A.U.
>  (L)    LOCAL PSEUDOPOTENTIAL ENERGY =          -65.96540035 A.U.
>  (N)      N-L PSEUDOPOTENTIAL ENERGY =           12.56202777 A.U.
>  (X)     EXCHANGE-CORRELATION ENERGY =          -40.79762548 A.U.
>           GRADIENT CORRECTION ENERGY =           -1.86569866 A.U.
>
>
> ODIIS without PRINT ON FORCES EIGENVALUES --------------------------------------
>
>  (K+E1+L+N+X)           TOTAL ENERGY =         -122.64331838 A.U.
>  (K)                  KINETIC ENERGY =           14.43425982 A.U.
>  (E1=A-S+R)     ELECTROSTATIC ENERGY =         -125.12751216 A.U.
>  (S)                           ESELF =          134.04460621 A.U.
>  (R)                             ESR =            3.42570289 A.U.
>  (L)    LOCAL PSEUDOPOTENTIAL ENERGY =          -65.96541815 A.U.
>  (N)      N-L PSEUDOPOTENTIAL ENERGY =           12.56201664 A.U.
>  (X)     EXCHANGE-CORRELATION ENERGY =          -40.79762829 A.U.
>           GRADIENT CORRECTION ENERGY =           -1.86569863 A.U.
>
> LANCZOS DIAGO ------------------------------------------------------------------
>
>  (B+E2+X-V)             TOTAL ENERGY =         -122.64331846 A.U.
>  (B)                     BAND ENERGY =          -21.22863427 A.U.
>  (E2=I-H-S+R)   ELECTROSTATIC ENERGY =         -113.55043258 A.U.
>  (S)                           ESELF =          134.04460621 A.U.
>  (R)                             ESR =            3.42570289 A.U.
>  (X)     EXCHANGE-CORRELATION ENERGY =          -40.79763615 A.U.
>  (V)     EXCHANGE-CORRELATION POTEN. =          -52.93338459 A.U.
>           GRADIENT CORRECTION ENERGY =           -1.86569859 A.U.
>
> --------------------------------------------------------------------------------
>
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://cpmd.org/mailman/listinfo/cpmd-list
>