[molpro-user] analytic gradients not used during geometry optimization when proc specified

[Werner Gyorffy]

Dear Benj,

By using numerical energy gradients, it is possible to perform a 
geometry optimization with another method or basis set used for the 
Hessian calculation. The keyword NUMERICAL must be added to the OPTG card.

Alternatively, you can compute the Hessian first, store it in a record, 
then use HESSREC=irecord.ifile to read the initial Hessian from the 
given record in the geometry optimization. This may be a good strategy 
in those cases when it is enough to compute the Hessian only once during 
the geometry optimizations (or, due to the flexible input language of 
Molpro, one can ``manually'' recompute the Hessian after each nth 
iteration step). This way, analytical energy derivatives can be also 
used in the geometry optimization if they are available.

The case you reported may be viewed as a limitation of the program 
rather than a bug. We however should certainly consider developments to 
make the program more consistent on this issue. Thank you for reporting 
the case.

Regards,

Werner.


On 05/22/2013 07:58 PM, Benj FitzPatrick wrote:
> Werner,
> Thank you for this suggestion. I should have mentioned that this was one
> of the permutations I tried. The problem is that after the first
> iteration optg defaults to the basis set defined in hessproc. This is
> most easily seen in the molpro.log file which shows the # of external
> orbitals for avtz as 33  21  21  10, and avdz is 17   9   9   4. This
> problem is what led me to try using proc and hessproc.
> Thanks,
> Benj
>
>
> On Wed, May 22, 2013 at 8:34 AM, Werner Gyorffy
> <gyorffy at theochem.uni-stuttgart.de
> <mailto:gyorffy at theochem.uni-stuttgart.de>> wrote:
>
>     Dear Benj,
>
>     This is a slightly modified version of your input which does the job
>     you wanted:
>
>
>     ***,h2o
>     memory,150,M
>     gthresh,oneint=1.d-14,twoint=__1.d-14,zero=1.d-14
>
>     oc2    =     1.400000
>
>     angstrom
>     symmetry,auto
>     geom={
>       c;
>       o,   1, oc2;
>     }
>     basis=avtz;
>
>     hf;
>     ccsd;
>     {optg,energy=1.d-5,gradient=1.__d-5,numhess=5,hessproc=__b3lyphess}
>
>     proc b3lyphess
>     basis=avdz; ! for example
>     hf;
>     ks,b3lyp;
>     endproc
>
>     ---
>
>     Regards,
>
>     Werner Gyorffy.
>
>
>     On 05/21/2013 11:38 PM, Benj FitzPatrick wrote:
>
>         Hello all,
>         I am trying to perform geometry optimizations using a hessian
>         procedure
>         with a smaller basis set, which then requires I use another
>         procedure
>         for the energy and gradient calculations (otherwise the hessian
>         basis
>         set becomes the default). The first point uses the analytic CCSD
>         gradients, but after the hessian calculation numerical CCSD
>         gradients
>         are used. Have I specified one of my procedures incorrectly or is
>         something more nefarious going on?
>         Thanks,
>         Benj FitzPatrick
>
>         ***,h2o
>         memory,150,M
>         gthresh,oneint=1.d-14,twoint=__1.d-14,zero=1.d-14
>
>         oc2    =     1.400000
>
>         angstrom
>         symmetry,auto
>         geom={
>            c;
>            o,   1, oc2;
>         }
>         basis=avtz;
>         {optg,root=1,energy=1.d-5,__gradient=1.d-5,procedure=__ccsdavtz,numhess=5,hessproc=__b3lyphess}
>
>         proc ccsdavtz
>         hf;
>         ccsd;
>         endproc
>
>         proc b3lyphess
>         hf;
>         ks,b3lyp;
>         endproc
>
>
>
>         _________________________________________________
>         Molpro-user mailing list
>         Molpro-user at molpro.net <mailto:Molpro-user at molpro.net>
>         http://www.molpro.net/mailman/__listinfo/molpro-user
>         <http://www.molpro.net/mailman/listinfo/molpro-user>
>
>     _________________________________________________
>     Molpro-user mailing list
>     Molpro-user at molpro.net <mailto:Molpro-user at molpro.net>
>     http://www.molpro.net/mailman/__listinfo/molpro-user
>     <http://www.molpro.net/mailman/listinfo/molpro-user>
>
>