[molpro-user] Optg and Numerical Hessian

[Tsuneo Hirano]

Thank you for the quick response.

As I guessed in the latter half of my previous communication, the 
check for the numerical Hessian for the multi-reference method or 
with symmetry constraint has been inserted to prevent the symmetry 
change during numerical evaluation of the Hessian.  I understand. 
However, this is too much.  We can write symmetry-adaptive z-metrix 
or xyz-type geometry definition which can maintain the symmetry 
against the geometrical change incurred in the numerical calculation 
steps of the Hessian.  What we need is the check or warning message 
for the case where the symmetry happens to be lowered during the 
numerical evaluation of the Hessian.

We, MOLPRO users, will appreciate the efforts to be done by Prof. Werner.

Best wishes,
Tsuneo Hirano

At 19:34 11/01/07, you wrote:
>This check was added purposely for the following reason:
>
>Computing a numerical Hessian requires in general displacements that
>lower the symmetry, and this may lead to problems in multireference
>calculations (e.g., core/closed/occ and state symmetries can change if the
>symmetry is lowered; also the MRCI energy can slightly change if the symmetry
>is lowered; this is due to the way in which the configuration spaces
>are generated; see, e.g., SYM directives in the manual)
>
>However, I agree that the check should not be made if the hessian is
>computed only in symmetrical displacement coordinates.
>I will try to implement a check for this case.
>
>Joachim Werner
>
>
>Am 07.01.2011 um 10:16 schrieb Tsuneo Hirano:
>
> > Dear MOLPRO developers (and users),
> >
> > When MOLPRO was upgraded from MOLPRO 2008 to MOLPRO 2009 and 
> above, optg calculation at the MRCI level with symmetry constraint 
> has become "llegal"and the relevant job will be aborted.  We 
> encountered this situation when we tried to optimize the geometry 
> at the MR-SDCI level based on the SA-MCSCF natural orbitals using 
> MOLPRO 2010.1 (Patch level: 8).  An example of the input file is 
> given as an attached file "co_cs_ts_Ap_opt_vdz_2010.inp." The error 
> message from the MOLPRO 2010.1 for this input is;
> > -------------------
> >  method=MRCI  nstsym= 2  istsym(1)=1  isyref=1
> >
> >  Numerical hessians for multireference methods can only be 
> computed without symmetry.
> >  Use NOSYM !
> >  ? Error
> >  ? Illegal use of symmetry
> >  ? The problem occurs in numhess_mppx
> >
> >  ERROR EXIT
> >  CURRENT STACK:      MAIN
> > -------------------
> > This error message came from the Lines 167-174 in the 
> "numhess_mppx.f " of MOLPRO 2010.1.
> >
> > The same input (with change in "symmetry,x; geometry={"to 
> "geometry = {x;") successfully ran to the end using MOLPRO 2006.1 
> (Patch level: 149).  The corresponding message is;
> > -------------------
> >  Numerically approximating hessian using central energy differences
> >
> >  Task list generated. Total number of displacements:     12
> > -------------------
> >
> > Generally speaking, numerical Hessian is simply the 2nd-order 
> geometrical derivatives of the energy, and it does not matter by 
> what methods the energy is calculated.  Therefore, with or without 
> symmetry constraint, or for multi-reference or single-reference 
> method, the numerical Hessian can be constructed once the energy 
> that is specified by the "variable" directive can be calculated for 
> the given geometrical changes.
> >
> > When symmetry constraint is imposed, the geometrical changes 
> for the numerical Hessian should be done to keep the given 
> symmetry.  This can be done by setting the z-matrix or xyz-type 
> geometry specification symmetry-adaptive.  It is the responsibility 
> of the user of MOLPRO to write a proper z-matrix or xyz-type 
> geometry specification to keep the symmetry for any change of the 
> geometry in the numerical evaluation steps for the Hessian.
> >
> > The one who upgraded the optg codes might misunderstand what 
> the numerical Hessian means or there might be some reason which I 
> cannot think out.
> > It should be noted that Optg using numerical Hessian at the 
> multi-reference level with symmetry constraint is very important.
> >
> > Best wishes,
> > Tsuneo Hirano

************************************************
Tsuneo Hirano
Emeritus Professor of Chemistry, Ochanomizu University

e-mail:  hirano at nccsk.com
************************************************