From owner-chemistry@ccl.net Sun Sep 10 23:49:00 2006
From: "Seth Olsen s.olsen1=uq.edu.au"
To: CCL
Subject: CCL: Energy convergence around conical intersection
Message-Id: <-32514-060910213919-26047-Mg8pfjGiQCzjNpv+Rkdt/g%%server.ccl.net>
X-Original-From: Seth Olsen
Content-Transfer-Encoding: 7bit
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Date: Mon, 11 Sep 2006 10:49:49 +1000
MIME-Version: 1.0
Sent to CCL by: Seth Olsen [s.olsen1|a|uq.edu.au]
Hi Sherin,
Changing the weighting of the states in a state-averaged CAS calculation
will give you convergence problems if the states are close together just
because it doesn't take much variation in the energy of either state in
order to flip the roots. I'm a bit confused as to what your objective
is, though, since any weighting scheme other than 0.5/0.5 would be
expected to overbias one state or the other. This would change the
energy and the position of the intersection, and it's not clear to me
that this would happen in a regular fashion. The question of what
weighting is 'right' to describe the intersection (in the sense of
getting the 'right answer for the right reason') does not seem to be
very relevant because the 'right' weighting will probably vary depending
on the state and the system. In this case, even if the 'right' answer
is achieved, the weighting has become just a tuning parameter and its
not clear that the it is 'right for the right reason'. When you vary
the weighting, are you then comparing to something more trustworthy
(like evenly-weighted MCQDPT or MRCI)?
I've often thought that there isn't that information about
state-averaging in the literature, given that it is the currently the
most popular method for computational photochemical modeling. The rate
of change of weighting of a state averaged CAS solution has been
suggested as a diagnostic for the quality of the wavefunction (Stalring
et al, Mol. Phys. v.2 pp.103-114 (2001)). In this work it is also
pointed out that only for an evenly-weighted wave function is the final
solution invariant to projections within the state-averaged subspace (in
addition to the usual CASSCF orbital rotation invariance). This in turn
has implications for the Lagrangian used to determine analytic gradients
w/ respect to geometry.
Good luck. :-)
Cheers,
Seth
Sherin Alfalah shireen.alfalah^yahoo.com wrote:
>Sent to CCL by: "Sherin Alfalah" [shireen.alfalah-*-yahoo.com]
>Dear CCL users,
>We are trying to run energy calculations for some points around a conical intersection. I am facing some problems in convergence for the excited state. To reach MCSCF convergence, we try to read some molecular orbitals of other close points or to run the energy calculations for the excited state with more weight of the ground state for example "0.1 or 0.2". In the conical intersection region, reading different vectors may lead to different stationary points with different energies. I am a bit confused about the most proper way to have convergence. Shall it be the choice of method that gave the lowest energy or what? How can I know that I am not over shooting the minimum?
>
>We are using GAMESS, I am wondering if the results we have are due to chemical reasons or some artificial results of GAMESS software.
>
>I think that having more weights of the ground state, is reasonable since the points are within the conical intersection area?
>
>I am wondering about the most proper way to obtain convergence? and also if some one has any experience or know some tricks that may be useful to obtain convergence? Also, any information or discussion for this issue would be highly appreciated.
>
>Thanks in advance.
>
>
>
>
>**************************************
>Sherin Alfalah
>PhD Student
>Theoretical Chemist
>Chemistry Department
>AlQuds University
>**************************************>
>
>
>
>