>thanks for the reply. Actually it might be a good idea to
>implement a state-specific solvation for MCSCF.
>An interesting case is when a system has an excited state with
>a charge-transfer character. In that case using an average density
>might give an unexpected result, but I might be wrong.
I do not agree with this. Consider SA-CASSCF over two states,
the ground state with no charge-transfer and the first excited
state of charge-transfer type. If one is using field generated
by the averaged density then both states are not described
very good. However, they are least approximately described
on an equal footing. If one is using reaction field from the
second state, at the first glance S1 state becomes much better
described. At the same time the ground state becomes described
very poorly. Even worse, because S0 is now not well described,
in SA-MCSCF procedure S1 state will tend to partly mimic S0 state
and thus, actually even S1 state will become not such a well described.
Perhaps if one is interested in describing S1 state only,
then this approach may work. However, I'd prefer just to
increase to the weight of S1 state in state-averaging.
>Concerning the gradients, is there a theoretical reason why
>they should be disabled? or is it just a technical problem?
Because at least at the first glance this modified MCSCF procedure
will not be longer compatible with our implementation of gradients
for SA-MCSCF (http://classic.chem.msu.su/gran/gamess/ss-gradients.pdf).
However, this may requre a more detailed analysis.