[molpro-user] About non-HF orbitals used in CCSD(T) calculation

[Gerald Knizia]

On Monday 18 January 2010 00:26, Chen Hui wrote:
> So does anybody know something about the second question?
> My question is that if CCSD(T)-F12 method can be used with the DFT
> orbital rather than the HF orbital.
>  From manual, it seems that this is not right. But I am trying this and
> there is no running error yet.

There are some potential problems. Namely, under certain conditions the F12 
programs use the Fock matrices stored by the SCF calculation as input; and if 
this Fock matrix is not actually a Fock matrix but a Kohn-Sham matrix, then 
things /will/ break. The closed-shell program may also assume that the Fock 
matrix is diagonal, but it might also check and fall back to an iterative 
algorithm if it is not.
In principle running these calculations on top of any kind of orbitals is 
possible, but there several different places where either the single programs 
or the synchronization between the open- and closed-shell programs might go 
awry.
I will look up what exactly is happening in both programs (closed- and 
open-shell) under various conditions, and then I'll tell you. 

That being said, I'd also like to point out that it is very probable that the 
F12 singles correction will just make something closely resembling a 
Hartree-Fock reference function out of whatever you use as input for the 
orbitals.
-- 
Gerald Knizia