[molpro-user] noncanonical orbitals in CCSD-F12

Gerald Knizia knizia at theochem.uni-stuttgart.de
Fri Mar 1 11:59:19 GMT 2013

On Thu, 2013-02-28 at 11:23 +0100, Mariusz Radon wrote:
> Dear Kirk,
> thank you very much for your reply.
> I have tried to perform calculation according to your suggestion (i.e.,
> something like: {ks,b3lyp}, then {hf;maxit,1}, followed by
> {ccsd(t)-f12;orbital,ignore_error}).
> Unfortunately, it seems that Molpro always performs at least 3 HF
> iterations (even if maxit is set to 1 or 0, or -1)....

I am rather sure that I fixed this a few years ago: {hf; maxit,1} should
now always perform exactly one iteration, just make a new Fock matrix,
and not update the orbitals. I am not sure in which released version
this first occurred, but it might be helpful to update your Molpro
binaries if you are using old ones. 

> But, intriguingly, when I carry out CCSD(T)-F12 right after DFT
> (w/o HF in between), the relative energies look very good, and rather
> close to CBS limit from extrapolated conventional calculations.
> I wonder whether F12 routine in Molpro can somehow complete the missing
> information, even if HF program was not called before F12?

I also put in some sanity checks, which would complete some of the
missing information, but I think this still needed the correct HF Fock
matrix (at least in the closed-shell program). There are two tests to
see if the CC-F12 is definitely wrong:

- If ccsd-f12 does a non-iterative MP2-F12 for non-canonical or non-HF
orbitals, the results will be wrong.

- If uccsd-f12 and ccsd-f12 give different results (uccsd always makes a
new Fock matrix and uses non-canonical formulas, and should work for any
orbitals, but ccsd-f12 has some optimizations which are not valid in the
non-HF case)

> Or am I obtaining "good" results by pure accident?

I would recommend not using the KS-Fock matrix directly. It will
actually be mixed with other Fock matrices (for the RI-space), and they
do not have the exchange-correlation contributions. It is quite likely
that at the very least you would get very wrong CABS singles energies.
Gerald Knizia

More information about the Molpro-user mailing list