[molpro-user] open-shell systems with MOLPRO

[Gerald Knizia]

On Tuesday 03 November 2009 16:49, Lubomir Rulisek wrote:
> I encountered following problem running MOLPRO for seemingly simple
> open-shell systems.
>
> My idea was to run few benchmark calculations for transition metal
> containing complexes.
> One of the simplest is the linear [(CH3S-)Cu(H2O)]1+, i.e. with copper
> (II) ion. As it is the doublet state, I had intention to run UCCSD(T) on
> top of UHF reference. I found out it is not possible with MOLPRO.

The reason for this is that RHF-based approaches are just as accurate as UHF 
based ones in well conditioned situations, but they can be done more 
efficiently and they don't suffer from (noteworthy) spin contamination.
RHF is the better choice in nearly all situations.

> ROHF is (in my hands) much more difficult, if not impossible, to
> converge.

Hm. One thing you could try is to first calculate an cation of the species you 
are trying to actually treat (e.g., something making the system closed 
shell), and then use its orbitals as input for the actual system. This way it 
is possible to force convergence in most of the cases.

If there is molecular symmetry, you might also need to check if you specified 
the correct wave function symmetry (or if necessary, if the open shells are 
in the correct symmetries). Current Molpro versions try to figure out the 
wave function symmetry on their own using the aufbau principle, but older 
ones assumed symmetry 1 (totally symmetric) if you didn't specify anything 
else. And this was usually wrong.

> Then I thought for a moment to run UHF first, use the orbitals as the
> starting guess for ROHF and run UCCSD(T) afterwards. This seemed to work
> out  pretty well until the moment when I compared two calcs where I got
> both ROHF/UHF and ROHF  ("from scratch") convergence. These two
> results differed significantly (30 kcal.mol-1 or so). I am a bit puzzled
> what to think about it.

These two calculations find different states, with the RHF-from-scratch one 
being better (lower in energy). The dipole moments are also quite different. 
Molpro is really much better at RHF than at UHF. If RHF does not converge, 
playing around with the level shifts (e.g., 
{rhf; shift,0,0}
) or using orbitals from a species with less electrons is probably more 
promising.
-- 
Gerald Knizia