[molpro-user] About spin contamination in open shell RCCSD(T) calculation

[Gerald Knizia]

On Sunday 20 February 2011 12:58, Chen Hui wrote:
> Dear molpro developers,
> I have a question concerning the spin contamination issue in RCCSD(T)
> method in molpro.
> I know that even if CCSD is based on ROHF reference, still spin
> contamination can be introduced. 

This is true, but the effect is generally very small then. Unlike perturbative 
methods, CCSD is intrinsically not prone to spin contamination, and if based 
on an ROHF reference the effect basically negligible for almost all 
applications.

> Thus my question is that whether 
> RCCSD(T) in molpro, which is explained in the manual with "In the RCCSD
> theory certain restrictions among the amplitudes are introduced, such
> that the linear part of the wavefunction becomes a spin eigenfunction",
> is free of spin contamination if " Spin contamination
> <S**2-Sz**2-Sz>     0.00000000" appears after CCSD converged iteration
> in the molpro output file. 

No, not really. Maybe that output should be clarified. The spin contamination 
printed there refers to the linear part of the wave function; the part which 
is eliminated in RCCSD (but not in UCCSD) and thus always zero.

> If the answer is not (which means still CCSD 
> wavefunction can be spin-contaminated), is there a way we know how large
> is the spin contamination?

No, unfortunately. You can get that information from UCCSD wave functions, 
however. You will usually see that in the UCCSD case the (linear) spin 
contamination is already very small, and in RCCSD it can basically only get 
better.
-- 
Gerald Knizia