[molpro-user] open-shell, orbital-invariant mp2

[Gerald Knizia]

On Friday 04 September 2009 00:08, Benj FitzPatrick wrote:
> I couldn't find any mention in the manual if rmp2 calls the same method
> that is used during an rccsd/uccsd calculation with an rohf reference.  In
> the ccsd calculations I see several iterations printed, but when I call
> rmp2 all by itself I do not, and this difference led me to wonder if they
> are the same.  

There are actually four different conventional rmp2 programs (don't ask), in 
principle all can calculate equivalent rmp2 values, so different energies are 
not supposed to be produced unless either density fitting or an additional 
spin projection is employed. 

Iirc, the programs are all orbital invariant in the sense of returning the 
same numbers regardless of what equivalent orbital set is input. In the 
non-iterative methods this is accomplished by calculating their own 
semi-canonical orbitals from the input orbitals.

Some of the programs set different output variables than others, and there are 
some additional caveeats:
  - the rmp2 run automatically by uccsd is really rmp2, but the rmp2 run by 
rccsd is not really rmp2 (despite claiming so). It employs an additional 
spin-projection on the amplitudes and thus gets slightly different results 
than rmp2 would.
  - the mp2 run by mp2-f12 or ccsd-f12 methods is density fitted if a fixed 
amplitude ansatz is employed, but not if that is not the case.
  - {rmp2,maxit=100} calls something similar, but not quite the same, as the 
rmp2 called from uccsd. It should produce the same numbers as "rmp2" and the 
rmp2 from uccsd.
-- 
Gerald Knizia