[molpro-user] {SPAM}? open shell SCF discrepancy with GAUSSIAN09

[Gerald Knizia]

On Thursday 21 January 2010 16:04, Jason Byrd wrote:
> I am calculating the structure of KRb2 and came across the following
> discrepancy.  Using the Def2-TZVPP (testing the following with different
> basis sets does not change the results) basis set I get the following
> discrepancy:
>
> SCF Done:  E(ROHF) =                  -646.751495036
> !RHF STATE 1.1 Energy               -646.750268500845
>
> Testing this further with GAMESS suggests that the G09 rhf value is
> correct.  Ive attached the gaussian and molpro output files.

All three programs are right. There are (at least) two different ROHF 
solutions, and the different programs find different ones (the solutions have 
quite different dipole moments; one might correspond to an excited state). 
You can get the one G09 found in Molpro with
   {rhf,accu=10,maxit=200,maxdis=8; shift,0,0; occ,15,4}
or
   {mcscf;maxit,200;occ,15,4;wf,sym=1,spin=1; orbital,2140.2}

If you just use "basis=def2-tzvpp" instead of giving the basis and ECP 
manually, there is also the other issue of the core potentials. By default, 
Turbomole associates its basis sets with modified versions of the 
Stuttgart-Koeln-Bonn ECPs, because Turbomole does not support some higher ECP 
projectors. In Molpro, these sets have been linked to the original versions 
of these ECPs. That leads to a tiny additional difference in the energy here:
def2-TZVPP & Turbomole's modified ECP28MWB:
 !RHF STATE 1.1 Energy               -646.751495035794
def2-TZVPP & original ECP28MWB:
 !RHF STATE 1.1 Energy               -646.751496648248
This is unlikely to cause any real-world trouble, apart from different 
programs giving slightly different answers in absolute energies.
-- 
Gerald Knizia