[molpro-user] Making SCF converge to the ground state

[Terry Frankcombe]

Hi folks

I've been running into a problem with my HF orbital occupations quite a
lot recently.  I'm trying to do calculations on a triplet system, but
the SCF program insists on converging to a state that isn't the ground
state.

The problem is illustrated by the following output from a dump to a
molden file:

 DUMP ORBITAL  1.1 AS ORBITAL    1  occ=    2.0000  eig=  -15.6686
GROUP=   2
 DUMP ORBITAL  2.1 AS ORBITAL    2  occ=    2.0000  eig=  -15.6633
GROUP=   2
 DUMP ORBITAL  3.1 AS ORBITAL    3  occ=    2.0000  eig=  -11.3487
GROUP=   2
 DUMP ORBITAL  4.1 AS ORBITAL    4  occ=    2.0000  eig=   -1.2977
GROUP=   2
 DUMP ORBITAL  5.1 AS ORBITAL    5  occ=    2.0000  eig=   -0.9600
GROUP=   2
 DUMP ORBITAL  6.1 AS ORBITAL    6  occ=    2.0000  eig=   -0.6882
GROUP=   2
 DUMP ORBITAL  7.1 AS ORBITAL    7  occ=    2.0000  eig=   -0.5551
GROUP=   2
 DUMP ORBITAL  8.1 AS ORBITAL    8  occ=    2.0000  eig=   -0.5302
GROUP=   2
 DUMP ORBITAL  1.2 AS ORBITAL    9  occ=    2.0000  eig=   -0.4495
GROUP=   2
 DUMP ORBITAL  2.2 AS ORBITAL   10  occ=    1.0000  eig=   -0.6223
GROUP=   1
 DUMP ORBITAL  9.1 AS ORBITAL   11  occ=    1.0000  eig=   -0.5463
GROUP=   1

The energy of the singly occupied orbital 2.2 is lower than the doubly
occupied orbitals 7.1, 8.1 and 1.2!  The chemistry agrees that this is
wrong and that the second unpaired spin should be orbital 1.2, mostly
localised to the same atom as 9.1 (on an incoming nitrogen atom).  The
calculated SCF energy is about 0.3 hartree higher than it should be.
CASSCF calculations give the correct state and energy, with the second
electron "excited" out of that 1.2 orbital.

I've played quite a bit with orbital shifts and the initial guess, to no
avail.  How can I convince Molpro that a singly occupied orbital deep in
the spectrum is wrong?

Regards
Terry

-- 
Dr Terry Frankcombe
Physical Chemistry, Department of Chemistry
Göteborgs Universitet
SE-412 96 Göteborg Sweden
Ph: +46 76 224 0887   Skype: terry.frankcombe
<terry at chem.gu.se>