CCL: Symmetry breaking during CASSCF

 Sent to CCL by: Wai-To Chan <chan(!)>
 Dear CCL Readers,
 I have a planar conjugated organic molecule that, on paper, has D2h
 symmetry. I am working with the hypothesis that the molecule is a
 biradical singlet. I want to perform a CASSCF(2,2) on this molecule
 to estimate the biradical character of this molecule. When I
 calculate the molecule with the constraint that the molecule has D2h
 symmetry, the calculation has severe convergence problems. If I
 remove this constraint, the molecule distorts to a C2v symmetry and I
 obtain a biradical character ~20%. I checked the orbitals being used
 for the CASSCF calculation and they look right (one nonbonding and
 another slightly antibonding).
 Is the situation described above correct? Can an organic molecule
 have such a symmetry breaking as a result of a CASSCF calculation? Is
 the molecule really a C2v system? Any suggestions are welcome.
    Thank you very much in advance for your help.
 		Sincerely yours,
 													Gustavo L.C. Moura
    I assume that you removed the D2h symmetry constraint by changing
 the original cartesian coordinates just small enough for the program to
 the C2v symmetry.
         If the distortion of the C2v structure from D2h is not too large
 I would repeat the optimization starting with the undistorted D2h geometry.
 Small deviation from ideal symmtry is not that unusual with CASSCF geometry
 even when it is properly executed.
 I would try to avoid the convergence difficulties by changing the symmetry
 of the wavefunction instead of the geometry. With GAMESS this can be done by
 specifying a lower symmetry than D2h in the $DET or $DRT input card.
 With Gaussian I am not sure how this can be done. Perhaps the nosymm keyword
 could do
 but I am not sure if it is effective in the CASSCF step. If the CASSCF is based
 UHF unrestricted orbitals I will try to break the symmetry of the UHF
 	  If your casscf-optimized C2V geometry is clearly distinctive from
 the D2h symmetry you perceive to be correct a possible explanation
 is that the active space in your MCSCF wavefunction is not sufficiently
 large. A 2-electrons-2-MOs active space is of the minimum size for
 diradicals. Your system seems to be more of a diradicaloid than an
 open-shell singlet diradical. It may sound surprising but partial diradical
 systems may require a larger active space than a pure diradical.
 With Gaussian I would design the size of the active space required by
 running a UHF calculation on the D2h structure first with guess=mix and
 specified. I would then examine the occupancy of the UHF-natural orbitals
 (with pop=naturalorbitals). If the fraction occupancies  of the natural orbitals
 beyond the HOMO and the LUMO are significant I would increase the active space
 size accordingly.
 Wai-To Chan