CCL:G: CMO Analysis of a roB3LYP calculation using Gaussian 09 and NBO 6

 Sent to CCL by: "Johannes  Straub" [johannes.straub()]
 I already posted this problem in the NBO forum, but still got no reply to my
 problem, so I thought I'll try it here on CCL.
 I'm working with Gaussian 09 D.01 and NBO 6.0.
 For my studies, I'd like to compare some particular orbital energies (d-Orbital
 set) of two isomeric Fe(IV)-oxo complexes, to see, if there are any significant
 (The two isomers show different reactivity in HAT reactions.)
 We figured out that the CMO Analysis feature of NBO 6.0 might be very promising
 for this task, because I can see which MOs are the ones built up by the
 important Fe d- and O p-Orbitals.
 I started using a normal uB3LYP calculation with Gaussian
 "#p sp pop=(full,nbo6read) scrf=(pcm,solvent=acetonitrile) gfinput
 gfoldprint 5d 7f def2tzvp ub3lyp scf=(tight,xqc)
 $NBO plot print=3 file=L1Fe-e2-s1-bs3-nd-s-CMO-1 CMO archive $END"
 I get the .31 and .37 files containing the NBOs and the corresponding CMO output
 in the Gaussian output file, as expected. However, I was not able to completely
 identify the d-Orbital set for both alpha- and beta-spin orbitals. (They don't
 have the same MO number and are built up by NBOs very differently).
 So, I thought, a similar roB3LYP calculation might help, because I would get
 only one set of orbital eigenvalues and therefore only one set of NBO orbitals,
 which makes it more easy to assign the important orbitals. However, after the
 restricted oben shell SCF calculation, the NBO 6 module seems to split the MO
 eigenvalues again into a different set of alpha and beta spin orbitals having
 new, different eigenvalues. (At least I was able to identify all the d-orbitals
 both in alpha and beta spin now)
 What exactly is the NBO module doing in this step? Are the new eigenvalues
 reliable for the comparison of orbital energies between two isomeric complexes?
 Thank you!