Multistate CASPT2 vs Density keyword

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: I computed the energies of the two lowest Bu excitations of bithiophene. State average CASSCF calculation (2 states) was followed by the CASPT2 procedure. : I did the CASPT2 calculations twice: initially I used the default CASPT2 settings, and then I used the keyword DENSITY, to check whether there would be any changes in the natural orbitals. : What I found, however, are the drastic changes in the off-diagonal elements of the effective Hamiltonian: : Without 'DENSITY': : 1 2 : 1 -0.71474674 -0.02998095 : 2 -0.02337891 -0.68139092 : With 'DENSITY': : 1 2 : 1 -0.71474674 0.02191993 : 2 -0.15510176 -0.68139092 : Easy to see, that the diagonal elements remain unchanged, whereas the off-diagonal ones undergo large changes. : I do not understand this behavior (could anyone explain it briefly?) and moreover, I am unsure of which of the two results are more credible and should be used for further analysis. The off diagonal elements obtained without the DENSITY keyword are smaller and similar to each other, soe they look better. On the other hand, the DENSITY keyword is used to produce the exact first-order density matrix, so shouldn't it in theory lead to more accurate results? : I'd be obliged for comments and explanations. : Best regards : Marcin