Re: A riddle for CASPT2 gurus!

Subject:

if B is 1s22s22p1, what is Li?

Passfield:

Comments:
: : Dear Evgeniy, : The fact that CASSCF and CASPT2 do not agree in locating excited states is : not an extraordinary occurrence, but the expected behavior. Remember that : CASSCF includes just a minimum amount of correlation energy and it CANNOT : be considered quantitative in any case, unlike CASPT2. Therefore, forget about : excitation energies at the CASSCF level! : Regarding your specific problem. When you mention two configurations I guess : you refer to two conformations or geometries. The first thing you should really : wonder is if this "a priori" evidence of S1 lower in the 2nd than in the 1st geometry : is really so clear (right experiment?, right model?, right environment?..., there are : so many aspects...). Second, if I were you my first check would be to compare : CASPT2, not MS-CASPT2 energies. There is no "a priori" reason to prefer MS-CASPT2 : to regular single-state CASPT2. Except for especific cases like valence-Rydberg : mixing or orthogonality problems between caspt2 states (close to surface crossings) : the MS-CASPT2 should not be used. The reason is that the coupling between CASPT2 : states leading to MS-CASPT2 is typically overshoot for normal active spaces. One : typically requires a much larger active space to get a reasonable coupling (which : means states splitting) at the MS-CASPT2 level. My advise: look at the SS-CASPT2 energies. : Finally, be sure that S1 is the SAME state at the two geometries by checking wave functions : (CI coefs and MOs), properties, etc. It may well happen that you are comparing different : states and even that your state is higher in energy (at the casscf level, even if caspt2 puts : it lower) and you are missing it. : Hope that this is useful to you. : Best : Luis : : Dear All, : : I have encountered an extraordinary occurrence in the CASPT2 energies for two different configurations of the same system. The data is the following: : : 1st configuration: : : CASSCF energy: : : S0 (ground) state: -1083.32199351 : : S1 (lowest excited) state: -1083.16358216 : : Multistate CASPT2 energy: : : S0 (ground) state: -1085.56591537 : : S1 (lowest excited) state: -1085.45365823 : : 2nd configuration: : : CASSCF energy: : : S0 (ground) state: -1083.25860324 : : S1 (lowest excited) state: -1083.19263076 : : Multistate CASPT2 energy: : : S0 (ground) state: -1085.52232322 : : S1 (lowest excited) state: -1085.45281629 : : As you can see CASSCF predicts that the S1 state at the 2nd configuration is lower in energy than at the 1st configuration. Surprisingly CASPT2 does not agree with CASSCF and finds the S1 state higher in energy at the 2nd configuration. It is a priori known that S1 is lower in energy at the 2nd configuration! : : The active space and all things are the same in all calculations, only the geometry is different. : : I wonder why CASPT2 gives obviously a wrong result in this case. Has someone encountered the same behaviour? : : Thanks! : : Best, : : Evgeniy