Igor Polyakov
polyakoviv@gmail.com
I'm not Alex, but I'll try to help. First of all, CASSCF state order is not guaranteed to be the correct one=)
Larger effective Hamiltonian dimension should yield more correct results, however for big systems it is simply impractical (high computational effort) to calculate Heff for more than 10-20 states. 2x2 seems to be a bit extreme in the low end because higher states (up to 5 or even more sometimes) can be of influence to the lowest ones. I would recommend to conduct a single large calculation to see the "EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN".
For example for one of my systems:
1 2 3 4 5
*******************************************************
1 -0.996755 -0.063006 -0.025667 0.000488 0.007272
2 0.068939 -0.930901 -0.324050 0.023948 0.052243
3 0.002769 0.257066 -0.672227 0.680566 -0.104661
4 0.004256 0.225480 -0.647910 -0.704316 0.166305
5 0.013621 0.100592 0.148076 0.039046 0.262885
6 -0.004291 -0.005923 -0.014240 -0.189029 -0.930557
7 -0.005277 0.019798 0.006718 -0.032537 0.111529
8 0.036419 -0.043997 -0.013186 -0.042064 -0.102410
9 -0.008000 -0.005780 0.016341 0.009414 0.020153
10 -0.008890 0.003653 0.006373 -0.004601 -0.015154
This means that to correctly describe state 3 it is crucial to include state 4. This can also be seen in the CAS-CI states, states 3 and 4 share some large weighted CSF-S
Best regards, Igor
On Fri Dec 23 '16 5:37pm, Maksim Shundalau wrote
------------------------------------------------
>Dear Alex,
>
>
>Let's pretend that I calculated two lower states (triplet and singlet) of transient molecule (or radical) at the CASSCF level of theory. I get triplet as a ground state and singlet is only a few tens of cm-1 higher than the triplet.
>How many triplets and singlets should I take in the XMCQDPT2 procedure to get the same sequence of states and trust the results? If I take 2 and 2 (or 10 and 10), I have triplet higher than singlet. Should I take 3 and 2 (or 5 and 2, for example) for triplet and singlet, respectively?
>
>
>Kind regards,
>Maksim Shundalau.
>
[ This message was edited on Sat Dec 24 '16 at 9:00pm by the author ]