## Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Re^2: Some questions about MCQDPT theory

Solntsev Pasha
solntsev@univ.kiev.ua

Alex,

thank you for time and answer.

>>Q1: Is there any difference between kstate(1)=1,1,1,0,1 and
>kstae(1)=1,1,1,1,0 for XMCQDPT. Since Heff has the same size, i am a
>little bit confused, especially after reading of the manual from
>GAMESS-US. I don't understand how can we use correctly kstate array,
>while situation with nstate absolutely clear (nstate == Heff)

>As to the question on differences between  kstate(1)=1,1,1,0,1 and
>kstae(1)=1,1,1,1,0 runs, they do differ. In the first case,
>the model space is formed by the first, second, third, and fifth
>CASCI root. The fourth CASCI state is computed at the very early
>stage of the run and only because it is required by Davidson
>diagonalization procedure to compute the fifth state. After this,
>the fourth state is discarded and CASCI states are renumbered to
>condense their labels. This means that the fifth CASCI state is
>referred to as the CASCI state #4 throughout the rest of computations.

ok, it sounds very interesting. So, if an order of states from CAS-CI part is 1,1,1,0,1 we can setup kstate(1)=1,1,1,0,1 and basically remove "undesired" state from PT2 part. Does it make sense? Is this equivalent to setting of the AVECOE and WSTATE (XMCQDPT) arrays to be AVECOE(1)=1,1,1,0,1 and WSTATE(1)=1,1,1,0,1. If not, what is the best strategy, to use kstate or avecoe/wstate? Or, maybe, we can just use nstate/wstate/avecoe combination and forget about kstate.

>>Q2: Is there any correlation between nstate (SA-CASSCF) and Hef.
>Thus, if Heff <= nstate then no problem at all. But if Heff > nstate
>i am confused about eigenvalues of Hef and nature of "extra" states
>correlation between nstate (CAS-CI) and Heff.

>The states used in XMCQDPT2 and MCQDPT2 are just the CASCI states
>computed with the (canonicalized) orbitals generated by the
>preceeding CASSCF or SA-CASSCF computations. For instance,
>one can use SA2-CASSCF to generate orbitals, then use 10 states to
>construct Heff. The first two CASCI states will be those of SA2-CASSCF
>roots, the extra states will be just the additional eigenvectors of
>CASCI Hamiltonian.

What if those additional eigenvectors contribute to some low lying states? Do we need repeat CASSCF part once again and include those extra states?

>>Q3: Is it good idea to do SA-CASSCF/XMCQDPT via singlets and triplets
>simultaneously in case of high spin-orbit coupling. Actually, i am
>going to check singlets only and singlets-triplets, but maybe it make
>no sense.

>I'd suggest you to use to use SA-CASSCF averaged over both singlet
>and triplet states to generate the same set of MOs to be used in PT2
>computations. Firefly v. 8.0.0 has some new features specifically
>tailored for these types of jobs which I'll describe in my next post