Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

sanya
sanya@photonics.ru

Of course, these are different arrays (although I'd propose to set the default value of wstate in $XMCQDPT group equal to wstate in CASSCF groups (GUGDM2 or $DET)). Nevertheless, careful inspection is needed anyway.

>Or more simpler, where and what should i look in an output file to find contribution of CASSCF states into XMCQDPT states?

You can find this in EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN part of XMCQDPT output, see below.

>Also, i saw a lot of posts about analysis of CASSCF part, but for me it is steel "pitfall". For example, for CASSCF(4,4) first state has main (coef~0.9) configuration 2200. But second 2020 (coef~0.9). Does it make sense for UV spectra? For me more natural configuration would be 2110 or 2101. Should we consider such state (2020 or 2002) for construction of wstate array for XMCQDPT? I would be very appreciate  for any explanation of my questions.

Such a situation indicates inadequate choice of the active space. Failure to obtain low-lying singly excited state shows that the chosen active orbitals do not properly describe the excitation. Instead, these orbitals serve to improve the description of the ground state. Doubly-excited states in this case are just by-products.

So, the remedy in this case is to enhance the active space with both occupied and virtual orbitals and look, which of them contribute to single excitations. Later, after obtaining adequate single excitations, you may try to remove unnecessary (non-contributing) orbitals from the active space.

Remember that the success of using QDPT depends on the quality of the reference CASSCF solution.

In addition, below is the citation from Alex's answer to my question concerning XMCQDPT:

sanya> As for the check of the overlap matrix of non-orthogonal eigenvectors, it's a good idea, thanks. So, if off-diagonal values are small and all eigenvalues of non-symmetric Heff are real, it's no need to further increase Heff, isn't it?

Alex> No, the idea is that if they are large and/or some eigenvalues
are complex, your XMCQDPT2 job is definitely running into problems.  

This means that you should look at the EIGENVECTORS OF THE EFFECTIVE HAMILTONIAN matrix to make sure that off-diagonal elements are no more than 0.1 and at EIGENVALUES OF THE NON-SYMMETRIC EFFECTIVE HAMILTONIAN to make sure that the values are real (at least, the imaginary part is no more than 1e-7).

Another citation:

sanya> What off-diagonal value should be considered as large: ~0.1, or ~0.01, or ~1?

Alex> I'd suggest to avoid off-diagonals larger than ca. 0.1-0.2
and safely ignore smaller values. Very large off-diagonal values
of overlap of two XMCQDPT2 states typically mean that these two
states attempt to describe the same real state. The most
likely reason of this is the poor choice of the model space
(e.g., due to some low-lying spurious CASSCF states entering
model space). Another reason could be intruder state problem.

sanya> and is it safe if the imaginary part of an eigenvalue of non-symmetric Heff is ~1.0e-16?

Alex> It is absolutely safe. These small values are artifacts of
accumulation of roundoff errors inside non-symmetric matrix
diagonalization code. Look at the output and you'll find that
complex eigenvalues are always coming in pairs of conjugated
values.