Can one obtain triplet excited state energies starting from the singlet, closed shell ground state?
If yes, what keywords should I use to specify the desired multiplicity of the excited state?
Thank you all for helping me out!
Marcin Andrzejak
Yes, you should be able to set the reference wave function to UHF for the closed-shell ground state, and then the EOM algorithms can find singlets or triplets. This was also addressed here: EOM-CCSD for triplets excited states
I saw your inquiry on the CFOUR mailing list. Will Psi4 do the job for you?
Sure, it might 
However, if you suggest for me to use UHF
wavefunction of the ground state as reference and analyse relative signs
of the amplitudes for the alpha and beta configurations, then it might
be too much
because of the size of the system I am working on. It is a tetracene
with two diacetyl-SiH3 substituents.
I cannot freely replace those substituents with hydrogen atoms, as the
substituents non-negligibly modify the electronic structure of the
chromophore and modify the nature of the excited states with respect to
pristine tetracene.
However, the size of the molecule and relatively low symmetry (C2v)
make the CC3 calculations very long. With the Dunning double-zeta basis
set the singlet excitations take approximately two to four weeks on 24
cores to converge - I am using Psi4 for this job and it works fine,
although sometimes there are convergence problems
and root flipping (noticed for unsubstituted tetracene, which I used as
a sort of reference).
Yet, if there is a possibility of using closed shell reference for
obtaining triplet ecitations, then I would be very interested in it,
because then I think that such calculations would still be feasible for
the system I am working on.
Yours sincerely!
Marcin Andrzejak
It’s true that we don’t have a fully spin-adapted implementation of triplets based on the RHF-CC reference, so a UHF-based approach for that size system is indeed probably going to be too expensive. I think DALTON may have such an implementation, but I don’t know how efficient their CC3 code is.
Thank you very much for the reply and the suggestion. Dalton has a good
variety of capabilities, but I have found this code to be strangely
inefficient. Most od the calculations I did with Dlalton went much
slower than analogous computations performed with Psi4, Turbomole or
Molpro. It seems to me that parallelization of the code is, or used to
be two years ago (when I last used Dalton), rather poorly done. But it
may also be that my admins did a poor job while compiling the code.
Yours sincerely!
Marcin Andrzejak