is it possible to use EOM-CC with DF, FNO or other technics that can accelerate the calculation and (especially) reduce the number of stored integrals? I am trying to investigate UV-VIS spectra of “push-pull” dyes such as 1-dimethylamino-4-nitrobenzene and larger analogues. TD-DFT fails hardly for these compounds due to charge-transfer nature of the corresponding bands (CAM-B3LYP results are more acceptable, but still not good); NEVPT2/CASSCF seems to be accessible (at least with RI optimizations implemented in ORCA), but leaves some ambiguity in CAS selection (and prohibitively large number of orbitals should be included to CAS in order to “be safe”). Conventional EOM-CCSD in vacuum with Psi4 with smaller molecules and cc-pvdz basis was ok, but for larger tasks the required size of tmp files exceeded 400 Gb. Is there any ways in current Psi4 code to deal with such systems?
EOM-CC might be overkill. CC2 or ADC(2) give usually reliable results for organic dyes.
Not sure what psi4 has to offer regarding their implementation details.
If you use ORCA, you can also try TD-B2GPPLYP (a double hybrid, it includes a CIS(D) calculation for the TD part).
Not solving the CT problem, but it can help and with ORCA’s rijcosx you can do pretty large molecules.
CC2 and ADC(2) in any case endlessly create .33 file (ADC after direct SCF Hartree-Fock dies because of missing 33 file).
I have tried TD-B2PLYP in ORCA, it indeed worked and gave results with quality similar to CAM-B3LYP (with another sign of error), probably I should screen other double hybrids implemented in ORCA including B2GP-PLYP as you have suggested.
[In the worst case, I will get at least some prompts about CAS selection…]
B2GP has more Fock exchange and behaves like BHLYP with improved excitation energies (from the CIS(D) correction). for TD-DFT. Transition moments are not improved! I would not screen other functionals. Too much Fock exchange (say larger 60%) is often harmful, less fock exchange should not work for CT states.
CC2 with water works for me using: property('eom-cc2',properties=['oscillator_strength'])
and a few days old psi4 version.
I mean that Psi4 creates a temporary file with extension .33 and intensively writes data to that file. If I use conventional SCF at HF stage, the file is created at the beginning of the run. If I use direct SCF, the file is not created at HF stage. So, I guess, it contains integrals. However, at the beginning of CC calculation the file appears in any case (within ADC(2) the file is not created, but Psi4 aborts). The problem is, that the size of the file is about 100 Gb in the case of 1-dimethylamino-4-nitrobenzene/cc-pVTZ, for more complicated analogues the size is over 450 Gb (I cannot say exactly because wasn’t actually able to reach the complete creation). At the same time, these systems are readily treated in direct SCF mode with rather low execution times. Thus, the question is, are there any possibilities to avoid such extensive disk usage or to lower in general the computational efforts for EOM – maybe I’ve missed something in the manual? Or the only way is to buy larger hard disks and be patient?
OK, this clarifies the situation for me. The problem isn’t really with EOM-CC per se, but with any post-HF method that requires the full integral list, including MPn, CI, and all other CC methods.
Out of curiosity, are you taking advantage of symmetry in this systems?
I dont see an option to avoid the full 2e-ints for eom-cc2 in psi4. You could tighten the integral screening to 1e-9, but it wont really help i guess. Btw, the actual eom-cc2 step will use even more disk space, unless there is a direct AO formulation somewhere (ao_basis direct ??).
If you really have no access to perhaps turbomole for ri-cc2/ri-adc2, you can reduce the basis. Go for def2-TZVP and delete the f and higher functions. maybe even the older def-TZVP. f and higher momentum basis function are not so important, for excitation energies at least. The high contraction count in the Dunning sets is pretty expensive as well.
Not ideal, but maybe affordable.
If the molecule has symmetry, it could help a great deal to reduce the computational costs: storage will drop by roughly a factor of the order of the point group and computational scaling by a factor of the square of the order of the point group.