TD-DFT calculation general question

[jake_burner1]

Hi there,

I wish to perform a TD-DFT calculation on an insulating metal-organic system (fundamental gap of ~5.5 eV) to obtain the optical gap of the material. I am using the DSH functional (parameterized according to: https://pubs.acs.org/doi/10.1021/acs.jpclett.8b00919). I have used the following INCAR for the ground state calculation:

EXX = 0.52
ALGO = Conjugate
EDIFF = 1e-06
HFSCREEN = 1.89
ISMEAR = 0
LHFCALC = .TRUE.
LMODELHF = .TRUE.
LORBIT = 11
LREAL = .FALSE.
NEDOS = 5000
NELM = 1000
NSW = 0
NWRITE = 1
POTIM = 0.4
PREC = Accurate
SIGMA = 0.1
SYMPREC = 0.001

I understand from the wiki that I should now do a separate calculation where I include more unoccupied states and LOPTICS = .TRUE. to obtain the WAVEDER file required for the following TDHF step (https://www.vasp.at/wiki/index.php/Time ... lculations). Then, after the ALGO = TDHF calculation, I should have a resulting optical spectrum. I wanted to get some clarification, since I noticed a separate tutorial (focused on comparing TDDDH with GW/BSE) an exact diagonalization step (ALGO = Exact) was required for the unoccupied states (https://www.vasp.at/tutorials/latest/bse/part1/). Is this just because the system is sufficiently small that exact diagonalization is easily tractable, or should I be including this step? Perhaps I am confused about why exact diagonalization is required in the first place. Any suggestions on the correct procedure considering my system and desired property would be really appreciated.

Thanks!
Jake

[alexey.tal]

Dear Jake,

Thank you for your question.

That is correct. If the system is sufficiently small we can use the exact diagonalization (ALGO=Exact). However, it is also possible to use iterative algorithms (Normal, Damped, etc.), but the convergence criteria in this case is based on the total energies, i.e., occupied states and the unoccupied states might be not as well converged. For large systems, where the exact algorithm is too costly, one can use iterative algorithms with more stringent convergence criteria and increase the number of SCF iterations to ensure that the empty states are well converged.

Best wishes,
Alexey