As for now, the resp module supports all pure functionals and simple hybrid functionals, but not including the LibXC functionals (B97 etc.). For range-separated hybrids, only properties that do not involve nuclear derivatives (e.g. dipole moments, polarizabilities, hyperpolarizabilities, SOC corrections, etc.) are supported (again, this does not include the LibXC range-separated hybrids, e.g. wB97, wB97X etc., which are not supported at all); we are working on extending the scope of applicability to properties that do involve nuclear derivatives (gradients, NACMEs). Double hybrids are not yet supported, and will not be supported in the near future.
scf
The following requirements are for finite-difference NAC only.
1. Tight convergence on density matrix is required (10-14).
2. sgnfix: fix adjacent sign of MOs during SCF iterations
3. iaufbau=3: fix ordering and sign with respect to the initial MOs.
For gradients and analytic NACs, the default SCF convergence criteria are likely sufficient. For numerical Hessians the default convergence criteria give frequencies that are accurate to a couple of cm^-1. In cases where spurious imaginary frequencies arise, it may be necessary to tighten the convergence criteria further.
tddft
1. Tight convergence on eigenvectors (10-6) and eigenvalues (10-8). The default criteria (10-5 and 10-7, respectively) are frequently insufficient.
2. When self-adaptive XC integration grid is used, it is recommended to tighten gridtol in $tddft, especially for open-shell systems. Recommended value is 10-7.
3. Keyword lefteig should be used for storing left eigenvectors in TD-DFT
4. Keyword istore speficify the file number of TD-DFT calculations