I am wondering what the common wisdom is regarding the calculation of harmonic vibrational frequencies with density fitted methods. Is the final calculated hessian sensitive to the approximations introduced through density fitting? Has a comparison between conventional and density fitted ccsd(t) frequencies been published anywhere?
This may be a question for @bozkaya, who wrote the code for our analytic DF-CCSD(T) gradients. We have numerical Hessians, but I’m not aware of any direct comparisons of their quality when using DF or not.
In our recent paper on the DF-CCSD gradient (JCP, 144, 174103 (2016); doi: 10.1063/1.4948318) there is a comparison of DF-CCSD bond lengths and frequencies w.r.t. CCSD. The mean average error introduced by DF, for the considered molecular set, was 10-4 angstrom for bond lengths and 0.5 cm-1 for harmonic frequencies. I have observed a similar situation in the case of DF-CCSD(T) vs. CCSD(T).
Our DF-CCSD(T) analytic gradients code has been completed, and will be available in the near future releases.