What is the state of the hybrid RI-DFT analytic 2nd derivative implementation in TURBOMOLE 5.9? I made some tests using RI-B3LYP and aoforce happily accepts the B3LYP functional in combination with RI. No warnings are given, except the usual note about RI implementation of aoforce being in preliminary state. However, the results do not look so good. Here are the wavenumbers (cm-1) for the three IR-active vibrations of Td-symmetric P4:
B3LYP RI-B3LYP RI-B3LYP
analytic analytic numeric
367.23 420.13 367.68
460.56 506.59 460.14
602.77 628.66 602.37
As can be seen, the analytic RI-B3LYP wavenumbers are very different from traditional B3LYP and numeric RI-B3LYP values. This difference then affects the zero-point energies and thermodynamical quantities. The non-RI and RI geometries are nearly identical, as are the relative energies when compared to for example P8. The similar comparison for pure functional (BP86) yielded the expected results (almost identical wavenumbers for RI and non-RI).
So should I just avoid calculating the RI-hydrid-DFT frequencies analytically?