Trouble with negative frequencies


Click here for full thread
Just Got Here
Dear developers and users of nwchem,
We are investigating small cluster systems via global optimisation on the DFT level (TPSS, PBE0, B3LYP, B3P86 with basis set def2-TZVPP or ccPVTZ-pp) of theorie with the goal to find the Global Minimum. After analysing the system, we often find that our predicted GM show negative frequencies. Even for a system with only three atoms the problems occur. Higher level QC calculations like CCSD(T) show the structure to be real minima with properties matching the experimental data. We have tried different methods to get rid of the negative frequencies like:

-elongation along the negative frequency modes
-elongation along positive real frequency modes
-random displacements and distortions
-with different driver trust settings, different grids and convergence criteria.
Unfortunately nothing of this helps.

By coincidence we found the other day, that it is also possible to calculate the frequencies numerically instead of analytically using the statement “task dft frequencies numerical”. With this numerical statement all frequencies are positive for a lot of found GM’s.
Sometimes other Isomers, for which we performed geometry optimisation, show purely negative frequencies with the analytical method, which should not be the case after geometry optimisation. We suspect that NWChem has issues with the derivatives needed for the analytic frequency analysis.

Is the analytic frequency analysis known to cause these kind of problems? Is the numerical calculations of the frequencies of equal quality to the analytical?

Thank you in advance!