CCL:G: Optimization failed, but frequencies are all real!?

 Sent to CCL by: Susi Lehtola []
 On 10/06/2016 06:55 AM, Igors Mihailovs igors.mihailovs0 .. wrote:
 Dear CCL followers,
 I have got something very unusual to me in a geometry optimization. The
 optimization with the correct force constants (/Opt=CalcAll/ in
 /Gaussian/) resulted in error because the forces were way above the
 threshold (0.025-0.04, threshold 0.00045 for max force) and
 displacements were also high, but /rms/ one was closer to the threshold.
 But what shocked me was that all the frequencies were real (shown
 positive)! All eigenvalues seem to be normal, virial theorem is also
 satisfied. So I conclude this is a local energy minimum (the energy is
 still higher than that for input geometry). But how can the forces be of
 such high magnitude in a minimum on PES? Or is this some feature of the
 implemented method of Hessian diagonalization or whatever else? I am
 absolutely confused, so I decided to write to the CCL for those having
 much more experience than me.
The computation of frequencies isn't physical if the gradient does not vanish. After all, physical frequencies are obtained in the case where you're sitting at the bottom of a parabola, and displacements in a direction result in a force opposing the displacement, which leads to periodic motion.
If you have a gradient, then there is no equilibrium about which the oscillations could happen. The "real frequencies" are just saying something about the local curvature of the potential energy surface.
 Mr. Susi Lehtola, PhD             Chemist Postdoctoral Fellow
 susi.lehtola|,|   Lawrence Berkeley National Laboratory  USA