TURBOMOLE Modules > Aoforce and Numforce

Stubborn imaginary frequency

(1/3) > >>

marand:
I am trying to compute the lowest singlet excited state vibrations at the CC2/def2-TZVPP level of theory for cis-planar conformer of bithiophene molecule.

The standard procedure I follow begins with geometry optimization for the excited state, then switching from C2v to C1 symmetry, having copied firs the gradient file
obtained during the geometry optimization into the submit directory.

The NumForce module (with level=cc2, central and ecnomic options) is then used and the calculations go smoothly till the hessian, aoforce output files are produced.
There is, however, one imaginary frequency that corresponds to the torsional motion of the rings along the central C-C bond. This normally indicates a saddle point, which is in fact present in the ground state (in which gosh-conformers are the optimum structures). In the S1 state, however, increase coupling between the pi electrons of different rings makes the inter-ring bond acquire a partly double character and rigidifies it, so the structure is expected to be actually planar. This has been reported in literature several times and my own experience also supports such a picture.

In order to check whether the imaginary frequency is some artifact of the numerical procedure or a genuine find, I have conducted DFT calculations (with B3LYP and cam-B3LYP functionals), which both returned all-real frequencies for the S1 excited states. Moreover, I have made a relaxed potential energy scan with respect to the inter-ring rotation and it also shows that the planar conformation corresponds to the minimum on the PES. Finally, I used screwer to distort geometry in the direction of the imaginary frequency normal mode and then reoptimized the structure without any symmetry constraints (C1). The outcome was - again - a planar structure and the same imaginary frequency obtained for it in the NumForce calculations.

It, I think, confirms that the imaginary frequency is some numerical artifact. I tried to remove it by 1. increasing the convergence criteria for the orbitals, 2. changing the size of the geometry distortions in NumForce, 3. switching from ecnomic to cartesian displacements. None of those changed anything.

Please, advise me what can be the reason for such behaviour and how can I possibly get rid of the imaginary frequency. I need to interpret a helium temperature spectra and so I need vg quality of normal modes, including the low-frequency ones.

Yours sincerely
Marcin Andrzejak

 

marand:
One more, but possibly important remark: I obtained the same results using the 7.0.1 and 7.5 versions of Turbomole.

antti_karttunen:
Hi,

you mention that you run NumForce with "central" and "ecnomic" options. But aren't these in a sense conflicting options and if you give central before ecnomic, you may end up with the polyedr option. Could you confirm that you really do get "central" and not "polyedr" type displacements? I recommend running with central differences.

How large is the imaginary mode, by the way?

Best,
Antti

marand:
Well, the number of displacements shows that these are in fact central displacements, but I have already tried running the calculations without the -ecnomic switch, and
the imaginary frequency is still there.

It is not large - i53 cm-1, but it is nonetheless important for my studies to have it removed from the calculations.

Yours!
Marcin

antti_karttunen:
Thanks for the clarification. Seems like a tough case. One question and one more comment:

1) You have used C2v point group symmetry during the excited state optimization. Is there some particular reason why you remove symmetry before running NumForce? Is this because NumForce anyway breaks symmetry and then $excitations keyword used foir C2v will not be valid? I'm just guessing here since I don't remember if I ever ran excited state frequency calculations with ricc2, but what if after the geometry optimization you just put in control file:

--- Code: ---$ricc2
geoopt model=cc2 state=(s1)
$excitations
irrep=a nexc=1
# or possibly few more states in nexc, whatever works best for your system
--- End code ---
And then run NumForce (for the C2v symmetric structure). I'm just thinking if symmetry could help to reduce some numerical noise here.

2) You mentioned increasing convergence criteria for orbitals, but what keywords did you exactly use? Did you try something like

--- Code: ---$scfconv 8
$denconv 1d-8
--- End code ---
or even tighter during the optimization? And then jobex with -gcart 5?

Navigation

[0] Message Index

[#] Next page