Vibrational Analysis

[flo]

Hi Ali,

could you post the part of your input file related to the vibrational
analysis?
>From the output, and what you write, it looks like you used a multiple
definition of the frequencies you are interested in. Using the mode
selective method, you have to chose what you are looking for:
Frequency means you are looking for the (nproc/nrep) frequencies
closest to the selected one.
Range converges all the frequencies in the Range.
Involved_atoms is slightly different: the eigenvectors are projected
on a unit vector constructed out of your chosen atoms. If you chose
many atoms, you will most likely end up with a low frequent mode
(since these are more delocalized, and match the constructed vector
the best). For to clarify since I expect you used this keyword. Using
four atoms won't give you the frequencies where one of them is
involved, it will give you the frequencies where all of them are
involved the most.
>From these three points you have to choose only one, since each
specifies a way how to create your residuum vector.

The output:
Since not only frequencies are present which are converged with
respect to the EPS valuesyou get a list of all the eigenvalues of the
approximate Hessian and how far they are converged. Since there are
two EPS values to specify you egt two values (and the last column
tells you if they are converged with respect to the EPS).

During iteration cycle you can follow which frequencies are the ones
used for updating the Davidson algorithm. And in the end which have
been the last ones chosen by the algorthm.

The output created (the VIBRATIONS.mol file uses molden format. Of
course one could implement different outputs, but since molden is a
freely available software I think it's convenient to use this format.
If you want to use a different Program, you will have to convert it to
a different format.

Of course, the way you select the modes in CP2K can still be improved,
but with the present methods you can cover most of the stuff you are
looking for.
I would suggest: just take one involved atom and use a multiprocessor
run (nproc/nrep=4 or more), in this way you will get the four modes in
which the atom moves most. Afterwards restart the calculation
(MS_RESTART) with the second,third, fourth atom.

Cheers
Flo