Vibrational Analysis

[Ali]

Dear CP2K Users,

I wanted to make sure I understood some data I am getting from a
vibrational analysis run type. I am doing the analysis on 4 carbon
atoms of my system and am searching for frequencies in the range of
1-3000. From my output I get the following at the end:

MS| ITERATION STEP 172
MS| TRACKED MODE 716.401860cm-1 NOT CONVERGED
MS| TRACKED MODE 631.347431cm-1 NOT CONVERGED
MS| DAVIDSON ALGORITHM CONVERGED
MS| TRACKED FREQUENCY (1) IS: 314.918925 cm-1
MS| TRACKED FREQUENCY (2) IS: 815.338097 cm-1


-------------------------------------------------------------------------------
- FREQUENCY AND CONVERGENCE
LIST -
FREQUENCY MAXVAL CRITERIA NORM CRITERIA
CONVERGENCE
5792.243 0.799E-04
0.550E-03 NO
3748.327 0.317E-04
0.241E-03 NO
2114.956 0.108E-04
0.581E-04 NO
2066.005 0.142E-04
0.760E-04 NO
2009.355 0.973E-05
0.632E-04 NO
1984.510 0.981E-05
0.613E-04 NO
......

Followed by a whole range of other frequencies all of which are not
converged. (A lot of them are outside the range I requested so I am
assuming thats the default). After the 172nd iteration the davidson
algorithm converges yielding 2 frequencies: 314 and 815 cm^-1. Are the
vibrations associated with these 2 frequencies, the normal modes
associated with those 4 Carbon atoms? These frequencies do not appear
in the Frequency and Convergence List later on though. Am I missing
something?

Any help will be appreciated. Is there any easy way to visualize the
normal modes associated with these frequencies (without having to use
Molden)?

Ali

[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

[Ali]

Hi Flo,

Below is a section of my input file. You are correct in that I am
conducting a vibrational analysis on 4 atoms together. From your email
it seems like you were saying that using Frequency, Range and Involved
atoms are mutually exclusive. I actually tried just doing the involved
atoms and I got an error stating that I needed to have range or
frequency in the input.

I am actually interested in determining a frequency that is associated
with all 4 atoms together which is why I chose 4 of them. I guess its
possible that if I did it seperately on the 4 different atoms that the
mode I get from all 4 together will appear in one of the top 4
seperately.

&VIBRATIONAL_ANALYSIS
FULLY_PERIODIC T
!NPROC_REP 2
&MODE_SELECTIVE
FREQUENCY 500
ATOMS 115 117 120 122
EPS_MAX_VAL 1.0E-07
&END

Ali

[flo]

Hi Ali,

yes you can get the frequency which is associated with these four
atoms but you forgot a keyword. maybe for to claryfy for everbody who
may read this topic I will just explain the different keywords in your
input file:

&VIBRATIONAL_ANALYSIS
! important!!! if you specify the keyword intensities you will have to
add the
! keyword for evaluating dipoles in the FORCE_EVAL section

FULLY_PERIODIC T
! Only important for a full vibrational analysis

!NPROC_REP 2
! Important for multiprocessor runs with n processors. It will
perform
! n(processors)/nproc_rep calculations at the same time.
! with mode_selective a block davidson algorithm is started.

&MODE_SELECTIVE
FREQUENCY 500
! frequency you want to track (as Ali said:
! using Frequency, Range and Involved
! atoms are mutually exclusive)
! in your case Ali, you will have to use involved_atoms 115 117 120
122
! to get the frequency associated with these atoms

ATOMS 115 117 120 122

! atom numbers for the initial guess (might help to converge faster).
! For the initial guess a random vector is construct only having
components
! on these atoms

EPS_MAX_VAL 1.0E-07
! convergence criterium for the Davidson algorithm.
! The default is already tight enough, since the second derivative is
evaluated analytically.

&END

For you it means jsut remove frequency and add INVOLVED_ATOMS 115 117
120 122.

One importtant point for your calculation might be:
Since I haven't removed the translations and rotations from the
algorithm (in this algorithm it's a bit more difficult), I restricted
the frequencies for the involved atoms part to values >400cm-1
(otherwise you will many times get the translations). Maybe your
frequency is lower. Just to be sure, you will get it, you can change
the code as follows:

grep '400' mode_selective.F
everywhere it occures just change the value to the threshold you want
and compile cp2k (should be fast, because it's a tiny change).

Hope this helps
Flo

[Ali]

Thanks Flo.

That was very helpful.

Ali