Vibrational analisys

55 views
Skip to first unread message

Lucas Lodeiro

unread,
Oct 11, 2022, 6:35:08 PM10/11/22
to cp...@googlegroups.com
Hello all,

I have a detailed doubt about the files and data printed for a vibrational analysis. AFAIK, the calculation is done by numerical differentiation of gradients (differentiation in both directions and average them) so for a molecule with N atoms it needs 6N calculations, which is clearly printed in the main output. Also, each replica/configurationn (6N) is computed and a particular output is printed for each one.
Up to this point everything looks fine, but when I see the particular replica output, there are 5 energy computations into the output... I would expect just one.
Why is this happening? Am I missing something?

Regards - Lucas

Marcella Iannuzzi

unread,
Oct 12, 2022, 3:51:37 AM10/12/22
to cp2k

Hi Lucas

The number of output files of the type ${projectname}-r-#.out
should be equal to REPLICA| layout of the replica grid, number of groups
and not to 6N

Kind regards
Marcella

Lucas Lodeiro

unread,
Oct 13, 2022, 11:38:28 AM10/13/22
to cp...@googlegroups.com
Hi Marcella,

I expressed it badly, I want to say that each displaced atomic configuration needs an energy+gradient calculation, this is 6N energy+gradient evaluations, plus the minimum energy geometry. Of course the number of output files is equal to the replica layout, but my problem is that the number of energy+gradient evaluations into all these replica outputs is not equal to 6N.
A practical example, I have an 8 atom isolated molecule, so it needs 48 energy+gradient evaluations which are printed in the main output correctly. I launched this calculation in a node with 20 cores, so the replica layout is 20 and there are 20 output files. When I see these output files I check there are 5 energy computations into each output, so there are 5*20=100 energy evaluations. This is the point which makes me worried. I do not understand why there are that huge amount of energy evaluations when 6N=48. 

Regards - Lucas

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/90776d8b-086e-4772-8683-8dd97b5afc8dn%40googlegroups.com.

Marcella Iannuzzi

unread,
Oct 13, 2022, 12:19:38 PM10/13/22
to cp2k
Dear Lucas

In the main output you should find how the calculations are distributed among the processors' groups. 
It is possible that some calculations are done more than once to avoid that processors remain idle. 
What happens if you run the 48 calculations with 24 cores? 
Kind regards
Marcella

Lucas Lodeiro

unread,
Oct 13, 2022, 1:07:15 PM10/13/22
to cp...@googlegroups.com
I launch the calculation with 12 nodes with 5 energy evaluations per replica output (12*5=60) and 24 nodes with 3 energy evaluations per replica output (24*3=72). Also the resulting frequencies differ in 0.1cm⁻¹ at most.

Reply all
Reply to author
Forward
0 new messages