Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

Re: negative frequencies after CI opt and huge PUNCH file

Alex Granovsky


On Thu Jun 24 '10 0:52am, Jonas Baltrusaitis wrote
>I have somewhat awkward problem: I did CI geometry optimization and frequency calculations. All of my comparable HF geometries had no negative frequencies, whereas all of the CI ones had one or two low negatives. Interestingly, all the SP upon the atom displacement are of higher energy than the initial one so I do not know how do these negatives come about...

This may be just an artifact caused by nvib=1.
However, the (semi-numerical) Hessian is formed using information
on how the energy gradient is changed along each ivib, not the
energy itself, and so strictly speaking having each of SP upon
the atom displacement of higher energy than the initial one
simply cannot guarantee that all the frequencies are positive.

>Anyway, I decided to do NVIB=2 and see if it's some accuracy error.

You can also try to increase overall accuracy of each relevant step:

1. 2-e integral evaluation ($contrl icut=10 or 11, inttyp=hondo)
2. SCF ($scf nconv=6 or better)
3. CIS Davidson diagonalization ($cis davcvg=1d-6)
4. CPHF ($cis cptol=1d-6, icutcp=12)

> The problem is that I have generated a 5 Gb PUNCH file and nothing can open it to extract data for MOREAD. Of course, IRCDATA provided $VIB of NVIB=1 for restart.

>So I have two questions:
>-what about those negative frequencies after CI opt? Numeric error and NVIB=2 would solve it or reoptimization is necessary?

Both cases are possible, see above.

>-how do I extract last orbitals for MOREAD form that gianormous file? Any Linux based commands/tricks? I will need it for either geometry reoptimization or NVIB=2 anyways

Under Linux, use "split" command, read "man split" first.
You can also use "tail".

For our Windows users, I can provide a small utility that saves
the desired part of the punch file starting from its very end.

Alex Granovsky



[ Previous ] [ Next ] [ Index ]           Mon Jun 28 '10 2:56pm
[ Reply ] [ Edit ] [ Delete ]           This message read 848 times