Molcas Forum

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Too many imaginary frequencies (and a potential bug?)

Dear Users,

I'm doing some geometry optimization with RASSCF, and I've noticed that it results in many imaginary frequencies, despite the geometry being optimized in the very same job.
The same procedure with SCF or CASPT2 results in no imaginary frequencies, or maybe just one (which is fine).

Please consider the examples.
https://www.dropbox.com/scl/fi/5rkrw6cy … vwh76&dl=0

In this particular case, there is also a problem occurring in Alaska (after the first CASPT2 step), with a "double free or corruption". I do not get this one in the bigger jobs, and thus I know that CASPT2 generally does fine.

Any ideas?

#8  0x47376b in dmma_free_1d_
    at /opt/OpenMolcas-ff0b5d1c/src/Include/mma_allo_template.fh:348
#9  0x40cb69 in drvg1_
    at /opt/OpenMolcas-ff0b5d1c/src/alaska/drvg1.F90:512
#10  0x408f9c in alaska_
    at /opt/OpenMolcas-ff0b5d1c/src/alaska/alaska.F90:221
#11  0x4062e7 in alaska_super_driver_
    at /opt/OpenMolcas-ff0b5d1c/src/alaska/alaska_super_driver.F90:224
#12  0x403d90 in MAIN__
    at /opt/OpenMolcas-ff0b5d1c/src/alaska/main.F90:28
#13  0x403d90 in main
    at /opt/OpenMolcas-ff0b5d1c/src/alaska/main.F90:18

Last edited by andrewshyichuk (2024-09-27 19:02:20)

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: Too many imaginary frequencies (and a potential bug?)

A geometry optimization does not guarantee that the result will be a minimum, that's why you compute the frequencies after. If you get imaginary frequencies, it may be because the optimized geometry is actually a saddle point. This happens often when your initial geometry has some forced symmetry, and the optimization cannot in general break it. You could use the SHAKE keyword in GATEWAY to break the symmetry at the beginning, or distort the optimized geometry along the imaginary-frequency modes and reoptimize. If the optimization wants to break the symmetry and it shouldn't... maybe your active space or state-average selection is not stable.

On the other hand, in most cases the frequency calculation is (semi)numerical, so the frequencies will not be extremely accurate, and you can find some spurious imaginary frequencies some times.

andrewshyichuk

Member

Registered: 2020-02-13

Posts: 86

Re: Too many imaginary frequencies (and a potential bug?)

In my case, the symmetry is the culprit. I optimize a system that is octahedral, but I do not use symmetry. The code calculates redundant modes and those likely end up imaginary. Be that as it may - a fun Friday evening project turned out to be quite a rabbit hole. I am open to a collab in case you want to help me digging.