Getting rid of imaginary frequencies

[Pfister]

Hi!

I made a couple of SCS-CC2 calculations with NumForce (I know, expensive, but it was necessary). In principle it works now, but in 1 calculation I have an imaginary frequency, which makes to us no sense. Normally (to my knowledge) I should either change my structure a little bit, optimise it again to get a different extremum, and aplly NumForce. AND/OR I switch off symmetry to provide a former forbidden relaxation path along the imaginary frequency.
OK, I don't have symmetry, and if I visualize the imaginary frequency it is mainly 1 atom, which moves. Therefore we decided to simply delete all calculated points in /KRAFTWERK and start NumForce again, because recalculating everything would be very expensive. This is what happened: He didn't recalculate the missing points. It simply assumed they were there, calculating nothing and giving me the same output as I had before. Then I deleted all files in the /numforce directory but kept /KRAFTWERK. Still no result, but this time the output told me, that I have no points at all:
"step xm01 started  (#1 of )
ls: lockhost.*: No such file or directory"

Any ideas?
Thank you in advance

[lasermichel]

Hi (Johannes?),

the best guess you can get for the real minimum comes from the screwer module. Just call it in your working directory and it lets you distort the molecule along the mode with the imaginary frequency. Mind the sign, which screwer proposes and don't select a too high temperature for the distortion. This depends on the size of the molecule. For 20-30 atoms I found temperatures fo 50 - 100 K appropriate. This is really the only systematic way to find the way dwon from the saddle point.

Note: NumForce transforms your orbitals into C1 symmetry and you should define your geometry in C1, since distorting along cartesian coordinates breaks your symmetry anyhow.

Good luck

Michael

[Pfister]

Just for those guys, who might face the same problem, here is the solution:
Go into your NumForce/Kraftwerk directory and type in "grep NaN *.log"
If you find something then these calculations are somehow corrupted. Delete all these files as well as their XXX.grad and XXX.dipole and recreate them, but do not write anything into it (like xm20.log with file size 0). The NumForce script then realises that there are still some points to be calculated. Rerun NumForce and hopefully everything works fine.

[doguapo]

Hello,

I am using Turbomole 6.1 to calculate the interaction energies of certain dimer complexes.  I'm having issues with the geometry optimizations:

Whenever I run aoforce for a variety of arene complexes, I almost ALWAYS have 3 negative eigenvalues.  I have tried running screwer/vibration along any of the modes - at one point, i was able to obtain only 2 negatives rather than 3 by doing this.  However, re-running vibration on the other modes hasn't proven successful (i always ran vibration at room temp).

I'm sure this is a systematic error, most likely due to my (so far) brief experience with Turbomole.  However, I'm pretty stuck and I haven't been able to find too much elaboration on vibration calculations in the manual.

Any help would be greatly appreciated.

PS: using TPSS functional and TZVPP basis set, in case this information may be relevant.

[Arnim]

Hi doguapo,

how large are your negative frequencies?
If they are small (<100 cm-1), then it could be an artifact of the numerical grids used in DFT calculations. To get rid of this it often helps to use a larger grid (like m4) and apply more strict converence criteria.

Hope that helps,
Arnim

[doguapo]

Arnim,

mode 1: -70.88 cm-1, mode 2: -53.28 cm-1, mode 3: -26.52 cm-1

Thank you very much for the suggestion.  I will try that now and post the results!

[doguapo]

Weird, it's still happening.  I've tried so many things, and I've so far had no luck correcting for this systematic error.  Any other ideas?

Edit: I also want to mention that I've taken the final xyz coordinates of this structure (containing 3 negative eigenvalues) and upon running a dscf + aoforce job, that output contains only 0 or positive eigenvalues.  No imaginary frequencies are produced.  Is this common?

[amavr78]

Dear doguapo ,
Numforce calculations are very sensitive on the Scf convergence criteria , grid and on the step size of the displacement.
I had used (scfconv=8, denconv=1E-8, grid=m5 and weight derivatives) for DFT calculations and I was still getting one imaginary frequency . (<-30 cm-1)
Aoforce was giving none.
When I changed step size to 0.01 au (default is 0.02au), then no negative frequency was calculated.
Same conclusions also for MP2 calculations. When I was running numforce with the "-d 0.01" option, I got no negative values, whereas with the default settings I got one negative, for exactly the same structure!

So you can try this  and see if the negative frequency disappears .

Andreas

[doguapo]

Andreas,

thank you for the tip - I will give it a shot.  one question, though - doesn't denconv only apply for tddft calculations?

thanks!

[amavr78]

Not necessary.
Denconv is an additional criterion for the Scf step. You define a convergence criterion for the density matrix.
It should be used when  second derivatives are to be calculated (like in numforce or escf) or for MP2.

Andreas

[doguapo]

Thanks for all the help, but I tried these suggestions and nothing has worked.  I'm only using aoforce after the optimizations.  Maybe it'd help to indicate that I'm adding "$disp" to the control file to take into account dispersion in my systems, but I'm sure I've mentioned that before.  Here are some parameters that I've included in case this may shed any light.  Thanks in advance!

scfdamp step=0.010, min=0.001, scfconv 8, denconv 1E-8, grid: m4, RI m: 3000, $disp, thre=1E-7 (from stp for tighter convergence criteria)

after optimization...
mode 1: -70.40 cm-1
mode 2: -51.71 cm-1
mode 3: -20.42 cm-1

[uwe]

Hi,

did you also try Turbomole 6.2 ? There have been some problems with DFT+D and aoforce, and I wonder if the latest update still shows the negative eigenvalues.

Regards,

Uwe

[doguapo]

Yes, I've run these calculations on Turbomole 6.2 on two separate clusters.  The problem still remains.

[uwe]

Hi,

in this case (and to check if there still is a problem with DFT+D or just a complicated structure) please contact the Turbomole support (turbomole at cosmologic dot de)!

Regards,

Uwe

[doguapo]

hi all,

just wanted to post the resolution to my problem as i finally figured it out...

it looks like aoforce with $disp in the control file (to take into account dispersive forces like pi stacking) doesn't cooperate.

1. optimize structure with $disp
2. run aoforce separately after removing $disp from control file.

thanks for all the input and ideas!