Hi Verma,

These are very small to be called imaginary vibration frequencies (these appear due to numerical errors). We get worried when we get imaginary freqs a bit larger. Of course it gives comfort if you get all positive freqs, however, in this specific case there should not be considerable problem on your vibrational analysis. I personally advise you to try to remove them by trying the following few tricks:

1) use tighter convergence criteria (possibly also use a fine grid), or 

2) visualize the vib freqs (may be using gaussview, chemcraft, ...). Identifying the imaginary frequencies. Then modify the bond length or angle related to these imaginary frequencies a little bit and save this modified geometry to rerun your calculation. Or chemcraft also allows to freeze these motions and then after you save this geometry for a rerun.

Since you have already optimized geometries, then optimization after these modifications will not take longer.

Good luck!

Hi Verma,

These are very small to be called imaginary vibration frequencies (these appear due to numerical errors). We get worried when we get imaginary freqs a bit larger. Of course it gives comfort if you get all positive freqs, however, in this specific case there should not be considerable problem on your vibrational analysis. I personally advise you to try to remove them by trying the following few tricks:

1) use tighter convergence criteria (possibly also use a fine grid), or 

2) visualize the vib freqs (may be using gaussview, chemcraft, ...). Identifying the imaginary frequencies. Then modify the bond length or angle related to these imaginary frequencies a little bit and save this modified geometry to rerun your calculation. Or chemcraft also allows to freeze these motions and then after you save this geometry for a rerun.

Since you have already optimized geometries, then optimization after these modifications will not take longer.

Good luck!

Thanks Dr. Taye.

I have had one structure and used opt+freq several times but it gave the same answer. However, latter on I tried using scf=xqc and scf=yqc but.....the same answer. Now I'm gonna try freeze option.

Dear Verma,

The best answer is already provided by Dr. Demissie and nothing to add. I think you don't get his first point (about tighter convergence). If you have doubt you can use opt=tight along with Int=ultrafine keyword in Gaussian09 input file. This will use finer grid along with tighter convergence criteria. If your structure is a true minima, you'll get rid of those small imaginary freq. All the Best.

Thanks for your answer Sir.

I meant that "prior to write this question on researchgate I tried those but didn't succeed.". Anyway thanks for the keywords. I'll try these. 

One question:

Does Gaussian 09 have any keyword by which we can get the imaginary frequency greater than some limit? let's say I want an imaginary frequency of greater than 200-300 cm-1 (I know it differs for different situation, e.g. C-H, C-O imaginary freq) because what happens is that for TS search it, many times, gives imaginary frequency of 40-150 cm-1 and these happens not to be useful for IRC calculation. 

To best of my knowledge there is no such keywords. Although I found sometimes the magnitude of imaginary freq depends on the level of theory involved (specially in DFT). If you are sure that imaginary freq lies within a higher range than what you're getting, you should start the optimization with different guess. Also, TS finding is not a blind process, there are methodical ways and diverse techniques to find out the right TS.

Hi Verma,

If you are using the 'TS' key word to search for TS structure, then it gives you any single imaginary freq (it depends on your luck in this case). It is always better to visualize the optimized TS structure how it looks like for the vibration of the imaginary freq. But still we have a keyword in gaussian to 'accurately' search the TS structure. Follow these instructions:

1) Optimize the reactants

2) Optimize the product/s

3) Prepare a TS guess structure based on your common sense (you can prepare it from the optimized products). 

In all cases, be careful about the charges and multiplicity. In addition, don't change the orders of the atoms in all geometries. Once you have these geometries, prepare the input as follows (be careful about the blank lines). I am simply putting examples, not real structures (change the CPU infos and if you are dealing with bond breaking-forming reactions use the unrestricted formalism (for eg. uB3LYP)):

%NProcShared=48

%mem=2000MB

#p opt=QST3 freq=noraman b3lyp/6-311+G(d,p) 

....give blank line here.....

Reactants

....give blank line here.....

0 1

N 1.957974581 3.842732356 1.025280837

N 1.299002581 1.833490356 2.105917837

C 2.207072581 2.760225356 1.774245837

C 0.704445581 3.997741356 0.586702837

....give blank line here.....

Products

....give blank line here.....

0 1

N 2.137089989 3.151552253 0.195878416

N 1.444826989 1.070085253 1.106660416

C 2.365650989 2.003814253 0.847809416

C 0.888594989 3.395472253 -0.228898584

....give blank line here.....

Transition state

....give blank line here.....

0 1

N 2.376011590 6.227357012 6.847501401

N 1.717039590 4.218115012 7.928138401

C 2.625109590 5.144850012 7.596466401

C 1.122482590 6.382366012 6.408923401

....give blank line here.....

....give blank line here.....

....give blank line here.....

Good luck!

QST methods (QST2 and QST3) are best methods to get TS. But there is no input command to get desired imaginary frequencies using Gaussian. If u r using OPT=TS, u will get a random TS. It may be your desired TS or not. You cannot assure of it. So, to get better TS go for QST methods as suggested by Dr.Demissie.