PC GAMESS/Firefly-related discussion club



Learn how to ask questions correctly


Re^3: False negative frequencies

Slawomir Janicki
slawomir.janicki@comcast.net


Alex checked my work (thank you, Alex!) and it turns out that the geometry optimization didn't reach the minimum. Because I was using PROJCT=.T. while checking for imaginary frequencies to test the geometry convergence I got a false positive.

Slawomir

On Mon Nov 23 '09 4:29am, Slawomir Janicki wrote
------------------------------------------------
>The system is the same, the Hessian calculations were done at the final geometry from the geometry optimization run.

>Visualization is difficult when I have hundreds of files to work up. I will try the other ideas first.

>Slawomir

>On Sun Nov 22 '09 8:34pm, Alex Granovsky wrote
>----------------------------------------------
>>Hi Slawomir,

>>Sometimes, one can get small negative (imaginary) frequencies
>>that actually correspond to rotations or translations. This is
>>not too unusual with numerical Hessians. This can be typically
>>avoided using nvib=2 with smaller vibsiz (e.g., 0.005 or so)
>>while increasing overall precision of calculations (more precise
>>integrals, DFT grids, tighter cutoffs throughout, etc...). However,
>>this usually does not seriously affect the computed values of "real"
>>vibrational frequencies. It is also very helpful to visualize
>>the vibration of question, and also examine T+R vibrations before
>>projection (i.e., with PROJCT=.f.).

>>However, what looks really strange are your numbers for other frequencies;

>>Analytic:
>>>            237.43
>>>            237.43
>>>            268.28
>>>            544.38

>>Numeric:
>>>            202.41
>>>            202.51
>>>            352.25
>>>            352.31

>>Is this exactly the same system? at the same geometry?
>>and exactly the same type of computations?

>>Regards,
>>Alex
>>

>>On Sun Nov 22 '09 5:38pm, Slawomir Janicki wrote
>>------------------------------------------------
>>>Hi,

>>>I was comparing analytical and numeric methods for hessian runs, and I found that in one case the numeric methods produced negative frequencies:

>>>geometry optimization run:
>>> $STATPT NSTEP=500 HSSEND=.F. TRMIN=0.01 METHOD=GDIIS NOREG=5
>>> OPTTOL=0.0000001 $END

>>>hessian runs:
>>>analytical:
>>> $FORCE METHOD=ANALYTIC PROJCT=.TRUE. VIBANL=.TRUE. PRTSCN=.TRUE. SCLFAC=1 $END
>>>gave:
>>>                          1           2           3           4           5
>>>       FREQUENCY:         0.00        0.00        0.00        0.00        0.00  

>>>and

>>>          FREQUENCY   ENTROPY   %-CONTRIBUTION
>>>          ---------   -------   --------------
>>>            237.43     1.822           26.50
>>>            237.43     1.822           26.50
>>>            268.28     1.607           23.38
>>>            544.38     0.556            8.09
>>>
>>>
>>>numeric 1 step:
>>> $FORCE PROJCT=.TRUE. VIBANL=.TRUE. PRTSCN=.TRUE. SCLFAC=1 METHOD=NUMERIC
>>> NVIB=1 $END
>>>gave:
>>>                          1           2           3           4           5
>>>       FREQUENCY:        14.32 I      0.00        0.00        0.00        0.00  

>>>and

>>>          FREQUENCY   ENTROPY   %-CONTRIBUTION
>>>          ---------   -------   --------------
>>>            207.78     2.063           29.17
>>>            212.70     2.020           28.56
>>>            355.72     1.141           16.13
>>>            355.99     1.140           16.11

>>>numeric 2 step:
>>> $FORCE PROJCT=.TRUE. VIBANL=.TRUE. PRTSCN=.TRUE. SCLFAC=1 NVIB=2
>>> METHOD=NUMERIC $END
>>>gave:
>>>                          1           2           3           4           5
>>>       FREQUENCY:        14.14 I      0.00        0.00        0.00        0.00  

>>>and

>>>          FREQUENCY   ENTROPY   %-CONTRIBUTION
>>>          ---------   -------   --------------
>>>            202.41     2.111           29.15
>>>            202.51     2.110           29.13
>>>            352.25     1.156           15.96
>>>            352.31     1.156           15.96

>>>Is there a way to avoid this? I need to rely on numeric requencies when I can't use analytical hessian.

>>>Slawomir


[ Previous ] [ Next ] [ Index ]           Mon Nov 23 '09 10:25pm
[ Reply ] [ Edit ] [ Delete ]           This message read 975 times