Ceilidh ... Re^9: How to get rid of Imaginary Frequencies.....

Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly

We are NATO-free zone

Re^9: How to get rid of Imaginary Frequencies.....

Dear Thomas,
Oh I am sorry, I didn't specify my question properly. I meant how to SEARCH for Imaginary freqs. from the HESSIAN OUTPUT file. And I am glad to say I have got the answer. I am able to get ALL the frequencies displayed in one shot, which makes it easier for me read "I".
Thomas Sir I mus thank you for your MOST VALUABLE idea of enhancing the DFT grid and NCONV one. Thanks so much. Could you provide me your personal email too, if u don't mind?

Regards,

On Mon Feb 18 '13 1:04pm, Thomas wrote
--------------------------------------
>Dear Siddheshwar,

>That information can only be obtained by a Hessian calculation. I realize this can be an expensive calculation, but it is the only way to know for certain that your geometry really is in a minimum.
>
>
>Kind regards,
>Thom
>
>
>
>On Sat Feb 16 '13 9:14am, Siddheshwar Chopra wrote
>--------------------------------------------------
>>Dear Thomas,
>>As you had suggested me to tryout a better DFT grid. I have done the same and got the output. Sir, IS there any command which I can add to the input file, to get the information WHETHER there is any IMAGINARY freq. or not? Or HOW do I determine In a easy way about the presence of IMAGINARY FREQS.???

>>Regards,
>>
>>
>>
>>On Fri Feb 15 '13 4:41pm, Thomas wrote
>>--------------------------------------
>>>Dear Siddheshwar,

>>>OPTTOL in $STATPT specifies the convergence criterion for the geometry optimization. NCONV in $SCF specifies the convergence criterion for the SCF procedure. If NCONV is not set tight enough, it could be that the wave function isn't optimized far enough for the gradient to be reliable. And when the gradient isn't reliable, your geometry optimization might not converge.

>>>The default value of NCONV is 6, which is good enough for most cases. In your case it might not be however, hence my suggestion to increase to 8.

>>>Note that these keywords are described in the manual as well.

>>>>Correction< The default value of NCONV is actually 5 (except for CI and MP2 gradients). For this reason, I expect that NCONV=7 is already accurate enough.
>>>
>>>
>>>Kind regards,
>>>Thom
>>>
>>>
>>>
>>>On Fri Feb 8 '13 12:01pm, Siddheshwar Chopra wrote
>>>--------------------------------------------------
>>>>Dear Thomas,
>>>>As suggested by you, I will definitely add these higher accuracy options. But Thomas DOES NCONV=8 mean the same as OPTTOL=10E-8??? or is different?? The second line as you said is for a better DFT grid.

>>>>Regards,

>>>>$SCF NCONV=8 $END
>>>>$DFT LMAX=41 NRAD=99 $END
>>>>
>>>>
>>>>
>>>>On Thu Feb 7 '13 6:36pm, Thomas wrote
>>>>-------------------------------------
>>>>>Dear Siddheshwar,

>>>>>I understood from your earlier post that you were able to find an equilibrium geometry, but if Firefly states that too many steps were take, it means that Firefly was not able to locate a minimum in the amount of steps that was specified with NSTEP. I am not familiar with your system, but I can get the geometry to converge with RM1, so I will assume that it should also be possible to do so with DFT. My suggestion is to increase the accuracy. I see you have already specified INTTYP=HONDO ICUT=11 ITOL=30, which is good. In addition, I would like to suggest:

>>>>>$SCF NCONV=8 $END
>>>>>$DFT LMAX=41 NRAD=99 $END

>>>>>The first line specifies a tighter SCF convergence, which should lead to better gradients. The second line specifies a better DFT grid.

>>>>>Hope this helps.
>>>>>
>>>>>
>>>>>Kind regards,
>>>>>Thom
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>On Thu Feb 7 '13 5:43pm, Siddheshwar Chopra wrote
>>>>>-------------------------------------------------
>>>>>>Dear Thomas,
>>>>>>Thank you for the information regarding optimization. I would like to inform you that I first find a local minimum using molecular mechanics with best tolerance=10^-6. Then i use this geometry as input to firefly with DFT and tolerance more tightly (10^-7/8). It is now in this state, I am unable to get global minimum. Hence i amm forced to change the nstep and opttol values, while keeping everything else same.
>>>>>>Yes Thomas I have used the optimized coordinates for the next run. :-)
>>>>>>Also Thomas when i use nstep=100 , i get the error.....too many steps taken... this problem is not even caught beforehand by checkrrun :-(..
>>>>>>Please help
>>>>>>Regards,
>>>>>>
>>>>>>
>>>>>>
>>>>>>On Thu Feb 7 '13 2:53pm, Thomas wrote
>>>>>>-------------------------------------
>>>>>>>Dear Siddheshwar,

>>>>>>>In the output of a calculation, the first few lines show what your input looked like. I copy-pasted those lines into a new input file together with the structure you performed your Hessian calculation on, changed the value of RUNTYP to GRADIENT, and then ran the job. Doing so, I saw a large gradient, which means that your geometry is not in a minimum. I could also have used RUNTYP=OPTIMIZE to see the same thing (as a geometry optimization always starts with a gradient calculation on your input geometry).

>>>>>>>Of course I believe you when you say that you did a geometry optimization. :-) However, did you really perform the geometry optimization and the Hessian calculation with the same functional, basis set, etc.?

>>>>>>>Also, and I would just like to ask this to rule out any possibilities (forgive me if it is a silly question); did you, after your geometry run completed, copy/paste the equilibrium geometry from the optimization output into the input file for the Hessian calculation?
>>>>>>>
>>>>>>>
>>>>>>>Kind regards,
>>>>>>>Thom

>>>>>>>PS I should point out that a RUNTYP=GRADIENT run is never a substitute for a full Hessian calculation, but it can be used as an... 'initial check'. Typically, there is no need to do such a check, except in confusing cases such as this one. :-)
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>On Thu Feb 7 '13 12:54pm, Siddheshwar Chopra wrote
>>>>>>>--------------------------------------------------
>>>>>>>>Dear Thom,
>>>>>>>>I am eager to know about your statement "When I start a geometry optimization form your structure, I find that the geometry you use in your Hessian calculation isn't a minimum geometry yet.".. Thomas how do you get an idea that it is a transition state BEFORE EVEN running hessian calculation? is there any method I can use to do know its local minimum nature before running a COSTLY hessian calc.?
>>>>>>>>Yes, the file has imaginary freqs. but I would add here that I DID run RUNTYP=OPTIMIZE, before going for HESSIAN calc. And when I completed the OPTIMIZATION job, I got the message, equilibrium geometry reached. What I knew is that I MUST run hessian calc. to know if it is a global minimum or not..
>>>>>>>>However, I read somewhere that we can somehow find the global/local minimum nature by looking at the structure.. PLEASE HELP..

>>>>>>>>Regards,
>>>>>>>>
>>>>>>>>
>>>>>>>>On Thu Feb 7 '13 11:58am, Thomas wrote
>>>>>>>>--------------------------------------
>>>>>>>>>Dear Siddheshwar,

>>>>>>>>>When I start a geometry optimization form your structure, I find that the geometry you use in your Hessian calculation isn't a minimum geometry yet. You must always first optimize your structure with a RUNTYP=OPTIMIZE job before you can obtain a Hessian free of imaginary frequencies.

>>>>>>>>>Note that you should always use the exact same settings during the optimization and Hessian calculation -- basis set, functional, and accuracy settings should be identical between the two. This is because different settings typically lead to different minima. A minimum calculated with pc-2 is different from a minimum calculated with 6-31G(d,p), and the Hessian will only be free of imaginary frequencies when the geometry really is in a minimum.

>>>>>>>>>Hope this helps.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>Kind regards,
>>>>>>>>>Thom
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>On Thu Feb 7 '13 6:54am, Siddheshwar Chopra wrote
>>>>>>>>>-------------------------------------------------
>>>>>>>>>>Dear Thom,
>>>>>>>>>>As asked by you, please find attached a sample file, which contains four imaginary frequencies..

>>>>>>>>>>Regards,
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>On Wed Feb 6 '13 11:36am, Thomas wrote
>>>>>>>>>>--------------------------------------
>>>>>>>>>>>Dear Siddheshwar,

>>>>>>>>>>>Could you please post the complete output of your Hessian calculation? This would allows us to see what kind of imaginary frequencies you're finding.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>Kind regards,
>>>>>>>>>>>Thom
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>On Wed Feb 6 '13 10:56am, Siddheshwar Chopra wrote
>>>>>>>>>>>--------------------------------------------------
>>>>>>>>>>>>Dear Sir,
>>>>>>>>>>>>This in regard to my problems with getting rid of imaginary frequencies in the vibrational data. I have tried changing the basis sets, changed the ITOL, ICUT values... Please suggest me what should I do to ge trid of imaginary frequencies?? Currently I am trying out PC-2 basis set. A sample input commands are as follows:

>>>>>>>>>>>> INPUT CARD> $SYSTEM MEMORY=12065831 aoints=dist TIMLIM=525600 KDIAG=1 $END
>>>>>>>>>>>> INPUT CARD> $CONTRL INTTYP=HONDO ICUT=11 ITOL=30 RUNTYP=HESSIAN SCFTYP=RHF $END
>>>>>>>>>>>> INPUT CARD> $CONTRL DFTTYP=B3LYP COORD=UNIQUE NZVAR=132 $END
>>>>>>>>>>>> INPUT CARD> $CONTRL MAXIT=10000 MOLPLT=.t. PLTORB=.t. $END
>>>>>>>>>>>> INPUT CARD>! $CONTRL EXETYP=CHECK $END
>>>>>>>>>>>> INPUT CARD> $BASIS GBASIS=N31 NGAUSS=6 NDFUNC=1 NPFUNC=1 $END
>>>>>>>>>>>> INPUT CARD> $STATPT opttol=10E-7 NSTEP=200 Method=GDIIS HSSEND=.TRUE. $END
>>>>>>>>>>>> INPUT CARD> $SMP HTTFIX=.f. $END
>>>>>>>>>>>> INPUT CARD> $p2p p2p=.t. mxbuf=2048 dlb=.t. $END
>>>>>>>>>>>> INPUT CARD> $SCF dirscf=.t. fdiff= .f. diis=.t. $END
>>>>>>>>>>>> INPUT CARD> $DFT HFX(1)=0.15 $END
>>>>>>>>>>>> INPUT CARD> $ZMAT DLC=.t. AUTO=.t. $END
>>>>>>>>>>>> INPUT CARD> $FORCE NVIB=2 $END

>>>>>>>>>>>>Please help me.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>Regards,
>>>>>>>>>>>>

Tue Feb 19 '13 7:57am

This message read 1080 times