Firefly and PC GAMESS-related discussion club

Learn how to ask questions correctly  
We are NATO-free zone

Re^5: Energy rises and oscillate at the end of optimization

Ilya Ioffe

Dear Oleg,

a) As was already pointed out, one can hardly obtain really tight gradients with traditional PCM. One can try the new variational CPCM technique recently implemented in Firefly - consult the recent manual for keywords. b) The problems might be rather due to a small gap between the ground and the excited states. If this is the case, the adequacy of DFT in itself may be not quite sufficient. Anyway, more output may be desirable, at least the orbitals.
c) You've set quite tight cutoffs but without requesting more precise integral routine (HONDO) and without improving SCF convergence, DFT integration grid, and PCM surface accuracy. You may consider enhancing all aspects of accuracy in a more consistent manner.
d) In principle, one can try reducing the allowed geometry change (trmax, dxmax) to damp the oscillations.
e) Finally, is the problem really severe? The oscillations are moderate, both in terms of energy and geometry, and are largely due to the torsions, which is quite expectable. Does one really need higher precision?

Best regards,


On Fri Sep 12 '14 10:34pm, Oleg Levitskiy wrote
>Thank you, Pavlo!
>Now I've started hessian computation though I don't expect that it will finish soon. Analytical calculation is precluded in this case...

>On Fri Sep 12 '14 5:30am, Pavlo Solntsev wrote
>>Dear Oleg.

>>It looks like you starting geometry very close to a transition state. I would recommend you to calculate hessian first and then perturb imaginary mode if it exists. Then try run optimization starting from exact hessian. Algorithms for transition state search and minimum are similar.

>>Hope this helps.

>>On Thu Sep 11 '14 8:48pm, Oleg Levitskiy wrote
>>>Dear Pavlo,
>>>optimization with corrected parameters did 9 steps and unfortunately almost all of them with energy encrease. It seems that convergence to the predefined gradient value can't be achieved in case of applying solvent modeling as you've supposed...
>>>Nonetheless I attached the resulting out in hope of appearing of any new ideas.

>>>Thanks again,

>>>On Thu Sep 11 '14 5:18am, Pavlo Solntsev wrote
>>>>I would recommend you try these parameters first:
>>>> $contrl scftyp=rhf dfttyp=pbe96
>>>> runtyp=OPTIMIZE icharg=-1 nprint=-5
>>>> maxit=100 ICUT=11      ITOL=30 NZVAR=168 $end
>>>> $system mwords=100 $end
>>>> $SCF DIRSCF=.true. maxdii=20 $end
>>>> $guess guess=moread $end
>>>> $ZMAT DLC=.T. AUTO=.T.$end
>>>> $p2p p2p=.t. dlb=.t. $end
>>>> $basis gbasis=TZV $end
>>>> $PCM PCMTYP=DPCM SOLVNT=input RSOLV=2.155 EPS=35.688 EPSINF=1.806874 $END
>>>> $PCMCAV RIN(58)=1.956 $END

>>>>Before apply a solvent correction, you need to optimize the geometry in a gas phase. Make sure you use a right functional. I am skeptical about pbe96. Do not expect very tight gradient for the geometry with the solvent correction. Usually 10^-4 is ok.

>>>>On Tue Sep 9 '14 3:30pm, Oleg Levitskiy wrote
>>>>>Dear Firefly Users,
>>>>>I've tryed to perform an optimization of anionic complex of Ni (DFT/PBE96, PCM) and faced a problem with convergence. Energy has unexpectedly risen and fluctuate spontaneousely from step to step. I've tryed to increase accuracy (ICUT, ITOL values and also DLCTOL and ORTTOL), to change optimization method (GDIIS, NR), to perform optimization in cartesian or delocalized coordinates, but the result was similar in all cases. Could you advise me something else to solve this problem.
>>>>>The beginning of my input is following:

>>>>> $contrl scftyp=rhf dfttyp=pbe96
>>>>>runtyp=OPTIMIZE icharg=-1 nprint=-5
>>>>>maxit=100 ICUT=20 ITOL=30 NZVAR=168 $end
>>>>> $system kdiag=-1 mwords=100 $end
>>>>> $SCF DIRSCF=.true. maxdii=20 $end
>>>>> $guess guess=moread $end
>>>>> $STATPT method=NR $end
>>>>> $ZMAT DLC=.T. AUTO=.T.
>>>>>DLCTOL=1D-7 ORTTOL=1D-7 $end
>>>>> $INTGRL PACKAO=.true. $end
>>>>> $p2p p2p=.t. dlb=.t. $end
>>>>> $basis gbasis=TZV $end
>>>>> $PCM PCMTYP=DPCM SOLVNT=input RSOLV=2.155 EPS=35.688 EPSINF=1.806874 $END
>>>>> $PCMCAV RIN(58)=1.956 $END

>>>>>Typical out-file is attached.

>>>>>Thank you,


[ Previous ] [ Next ] [ Index ]           Fri Sep 26 '14 1:07pm
[ Reply ] [ Edit ] [ Delete ]           This message read 1203 times