Transition state searches with Dimer method

[Tobias Kraemer]

Hi all,

I would like to come back to a question regarding the dimer method for TS searches. I have located a number of TS by the CI-NEB method, comfirmed by analysis of their harmonic frequencies. Some contain spurious imaginary modes, which I would quite like 

to get rid of, some are well converged to a 'real' TS. Either way, I want to re-optimize these TS with the dimer method, to somewhat tighter geometrical convergence criteria (for reasons I have discussed on this forum before, I had to use somewhat sloppy thresholds

for SCF and geometries in the CI-NEB calculations, but the obtained TS are 'good guesses' nonetheless). In a number of cases this was quite successful, leading to TS as expected, even thought the dimer method would take quite a large number of steps. I am using 

vectors corresponding to the reaction path obtained from the frequency analysis by the way, in order to guide the calculation to the true transition state structure. However, I am having some significant trouble locating some of these TS (even though the NEB energy profile

looks very reasonable, and the imaginary vibrational frequencies modes look ok). Even after 1000 steps, there is no convergence in sight, and in some cases (by visual inspection) the geometry completely escapes the TS region and goes back to what appears to be the ground state. I am wondering if there is anything I can do about the parameters for the dimer method, to (a) prevent this from happening and (b) get faster convergence. These calcualtions take ages to be honest, and sometimes don't yield the desired result even. Continuous restarts and hoping for the best are not the solution I feel. See below the relevant section from my input. Any suggestions are welcome.

&MOTION

  &GEO_OPT

    TYPE TRANSITION_STATE

     &TRANSITION_STATE

       METHOD DIMER

        &DIMER

         DR [angstrom] 0.01

!         ANGLE_TOLERANCE 1.5

         INTERPOLATE_GRADIENT T

         &ROT_OPT

          OPTIMIZER CG

          MAX_ITER 200

          &CG

          MAX_STEEP_STEPS 0

          &LINE_SEARCH

            TYPE 2PNT

           &END LINE_SEARCH

          &END CG

         &END ROT_OPT

         &DIMER_VECTOR

          @INCLUDE 'vector'

         &END DIMER_VECTOR

        &END DIMER

     &END TRANSITION_STATE

    OPTIMIZER CG

      &CG

       &LINE_SEARCH

         TYPE 2PNT

       &END LINE_SEARCH

      &END CG

    MAX_FORCE 1.0E-4

    MAX_ITER 1000

  &END GEO_OPT

  &PRINT

     &TRAJECTORY

      &EACH

         GEO_OPT 1

      &END EACH

    &END TRAJECTORY

  &END

&END MOTION

Tobias

[Matt W]

Hi Tobias,

I don't have any answer. But maybe what we need to get is a good PES to run the dimer on cheaply. Perhaps some sort of classical calculation (vacancy diffusion or similar in a solid) could be setup? That would be much faster to run and debug and decouple any issues about numerical noise in QS any issues with the optimization algorithms.

My 2cs,

Matt

[Jörg Saßmannshausen]

Hi Tobias,

jumping into this thread here: I got similar problems like you. I am using the
SM Band type and I get quite a number of imaginary frequencies next to the one
imaginary frequency for the transition state when I am doing a frequency
analysis of the suggested TS. Thus, like you I want to obtain a better result
and get rid off these spurious imaginary frequencies. I had a look at them and
most of them appear to be movements of an atom within the caged framework of
my compound.

So there seems to be a more common problem with NEB here. Unless we both do it
consistantly wrong it does produce unwanted imaginary frequencies.

As you are using the dimer methode: Can you actually read in the hessian of
the TS so the program has a good starting point to find t he true TS? I know
that is the common practice in other DFT codes.

Also: could somebody point me in the direction of how to set up a dimer
calculation properly? Maybe that would solve Tobias and my problem.

All the best from a cold but dry London

Jörg

--
*************************************************************
Dr. Jörg Saßmannshausen, MRSC
University College London
Department of Chemistry
20 Gordon Street
London
WC1H 0AJ

email: j.sassma...@ucl.ac.uk
web: http://sassy.formativ.net

Please avoid sending me Word or PowerPoint attachments.
See http://www.gnu.org/philosophy/no-word-attachments.html