Transition state searches with Dimer method

[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

On Thursday, April 14, 2016 at 12:20:49 PM UTC+1, Tobias Kraemer wrote:
>
> 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
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20160415/271a24f7/attachment.htm>