Re-optimising transition state using dimer method

Tobias Kraemer 161brun... at
Thu Sep 10 13:39:27 UTC 2015

Dear SL and Luca,

I am aware of the numerical noise problem in vibrational analysis. I have 
worked on this issue for a while, in order to sort out a reliable protocol 
for my systems. 
I have seen many examples where vibrational analysis yielded >50 imaginary 
modes. Presently I am using EPS_SCF of 10E-6 in combination with MAX_FORCE 
8E-4 for transition states, while the vibrational calculations uses a 
tighter EPS_SCF of 10E-8 in combination with DX 0.01. Optimising to minima 
a use somewhat tighter criteria than in the TS searches. I've come to 
realise that it is not easy to get good convergence in NEB calculations 
(don't ask me for the reason, it is my simple observation) and if the SCF 
convergence fails, it is impossible to converge to some decent structure in 
reasonable time. However, for a good number of optimisations (TS and 
minima) on the same system, this protocol has proven quite robust. I have 
heard from other people that they have experience similar issues with NEB, 
and I come to the conclusion that this is far from being straightforward. 

In view of my experience with the system so far, the presence of one small 
(~i25 cm-1) mode indicates that it is not numerical noise. I will try 
however to increase the cutoff (I am using 500 Ry indeed at the moment), to 
see if this changes anything. I have tested this in the past, and it did 
not influence the outcome of the result. Same for DX (increased to 0.02) So 
perhaps the answer lies in converging the dimer method to a tighter cutoff, 
in order to get the desired geometry. 



On Tuesday, September 1, 2015 at 5:11:13 PM UTC+1, Tobias Kraemer wrote:
> Hi all,
> I am trying to re-optimise a transition state structure which has a 
> spurious (small) imaginary mode on top of the large imaginary mode which 
> corresponds to the transition state. I've also tested doing a constrained 
> optimisation fixing selected atoms involved in the transition state, which 
> also worked in this particular case. Howeve, I would also like to try our 
> the dimer method for the purpose of re-optimising a transition state 
> structure. To this end, I have composed an input, however, the job crashes 
> and I keep getting an error message about certain line Search type not 
> being implemented (looks quite obvious from the message). However, I can't 
> seem to find what I need to change to make it work. Your advice would be 
> much appreciated. The motion section can be found below, full input file 
> also uploaded. 
> Any ideas on how to correctly set up a calculation using the dimer method 
> are welcome
> Thanks for your help
> Tobias
>  ************************************************************************
>  *** 16:48:48 ERRORL2 in cg_utils:cg_linmin processor 0  :: err=-1000 ***
>  *** UNIMPLEMENTED, Line Search type not yet implemented in CG for TS ***
>  *** search.                                                          ***
>  ************************************************************************
>  ===== Routine Calling Stack =====
>            10 cg_linmin
>             9 cp_cg_main
>             8 geoopt_cg
>             7 cp_rot_opt
>             6 cp_eval_at_ts
>             5 cp_eval_at
>             4 cp_cg_main
>             3 geoopt_cg
>             2 cp_geo_opt
>             1 CP2K
>   &GEO_OPT
>         &DIMER
>         &ROT_OPT
>           OPTIMIZER CG
>         &END ROT_OPT
>        &END DIMER
>     MAX_FORCE 8.0E-4
>     MAX_ITER 500
>   &PRINT
>       &EACH
>          GEO_OPT 1
>       &END EACH
>   &END
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list