SCF convergence in NEB calculation

[Tobias Kraemer]

Hi all,


I am running CI-NEB calculations in order to locate transition states along a reaction profile involving organometallic species in the solid state. More precisely, this is a hydrogenation reaction of double bonds, i.e. addition of H2 to an alkene. This should be a straightforward task. Anyway, for some cases it seems the SCF encounters problems to converge, or converges to the wrong electronic state. In particular, those replicas are affected which lead to the  transition state. I can see that the electronic structure may be a bit tricky, since in a TS notable rearrangement of the electrons takes place. It goes without saying, that having a wrong electronic structure in the TS leads to unreasonably high barriers. I should add that these type of calculations haven't imposed any problems so far using other codes. What causes some additional puzzlement is the fact that for the same TS search slightly different input  geometries lead to very different behaviour of the SCF. In one case there is reasonable convergence to a decent electronic structure, in the other case the SCF goes completely havoc. 

So my question really is, how can I control the SCF convergence for a particular replica? Can this be done in the main input file?

your comments are appreciated


Tobi

[hut...@chem.uzh.ch]

Hi Tobias

I would start using a very conservative setting for the SCF
optimizer. For example

&SCF
MAX_SCF 20
SCF_GUESS ATOMIC
EPS_SCF 1.e-06
&OT ON
MINIMIZER CG
ENERGY_GAP 0.002
LINESEARCH 3PNT
PRECONDITIONER FULL_ALL
&END OT
&OUTER_SCF
MAX_SCF 20
EPS_SCF 1.e-06
&END OUTER_SCF
&END SCF

But maybe you already do this?

Juerg

--------------------------------------------------------------
Juerg Hutter                         Phone : ++41 44 635 4491
Institut für Chemie C                FAX   : ++41 44 635 6838
Universität Zürich                   E-mail: hut...@chem.uzh.ch
Winterthurerstrasse 190
CH-8057 Zürich, Switzerland
---------------------------------------------------------------

-----cp...@googlegroups.com wrote: -----To: cp2k <cp...@googlegroups.com>
From: Tobias Kraemer <161brun...@gmail.com>
Sent by: cp...@googlegroups.com
Date: 07/28/2015 12:44PM
Subject: [CP2K:6828] SCF convergence in NEB calculation

--

You received this message because you are subscribed to the Google Groups "cp2k" group.

To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.

To post to this group, send email to cp...@googlegroups.com.

Visit this group at http://groups.google.com/group/cp2k.

For more options, visit https://groups.google.com/d/optout.

[Tobias Kraemer]

Dear Juerg,


thanks for your answer. Whilst I am using an SCF convergence of 10e-6 as you suggest, I am not using the conjugate gradient OT minimizer but DIIS instead. Perhaps something to try out, I will give it a go. What I was wondering about is, if there is a possibility to control these sort of parameters for individual replicas.
The electronic structure will change quite a bit as one traverses along the profile between starting and last point, hence I wouldn't expect too much of a problem with the first point, but more towards the transition state. Anyways, I guess it is more reasonable to control the SCF globally for all points.

Another question I have in this context, is there a rule of thumb for how many intermediate structures for NEB should be provided? At the moment I am using a starting structure, a guess for the TS plus a final structure. With 8 replicas defined, the algorithm should interpolate between these. I was successful to localise
one transition state with this already, however, for other transition states things are more complicated. Also, in cases I observe rather strange energy profiles, in which the first (sometimes also the last) point lies higher in energy then the remaining points (see attachment). This is somewhat surprising, since the first point corresponds to an optimised structure (I am assuming that the first replica correlates with my first starting structure). Inspection of the movie.xyz generated from the individual BAND files reassures me that I am looking at a reasonable trajectory. Is is worth waiting for this calculation to converge to some stationary point, or should I come up with a better setup in terms of structures?   

Thanks for the advice


Tobias

[Marcella Iannuzzi]

Hi Tobias, 

The replicas along the band should not be too distant. The algorithm calculates the distant in terms of RMSD, and tries to evenly distribute them along the path. Ideally the RMSD should be around 1-2 Å. 

The energy profile that you sent does not look very promising indeed. You should try to understand why the extrema are higher in energy.

Is it due to problems in the SCF, or it is true that the structures in the middle are lower in energy?

How does the structure change at each step of the NEB?

What happens if you extract from the band some of those structures and you start a standard GEO_OPT?

ciao

Marcella