[CPMD-list] transition state of a molecule

[Martin Konopka]

Dear Salomon (and also Erik),

On Fri, 15 Jul 2005, Salomon Billeter wrote:

> Hi Martin,
>
> > and a potential energy barrier between them. I am mostly starting from
> > the slightly randomized cis structure and trying to locate the top
> > of the barrier. I tried several choices for reaction core and mode to be
>
> starting very near a minimum can be problematic for P-RFO:
> - The gradient is small - also in the "upward" direction. The update of the
>   Hessian will not be very accurate, especially in plane-wave DFT due to
> the
>   ripple in the XC potential.
> - It can be difficult to find the appropriate shifts in the denominator of
>   the rational functions.
> - It can be more difficult to follow a mode (based on overlap) because of
>   very considerable change of shape. This is especially the case with pure
>   P-RFO (rather than microiterative TS search) when the mode to be followed
>   becomes near-degenerate with the rotational / translational modes (see
>   below).
>

Thanks, I will try to use a better starting configuration. (I also tried
it once but it did not work in that particular case.) I would know to
locate the TS approximately also without using P-RFO, just using
constrained geometry optimization (since the minimum-energy path is
very likely CNNC dihedral rotation) but I want in this case to use
P-RFO anyway.

> > followed. Regardless of setting the mode, the result remains the same and
> > unsatisfactory. In some cases the atoms in the reaction core do not move
> > at all (for example in a special case when I started from a
> > hand-modified configuration presumably closer to the TS.)
>
> This should work with the right mode and the right partitioning (the input
> you sent certainly chooses a good reaction core). Here are a couple of
> questions which I hope help lead to a solution:
> - Could there be a problem with the electrons? Do you need CONVERGENCE
> RELAX
>   because of the conical intersection? This could leave the electrons in
> the
>   wrong state.

I do not necessarily need CONVERGENCE RELAX. I tried to set it so that
I typically got electronic gradients below 1.0D-6 (it looked to me
sufficient in the present case; may be it is not) and the purpose was
just to save CPU time in a hope that wavefunctions are sufficiently
relaxed. There may be a problem with the conical intersection but I do
not know the PESs accuratelly yet. I would like to work on the lowest
singlet state for all geometries. If not, then separately on both S0 and
S1 PESs.

> - Why do you use LSD and MULTIPLICITY 1? For the ground state, LDA should
> do
>   just fine, and for an excited singlet state, you'll need a restricted
>   open-shell scheme. Besides the additional computational effort of LSD,
>   there could be a problem if you use CPMD <= 3.9.1 (PBE with LSD had a
>   small problem at low densities - why is your GC-cutoff so large?).
>
I know that NN bond at some dihedral angles (also around 90 deg) is not
correctly described without spin polarization - for example also in HNNH
molecule. The energies of spin-unpolarized ground state are simply much
higher compared to spin-polarized one.  It does not mean that using LSD
gives correct results but since the energy is significantly lower...
Explicitly specifying MULTIPLICITY 1 is of course just useless in this
case but sometime I may calculate also a triplet so that I put the
MULTIPLICITY keyword into the file.

I am using CPMD 3.9.2 and  Vanderbilt pseudopotentials for which
GC-cutoff of 5.0D-5 should be good. (I can test to lower it down but
I do not expect significant effects.)

Finally (to comment Erik's email) I have known about the diradical problem
and was just curious to see how it behaves in my own calculation and
perhaps to compare with some quantum chemistry results.

> >
> > I would greatly appreciate some hint to this problem.
> > Below is my input file and modes from vibrational analysis.
> >
> > I tried the 4th, 7th, 10th mode and also the 1st one, which is the
> > default. I also have a question: If I use, for example
> > PRFO MODE
> >  10
> > does it really mean that the mode corresponding to the 10th frequency of
> > vibrational analysis is followed? If yes, then using PRFO MODE less than
> > 7 (or 6 in special cases), which is the default, would have no sense
> since
> > it corresponds to overall translations or rotations. Or do I
> > misunderstand something?
>
> Yes, this is true for "conventional" (not microiterative) TS search if you
> have no imaginary frequency. If there is already an imaginary frequency, it
> likely corresponds the mode to be followed, and in a partitioned system,
> there is no translational / rotational invariance of the reaction core
> because of the surrounding atoms.
>
> Best regards,
> Salomon
>
>

Thanks again and best regards
Martin.


------------------------------------------------------------------------
Dr. Martin Konopka             http://www.ccms.elf.stuba.sk/konopka.html
Department of Physics, CCMS                      tel: +421 (0)2 60291714
Slovak University of Technology (FEI STU)        fax: +421 (0)2 65411483
Ilkovicova 3, 812 19 Bratislava, Slovakia
------------------------------------------------------------------------