[CPMD-list] P-RFO

[Erik Santiso]

Hi again,

I've tried this, and it works in several of the cases I'm looking at.
However, for the formaldehyde example (the one I sent originally) I'm
still having trouble. It seems that, no matter what I try (i.e. specifying
different modes with PRFO mode, or moving the atoms along an eigenmode and
start the optimization), P-RFO always wants to go to the minimum instead
of the transition state. Here's the input file I'm using now:

&CPMD

 GEOMETRY OPTIMIZATION xyz
 ISOLATED MOLECULE
 MEMORY BIG
 CONVERGENCE ENERGY
  0.05
 RESTART WAVEFUNCTION COORDINATES HESSIAN LATEST
 PRFO
 PRFO PRJHES
 PRFO MODE
  1
 PRFO HESSTYPE
  1

&END

&SYSTEM

SYMMETRY
 0
 CELL
 15.0000 1.0 1.0 0.0 0.0 0.0
 CUTOFF
 35

&END

&ATOMS

*C_BLYP.uspp FORMATTED NEWF
 LMAX=P
 1
  2.096865   1.778271   2.100213

*O_BLYP.uspp FORMATTED NEWF
 LMAX=P
 1
  3.098650   1.778271   2.100213

*H_BLYP.uspp FORMATTED NEWF
 LMAX=S
 2
  1.658508   1.490827   2.100130
  2.655409   1.446718   2.100386
&END

&DFT

  FUNCTIONAL BLYP
  GC-CUTOFF
   5.D-5

&END

This is after running a geometry optimization on a molecule where the
atoms where displaced along the reactive mode (the same that is now mode
1). I've also tried using a transition state from the literature (JPCA
102,10805) as a starting point, running a vibrational analysis on it, and
following the reactive mode, but CPMD converges to the minimum structure
again. Is there anything that looks wrong on my input file? I can send the
GEOMETRY, HESSIAN, etc. files if necessary if that may help pinpointing
what the problem is.

Thanks a lot for your help again,

Erik.

-------------------------------------
Vir prudens non contra ventum mingit.

>
>
>
>
> Hi,
>
> before the transition-state search, it is very helpful to shift the atoms
> most directly involved in the reaction towards the products. This has
> mainly two reasons:
>
> - Avoid starting from near a stationary point. The convergence criteria
> are
> already met. If the system leaves the initial "stationary point" at all,
> the direction you go is co-determined by the (random) residual gradient.
> - At a reactant (or product) configuration, you cannot expect the
> curvature
> along a reaction coordinate to be particularly small; if bond breaking is
> involved, it's even quite unlikely that a reaction coordinate has a low
> frequency there (stretching vs. bending modes). The coordinate you are
> looking for could be almost anywhere in the spectrum.
>
> Otherwise, your approach is fine for smaller systems:
> - Do a vibrational analysis
> - Determine the mode to follow
> - Re-use the Hessian
>
> A small technical hint: it's not enough to specify
>
> RESTART HESSIAN
>
> This will only affect the system Hessian which is different from the
> 'partial Hessian' used by P-RFO. This is required for the linear-scaling
> microiterative scheme (variational decoupling: P-RFO for core / L-BFGS for
> environment) where P-RFO handles only a fraction of the degrees of
> freedom.
> However, you can copy the relevant part of the system Hessian to the
> partial Hessian using
>
> PRFO HESSTYPE
> 1
>
> Also the opposite is possible if you want to use the Powell-updated
> Hessian
> for subsequent vibrational analysis (partial Hessian to Hessian):
>
> VIBRATIONAL ANALYSIS IN
> RESTART PHESS
> HESSIAN PARTIAL [UNIT,SCHLEGEL,DISCO]
>
> Hope this helps.
>
> Best regards
> Salomon
>
>