[molpro-user] Dispersion corrected Geometry Optimizations

[Taube, Andrew]

Dear Andreas (or anyone else),
Performing fully numerical derivatives with the dispersion corrected DFT is straightforward, however it seems like a huge waste; since there are analytical deratives for DFT, it would be nice to add the analytical DFT derivative to a numerical dispersion correction. I tried input formats like the following:

(The system is just a dummy, so don't worry that it isn't one where you would normally want that correction)

***,H2O
memory,400,M
basis=avdz
thrdscf=1e-12,thrdscf_max=1e-9
gthresh,throvl=1e-9,throrth=1e-7
gdirect
geomtyp=xyz
symmetry,nosym
orient,noorient
geometry={
 O          0.0000000000        0.0417232258        0.0323055610
 H          0.0000000000        0.0372295904        0.9974745462
 H          0.0000000000        0.9750241937       -0.2137039261
}

label:

df-ks,b3lyp

dispcorr
forces,numerical,variable=edisp,startcmd=dispcorr

df-ks,b3lyp
eks=energy
forces;
add,1
energy=eks+edisp

optg,startcmd=label:

A couple of notes:
(1) You need the initial DFT call to let dispcorr know what parameters to use
(2) You must calculate the numerical derivative first, because numerical derivatives don't understand the "add" command, while analytical derivatives do
(3) This correctly calculates the force on the first step and optg takes what appears to be the correct first optimization step, however, after that it crashes. From what I can tell, it (for some reason) skips the numerical derivative of the dispersion correction on the second iteration AND double calculates the ks derivative.

Alternatively, one could imagine moving the first df-ks,b3lyp outside of the label: block, however then the code doesn't seem to properly set the density fitting basis when it calls Alaska a second time (and it still skips the dispcorr gradient on the second time through).

Any ideas?
Thanks,
Andrew


Dear Andrew,

in the current user version of Molpro there are no gradients for the
dispersion correction implemented. But you can probably use the
scheme described in section 40.2 in the manual to calculate
numerical gradients (save the total sum of the DFT energy and the
dispersion correction in variable 'energy').

Wishes,
Andreas


On Monday 19 July 2010 21:36, Taube, Andrew wrote:
> Is there a way to use the empirical damped dispersion correction with DFT
> to do (analytical) geometry optimizations? Or does anyone have a slick way
> to calculate the numerical derivative of the dispersion correction and
> combine it with analytic DFT gradients? Thanks,
> Andrew
> ---
> Andrew G. Taube
> D. E. Shaw Research
> email: Andrew.Taube at DEShawResearch.com<mailto:Andrew.Taube at DEShawResearch.com>
> Tel: +1-212-478-0118
> Fax: +1-212-845-1118



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20100721/7237e9fe/attachment.html>